]>
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 | |
ad7fa852 TH |
326 | /** |
327 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
328 | * @page: page of interest | |
329 | * | |
330 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
331 | * with @page is returned. The returned css remains associated with @page | |
332 | * until it is released. | |
333 | * | |
334 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
335 | * is returned. | |
ad7fa852 TH |
336 | */ |
337 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
338 | { | |
339 | struct mem_cgroup *memcg; | |
340 | ||
ad7fa852 TH |
341 | memcg = page->mem_cgroup; |
342 | ||
9e10a130 | 343 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
344 | memcg = root_mem_cgroup; |
345 | ||
ad7fa852 TH |
346 | return &memcg->css; |
347 | } | |
348 | ||
2fc04524 VD |
349 | /** |
350 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
351 | * @page: the page | |
352 | * | |
353 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
354 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
355 | * is safe to call this function without holding a reference to @page. | |
356 | * | |
357 | * Note, this function is inherently racy, because there is nothing to prevent | |
358 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
359 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
360 | * do not care (such as procfs interfaces). | |
361 | */ | |
362 | ino_t page_cgroup_ino(struct page *page) | |
363 | { | |
364 | struct mem_cgroup *memcg; | |
365 | unsigned long ino = 0; | |
366 | ||
367 | rcu_read_lock(); | |
368 | memcg = READ_ONCE(page->mem_cgroup); | |
369 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
370 | memcg = parent_mem_cgroup(memcg); | |
371 | if (memcg) | |
372 | ino = cgroup_ino(memcg->css.cgroup); | |
373 | rcu_read_unlock(); | |
374 | return ino; | |
375 | } | |
376 | ||
f64c3f54 | 377 | static struct mem_cgroup_per_zone * |
e231875b | 378 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 379 | { |
97a6c37b JW |
380 | int nid = page_to_nid(page); |
381 | int zid = page_zonenum(page); | |
f64c3f54 | 382 | |
e231875b | 383 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
384 | } |
385 | ||
bb4cc1a8 AM |
386 | static struct mem_cgroup_tree_per_zone * |
387 | soft_limit_tree_node_zone(int nid, int zid) | |
388 | { | |
389 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
390 | } | |
391 | ||
392 | static struct mem_cgroup_tree_per_zone * | |
393 | soft_limit_tree_from_page(struct page *page) | |
394 | { | |
395 | int nid = page_to_nid(page); | |
396 | int zid = page_zonenum(page); | |
397 | ||
398 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
399 | } | |
400 | ||
cf2c8127 JW |
401 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
402 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 403 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
404 | { |
405 | struct rb_node **p = &mctz->rb_root.rb_node; | |
406 | struct rb_node *parent = NULL; | |
407 | struct mem_cgroup_per_zone *mz_node; | |
408 | ||
409 | if (mz->on_tree) | |
410 | return; | |
411 | ||
412 | mz->usage_in_excess = new_usage_in_excess; | |
413 | if (!mz->usage_in_excess) | |
414 | return; | |
415 | while (*p) { | |
416 | parent = *p; | |
417 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
418 | tree_node); | |
419 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
420 | p = &(*p)->rb_left; | |
421 | /* | |
422 | * We can't avoid mem cgroups that are over their soft | |
423 | * limit by the same amount | |
424 | */ | |
425 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
426 | p = &(*p)->rb_right; | |
427 | } | |
428 | rb_link_node(&mz->tree_node, parent, p); | |
429 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
430 | mz->on_tree = true; | |
431 | } | |
432 | ||
cf2c8127 JW |
433 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
434 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
435 | { |
436 | if (!mz->on_tree) | |
437 | return; | |
438 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
439 | mz->on_tree = false; | |
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 | 444 | { |
0a31bc97 JW |
445 | unsigned long flags; |
446 | ||
447 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 448 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 449 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
450 | } |
451 | ||
3e32cb2e JW |
452 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
453 | { | |
454 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 455 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
456 | unsigned long excess = 0; |
457 | ||
458 | if (nr_pages > soft_limit) | |
459 | excess = nr_pages - soft_limit; | |
460 | ||
461 | return excess; | |
462 | } | |
bb4cc1a8 AM |
463 | |
464 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
465 | { | |
3e32cb2e | 466 | unsigned long excess; |
bb4cc1a8 AM |
467 | struct mem_cgroup_per_zone *mz; |
468 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 469 | |
e231875b | 470 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
471 | /* |
472 | * Necessary to update all ancestors when hierarchy is used. | |
473 | * because their event counter is not touched. | |
474 | */ | |
475 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 476 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 477 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
478 | /* |
479 | * We have to update the tree if mz is on RB-tree or | |
480 | * mem is over its softlimit. | |
481 | */ | |
482 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
483 | unsigned long flags; |
484 | ||
485 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
486 | /* if on-tree, remove it */ |
487 | if (mz->on_tree) | |
cf2c8127 | 488 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
489 | /* |
490 | * Insert again. mz->usage_in_excess will be updated. | |
491 | * If excess is 0, no tree ops. | |
492 | */ | |
cf2c8127 | 493 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 494 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
495 | } |
496 | } | |
497 | } | |
498 | ||
499 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
500 | { | |
bb4cc1a8 | 501 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
502 | struct mem_cgroup_per_zone *mz; |
503 | int nid, zid; | |
bb4cc1a8 | 504 | |
e231875b JZ |
505 | for_each_node(nid) { |
506 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
507 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
508 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 509 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
510 | } |
511 | } | |
512 | } | |
513 | ||
514 | static struct mem_cgroup_per_zone * | |
515 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
516 | { | |
517 | struct rb_node *rightmost = NULL; | |
518 | struct mem_cgroup_per_zone *mz; | |
519 | ||
520 | retry: | |
521 | mz = NULL; | |
522 | rightmost = rb_last(&mctz->rb_root); | |
523 | if (!rightmost) | |
524 | goto done; /* Nothing to reclaim from */ | |
525 | ||
526 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
527 | /* | |
528 | * Remove the node now but someone else can add it back, | |
529 | * we will to add it back at the end of reclaim to its correct | |
530 | * position in the tree. | |
531 | */ | |
cf2c8127 | 532 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 533 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 534 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
535 | goto retry; |
536 | done: | |
537 | return mz; | |
538 | } | |
539 | ||
540 | static struct mem_cgroup_per_zone * | |
541 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
542 | { | |
543 | struct mem_cgroup_per_zone *mz; | |
544 | ||
0a31bc97 | 545 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 546 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 547 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
548 | return mz; |
549 | } | |
550 | ||
711d3d2c | 551 | /* |
484ebb3b GT |
552 | * Return page count for single (non recursive) @memcg. |
553 | * | |
711d3d2c KH |
554 | * Implementation Note: reading percpu statistics for memcg. |
555 | * | |
556 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
557 | * synchronization to implement "quick" read. There are trade-off between | |
558 | * reading cost and precision of value. Then, we may have a chance to implement | |
484ebb3b | 559 | * a periodic synchronization of counter in memcg's counter. |
711d3d2c KH |
560 | * |
561 | * But this _read() function is used for user interface now. The user accounts | |
562 | * memory usage by memory cgroup and he _always_ requires exact value because | |
563 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
564 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
565 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
566 | * | |
567 | * If there are kernel internal actions which can make use of some not-exact | |
568 | * value, and reading all cpu value can be performance bottleneck in some | |
484ebb3b | 569 | * common workload, threshold and synchronization as vmstat[] should be |
711d3d2c KH |
570 | * implemented. |
571 | */ | |
484ebb3b GT |
572 | static unsigned long |
573 | mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx) | |
c62b1a3b | 574 | { |
7a159cc9 | 575 | long val = 0; |
c62b1a3b | 576 | int cpu; |
c62b1a3b | 577 | |
484ebb3b | 578 | /* Per-cpu values can be negative, use a signed accumulator */ |
733a572e | 579 | for_each_possible_cpu(cpu) |
c0ff4b85 | 580 | val += per_cpu(memcg->stat->count[idx], cpu); |
484ebb3b GT |
581 | /* |
582 | * Summing races with updates, so val may be negative. Avoid exposing | |
583 | * transient negative values. | |
584 | */ | |
585 | if (val < 0) | |
586 | val = 0; | |
c62b1a3b KH |
587 | return val; |
588 | } | |
589 | ||
c0ff4b85 | 590 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
591 | enum mem_cgroup_events_index idx) |
592 | { | |
593 | unsigned long val = 0; | |
594 | int cpu; | |
595 | ||
733a572e | 596 | for_each_possible_cpu(cpu) |
c0ff4b85 | 597 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f JW |
598 | return val; |
599 | } | |
600 | ||
c0ff4b85 | 601 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 602 | struct page *page, |
f627c2f5 | 603 | bool compound, int nr_pages) |
d52aa412 | 604 | { |
b2402857 KH |
605 | /* |
606 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
607 | * counted as CACHE even if it's on ANON LRU. | |
608 | */ | |
0a31bc97 | 609 | if (PageAnon(page)) |
b2402857 | 610 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 611 | nr_pages); |
d52aa412 | 612 | else |
b2402857 | 613 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 614 | nr_pages); |
55e462b0 | 615 | |
f627c2f5 KS |
616 | if (compound) { |
617 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
b070e65c DR |
618 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
619 | nr_pages); | |
f627c2f5 | 620 | } |
b070e65c | 621 | |
e401f176 KH |
622 | /* pagein of a big page is an event. So, ignore page size */ |
623 | if (nr_pages > 0) | |
c0ff4b85 | 624 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 625 | else { |
c0ff4b85 | 626 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
627 | nr_pages = -nr_pages; /* for event */ |
628 | } | |
e401f176 | 629 | |
13114716 | 630 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
631 | } |
632 | ||
0a6b76dd VD |
633 | unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
634 | int nid, unsigned int lru_mask) | |
bb2a0de9 | 635 | { |
e231875b | 636 | unsigned long nr = 0; |
889976db YH |
637 | int zid; |
638 | ||
e231875b | 639 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 640 | |
e231875b JZ |
641 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
642 | struct mem_cgroup_per_zone *mz; | |
643 | enum lru_list lru; | |
644 | ||
645 | for_each_lru(lru) { | |
646 | if (!(BIT(lru) & lru_mask)) | |
647 | continue; | |
648 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
649 | nr += mz->lru_size[lru]; | |
650 | } | |
651 | } | |
652 | return nr; | |
889976db | 653 | } |
bb2a0de9 | 654 | |
c0ff4b85 | 655 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 656 | unsigned int lru_mask) |
6d12e2d8 | 657 | { |
e231875b | 658 | unsigned long nr = 0; |
889976db | 659 | int nid; |
6d12e2d8 | 660 | |
31aaea4a | 661 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
662 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
663 | return nr; | |
d52aa412 KH |
664 | } |
665 | ||
f53d7ce3 JW |
666 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
667 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
668 | { |
669 | unsigned long val, next; | |
670 | ||
13114716 | 671 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 672 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 673 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
674 | if ((long)next - (long)val < 0) { |
675 | switch (target) { | |
676 | case MEM_CGROUP_TARGET_THRESH: | |
677 | next = val + THRESHOLDS_EVENTS_TARGET; | |
678 | break; | |
bb4cc1a8 AM |
679 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
680 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
681 | break; | |
f53d7ce3 JW |
682 | case MEM_CGROUP_TARGET_NUMAINFO: |
683 | next = val + NUMAINFO_EVENTS_TARGET; | |
684 | break; | |
685 | default: | |
686 | break; | |
687 | } | |
688 | __this_cpu_write(memcg->stat->targets[target], next); | |
689 | return true; | |
7a159cc9 | 690 | } |
f53d7ce3 | 691 | return false; |
d2265e6f KH |
692 | } |
693 | ||
694 | /* | |
695 | * Check events in order. | |
696 | * | |
697 | */ | |
c0ff4b85 | 698 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
699 | { |
700 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
701 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
702 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 703 | bool do_softlimit; |
82b3f2a7 | 704 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 705 | |
bb4cc1a8 AM |
706 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
707 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
708 | #if MAX_NUMNODES > 1 |
709 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
710 | MEM_CGROUP_TARGET_NUMAINFO); | |
711 | #endif | |
c0ff4b85 | 712 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
713 | if (unlikely(do_softlimit)) |
714 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 715 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 716 | if (unlikely(do_numainfo)) |
c0ff4b85 | 717 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 718 | #endif |
0a31bc97 | 719 | } |
d2265e6f KH |
720 | } |
721 | ||
cf475ad2 | 722 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 723 | { |
31a78f23 BS |
724 | /* |
725 | * mm_update_next_owner() may clear mm->owner to NULL | |
726 | * if it races with swapoff, page migration, etc. | |
727 | * So this can be called with p == NULL. | |
728 | */ | |
729 | if (unlikely(!p)) | |
730 | return NULL; | |
731 | ||
073219e9 | 732 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 733 | } |
33398cf2 | 734 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 735 | |
df381975 | 736 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 737 | { |
c0ff4b85 | 738 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 739 | |
54595fe2 KH |
740 | rcu_read_lock(); |
741 | do { | |
6f6acb00 MH |
742 | /* |
743 | * Page cache insertions can happen withou an | |
744 | * actual mm context, e.g. during disk probing | |
745 | * on boot, loopback IO, acct() writes etc. | |
746 | */ | |
747 | if (unlikely(!mm)) | |
df381975 | 748 | memcg = root_mem_cgroup; |
6f6acb00 MH |
749 | else { |
750 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
751 | if (unlikely(!memcg)) | |
752 | memcg = root_mem_cgroup; | |
753 | } | |
ec903c0c | 754 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 755 | rcu_read_unlock(); |
c0ff4b85 | 756 | return memcg; |
54595fe2 KH |
757 | } |
758 | ||
5660048c JW |
759 | /** |
760 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
761 | * @root: hierarchy root | |
762 | * @prev: previously returned memcg, NULL on first invocation | |
763 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
764 | * | |
765 | * Returns references to children of the hierarchy below @root, or | |
766 | * @root itself, or %NULL after a full round-trip. | |
767 | * | |
768 | * Caller must pass the return value in @prev on subsequent | |
769 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
770 | * to cancel a hierarchy walk before the round-trip is complete. | |
771 | * | |
772 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
773 | * divide up the memcgs in the hierarchy among all concurrent | |
774 | * reclaimers operating on the same zone and priority. | |
775 | */ | |
694fbc0f | 776 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 777 | struct mem_cgroup *prev, |
694fbc0f | 778 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 779 | { |
33398cf2 | 780 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 781 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 782 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 783 | struct mem_cgroup *pos = NULL; |
711d3d2c | 784 | |
694fbc0f AM |
785 | if (mem_cgroup_disabled()) |
786 | return NULL; | |
5660048c | 787 | |
9f3a0d09 JW |
788 | if (!root) |
789 | root = root_mem_cgroup; | |
7d74b06f | 790 | |
9f3a0d09 | 791 | if (prev && !reclaim) |
5ac8fb31 | 792 | pos = prev; |
14067bb3 | 793 | |
9f3a0d09 JW |
794 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
795 | if (prev) | |
5ac8fb31 | 796 | goto out; |
694fbc0f | 797 | return root; |
9f3a0d09 | 798 | } |
14067bb3 | 799 | |
542f85f9 | 800 | rcu_read_lock(); |
5f578161 | 801 | |
5ac8fb31 JW |
802 | if (reclaim) { |
803 | struct mem_cgroup_per_zone *mz; | |
804 | ||
805 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
806 | iter = &mz->iter[reclaim->priority]; | |
807 | ||
808 | if (prev && reclaim->generation != iter->generation) | |
809 | goto out_unlock; | |
810 | ||
6df38689 | 811 | while (1) { |
4db0c3c2 | 812 | pos = READ_ONCE(iter->position); |
6df38689 VD |
813 | if (!pos || css_tryget(&pos->css)) |
814 | break; | |
5ac8fb31 | 815 | /* |
6df38689 VD |
816 | * css reference reached zero, so iter->position will |
817 | * be cleared by ->css_released. However, we should not | |
818 | * rely on this happening soon, because ->css_released | |
819 | * is called from a work queue, and by busy-waiting we | |
820 | * might block it. So we clear iter->position right | |
821 | * away. | |
5ac8fb31 | 822 | */ |
6df38689 VD |
823 | (void)cmpxchg(&iter->position, pos, NULL); |
824 | } | |
5ac8fb31 JW |
825 | } |
826 | ||
827 | if (pos) | |
828 | css = &pos->css; | |
829 | ||
830 | for (;;) { | |
831 | css = css_next_descendant_pre(css, &root->css); | |
832 | if (!css) { | |
833 | /* | |
834 | * Reclaimers share the hierarchy walk, and a | |
835 | * new one might jump in right at the end of | |
836 | * the hierarchy - make sure they see at least | |
837 | * one group and restart from the beginning. | |
838 | */ | |
839 | if (!prev) | |
840 | continue; | |
841 | break; | |
527a5ec9 | 842 | } |
7d74b06f | 843 | |
5ac8fb31 JW |
844 | /* |
845 | * Verify the css and acquire a reference. The root | |
846 | * is provided by the caller, so we know it's alive | |
847 | * and kicking, and don't take an extra reference. | |
848 | */ | |
849 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 850 | |
5ac8fb31 JW |
851 | if (css == &root->css) |
852 | break; | |
14067bb3 | 853 | |
0b8f73e1 JW |
854 | if (css_tryget(css)) |
855 | break; | |
9f3a0d09 | 856 | |
5ac8fb31 | 857 | memcg = NULL; |
9f3a0d09 | 858 | } |
5ac8fb31 JW |
859 | |
860 | if (reclaim) { | |
5ac8fb31 | 861 | /* |
6df38689 VD |
862 | * The position could have already been updated by a competing |
863 | * thread, so check that the value hasn't changed since we read | |
864 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 865 | */ |
6df38689 VD |
866 | (void)cmpxchg(&iter->position, pos, memcg); |
867 | ||
5ac8fb31 JW |
868 | if (pos) |
869 | css_put(&pos->css); | |
870 | ||
871 | if (!memcg) | |
872 | iter->generation++; | |
873 | else if (!prev) | |
874 | reclaim->generation = iter->generation; | |
9f3a0d09 | 875 | } |
5ac8fb31 | 876 | |
542f85f9 MH |
877 | out_unlock: |
878 | rcu_read_unlock(); | |
5ac8fb31 | 879 | out: |
c40046f3 MH |
880 | if (prev && prev != root) |
881 | css_put(&prev->css); | |
882 | ||
9f3a0d09 | 883 | return memcg; |
14067bb3 | 884 | } |
7d74b06f | 885 | |
5660048c JW |
886 | /** |
887 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
888 | * @root: hierarchy root | |
889 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
890 | */ | |
891 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
892 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
893 | { |
894 | if (!root) | |
895 | root = root_mem_cgroup; | |
896 | if (prev && prev != root) | |
897 | css_put(&prev->css); | |
898 | } | |
7d74b06f | 899 | |
6df38689 VD |
900 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
901 | { | |
902 | struct mem_cgroup *memcg = dead_memcg; | |
903 | struct mem_cgroup_reclaim_iter *iter; | |
904 | struct mem_cgroup_per_zone *mz; | |
905 | int nid, zid; | |
906 | int i; | |
907 | ||
908 | while ((memcg = parent_mem_cgroup(memcg))) { | |
909 | for_each_node(nid) { | |
910 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
911 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
912 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
913 | iter = &mz->iter[i]; | |
914 | cmpxchg(&iter->position, | |
915 | dead_memcg, NULL); | |
916 | } | |
917 | } | |
918 | } | |
919 | } | |
920 | } | |
921 | ||
9f3a0d09 JW |
922 | /* |
923 | * Iteration constructs for visiting all cgroups (under a tree). If | |
924 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
925 | * be used for reference counting. | |
926 | */ | |
927 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 928 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 929 | iter != NULL; \ |
527a5ec9 | 930 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 931 | |
9f3a0d09 | 932 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 933 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 934 | iter != NULL; \ |
527a5ec9 | 935 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 936 | |
925b7673 | 937 | /** |
dfe0e773 | 938 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 939 | * @page: the page |
fa9add64 | 940 | * @zone: zone of the page |
dfe0e773 JW |
941 | * |
942 | * This function is only safe when following the LRU page isolation | |
943 | * and putback protocol: the LRU lock must be held, and the page must | |
944 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 945 | */ |
599d0c95 | 946 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 947 | { |
08e552c6 | 948 | struct mem_cgroup_per_zone *mz; |
925b7673 | 949 | struct mem_cgroup *memcg; |
bea8c150 | 950 | struct lruvec *lruvec; |
6d12e2d8 | 951 | |
bea8c150 | 952 | if (mem_cgroup_disabled()) { |
599d0c95 | 953 | lruvec = &pgdat->lruvec; |
bea8c150 HD |
954 | goto out; |
955 | } | |
925b7673 | 956 | |
1306a85a | 957 | memcg = page->mem_cgroup; |
7512102c | 958 | /* |
dfe0e773 | 959 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 960 | * possibly migrated - before they are charged. |
7512102c | 961 | */ |
29833315 JW |
962 | if (!memcg) |
963 | memcg = root_mem_cgroup; | |
7512102c | 964 | |
e231875b | 965 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
966 | lruvec = &mz->lruvec; |
967 | out: | |
968 | /* | |
969 | * Since a node can be onlined after the mem_cgroup was created, | |
970 | * we have to be prepared to initialize lruvec->zone here; | |
971 | * and if offlined then reonlined, we need to reinitialize it. | |
972 | */ | |
599d0c95 MG |
973 | if (unlikely(lruvec->pgdat != pgdat)) |
974 | lruvec->pgdat = pgdat; | |
bea8c150 | 975 | return lruvec; |
08e552c6 | 976 | } |
b69408e8 | 977 | |
925b7673 | 978 | /** |
fa9add64 HD |
979 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
980 | * @lruvec: mem_cgroup per zone lru vector | |
981 | * @lru: index of lru list the page is sitting on | |
599d0c95 | 982 | * @zid: Zone ID of the zone pages have been added to |
fa9add64 | 983 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 984 | * |
ca707239 HD |
985 | * This function must be called under lru_lock, just before a page is added |
986 | * to or just after a page is removed from an lru list (that ordering being | |
987 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 988 | */ |
fa9add64 | 989 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
599d0c95 | 990 | enum zone_type zid, int nr_pages) |
3f58a829 MK |
991 | { |
992 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 993 | unsigned long *lru_size; |
ca707239 HD |
994 | long size; |
995 | bool empty; | |
3f58a829 | 996 | |
599d0c95 | 997 | __update_lru_size(lruvec, lru, zid, nr_pages); |
9d5e6a9f | 998 | |
3f58a829 MK |
999 | if (mem_cgroup_disabled()) |
1000 | return; | |
1001 | ||
fa9add64 HD |
1002 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1003 | lru_size = mz->lru_size + lru; | |
ca707239 HD |
1004 | empty = list_empty(lruvec->lists + lru); |
1005 | ||
1006 | if (nr_pages < 0) | |
1007 | *lru_size += nr_pages; | |
1008 | ||
1009 | size = *lru_size; | |
1010 | if (WARN_ONCE(size < 0 || empty != !size, | |
1011 | "%s(%p, %d, %d): lru_size %ld but %sempty\n", | |
1012 | __func__, lruvec, lru, nr_pages, size, empty ? "" : "not ")) { | |
1013 | VM_BUG_ON(1); | |
1014 | *lru_size = 0; | |
1015 | } | |
1016 | ||
1017 | if (nr_pages > 0) | |
1018 | *lru_size += nr_pages; | |
08e552c6 | 1019 | } |
544122e5 | 1020 | |
2314b42d | 1021 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1022 | { |
2314b42d | 1023 | struct mem_cgroup *task_memcg; |
158e0a2d | 1024 | struct task_struct *p; |
ffbdccf5 | 1025 | bool ret; |
4c4a2214 | 1026 | |
158e0a2d | 1027 | p = find_lock_task_mm(task); |
de077d22 | 1028 | if (p) { |
2314b42d | 1029 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1030 | task_unlock(p); |
1031 | } else { | |
1032 | /* | |
1033 | * All threads may have already detached their mm's, but the oom | |
1034 | * killer still needs to detect if they have already been oom | |
1035 | * killed to prevent needlessly killing additional tasks. | |
1036 | */ | |
ffbdccf5 | 1037 | rcu_read_lock(); |
2314b42d JW |
1038 | task_memcg = mem_cgroup_from_task(task); |
1039 | css_get(&task_memcg->css); | |
ffbdccf5 | 1040 | rcu_read_unlock(); |
de077d22 | 1041 | } |
2314b42d JW |
1042 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1043 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1044 | return ret; |
1045 | } | |
1046 | ||
19942822 | 1047 | /** |
9d11ea9f | 1048 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1049 | * @memcg: the memory cgroup |
19942822 | 1050 | * |
9d11ea9f | 1051 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1052 | * pages. |
19942822 | 1053 | */ |
c0ff4b85 | 1054 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1055 | { |
3e32cb2e JW |
1056 | unsigned long margin = 0; |
1057 | unsigned long count; | |
1058 | unsigned long limit; | |
9d11ea9f | 1059 | |
3e32cb2e | 1060 | count = page_counter_read(&memcg->memory); |
4db0c3c2 | 1061 | limit = READ_ONCE(memcg->memory.limit); |
3e32cb2e JW |
1062 | if (count < limit) |
1063 | margin = limit - count; | |
1064 | ||
7941d214 | 1065 | if (do_memsw_account()) { |
3e32cb2e | 1066 | count = page_counter_read(&memcg->memsw); |
4db0c3c2 | 1067 | limit = READ_ONCE(memcg->memsw.limit); |
3e32cb2e JW |
1068 | if (count <= limit) |
1069 | margin = min(margin, limit - count); | |
cbedbac3 LR |
1070 | else |
1071 | margin = 0; | |
3e32cb2e JW |
1072 | } |
1073 | ||
1074 | return margin; | |
19942822 JW |
1075 | } |
1076 | ||
32047e2a | 1077 | /* |
bdcbb659 | 1078 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1079 | * |
bdcbb659 QH |
1080 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1081 | * moving cgroups. This is for waiting at high-memory pressure | |
1082 | * caused by "move". | |
32047e2a | 1083 | */ |
c0ff4b85 | 1084 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1085 | { |
2bd9bb20 KH |
1086 | struct mem_cgroup *from; |
1087 | struct mem_cgroup *to; | |
4b534334 | 1088 | bool ret = false; |
2bd9bb20 KH |
1089 | /* |
1090 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1091 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1092 | */ | |
1093 | spin_lock(&mc.lock); | |
1094 | from = mc.from; | |
1095 | to = mc.to; | |
1096 | if (!from) | |
1097 | goto unlock; | |
3e92041d | 1098 | |
2314b42d JW |
1099 | ret = mem_cgroup_is_descendant(from, memcg) || |
1100 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1101 | unlock: |
1102 | spin_unlock(&mc.lock); | |
4b534334 KH |
1103 | return ret; |
1104 | } | |
1105 | ||
c0ff4b85 | 1106 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1107 | { |
1108 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1109 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1110 | DEFINE_WAIT(wait); |
1111 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1112 | /* moving charge context might have finished. */ | |
1113 | if (mc.moving_task) | |
1114 | schedule(); | |
1115 | finish_wait(&mc.waitq, &wait); | |
1116 | return true; | |
1117 | } | |
1118 | } | |
1119 | return false; | |
1120 | } | |
1121 | ||
58cf188e | 1122 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1123 | /** |
58cf188e | 1124 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1125 | * @memcg: The memory cgroup that went over limit |
1126 | * @p: Task that is going to be killed | |
1127 | * | |
1128 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1129 | * enabled | |
1130 | */ | |
1131 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1132 | { | |
58cf188e SZ |
1133 | struct mem_cgroup *iter; |
1134 | unsigned int i; | |
e222432b | 1135 | |
e222432b BS |
1136 | rcu_read_lock(); |
1137 | ||
2415b9f5 BV |
1138 | if (p) { |
1139 | pr_info("Task in "); | |
1140 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1141 | pr_cont(" killed as a result of limit of "); | |
1142 | } else { | |
1143 | pr_info("Memory limit reached of cgroup "); | |
1144 | } | |
1145 | ||
e61734c5 | 1146 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1147 | pr_cont("\n"); |
e222432b | 1148 | |
e222432b BS |
1149 | rcu_read_unlock(); |
1150 | ||
3e32cb2e JW |
1151 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1152 | K((u64)page_counter_read(&memcg->memory)), | |
1153 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1154 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1155 | K((u64)page_counter_read(&memcg->memsw)), | |
1156 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1157 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1158 | K((u64)page_counter_read(&memcg->kmem)), | |
1159 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1160 | |
1161 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1162 | pr_info("Memory cgroup stats for "); |
1163 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1164 | pr_cont(":"); |
1165 | ||
1166 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
37e84351 | 1167 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
58cf188e | 1168 | continue; |
484ebb3b | 1169 | pr_cont(" %s:%luKB", mem_cgroup_stat_names[i], |
58cf188e SZ |
1170 | K(mem_cgroup_read_stat(iter, i))); |
1171 | } | |
1172 | ||
1173 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1174 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1175 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1176 | ||
1177 | pr_cont("\n"); | |
1178 | } | |
e222432b BS |
1179 | } |
1180 | ||
81d39c20 KH |
1181 | /* |
1182 | * This function returns the number of memcg under hierarchy tree. Returns | |
1183 | * 1(self count) if no children. | |
1184 | */ | |
c0ff4b85 | 1185 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1186 | { |
1187 | int num = 0; | |
7d74b06f KH |
1188 | struct mem_cgroup *iter; |
1189 | ||
c0ff4b85 | 1190 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1191 | num++; |
81d39c20 KH |
1192 | return num; |
1193 | } | |
1194 | ||
a63d83f4 DR |
1195 | /* |
1196 | * Return the memory (and swap, if configured) limit for a memcg. | |
1197 | */ | |
3e32cb2e | 1198 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1199 | { |
3e32cb2e | 1200 | unsigned long limit; |
f3e8eb70 | 1201 | |
3e32cb2e | 1202 | limit = memcg->memory.limit; |
9a5a8f19 | 1203 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1204 | unsigned long memsw_limit; |
37e84351 | 1205 | unsigned long swap_limit; |
9a5a8f19 | 1206 | |
3e32cb2e | 1207 | memsw_limit = memcg->memsw.limit; |
37e84351 VD |
1208 | swap_limit = memcg->swap.limit; |
1209 | swap_limit = min(swap_limit, (unsigned long)total_swap_pages); | |
1210 | limit = min(limit + swap_limit, memsw_limit); | |
9a5a8f19 | 1211 | } |
9a5a8f19 | 1212 | return limit; |
a63d83f4 DR |
1213 | } |
1214 | ||
b6e6edcf | 1215 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1216 | int order) |
9cbb78bb | 1217 | { |
6e0fc46d DR |
1218 | struct oom_control oc = { |
1219 | .zonelist = NULL, | |
1220 | .nodemask = NULL, | |
2a966b77 | 1221 | .memcg = memcg, |
6e0fc46d DR |
1222 | .gfp_mask = gfp_mask, |
1223 | .order = order, | |
6e0fc46d | 1224 | }; |
9cbb78bb DR |
1225 | struct mem_cgroup *iter; |
1226 | unsigned long chosen_points = 0; | |
1227 | unsigned long totalpages; | |
1228 | unsigned int points = 0; | |
1229 | struct task_struct *chosen = NULL; | |
1230 | ||
dc56401f JW |
1231 | mutex_lock(&oom_lock); |
1232 | ||
876aafbf | 1233 | /* |
465adcf1 DR |
1234 | * If current has a pending SIGKILL or is exiting, then automatically |
1235 | * select it. The goal is to allow it to allocate so that it may | |
1236 | * quickly exit and free its memory. | |
876aafbf | 1237 | */ |
1af8bb43 | 1238 | if (task_will_free_mem(current)) { |
16e95196 | 1239 | mark_oom_victim(current); |
1af8bb43 | 1240 | wake_oom_reaper(current); |
dc56401f | 1241 | goto unlock; |
876aafbf DR |
1242 | } |
1243 | ||
2a966b77 | 1244 | check_panic_on_oom(&oc, CONSTRAINT_MEMCG); |
3e32cb2e | 1245 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1246 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1247 | struct css_task_iter it; |
9cbb78bb DR |
1248 | struct task_struct *task; |
1249 | ||
72ec7029 TH |
1250 | css_task_iter_start(&iter->css, &it); |
1251 | while ((task = css_task_iter_next(&it))) { | |
fbe84a09 | 1252 | switch (oom_scan_process_thread(&oc, task)) { |
9cbb78bb DR |
1253 | case OOM_SCAN_SELECT: |
1254 | if (chosen) | |
1255 | put_task_struct(chosen); | |
1256 | chosen = task; | |
1257 | chosen_points = ULONG_MAX; | |
1258 | get_task_struct(chosen); | |
1259 | /* fall through */ | |
1260 | case OOM_SCAN_CONTINUE: | |
1261 | continue; | |
1262 | case OOM_SCAN_ABORT: | |
72ec7029 | 1263 | css_task_iter_end(&it); |
9cbb78bb DR |
1264 | mem_cgroup_iter_break(memcg, iter); |
1265 | if (chosen) | |
1266 | put_task_struct(chosen); | |
1ebab2db TH |
1267 | /* Set a dummy value to return "true". */ |
1268 | chosen = (void *) 1; | |
dc56401f | 1269 | goto unlock; |
9cbb78bb DR |
1270 | case OOM_SCAN_OK: |
1271 | break; | |
1272 | }; | |
1273 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1274 | if (!points || points < chosen_points) |
1275 | continue; | |
1276 | /* Prefer thread group leaders for display purposes */ | |
1277 | if (points == chosen_points && | |
1278 | thread_group_leader(chosen)) | |
1279 | continue; | |
1280 | ||
1281 | if (chosen) | |
1282 | put_task_struct(chosen); | |
1283 | chosen = task; | |
1284 | chosen_points = points; | |
1285 | get_task_struct(chosen); | |
9cbb78bb | 1286 | } |
72ec7029 | 1287 | css_task_iter_end(&it); |
9cbb78bb DR |
1288 | } |
1289 | ||
dc56401f JW |
1290 | if (chosen) { |
1291 | points = chosen_points * 1000 / totalpages; | |
2a966b77 | 1292 | oom_kill_process(&oc, chosen, points, totalpages, |
6e0fc46d | 1293 | "Memory cgroup out of memory"); |
dc56401f JW |
1294 | } |
1295 | unlock: | |
1296 | mutex_unlock(&oom_lock); | |
b6e6edcf | 1297 | return chosen; |
9cbb78bb DR |
1298 | } |
1299 | ||
ae6e71d3 MC |
1300 | #if MAX_NUMNODES > 1 |
1301 | ||
4d0c066d KH |
1302 | /** |
1303 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1304 | * @memcg: the target memcg |
4d0c066d KH |
1305 | * @nid: the node ID to be checked. |
1306 | * @noswap : specify true here if the user wants flle only information. | |
1307 | * | |
1308 | * This function returns whether the specified memcg contains any | |
1309 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1310 | * pages in the node. | |
1311 | */ | |
c0ff4b85 | 1312 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1313 | int nid, bool noswap) |
1314 | { | |
c0ff4b85 | 1315 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1316 | return true; |
1317 | if (noswap || !total_swap_pages) | |
1318 | return false; | |
c0ff4b85 | 1319 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1320 | return true; |
1321 | return false; | |
1322 | ||
1323 | } | |
889976db YH |
1324 | |
1325 | /* | |
1326 | * Always updating the nodemask is not very good - even if we have an empty | |
1327 | * list or the wrong list here, we can start from some node and traverse all | |
1328 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1329 | * | |
1330 | */ | |
c0ff4b85 | 1331 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1332 | { |
1333 | int nid; | |
453a9bf3 KH |
1334 | /* |
1335 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1336 | * pagein/pageout changes since the last update. | |
1337 | */ | |
c0ff4b85 | 1338 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1339 | return; |
c0ff4b85 | 1340 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1341 | return; |
1342 | ||
889976db | 1343 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1344 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1345 | |
31aaea4a | 1346 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1347 | |
c0ff4b85 R |
1348 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1349 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1350 | } |
453a9bf3 | 1351 | |
c0ff4b85 R |
1352 | atomic_set(&memcg->numainfo_events, 0); |
1353 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1354 | } |
1355 | ||
1356 | /* | |
1357 | * Selecting a node where we start reclaim from. Because what we need is just | |
1358 | * reducing usage counter, start from anywhere is O,K. Considering | |
1359 | * memory reclaim from current node, there are pros. and cons. | |
1360 | * | |
1361 | * Freeing memory from current node means freeing memory from a node which | |
1362 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1363 | * hit limits, it will see a contention on a node. But freeing from remote | |
1364 | * node means more costs for memory reclaim because of memory latency. | |
1365 | * | |
1366 | * Now, we use round-robin. Better algorithm is welcomed. | |
1367 | */ | |
c0ff4b85 | 1368 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1369 | { |
1370 | int node; | |
1371 | ||
c0ff4b85 R |
1372 | mem_cgroup_may_update_nodemask(memcg); |
1373 | node = memcg->last_scanned_node; | |
889976db | 1374 | |
0edaf86c | 1375 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1376 | /* |
fda3d69b MH |
1377 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1378 | * last time it really checked all the LRUs due to rate limiting. | |
1379 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1380 | */ |
1381 | if (unlikely(node == MAX_NUMNODES)) | |
1382 | node = numa_node_id(); | |
1383 | ||
c0ff4b85 | 1384 | memcg->last_scanned_node = node; |
889976db YH |
1385 | return node; |
1386 | } | |
889976db | 1387 | #else |
c0ff4b85 | 1388 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1389 | { |
1390 | return 0; | |
1391 | } | |
1392 | #endif | |
1393 | ||
0608f43d AM |
1394 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1395 | struct zone *zone, | |
1396 | gfp_t gfp_mask, | |
1397 | unsigned long *total_scanned) | |
1398 | { | |
1399 | struct mem_cgroup *victim = NULL; | |
1400 | int total = 0; | |
1401 | int loop = 0; | |
1402 | unsigned long excess; | |
1403 | unsigned long nr_scanned; | |
1404 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1405 | .zone = zone, | |
1406 | .priority = 0, | |
1407 | }; | |
1408 | ||
3e32cb2e | 1409 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1410 | |
1411 | while (1) { | |
1412 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1413 | if (!victim) { | |
1414 | loop++; | |
1415 | if (loop >= 2) { | |
1416 | /* | |
1417 | * If we have not been able to reclaim | |
1418 | * anything, it might because there are | |
1419 | * no reclaimable pages under this hierarchy | |
1420 | */ | |
1421 | if (!total) | |
1422 | break; | |
1423 | /* | |
1424 | * We want to do more targeted reclaim. | |
1425 | * excess >> 2 is not to excessive so as to | |
1426 | * reclaim too much, nor too less that we keep | |
1427 | * coming back to reclaim from this cgroup | |
1428 | */ | |
1429 | if (total >= (excess >> 2) || | |
1430 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1431 | break; | |
1432 | } | |
1433 | continue; | |
1434 | } | |
0608f43d AM |
1435 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1436 | zone, &nr_scanned); | |
1437 | *total_scanned += nr_scanned; | |
3e32cb2e | 1438 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1439 | break; |
6d61ef40 | 1440 | } |
0608f43d AM |
1441 | mem_cgroup_iter_break(root_memcg, victim); |
1442 | return total; | |
6d61ef40 BS |
1443 | } |
1444 | ||
0056f4e6 JW |
1445 | #ifdef CONFIG_LOCKDEP |
1446 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1447 | .name = "memcg_oom_lock", | |
1448 | }; | |
1449 | #endif | |
1450 | ||
fb2a6fc5 JW |
1451 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1452 | ||
867578cb KH |
1453 | /* |
1454 | * Check OOM-Killer is already running under our hierarchy. | |
1455 | * If someone is running, return false. | |
1456 | */ | |
fb2a6fc5 | 1457 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1458 | { |
79dfdacc | 1459 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1460 | |
fb2a6fc5 JW |
1461 | spin_lock(&memcg_oom_lock); |
1462 | ||
9f3a0d09 | 1463 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1464 | if (iter->oom_lock) { |
79dfdacc MH |
1465 | /* |
1466 | * this subtree of our hierarchy is already locked | |
1467 | * so we cannot give a lock. | |
1468 | */ | |
79dfdacc | 1469 | failed = iter; |
9f3a0d09 JW |
1470 | mem_cgroup_iter_break(memcg, iter); |
1471 | break; | |
23751be0 JW |
1472 | } else |
1473 | iter->oom_lock = true; | |
7d74b06f | 1474 | } |
867578cb | 1475 | |
fb2a6fc5 JW |
1476 | if (failed) { |
1477 | /* | |
1478 | * OK, we failed to lock the whole subtree so we have | |
1479 | * to clean up what we set up to the failing subtree | |
1480 | */ | |
1481 | for_each_mem_cgroup_tree(iter, memcg) { | |
1482 | if (iter == failed) { | |
1483 | mem_cgroup_iter_break(memcg, iter); | |
1484 | break; | |
1485 | } | |
1486 | iter->oom_lock = false; | |
79dfdacc | 1487 | } |
0056f4e6 JW |
1488 | } else |
1489 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1490 | |
1491 | spin_unlock(&memcg_oom_lock); | |
1492 | ||
1493 | return !failed; | |
a636b327 | 1494 | } |
0b7f569e | 1495 | |
fb2a6fc5 | 1496 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1497 | { |
7d74b06f KH |
1498 | struct mem_cgroup *iter; |
1499 | ||
fb2a6fc5 | 1500 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1501 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1502 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1503 | iter->oom_lock = false; |
fb2a6fc5 | 1504 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1505 | } |
1506 | ||
c0ff4b85 | 1507 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1508 | { |
1509 | struct mem_cgroup *iter; | |
1510 | ||
c2b42d3c | 1511 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1512 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1513 | iter->under_oom++; |
1514 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1515 | } |
1516 | ||
c0ff4b85 | 1517 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1518 | { |
1519 | struct mem_cgroup *iter; | |
1520 | ||
867578cb KH |
1521 | /* |
1522 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1523 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1524 | */ |
c2b42d3c | 1525 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1526 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1527 | if (iter->under_oom > 0) |
1528 | iter->under_oom--; | |
1529 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1530 | } |
1531 | ||
867578cb KH |
1532 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1533 | ||
dc98df5a | 1534 | struct oom_wait_info { |
d79154bb | 1535 | struct mem_cgroup *memcg; |
dc98df5a KH |
1536 | wait_queue_t wait; |
1537 | }; | |
1538 | ||
1539 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1540 | unsigned mode, int sync, void *arg) | |
1541 | { | |
d79154bb HD |
1542 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1543 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1544 | struct oom_wait_info *oom_wait_info; |
1545 | ||
1546 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1547 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1548 | |
2314b42d JW |
1549 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1550 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1551 | return 0; |
dc98df5a KH |
1552 | return autoremove_wake_function(wait, mode, sync, arg); |
1553 | } | |
1554 | ||
c0ff4b85 | 1555 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1556 | { |
c2b42d3c TH |
1557 | /* |
1558 | * For the following lockless ->under_oom test, the only required | |
1559 | * guarantee is that it must see the state asserted by an OOM when | |
1560 | * this function is called as a result of userland actions | |
1561 | * triggered by the notification of the OOM. This is trivially | |
1562 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1563 | * triggering notification. | |
1564 | */ | |
1565 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1566 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1567 | } |
1568 | ||
3812c8c8 | 1569 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1570 | { |
d0db7afa | 1571 | if (!current->memcg_may_oom) |
3812c8c8 | 1572 | return; |
867578cb | 1573 | /* |
49426420 JW |
1574 | * We are in the middle of the charge context here, so we |
1575 | * don't want to block when potentially sitting on a callstack | |
1576 | * that holds all kinds of filesystem and mm locks. | |
1577 | * | |
1578 | * Also, the caller may handle a failed allocation gracefully | |
1579 | * (like optional page cache readahead) and so an OOM killer | |
1580 | * invocation might not even be necessary. | |
1581 | * | |
1582 | * That's why we don't do anything here except remember the | |
1583 | * OOM context and then deal with it at the end of the page | |
1584 | * fault when the stack is unwound, the locks are released, | |
1585 | * and when we know whether the fault was overall successful. | |
867578cb | 1586 | */ |
49426420 | 1587 | css_get(&memcg->css); |
626ebc41 TH |
1588 | current->memcg_in_oom = memcg; |
1589 | current->memcg_oom_gfp_mask = mask; | |
1590 | current->memcg_oom_order = order; | |
3812c8c8 JW |
1591 | } |
1592 | ||
1593 | /** | |
1594 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1595 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1596 | * |
49426420 JW |
1597 | * This has to be called at the end of a page fault if the memcg OOM |
1598 | * handler was enabled. | |
3812c8c8 | 1599 | * |
49426420 | 1600 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1601 | * sleep on a waitqueue until the userspace task resolves the |
1602 | * situation. Sleeping directly in the charge context with all kinds | |
1603 | * of locks held is not a good idea, instead we remember an OOM state | |
1604 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1605 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1606 | * |
1607 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1608 | * completed, %false otherwise. |
3812c8c8 | 1609 | */ |
49426420 | 1610 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1611 | { |
626ebc41 | 1612 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1613 | struct oom_wait_info owait; |
49426420 | 1614 | bool locked; |
3812c8c8 JW |
1615 | |
1616 | /* OOM is global, do not handle */ | |
3812c8c8 | 1617 | if (!memcg) |
49426420 | 1618 | return false; |
3812c8c8 | 1619 | |
c32b3cbe | 1620 | if (!handle || oom_killer_disabled) |
49426420 | 1621 | goto cleanup; |
3812c8c8 JW |
1622 | |
1623 | owait.memcg = memcg; | |
1624 | owait.wait.flags = 0; | |
1625 | owait.wait.func = memcg_oom_wake_function; | |
1626 | owait.wait.private = current; | |
1627 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1628 | |
3812c8c8 | 1629 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1630 | mem_cgroup_mark_under_oom(memcg); |
1631 | ||
1632 | locked = mem_cgroup_oom_trylock(memcg); | |
1633 | ||
1634 | if (locked) | |
1635 | mem_cgroup_oom_notify(memcg); | |
1636 | ||
1637 | if (locked && !memcg->oom_kill_disable) { | |
1638 | mem_cgroup_unmark_under_oom(memcg); | |
1639 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1640 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1641 | current->memcg_oom_order); | |
49426420 | 1642 | } else { |
3812c8c8 | 1643 | schedule(); |
49426420 JW |
1644 | mem_cgroup_unmark_under_oom(memcg); |
1645 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1646 | } | |
1647 | ||
1648 | if (locked) { | |
fb2a6fc5 JW |
1649 | mem_cgroup_oom_unlock(memcg); |
1650 | /* | |
1651 | * There is no guarantee that an OOM-lock contender | |
1652 | * sees the wakeups triggered by the OOM kill | |
1653 | * uncharges. Wake any sleepers explicitely. | |
1654 | */ | |
1655 | memcg_oom_recover(memcg); | |
1656 | } | |
49426420 | 1657 | cleanup: |
626ebc41 | 1658 | current->memcg_in_oom = NULL; |
3812c8c8 | 1659 | css_put(&memcg->css); |
867578cb | 1660 | return true; |
0b7f569e KH |
1661 | } |
1662 | ||
d7365e78 | 1663 | /** |
81f8c3a4 JW |
1664 | * lock_page_memcg - lock a page->mem_cgroup binding |
1665 | * @page: the page | |
32047e2a | 1666 | * |
81f8c3a4 JW |
1667 | * This function protects unlocked LRU pages from being moved to |
1668 | * another cgroup and stabilizes their page->mem_cgroup binding. | |
d69b042f | 1669 | */ |
62cccb8c | 1670 | void lock_page_memcg(struct page *page) |
89c06bd5 KH |
1671 | { |
1672 | struct mem_cgroup *memcg; | |
6de22619 | 1673 | unsigned long flags; |
89c06bd5 | 1674 | |
6de22619 JW |
1675 | /* |
1676 | * The RCU lock is held throughout the transaction. The fast | |
1677 | * path can get away without acquiring the memcg->move_lock | |
1678 | * because page moving starts with an RCU grace period. | |
6de22619 | 1679 | */ |
d7365e78 JW |
1680 | rcu_read_lock(); |
1681 | ||
1682 | if (mem_cgroup_disabled()) | |
62cccb8c | 1683 | return; |
89c06bd5 | 1684 | again: |
1306a85a | 1685 | memcg = page->mem_cgroup; |
29833315 | 1686 | if (unlikely(!memcg)) |
62cccb8c | 1687 | return; |
d7365e78 | 1688 | |
bdcbb659 | 1689 | if (atomic_read(&memcg->moving_account) <= 0) |
62cccb8c | 1690 | return; |
89c06bd5 | 1691 | |
6de22619 | 1692 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1693 | if (memcg != page->mem_cgroup) { |
6de22619 | 1694 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1695 | goto again; |
1696 | } | |
6de22619 JW |
1697 | |
1698 | /* | |
1699 | * When charge migration first begins, we can have locked and | |
1700 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 1701 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
1702 | */ |
1703 | memcg->move_lock_task = current; | |
1704 | memcg->move_lock_flags = flags; | |
d7365e78 | 1705 | |
62cccb8c | 1706 | return; |
89c06bd5 | 1707 | } |
81f8c3a4 | 1708 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 1709 | |
d7365e78 | 1710 | /** |
81f8c3a4 | 1711 | * unlock_page_memcg - unlock a page->mem_cgroup binding |
62cccb8c | 1712 | * @page: the page |
d7365e78 | 1713 | */ |
62cccb8c | 1714 | void unlock_page_memcg(struct page *page) |
89c06bd5 | 1715 | { |
62cccb8c JW |
1716 | struct mem_cgroup *memcg = page->mem_cgroup; |
1717 | ||
6de22619 JW |
1718 | if (memcg && memcg->move_lock_task == current) { |
1719 | unsigned long flags = memcg->move_lock_flags; | |
1720 | ||
1721 | memcg->move_lock_task = NULL; | |
1722 | memcg->move_lock_flags = 0; | |
1723 | ||
1724 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1725 | } | |
89c06bd5 | 1726 | |
d7365e78 | 1727 | rcu_read_unlock(); |
89c06bd5 | 1728 | } |
81f8c3a4 | 1729 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 1730 | |
cdec2e42 KH |
1731 | /* |
1732 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1733 | * TODO: maybe necessary to use big numbers in big irons. | |
1734 | */ | |
7ec99d62 | 1735 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1736 | struct memcg_stock_pcp { |
1737 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1738 | unsigned int nr_pages; |
cdec2e42 | 1739 | struct work_struct work; |
26fe6168 | 1740 | unsigned long flags; |
a0db00fc | 1741 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1742 | }; |
1743 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1744 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1745 | |
a0956d54 SS |
1746 | /** |
1747 | * consume_stock: Try to consume stocked charge on this cpu. | |
1748 | * @memcg: memcg to consume from. | |
1749 | * @nr_pages: how many pages to charge. | |
1750 | * | |
1751 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1752 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1753 | * service an allocation will refill the stock. | |
1754 | * | |
1755 | * returns true if successful, false otherwise. | |
cdec2e42 | 1756 | */ |
a0956d54 | 1757 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1758 | { |
1759 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 1760 | bool ret = false; |
cdec2e42 | 1761 | |
a0956d54 | 1762 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 1763 | return ret; |
a0956d54 | 1764 | |
cdec2e42 | 1765 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 1766 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1767 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1768 | ret = true; |
1769 | } | |
cdec2e42 KH |
1770 | put_cpu_var(memcg_stock); |
1771 | return ret; | |
1772 | } | |
1773 | ||
1774 | /* | |
3e32cb2e | 1775 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1776 | */ |
1777 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1778 | { | |
1779 | struct mem_cgroup *old = stock->cached; | |
1780 | ||
11c9ea4e | 1781 | if (stock->nr_pages) { |
3e32cb2e | 1782 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 1783 | if (do_memsw_account()) |
3e32cb2e | 1784 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1785 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1786 | stock->nr_pages = 0; |
cdec2e42 KH |
1787 | } |
1788 | stock->cached = NULL; | |
cdec2e42 KH |
1789 | } |
1790 | ||
1791 | /* | |
1792 | * This must be called under preempt disabled or must be called by | |
1793 | * a thread which is pinned to local cpu. | |
1794 | */ | |
1795 | static void drain_local_stock(struct work_struct *dummy) | |
1796 | { | |
7c8e0181 | 1797 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 1798 | drain_stock(stock); |
26fe6168 | 1799 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
1800 | } |
1801 | ||
1802 | /* | |
3e32cb2e | 1803 | * Cache charges(val) to local per_cpu area. |
320cc51d | 1804 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1805 | */ |
c0ff4b85 | 1806 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1807 | { |
1808 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1809 | ||
c0ff4b85 | 1810 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1811 | drain_stock(stock); |
c0ff4b85 | 1812 | stock->cached = memcg; |
cdec2e42 | 1813 | } |
11c9ea4e | 1814 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
1815 | put_cpu_var(memcg_stock); |
1816 | } | |
1817 | ||
1818 | /* | |
c0ff4b85 | 1819 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 1820 | * of the hierarchy under it. |
cdec2e42 | 1821 | */ |
6d3d6aa2 | 1822 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 1823 | { |
26fe6168 | 1824 | int cpu, curcpu; |
d38144b7 | 1825 | |
6d3d6aa2 JW |
1826 | /* If someone's already draining, avoid adding running more workers. */ |
1827 | if (!mutex_trylock(&percpu_charge_mutex)) | |
1828 | return; | |
cdec2e42 | 1829 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 1830 | get_online_cpus(); |
5af12d0e | 1831 | curcpu = get_cpu(); |
cdec2e42 KH |
1832 | for_each_online_cpu(cpu) { |
1833 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1834 | struct mem_cgroup *memcg; |
26fe6168 | 1835 | |
c0ff4b85 R |
1836 | memcg = stock->cached; |
1837 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 1838 | continue; |
2314b42d | 1839 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 1840 | continue; |
d1a05b69 MH |
1841 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
1842 | if (cpu == curcpu) | |
1843 | drain_local_stock(&stock->work); | |
1844 | else | |
1845 | schedule_work_on(cpu, &stock->work); | |
1846 | } | |
cdec2e42 | 1847 | } |
5af12d0e | 1848 | put_cpu(); |
f894ffa8 | 1849 | put_online_cpus(); |
9f50fad6 | 1850 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
1851 | } |
1852 | ||
0db0628d | 1853 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
1854 | unsigned long action, |
1855 | void *hcpu) | |
1856 | { | |
1857 | int cpu = (unsigned long)hcpu; | |
1858 | struct memcg_stock_pcp *stock; | |
1859 | ||
619d094b | 1860 | if (action == CPU_ONLINE) |
1489ebad | 1861 | return NOTIFY_OK; |
1489ebad | 1862 | |
d833049b | 1863 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 1864 | return NOTIFY_OK; |
711d3d2c | 1865 | |
cdec2e42 KH |
1866 | stock = &per_cpu(memcg_stock, cpu); |
1867 | drain_stock(stock); | |
1868 | return NOTIFY_OK; | |
1869 | } | |
1870 | ||
f7e1cb6e JW |
1871 | static void reclaim_high(struct mem_cgroup *memcg, |
1872 | unsigned int nr_pages, | |
1873 | gfp_t gfp_mask) | |
1874 | { | |
1875 | do { | |
1876 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
1877 | continue; | |
1878 | mem_cgroup_events(memcg, MEMCG_HIGH, 1); | |
1879 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); | |
1880 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1881 | } | |
1882 | ||
1883 | static void high_work_func(struct work_struct *work) | |
1884 | { | |
1885 | struct mem_cgroup *memcg; | |
1886 | ||
1887 | memcg = container_of(work, struct mem_cgroup, high_work); | |
1888 | reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL); | |
1889 | } | |
1890 | ||
b23afb93 TH |
1891 | /* |
1892 | * Scheduled by try_charge() to be executed from the userland return path | |
1893 | * and reclaims memory over the high limit. | |
1894 | */ | |
1895 | void mem_cgroup_handle_over_high(void) | |
1896 | { | |
1897 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 1898 | struct mem_cgroup *memcg; |
b23afb93 TH |
1899 | |
1900 | if (likely(!nr_pages)) | |
1901 | return; | |
1902 | ||
f7e1cb6e JW |
1903 | memcg = get_mem_cgroup_from_mm(current->mm); |
1904 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
1905 | css_put(&memcg->css); |
1906 | current->memcg_nr_pages_over_high = 0; | |
1907 | } | |
1908 | ||
00501b53 JW |
1909 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1910 | unsigned int nr_pages) | |
8a9f3ccd | 1911 | { |
7ec99d62 | 1912 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 1913 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 1914 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 1915 | struct page_counter *counter; |
6539cc05 | 1916 | unsigned long nr_reclaimed; |
b70a2a21 JW |
1917 | bool may_swap = true; |
1918 | bool drained = false; | |
a636b327 | 1919 | |
ce00a967 | 1920 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 1921 | return 0; |
6539cc05 | 1922 | retry: |
b6b6cc72 | 1923 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 1924 | return 0; |
8a9f3ccd | 1925 | |
7941d214 | 1926 | if (!do_memsw_account() || |
6071ca52 JW |
1927 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
1928 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 1929 | goto done_restock; |
7941d214 | 1930 | if (do_memsw_account()) |
3e32cb2e JW |
1931 | page_counter_uncharge(&memcg->memsw, batch); |
1932 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 1933 | } else { |
3e32cb2e | 1934 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 1935 | may_swap = false; |
3fbe7244 | 1936 | } |
7a81b88c | 1937 | |
6539cc05 JW |
1938 | if (batch > nr_pages) { |
1939 | batch = nr_pages; | |
1940 | goto retry; | |
1941 | } | |
6d61ef40 | 1942 | |
06b078fc JW |
1943 | /* |
1944 | * Unlike in global OOM situations, memcg is not in a physical | |
1945 | * memory shortage. Allow dying and OOM-killed tasks to | |
1946 | * bypass the last charges so that they can exit quickly and | |
1947 | * free their memory. | |
1948 | */ | |
1949 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
1950 | fatal_signal_pending(current) || | |
1951 | current->flags & PF_EXITING)) | |
10d53c74 | 1952 | goto force; |
06b078fc JW |
1953 | |
1954 | if (unlikely(task_in_memcg_oom(current))) | |
1955 | goto nomem; | |
1956 | ||
d0164adc | 1957 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 1958 | goto nomem; |
4b534334 | 1959 | |
241994ed JW |
1960 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
1961 | ||
b70a2a21 JW |
1962 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
1963 | gfp_mask, may_swap); | |
6539cc05 | 1964 | |
61e02c74 | 1965 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 1966 | goto retry; |
28c34c29 | 1967 | |
b70a2a21 | 1968 | if (!drained) { |
6d3d6aa2 | 1969 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
1970 | drained = true; |
1971 | goto retry; | |
1972 | } | |
1973 | ||
28c34c29 JW |
1974 | if (gfp_mask & __GFP_NORETRY) |
1975 | goto nomem; | |
6539cc05 JW |
1976 | /* |
1977 | * Even though the limit is exceeded at this point, reclaim | |
1978 | * may have been able to free some pages. Retry the charge | |
1979 | * before killing the task. | |
1980 | * | |
1981 | * Only for regular pages, though: huge pages are rather | |
1982 | * unlikely to succeed so close to the limit, and we fall back | |
1983 | * to regular pages anyway in case of failure. | |
1984 | */ | |
61e02c74 | 1985 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
1986 | goto retry; |
1987 | /* | |
1988 | * At task move, charge accounts can be doubly counted. So, it's | |
1989 | * better to wait until the end of task_move if something is going on. | |
1990 | */ | |
1991 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
1992 | goto retry; | |
1993 | ||
9b130619 JW |
1994 | if (nr_retries--) |
1995 | goto retry; | |
1996 | ||
06b078fc | 1997 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 1998 | goto force; |
06b078fc | 1999 | |
6539cc05 | 2000 | if (fatal_signal_pending(current)) |
10d53c74 | 2001 | goto force; |
6539cc05 | 2002 | |
241994ed JW |
2003 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
2004 | ||
3608de07 JM |
2005 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
2006 | get_order(nr_pages * PAGE_SIZE)); | |
7a81b88c | 2007 | nomem: |
6d1fdc48 | 2008 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2009 | return -ENOMEM; |
10d53c74 TH |
2010 | force: |
2011 | /* | |
2012 | * The allocation either can't fail or will lead to more memory | |
2013 | * being freed very soon. Allow memory usage go over the limit | |
2014 | * temporarily by force charging it. | |
2015 | */ | |
2016 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2017 | if (do_memsw_account()) |
10d53c74 TH |
2018 | page_counter_charge(&memcg->memsw, nr_pages); |
2019 | css_get_many(&memcg->css, nr_pages); | |
2020 | ||
2021 | return 0; | |
6539cc05 JW |
2022 | |
2023 | done_restock: | |
e8ea14cc | 2024 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2025 | if (batch > nr_pages) |
2026 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2027 | |
241994ed | 2028 | /* |
b23afb93 TH |
2029 | * If the hierarchy is above the normal consumption range, schedule |
2030 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2031 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2032 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2033 | * not recorded as it most likely matches current's and won't | |
2034 | * change in the meantime. As high limit is checked again before | |
2035 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2036 | */ |
2037 | do { | |
b23afb93 | 2038 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2039 | /* Don't bother a random interrupted task */ |
2040 | if (in_interrupt()) { | |
2041 | schedule_work(&memcg->high_work); | |
2042 | break; | |
2043 | } | |
9516a18a | 2044 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2045 | set_notify_resume(current); |
2046 | break; | |
2047 | } | |
241994ed | 2048 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2049 | |
2050 | return 0; | |
7a81b88c | 2051 | } |
8a9f3ccd | 2052 | |
00501b53 | 2053 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2054 | { |
ce00a967 JW |
2055 | if (mem_cgroup_is_root(memcg)) |
2056 | return; | |
2057 | ||
3e32cb2e | 2058 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2059 | if (do_memsw_account()) |
3e32cb2e | 2060 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2061 | |
e8ea14cc | 2062 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2063 | } |
2064 | ||
0a31bc97 JW |
2065 | static void lock_page_lru(struct page *page, int *isolated) |
2066 | { | |
2067 | struct zone *zone = page_zone(page); | |
2068 | ||
a52633d8 | 2069 | spin_lock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2070 | if (PageLRU(page)) { |
2071 | struct lruvec *lruvec; | |
2072 | ||
599d0c95 | 2073 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2074 | ClearPageLRU(page); |
2075 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2076 | *isolated = 1; | |
2077 | } else | |
2078 | *isolated = 0; | |
2079 | } | |
2080 | ||
2081 | static void unlock_page_lru(struct page *page, int isolated) | |
2082 | { | |
2083 | struct zone *zone = page_zone(page); | |
2084 | ||
2085 | if (isolated) { | |
2086 | struct lruvec *lruvec; | |
2087 | ||
599d0c95 | 2088 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2089 | VM_BUG_ON_PAGE(PageLRU(page), page); |
2090 | SetPageLRU(page); | |
2091 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2092 | } | |
a52633d8 | 2093 | spin_unlock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2094 | } |
2095 | ||
00501b53 | 2096 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2097 | bool lrucare) |
7a81b88c | 2098 | { |
0a31bc97 | 2099 | int isolated; |
9ce70c02 | 2100 | |
1306a85a | 2101 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2102 | |
2103 | /* | |
2104 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2105 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2106 | */ | |
0a31bc97 JW |
2107 | if (lrucare) |
2108 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2109 | |
0a31bc97 JW |
2110 | /* |
2111 | * Nobody should be changing or seriously looking at | |
1306a85a | 2112 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2113 | * |
2114 | * - the page is uncharged | |
2115 | * | |
2116 | * - the page is off-LRU | |
2117 | * | |
2118 | * - an anonymous fault has exclusive page access, except for | |
2119 | * a locked page table | |
2120 | * | |
2121 | * - a page cache insertion, a swapin fault, or a migration | |
2122 | * have the page locked | |
2123 | */ | |
1306a85a | 2124 | page->mem_cgroup = memcg; |
9ce70c02 | 2125 | |
0a31bc97 JW |
2126 | if (lrucare) |
2127 | unlock_page_lru(page, isolated); | |
7a81b88c | 2128 | } |
66e1707b | 2129 | |
127424c8 | 2130 | #ifndef CONFIG_SLOB |
f3bb3043 | 2131 | static int memcg_alloc_cache_id(void) |
55007d84 | 2132 | { |
f3bb3043 VD |
2133 | int id, size; |
2134 | int err; | |
2135 | ||
dbcf73e2 | 2136 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2137 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2138 | if (id < 0) | |
2139 | return id; | |
55007d84 | 2140 | |
dbcf73e2 | 2141 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2142 | return id; |
2143 | ||
2144 | /* | |
2145 | * There's no space for the new id in memcg_caches arrays, | |
2146 | * so we have to grow them. | |
2147 | */ | |
05257a1a | 2148 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2149 | |
2150 | size = 2 * (id + 1); | |
55007d84 GC |
2151 | if (size < MEMCG_CACHES_MIN_SIZE) |
2152 | size = MEMCG_CACHES_MIN_SIZE; | |
2153 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2154 | size = MEMCG_CACHES_MAX_SIZE; | |
2155 | ||
f3bb3043 | 2156 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2157 | if (!err) |
2158 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2159 | if (!err) |
2160 | memcg_nr_cache_ids = size; | |
2161 | ||
2162 | up_write(&memcg_cache_ids_sem); | |
2163 | ||
f3bb3043 | 2164 | if (err) { |
dbcf73e2 | 2165 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2166 | return err; |
2167 | } | |
2168 | return id; | |
2169 | } | |
2170 | ||
2171 | static void memcg_free_cache_id(int id) | |
2172 | { | |
dbcf73e2 | 2173 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2174 | } |
2175 | ||
d5b3cf71 | 2176 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2177 | struct mem_cgroup *memcg; |
2178 | struct kmem_cache *cachep; | |
2179 | struct work_struct work; | |
2180 | }; | |
2181 | ||
d5b3cf71 | 2182 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2183 | { |
d5b3cf71 VD |
2184 | struct memcg_kmem_cache_create_work *cw = |
2185 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2186 | struct mem_cgroup *memcg = cw->memcg; |
2187 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2188 | |
d5b3cf71 | 2189 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2190 | |
5722d094 | 2191 | css_put(&memcg->css); |
d7f25f8a GC |
2192 | kfree(cw); |
2193 | } | |
2194 | ||
2195 | /* | |
2196 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2197 | */ |
d5b3cf71 VD |
2198 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2199 | struct kmem_cache *cachep) | |
d7f25f8a | 2200 | { |
d5b3cf71 | 2201 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2202 | |
776ed0f0 | 2203 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2204 | if (!cw) |
d7f25f8a | 2205 | return; |
8135be5a VD |
2206 | |
2207 | css_get(&memcg->css); | |
d7f25f8a GC |
2208 | |
2209 | cw->memcg = memcg; | |
2210 | cw->cachep = cachep; | |
d5b3cf71 | 2211 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2212 | |
d7f25f8a GC |
2213 | schedule_work(&cw->work); |
2214 | } | |
2215 | ||
d5b3cf71 VD |
2216 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2217 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2218 | { |
2219 | /* | |
2220 | * We need to stop accounting when we kmalloc, because if the | |
2221 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2222 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2223 | * |
2224 | * However, it is better to enclose the whole function. Depending on | |
2225 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2226 | * trigger an allocation. This too, will make us recurse. Because at | |
2227 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2228 | * the safest choice is to do it like this, wrapping the whole function. | |
2229 | */ | |
6f185c29 | 2230 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2231 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2232 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2233 | } |
c67a8a68 | 2234 | |
45264778 VD |
2235 | static inline bool memcg_kmem_bypass(void) |
2236 | { | |
2237 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
2238 | return true; | |
2239 | return false; | |
2240 | } | |
2241 | ||
2242 | /** | |
2243 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation | |
2244 | * @cachep: the original global kmem cache | |
2245 | * | |
d7f25f8a GC |
2246 | * Return the kmem_cache we're supposed to use for a slab allocation. |
2247 | * We try to use the current memcg's version of the cache. | |
2248 | * | |
45264778 VD |
2249 | * If the cache does not exist yet, if we are the first user of it, we |
2250 | * create it asynchronously in a workqueue and let the current allocation | |
2251 | * go through with the original cache. | |
d7f25f8a | 2252 | * |
45264778 VD |
2253 | * This function takes a reference to the cache it returns to assure it |
2254 | * won't get destroyed while we are working with it. Once the caller is | |
2255 | * done with it, memcg_kmem_put_cache() must be called to release the | |
2256 | * reference. | |
d7f25f8a | 2257 | */ |
45264778 | 2258 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2259 | { |
2260 | struct mem_cgroup *memcg; | |
959c8963 | 2261 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2262 | int kmemcg_id; |
d7f25f8a | 2263 | |
f7ce3190 | 2264 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2265 | |
45264778 | 2266 | if (memcg_kmem_bypass()) |
230e9fc2 VD |
2267 | return cachep; |
2268 | ||
9d100c5e | 2269 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2270 | return cachep; |
2271 | ||
8135be5a | 2272 | memcg = get_mem_cgroup_from_mm(current->mm); |
4db0c3c2 | 2273 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2274 | if (kmemcg_id < 0) |
ca0dde97 | 2275 | goto out; |
d7f25f8a | 2276 | |
2a4db7eb | 2277 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2278 | if (likely(memcg_cachep)) |
2279 | return memcg_cachep; | |
ca0dde97 LZ |
2280 | |
2281 | /* | |
2282 | * If we are in a safe context (can wait, and not in interrupt | |
2283 | * context), we could be be predictable and return right away. | |
2284 | * This would guarantee that the allocation being performed | |
2285 | * already belongs in the new cache. | |
2286 | * | |
2287 | * However, there are some clashes that can arrive from locking. | |
2288 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2289 | * memcg_create_kmem_cache, this means no further allocation |
2290 | * could happen with the slab_mutex held. So it's better to | |
2291 | * defer everything. | |
ca0dde97 | 2292 | */ |
d5b3cf71 | 2293 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2294 | out: |
8135be5a | 2295 | css_put(&memcg->css); |
ca0dde97 | 2296 | return cachep; |
d7f25f8a | 2297 | } |
d7f25f8a | 2298 | |
45264778 VD |
2299 | /** |
2300 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache | |
2301 | * @cachep: the cache returned by memcg_kmem_get_cache | |
2302 | */ | |
2303 | void memcg_kmem_put_cache(struct kmem_cache *cachep) | |
8135be5a VD |
2304 | { |
2305 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2306 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2307 | } |
2308 | ||
45264778 VD |
2309 | /** |
2310 | * memcg_kmem_charge: charge a kmem page | |
2311 | * @page: page to charge | |
2312 | * @gfp: reclaim mode | |
2313 | * @order: allocation order | |
2314 | * @memcg: memory cgroup to charge | |
2315 | * | |
2316 | * Returns 0 on success, an error code on failure. | |
2317 | */ | |
2318 | int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, | |
2319 | struct mem_cgroup *memcg) | |
7ae1e1d0 | 2320 | { |
f3ccb2c4 VD |
2321 | unsigned int nr_pages = 1 << order; |
2322 | struct page_counter *counter; | |
7ae1e1d0 GC |
2323 | int ret; |
2324 | ||
f3ccb2c4 | 2325 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2326 | if (ret) |
f3ccb2c4 | 2327 | return ret; |
52c29b04 JW |
2328 | |
2329 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2330 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2331 | cancel_charge(memcg, nr_pages); | |
2332 | return -ENOMEM; | |
7ae1e1d0 GC |
2333 | } |
2334 | ||
f3ccb2c4 | 2335 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2336 | |
f3ccb2c4 | 2337 | return 0; |
7ae1e1d0 GC |
2338 | } |
2339 | ||
45264778 VD |
2340 | /** |
2341 | * memcg_kmem_charge: charge a kmem page to the current memory cgroup | |
2342 | * @page: page to charge | |
2343 | * @gfp: reclaim mode | |
2344 | * @order: allocation order | |
2345 | * | |
2346 | * Returns 0 on success, an error code on failure. | |
2347 | */ | |
2348 | int memcg_kmem_charge(struct page *page, gfp_t gfp, int order) | |
7ae1e1d0 | 2349 | { |
f3ccb2c4 | 2350 | struct mem_cgroup *memcg; |
fcff7d7e | 2351 | int ret = 0; |
7ae1e1d0 | 2352 | |
45264778 VD |
2353 | if (memcg_kmem_bypass()) |
2354 | return 0; | |
2355 | ||
f3ccb2c4 | 2356 | memcg = get_mem_cgroup_from_mm(current->mm); |
b6ecd2de | 2357 | if (!mem_cgroup_is_root(memcg)) |
45264778 | 2358 | ret = memcg_kmem_charge_memcg(page, gfp, order, memcg); |
7ae1e1d0 | 2359 | css_put(&memcg->css); |
d05e83a6 | 2360 | return ret; |
7ae1e1d0 | 2361 | } |
45264778 VD |
2362 | /** |
2363 | * memcg_kmem_uncharge: uncharge a kmem page | |
2364 | * @page: page to uncharge | |
2365 | * @order: allocation order | |
2366 | */ | |
2367 | void memcg_kmem_uncharge(struct page *page, int order) | |
7ae1e1d0 | 2368 | { |
1306a85a | 2369 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2370 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2371 | |
7ae1e1d0 GC |
2372 | if (!memcg) |
2373 | return; | |
2374 | ||
309381fe | 2375 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2376 | |
52c29b04 JW |
2377 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2378 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2379 | ||
f3ccb2c4 | 2380 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2381 | if (do_memsw_account()) |
f3ccb2c4 | 2382 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2383 | |
1306a85a | 2384 | page->mem_cgroup = NULL; |
f3ccb2c4 | 2385 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2386 | } |
127424c8 | 2387 | #endif /* !CONFIG_SLOB */ |
7ae1e1d0 | 2388 | |
ca3e0214 KH |
2389 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2390 | ||
ca3e0214 KH |
2391 | /* |
2392 | * Because tail pages are not marked as "used", set it. We're under | |
a52633d8 | 2393 | * zone_lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2394 | */ |
e94c8a9c | 2395 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2396 | { |
e94c8a9c | 2397 | int i; |
ca3e0214 | 2398 | |
3d37c4a9 KH |
2399 | if (mem_cgroup_disabled()) |
2400 | return; | |
b070e65c | 2401 | |
29833315 | 2402 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2403 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2404 | |
1306a85a | 2405 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2406 | HPAGE_PMD_NR); |
ca3e0214 | 2407 | } |
12d27107 | 2408 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2409 | |
c255a458 | 2410 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2411 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2412 | bool charge) | |
d13d1443 | 2413 | { |
0a31bc97 JW |
2414 | int val = (charge) ? 1 : -1; |
2415 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2416 | } |
02491447 DN |
2417 | |
2418 | /** | |
2419 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2420 | * @entry: swap entry to be moved | |
2421 | * @from: mem_cgroup which the entry is moved from | |
2422 | * @to: mem_cgroup which the entry is moved to | |
2423 | * | |
2424 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2425 | * as the mem_cgroup's id of @from. | |
2426 | * | |
2427 | * Returns 0 on success, -EINVAL on failure. | |
2428 | * | |
3e32cb2e | 2429 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2430 | * both res and memsw, and called css_get(). |
2431 | */ | |
2432 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2433 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2434 | { |
2435 | unsigned short old_id, new_id; | |
2436 | ||
34c00c31 LZ |
2437 | old_id = mem_cgroup_id(from); |
2438 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2439 | |
2440 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2441 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2442 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2443 | return 0; |
2444 | } | |
2445 | return -EINVAL; | |
2446 | } | |
2447 | #else | |
2448 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2449 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2450 | { |
2451 | return -EINVAL; | |
2452 | } | |
8c7c6e34 | 2453 | #endif |
d13d1443 | 2454 | |
3e32cb2e | 2455 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2456 | |
d38d2a75 | 2457 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2458 | unsigned long limit) |
628f4235 | 2459 | { |
3e32cb2e JW |
2460 | unsigned long curusage; |
2461 | unsigned long oldusage; | |
2462 | bool enlarge = false; | |
81d39c20 | 2463 | int retry_count; |
3e32cb2e | 2464 | int ret; |
81d39c20 KH |
2465 | |
2466 | /* | |
2467 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2468 | * is depends on callers. We set our retry-count to be function | |
2469 | * of # of children which we should visit in this loop. | |
2470 | */ | |
3e32cb2e JW |
2471 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2472 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2473 | |
3e32cb2e | 2474 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2475 | |
3e32cb2e | 2476 | do { |
628f4235 KH |
2477 | if (signal_pending(current)) { |
2478 | ret = -EINTR; | |
2479 | break; | |
2480 | } | |
3e32cb2e JW |
2481 | |
2482 | mutex_lock(&memcg_limit_mutex); | |
2483 | if (limit > memcg->memsw.limit) { | |
2484 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2485 | ret = -EINVAL; |
628f4235 KH |
2486 | break; |
2487 | } | |
3e32cb2e JW |
2488 | if (limit > memcg->memory.limit) |
2489 | enlarge = true; | |
2490 | ret = page_counter_limit(&memcg->memory, limit); | |
2491 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2492 | |
2493 | if (!ret) | |
2494 | break; | |
2495 | ||
b70a2a21 JW |
2496 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2497 | ||
3e32cb2e | 2498 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2499 | /* Usage is reduced ? */ |
f894ffa8 | 2500 | if (curusage >= oldusage) |
81d39c20 KH |
2501 | retry_count--; |
2502 | else | |
2503 | oldusage = curusage; | |
3e32cb2e JW |
2504 | } while (retry_count); |
2505 | ||
3c11ecf4 KH |
2506 | if (!ret && enlarge) |
2507 | memcg_oom_recover(memcg); | |
14797e23 | 2508 | |
8c7c6e34 KH |
2509 | return ret; |
2510 | } | |
2511 | ||
338c8431 | 2512 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2513 | unsigned long limit) |
8c7c6e34 | 2514 | { |
3e32cb2e JW |
2515 | unsigned long curusage; |
2516 | unsigned long oldusage; | |
2517 | bool enlarge = false; | |
81d39c20 | 2518 | int retry_count; |
3e32cb2e | 2519 | int ret; |
8c7c6e34 | 2520 | |
81d39c20 | 2521 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2522 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2523 | mem_cgroup_count_children(memcg); | |
2524 | ||
2525 | oldusage = page_counter_read(&memcg->memsw); | |
2526 | ||
2527 | do { | |
8c7c6e34 KH |
2528 | if (signal_pending(current)) { |
2529 | ret = -EINTR; | |
2530 | break; | |
2531 | } | |
3e32cb2e JW |
2532 | |
2533 | mutex_lock(&memcg_limit_mutex); | |
2534 | if (limit < memcg->memory.limit) { | |
2535 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2536 | ret = -EINVAL; |
8c7c6e34 KH |
2537 | break; |
2538 | } | |
3e32cb2e JW |
2539 | if (limit > memcg->memsw.limit) |
2540 | enlarge = true; | |
2541 | ret = page_counter_limit(&memcg->memsw, limit); | |
2542 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2543 | |
2544 | if (!ret) | |
2545 | break; | |
2546 | ||
b70a2a21 JW |
2547 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
2548 | ||
3e32cb2e | 2549 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 2550 | /* Usage is reduced ? */ |
8c7c6e34 | 2551 | if (curusage >= oldusage) |
628f4235 | 2552 | retry_count--; |
81d39c20 KH |
2553 | else |
2554 | oldusage = curusage; | |
3e32cb2e JW |
2555 | } while (retry_count); |
2556 | ||
3c11ecf4 KH |
2557 | if (!ret && enlarge) |
2558 | memcg_oom_recover(memcg); | |
3e32cb2e | 2559 | |
628f4235 KH |
2560 | return ret; |
2561 | } | |
2562 | ||
0608f43d AM |
2563 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2564 | gfp_t gfp_mask, | |
2565 | unsigned long *total_scanned) | |
2566 | { | |
2567 | unsigned long nr_reclaimed = 0; | |
2568 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2569 | unsigned long reclaimed; | |
2570 | int loop = 0; | |
2571 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 2572 | unsigned long excess; |
0608f43d AM |
2573 | unsigned long nr_scanned; |
2574 | ||
2575 | if (order > 0) | |
2576 | return 0; | |
2577 | ||
2578 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
2579 | /* | |
2580 | * This loop can run a while, specially if mem_cgroup's continuously | |
2581 | * keep exceeding their soft limit and putting the system under | |
2582 | * pressure | |
2583 | */ | |
2584 | do { | |
2585 | if (next_mz) | |
2586 | mz = next_mz; | |
2587 | else | |
2588 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2589 | if (!mz) | |
2590 | break; | |
2591 | ||
2592 | nr_scanned = 0; | |
2593 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
2594 | gfp_mask, &nr_scanned); | |
2595 | nr_reclaimed += reclaimed; | |
2596 | *total_scanned += nr_scanned; | |
0a31bc97 | 2597 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2598 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2599 | |
2600 | /* | |
2601 | * If we failed to reclaim anything from this memory cgroup | |
2602 | * it is time to move on to the next cgroup | |
2603 | */ | |
2604 | next_mz = NULL; | |
bc2f2e7f VD |
2605 | if (!reclaimed) |
2606 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2607 | ||
3e32cb2e | 2608 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2609 | /* |
2610 | * One school of thought says that we should not add | |
2611 | * back the node to the tree if reclaim returns 0. | |
2612 | * But our reclaim could return 0, simply because due | |
2613 | * to priority we are exposing a smaller subset of | |
2614 | * memory to reclaim from. Consider this as a longer | |
2615 | * term TODO. | |
2616 | */ | |
2617 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2618 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2619 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2620 | css_put(&mz->memcg->css); |
2621 | loop++; | |
2622 | /* | |
2623 | * Could not reclaim anything and there are no more | |
2624 | * mem cgroups to try or we seem to be looping without | |
2625 | * reclaiming anything. | |
2626 | */ | |
2627 | if (!nr_reclaimed && | |
2628 | (next_mz == NULL || | |
2629 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2630 | break; | |
2631 | } while (!nr_reclaimed); | |
2632 | if (next_mz) | |
2633 | css_put(&next_mz->memcg->css); | |
2634 | return nr_reclaimed; | |
2635 | } | |
2636 | ||
ea280e7b TH |
2637 | /* |
2638 | * Test whether @memcg has children, dead or alive. Note that this | |
2639 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2640 | * returns %true if there are child csses according to the cgroup | |
2641 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2642 | */ | |
b5f99b53 GC |
2643 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2644 | { | |
ea280e7b TH |
2645 | bool ret; |
2646 | ||
ea280e7b TH |
2647 | rcu_read_lock(); |
2648 | ret = css_next_child(NULL, &memcg->css); | |
2649 | rcu_read_unlock(); | |
2650 | return ret; | |
b5f99b53 GC |
2651 | } |
2652 | ||
c26251f9 | 2653 | /* |
51038171 | 2654 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
2655 | * |
2656 | * Caller is responsible for holding css reference for memcg. | |
2657 | */ | |
2658 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2659 | { | |
2660 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2661 | |
c1e862c1 KH |
2662 | /* we call try-to-free pages for make this cgroup empty */ |
2663 | lru_add_drain_all(); | |
f817ed48 | 2664 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2665 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2666 | int progress; |
c1e862c1 | 2667 | |
c26251f9 MH |
2668 | if (signal_pending(current)) |
2669 | return -EINTR; | |
2670 | ||
b70a2a21 JW |
2671 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2672 | GFP_KERNEL, true); | |
c1e862c1 | 2673 | if (!progress) { |
f817ed48 | 2674 | nr_retries--; |
c1e862c1 | 2675 | /* maybe some writeback is necessary */ |
8aa7e847 | 2676 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2677 | } |
f817ed48 KH |
2678 | |
2679 | } | |
ab5196c2 MH |
2680 | |
2681 | return 0; | |
cc847582 KH |
2682 | } |
2683 | ||
6770c64e TH |
2684 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2685 | char *buf, size_t nbytes, | |
2686 | loff_t off) | |
c1e862c1 | 2687 | { |
6770c64e | 2688 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2689 | |
d8423011 MH |
2690 | if (mem_cgroup_is_root(memcg)) |
2691 | return -EINVAL; | |
6770c64e | 2692 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2693 | } |
2694 | ||
182446d0 TH |
2695 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2696 | struct cftype *cft) | |
18f59ea7 | 2697 | { |
182446d0 | 2698 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2699 | } |
2700 | ||
182446d0 TH |
2701 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2702 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2703 | { |
2704 | int retval = 0; | |
182446d0 | 2705 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2706 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2707 | |
567fb435 | 2708 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 2709 | return 0; |
567fb435 | 2710 | |
18f59ea7 | 2711 | /* |
af901ca1 | 2712 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2713 | * in the child subtrees. If it is unset, then the change can |
2714 | * occur, provided the current cgroup has no children. | |
2715 | * | |
2716 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2717 | * set if there are no children. | |
2718 | */ | |
c0ff4b85 | 2719 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2720 | (val == 1 || val == 0)) { |
ea280e7b | 2721 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2722 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2723 | else |
2724 | retval = -EBUSY; | |
2725 | } else | |
2726 | retval = -EINVAL; | |
567fb435 | 2727 | |
18f59ea7 BS |
2728 | return retval; |
2729 | } | |
2730 | ||
72b54e73 | 2731 | static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat) |
ce00a967 JW |
2732 | { |
2733 | struct mem_cgroup *iter; | |
72b54e73 | 2734 | int i; |
ce00a967 | 2735 | |
72b54e73 | 2736 | memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT); |
ce00a967 | 2737 | |
72b54e73 VD |
2738 | for_each_mem_cgroup_tree(iter, memcg) { |
2739 | for (i = 0; i < MEMCG_NR_STAT; i++) | |
2740 | stat[i] += mem_cgroup_read_stat(iter, i); | |
2741 | } | |
ce00a967 JW |
2742 | } |
2743 | ||
72b54e73 | 2744 | static void tree_events(struct mem_cgroup *memcg, unsigned long *events) |
587d9f72 JW |
2745 | { |
2746 | struct mem_cgroup *iter; | |
72b54e73 | 2747 | int i; |
587d9f72 | 2748 | |
72b54e73 | 2749 | memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS); |
587d9f72 | 2750 | |
72b54e73 VD |
2751 | for_each_mem_cgroup_tree(iter, memcg) { |
2752 | for (i = 0; i < MEMCG_NR_EVENTS; i++) | |
2753 | events[i] += mem_cgroup_read_events(iter, i); | |
2754 | } | |
587d9f72 JW |
2755 | } |
2756 | ||
6f646156 | 2757 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2758 | { |
72b54e73 | 2759 | unsigned long val = 0; |
ce00a967 | 2760 | |
3e32cb2e | 2761 | if (mem_cgroup_is_root(memcg)) { |
72b54e73 VD |
2762 | struct mem_cgroup *iter; |
2763 | ||
2764 | for_each_mem_cgroup_tree(iter, memcg) { | |
2765 | val += mem_cgroup_read_stat(iter, | |
2766 | MEM_CGROUP_STAT_CACHE); | |
2767 | val += mem_cgroup_read_stat(iter, | |
2768 | MEM_CGROUP_STAT_RSS); | |
2769 | if (swap) | |
2770 | val += mem_cgroup_read_stat(iter, | |
2771 | MEM_CGROUP_STAT_SWAP); | |
2772 | } | |
3e32cb2e | 2773 | } else { |
ce00a967 | 2774 | if (!swap) |
3e32cb2e | 2775 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2776 | else |
3e32cb2e | 2777 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2778 | } |
c12176d3 | 2779 | return val; |
ce00a967 JW |
2780 | } |
2781 | ||
3e32cb2e JW |
2782 | enum { |
2783 | RES_USAGE, | |
2784 | RES_LIMIT, | |
2785 | RES_MAX_USAGE, | |
2786 | RES_FAILCNT, | |
2787 | RES_SOFT_LIMIT, | |
2788 | }; | |
ce00a967 | 2789 | |
791badbd | 2790 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2791 | struct cftype *cft) |
8cdea7c0 | 2792 | { |
182446d0 | 2793 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2794 | struct page_counter *counter; |
af36f906 | 2795 | |
3e32cb2e | 2796 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 2797 | case _MEM: |
3e32cb2e JW |
2798 | counter = &memcg->memory; |
2799 | break; | |
8c7c6e34 | 2800 | case _MEMSWAP: |
3e32cb2e JW |
2801 | counter = &memcg->memsw; |
2802 | break; | |
510fc4e1 | 2803 | case _KMEM: |
3e32cb2e | 2804 | counter = &memcg->kmem; |
510fc4e1 | 2805 | break; |
d55f90bf | 2806 | case _TCP: |
0db15298 | 2807 | counter = &memcg->tcpmem; |
d55f90bf | 2808 | break; |
8c7c6e34 KH |
2809 | default: |
2810 | BUG(); | |
8c7c6e34 | 2811 | } |
3e32cb2e JW |
2812 | |
2813 | switch (MEMFILE_ATTR(cft->private)) { | |
2814 | case RES_USAGE: | |
2815 | if (counter == &memcg->memory) | |
c12176d3 | 2816 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 2817 | if (counter == &memcg->memsw) |
c12176d3 | 2818 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
2819 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
2820 | case RES_LIMIT: | |
2821 | return (u64)counter->limit * PAGE_SIZE; | |
2822 | case RES_MAX_USAGE: | |
2823 | return (u64)counter->watermark * PAGE_SIZE; | |
2824 | case RES_FAILCNT: | |
2825 | return counter->failcnt; | |
2826 | case RES_SOFT_LIMIT: | |
2827 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
2828 | default: | |
2829 | BUG(); | |
2830 | } | |
8cdea7c0 | 2831 | } |
510fc4e1 | 2832 | |
127424c8 | 2833 | #ifndef CONFIG_SLOB |
567e9ab2 | 2834 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 2835 | { |
d6441637 VD |
2836 | int memcg_id; |
2837 | ||
b313aeee VD |
2838 | if (cgroup_memory_nokmem) |
2839 | return 0; | |
2840 | ||
2a4db7eb | 2841 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 2842 | BUG_ON(memcg->kmem_state); |
d6441637 | 2843 | |
f3bb3043 | 2844 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
2845 | if (memcg_id < 0) |
2846 | return memcg_id; | |
d6441637 | 2847 | |
ef12947c | 2848 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 2849 | /* |
567e9ab2 | 2850 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 2851 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
2852 | * guarantee no one starts accounting before all call sites are |
2853 | * patched. | |
2854 | */ | |
900a38f0 | 2855 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 2856 | memcg->kmem_state = KMEM_ONLINE; |
0b8f73e1 JW |
2857 | |
2858 | return 0; | |
d6441637 VD |
2859 | } |
2860 | ||
8e0a8912 JW |
2861 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
2862 | { | |
2863 | struct cgroup_subsys_state *css; | |
2864 | struct mem_cgroup *parent, *child; | |
2865 | int kmemcg_id; | |
2866 | ||
2867 | if (memcg->kmem_state != KMEM_ONLINE) | |
2868 | return; | |
2869 | /* | |
2870 | * Clear the online state before clearing memcg_caches array | |
2871 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
2872 | * guarantees that no cache will be created for this cgroup | |
2873 | * after we are done (see memcg_create_kmem_cache()). | |
2874 | */ | |
2875 | memcg->kmem_state = KMEM_ALLOCATED; | |
2876 | ||
2877 | memcg_deactivate_kmem_caches(memcg); | |
2878 | ||
2879 | kmemcg_id = memcg->kmemcg_id; | |
2880 | BUG_ON(kmemcg_id < 0); | |
2881 | ||
2882 | parent = parent_mem_cgroup(memcg); | |
2883 | if (!parent) | |
2884 | parent = root_mem_cgroup; | |
2885 | ||
2886 | /* | |
2887 | * Change kmemcg_id of this cgroup and all its descendants to the | |
2888 | * parent's id, and then move all entries from this cgroup's list_lrus | |
2889 | * to ones of the parent. After we have finished, all list_lrus | |
2890 | * corresponding to this cgroup are guaranteed to remain empty. The | |
2891 | * ordering is imposed by list_lru_node->lock taken by | |
2892 | * memcg_drain_all_list_lrus(). | |
2893 | */ | |
3a06bb78 | 2894 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
2895 | css_for_each_descendant_pre(css, &memcg->css) { |
2896 | child = mem_cgroup_from_css(css); | |
2897 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
2898 | child->kmemcg_id = parent->kmemcg_id; | |
2899 | if (!memcg->use_hierarchy) | |
2900 | break; | |
2901 | } | |
3a06bb78 TH |
2902 | rcu_read_unlock(); |
2903 | ||
8e0a8912 JW |
2904 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); |
2905 | ||
2906 | memcg_free_cache_id(kmemcg_id); | |
2907 | } | |
2908 | ||
2909 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2910 | { | |
0b8f73e1 JW |
2911 | /* css_alloc() failed, offlining didn't happen */ |
2912 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
2913 | memcg_offline_kmem(memcg); | |
2914 | ||
8e0a8912 JW |
2915 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
2916 | memcg_destroy_kmem_caches(memcg); | |
2917 | static_branch_dec(&memcg_kmem_enabled_key); | |
2918 | WARN_ON(page_counter_read(&memcg->kmem)); | |
2919 | } | |
8e0a8912 | 2920 | } |
d6441637 | 2921 | #else |
0b8f73e1 | 2922 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
2923 | { |
2924 | return 0; | |
2925 | } | |
2926 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
2927 | { | |
2928 | } | |
2929 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2930 | { | |
2931 | } | |
2932 | #endif /* !CONFIG_SLOB */ | |
2933 | ||
d6441637 | 2934 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2935 | unsigned long limit) |
d6441637 | 2936 | { |
b313aeee | 2937 | int ret; |
127424c8 JW |
2938 | |
2939 | mutex_lock(&memcg_limit_mutex); | |
127424c8 | 2940 | ret = page_counter_limit(&memcg->kmem, limit); |
127424c8 JW |
2941 | mutex_unlock(&memcg_limit_mutex); |
2942 | return ret; | |
d6441637 | 2943 | } |
510fc4e1 | 2944 | |
d55f90bf VD |
2945 | static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit) |
2946 | { | |
2947 | int ret; | |
2948 | ||
2949 | mutex_lock(&memcg_limit_mutex); | |
2950 | ||
0db15298 | 2951 | ret = page_counter_limit(&memcg->tcpmem, limit); |
d55f90bf VD |
2952 | if (ret) |
2953 | goto out; | |
2954 | ||
0db15298 | 2955 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
2956 | /* |
2957 | * The active flag needs to be written after the static_key | |
2958 | * update. This is what guarantees that the socket activation | |
2959 | * function is the last one to run. See sock_update_memcg() for | |
2960 | * details, and note that we don't mark any socket as belonging | |
2961 | * to this memcg until that flag is up. | |
2962 | * | |
2963 | * We need to do this, because static_keys will span multiple | |
2964 | * sites, but we can't control their order. If we mark a socket | |
2965 | * as accounted, but the accounting functions are not patched in | |
2966 | * yet, we'll lose accounting. | |
2967 | * | |
2968 | * We never race with the readers in sock_update_memcg(), | |
2969 | * because when this value change, the code to process it is not | |
2970 | * patched in yet. | |
2971 | */ | |
2972 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 2973 | memcg->tcpmem_active = true; |
d55f90bf VD |
2974 | } |
2975 | out: | |
2976 | mutex_unlock(&memcg_limit_mutex); | |
2977 | return ret; | |
2978 | } | |
d55f90bf | 2979 | |
628f4235 KH |
2980 | /* |
2981 | * The user of this function is... | |
2982 | * RES_LIMIT. | |
2983 | */ | |
451af504 TH |
2984 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
2985 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 2986 | { |
451af504 | 2987 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 2988 | unsigned long nr_pages; |
628f4235 KH |
2989 | int ret; |
2990 | ||
451af504 | 2991 | buf = strstrip(buf); |
650c5e56 | 2992 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
2993 | if (ret) |
2994 | return ret; | |
af36f906 | 2995 | |
3e32cb2e | 2996 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 2997 | case RES_LIMIT: |
4b3bde4c BS |
2998 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
2999 | ret = -EINVAL; | |
3000 | break; | |
3001 | } | |
3e32cb2e JW |
3002 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3003 | case _MEM: | |
3004 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3005 | break; |
3e32cb2e JW |
3006 | case _MEMSWAP: |
3007 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3008 | break; |
3e32cb2e JW |
3009 | case _KMEM: |
3010 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3011 | break; | |
d55f90bf VD |
3012 | case _TCP: |
3013 | ret = memcg_update_tcp_limit(memcg, nr_pages); | |
3014 | break; | |
3e32cb2e | 3015 | } |
296c81d8 | 3016 | break; |
3e32cb2e JW |
3017 | case RES_SOFT_LIMIT: |
3018 | memcg->soft_limit = nr_pages; | |
3019 | ret = 0; | |
628f4235 KH |
3020 | break; |
3021 | } | |
451af504 | 3022 | return ret ?: nbytes; |
8cdea7c0 BS |
3023 | } |
3024 | ||
6770c64e TH |
3025 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3026 | size_t nbytes, loff_t off) | |
c84872e1 | 3027 | { |
6770c64e | 3028 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3029 | struct page_counter *counter; |
c84872e1 | 3030 | |
3e32cb2e JW |
3031 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3032 | case _MEM: | |
3033 | counter = &memcg->memory; | |
3034 | break; | |
3035 | case _MEMSWAP: | |
3036 | counter = &memcg->memsw; | |
3037 | break; | |
3038 | case _KMEM: | |
3039 | counter = &memcg->kmem; | |
3040 | break; | |
d55f90bf | 3041 | case _TCP: |
0db15298 | 3042 | counter = &memcg->tcpmem; |
d55f90bf | 3043 | break; |
3e32cb2e JW |
3044 | default: |
3045 | BUG(); | |
3046 | } | |
af36f906 | 3047 | |
3e32cb2e | 3048 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3049 | case RES_MAX_USAGE: |
3e32cb2e | 3050 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3051 | break; |
3052 | case RES_FAILCNT: | |
3e32cb2e | 3053 | counter->failcnt = 0; |
29f2a4da | 3054 | break; |
3e32cb2e JW |
3055 | default: |
3056 | BUG(); | |
29f2a4da | 3057 | } |
f64c3f54 | 3058 | |
6770c64e | 3059 | return nbytes; |
c84872e1 PE |
3060 | } |
3061 | ||
182446d0 | 3062 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3063 | struct cftype *cft) |
3064 | { | |
182446d0 | 3065 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3066 | } |
3067 | ||
02491447 | 3068 | #ifdef CONFIG_MMU |
182446d0 | 3069 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3070 | struct cftype *cft, u64 val) |
3071 | { | |
182446d0 | 3072 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3073 | |
1dfab5ab | 3074 | if (val & ~MOVE_MASK) |
7dc74be0 | 3075 | return -EINVAL; |
ee5e8472 | 3076 | |
7dc74be0 | 3077 | /* |
ee5e8472 GC |
3078 | * No kind of locking is needed in here, because ->can_attach() will |
3079 | * check this value once in the beginning of the process, and then carry | |
3080 | * on with stale data. This means that changes to this value will only | |
3081 | * affect task migrations starting after the change. | |
7dc74be0 | 3082 | */ |
c0ff4b85 | 3083 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3084 | return 0; |
3085 | } | |
02491447 | 3086 | #else |
182446d0 | 3087 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3088 | struct cftype *cft, u64 val) |
3089 | { | |
3090 | return -ENOSYS; | |
3091 | } | |
3092 | #endif | |
7dc74be0 | 3093 | |
406eb0c9 | 3094 | #ifdef CONFIG_NUMA |
2da8ca82 | 3095 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3096 | { |
25485de6 GT |
3097 | struct numa_stat { |
3098 | const char *name; | |
3099 | unsigned int lru_mask; | |
3100 | }; | |
3101 | ||
3102 | static const struct numa_stat stats[] = { | |
3103 | { "total", LRU_ALL }, | |
3104 | { "file", LRU_ALL_FILE }, | |
3105 | { "anon", LRU_ALL_ANON }, | |
3106 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3107 | }; | |
3108 | const struct numa_stat *stat; | |
406eb0c9 | 3109 | int nid; |
25485de6 | 3110 | unsigned long nr; |
2da8ca82 | 3111 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3112 | |
25485de6 GT |
3113 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3114 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3115 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3116 | for_each_node_state(nid, N_MEMORY) { | |
3117 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3118 | stat->lru_mask); | |
3119 | seq_printf(m, " N%d=%lu", nid, nr); | |
3120 | } | |
3121 | seq_putc(m, '\n'); | |
406eb0c9 | 3122 | } |
406eb0c9 | 3123 | |
071aee13 YH |
3124 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3125 | struct mem_cgroup *iter; | |
3126 | ||
3127 | nr = 0; | |
3128 | for_each_mem_cgroup_tree(iter, memcg) | |
3129 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3130 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3131 | for_each_node_state(nid, N_MEMORY) { | |
3132 | nr = 0; | |
3133 | for_each_mem_cgroup_tree(iter, memcg) | |
3134 | nr += mem_cgroup_node_nr_lru_pages( | |
3135 | iter, nid, stat->lru_mask); | |
3136 | seq_printf(m, " N%d=%lu", nid, nr); | |
3137 | } | |
3138 | seq_putc(m, '\n'); | |
406eb0c9 | 3139 | } |
406eb0c9 | 3140 | |
406eb0c9 YH |
3141 | return 0; |
3142 | } | |
3143 | #endif /* CONFIG_NUMA */ | |
3144 | ||
2da8ca82 | 3145 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3146 | { |
2da8ca82 | 3147 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3148 | unsigned long memory, memsw; |
af7c4b0e JW |
3149 | struct mem_cgroup *mi; |
3150 | unsigned int i; | |
406eb0c9 | 3151 | |
0ca44b14 GT |
3152 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3153 | MEM_CGROUP_STAT_NSTATS); | |
3154 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3155 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3156 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3157 | ||
af7c4b0e | 3158 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
7941d214 | 3159 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3160 | continue; |
484ebb3b | 3161 | seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i], |
af7c4b0e | 3162 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
1dd3a273 | 3163 | } |
7b854121 | 3164 | |
af7c4b0e JW |
3165 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3166 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3167 | mem_cgroup_read_events(memcg, i)); | |
3168 | ||
3169 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3170 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3171 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3172 | ||
14067bb3 | 3173 | /* Hierarchical information */ |
3e32cb2e JW |
3174 | memory = memsw = PAGE_COUNTER_MAX; |
3175 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3176 | memory = min(memory, mi->memory.limit); | |
3177 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3178 | } |
3e32cb2e JW |
3179 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3180 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3181 | if (do_memsw_account()) |
3e32cb2e JW |
3182 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3183 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3184 | |
af7c4b0e | 3185 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
484ebb3b | 3186 | unsigned long long val = 0; |
af7c4b0e | 3187 | |
7941d214 | 3188 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3189 | continue; |
af7c4b0e JW |
3190 | for_each_mem_cgroup_tree(mi, memcg) |
3191 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
484ebb3b | 3192 | seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val); |
af7c4b0e JW |
3193 | } |
3194 | ||
3195 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3196 | unsigned long long val = 0; | |
3197 | ||
3198 | for_each_mem_cgroup_tree(mi, memcg) | |
3199 | val += mem_cgroup_read_events(mi, i); | |
3200 | seq_printf(m, "total_%s %llu\n", | |
3201 | mem_cgroup_events_names[i], val); | |
3202 | } | |
3203 | ||
3204 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3205 | unsigned long long val = 0; | |
3206 | ||
3207 | for_each_mem_cgroup_tree(mi, memcg) | |
3208 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3209 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3210 | } |
14067bb3 | 3211 | |
7f016ee8 | 3212 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3213 | { |
3214 | int nid, zid; | |
3215 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3216 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3217 | unsigned long recent_rotated[2] = {0, 0}; |
3218 | unsigned long recent_scanned[2] = {0, 0}; | |
3219 | ||
3220 | for_each_online_node(nid) | |
3221 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3222 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3223 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3224 | |
89abfab1 HD |
3225 | recent_rotated[0] += rstat->recent_rotated[0]; |
3226 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3227 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3228 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3229 | } |
78ccf5b5 JW |
3230 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3231 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3232 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3233 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3234 | } |
3235 | #endif | |
3236 | ||
d2ceb9b7 KH |
3237 | return 0; |
3238 | } | |
3239 | ||
182446d0 TH |
3240 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3241 | struct cftype *cft) | |
a7885eb8 | 3242 | { |
182446d0 | 3243 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3244 | |
1f4c025b | 3245 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3246 | } |
3247 | ||
182446d0 TH |
3248 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3249 | struct cftype *cft, u64 val) | |
a7885eb8 | 3250 | { |
182446d0 | 3251 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3252 | |
3dae7fec | 3253 | if (val > 100) |
a7885eb8 KM |
3254 | return -EINVAL; |
3255 | ||
14208b0e | 3256 | if (css->parent) |
3dae7fec JW |
3257 | memcg->swappiness = val; |
3258 | else | |
3259 | vm_swappiness = val; | |
068b38c1 | 3260 | |
a7885eb8 KM |
3261 | return 0; |
3262 | } | |
3263 | ||
2e72b634 KS |
3264 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3265 | { | |
3266 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3267 | unsigned long usage; |
2e72b634 KS |
3268 | int i; |
3269 | ||
3270 | rcu_read_lock(); | |
3271 | if (!swap) | |
2c488db2 | 3272 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3273 | else |
2c488db2 | 3274 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3275 | |
3276 | if (!t) | |
3277 | goto unlock; | |
3278 | ||
ce00a967 | 3279 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3280 | |
3281 | /* | |
748dad36 | 3282 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3283 | * If it's not true, a threshold was crossed after last |
3284 | * call of __mem_cgroup_threshold(). | |
3285 | */ | |
5407a562 | 3286 | i = t->current_threshold; |
2e72b634 KS |
3287 | |
3288 | /* | |
3289 | * Iterate backward over array of thresholds starting from | |
3290 | * current_threshold and check if a threshold is crossed. | |
3291 | * If none of thresholds below usage is crossed, we read | |
3292 | * only one element of the array here. | |
3293 | */ | |
3294 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3295 | eventfd_signal(t->entries[i].eventfd, 1); | |
3296 | ||
3297 | /* i = current_threshold + 1 */ | |
3298 | i++; | |
3299 | ||
3300 | /* | |
3301 | * Iterate forward over array of thresholds starting from | |
3302 | * current_threshold+1 and check if a threshold is crossed. | |
3303 | * If none of thresholds above usage is crossed, we read | |
3304 | * only one element of the array here. | |
3305 | */ | |
3306 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3307 | eventfd_signal(t->entries[i].eventfd, 1); | |
3308 | ||
3309 | /* Update current_threshold */ | |
5407a562 | 3310 | t->current_threshold = i - 1; |
2e72b634 KS |
3311 | unlock: |
3312 | rcu_read_unlock(); | |
3313 | } | |
3314 | ||
3315 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3316 | { | |
ad4ca5f4 KS |
3317 | while (memcg) { |
3318 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3319 | if (do_memsw_account()) |
ad4ca5f4 KS |
3320 | __mem_cgroup_threshold(memcg, true); |
3321 | ||
3322 | memcg = parent_mem_cgroup(memcg); | |
3323 | } | |
2e72b634 KS |
3324 | } |
3325 | ||
3326 | static int compare_thresholds(const void *a, const void *b) | |
3327 | { | |
3328 | const struct mem_cgroup_threshold *_a = a; | |
3329 | const struct mem_cgroup_threshold *_b = b; | |
3330 | ||
2bff24a3 GT |
3331 | if (_a->threshold > _b->threshold) |
3332 | return 1; | |
3333 | ||
3334 | if (_a->threshold < _b->threshold) | |
3335 | return -1; | |
3336 | ||
3337 | return 0; | |
2e72b634 KS |
3338 | } |
3339 | ||
c0ff4b85 | 3340 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3341 | { |
3342 | struct mem_cgroup_eventfd_list *ev; | |
3343 | ||
2bcf2e92 MH |
3344 | spin_lock(&memcg_oom_lock); |
3345 | ||
c0ff4b85 | 3346 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3347 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3348 | |
3349 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3350 | return 0; |
3351 | } | |
3352 | ||
c0ff4b85 | 3353 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3354 | { |
7d74b06f KH |
3355 | struct mem_cgroup *iter; |
3356 | ||
c0ff4b85 | 3357 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3358 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3359 | } |
3360 | ||
59b6f873 | 3361 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3362 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3363 | { |
2c488db2 KS |
3364 | struct mem_cgroup_thresholds *thresholds; |
3365 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3366 | unsigned long threshold; |
3367 | unsigned long usage; | |
2c488db2 | 3368 | int i, size, ret; |
2e72b634 | 3369 | |
650c5e56 | 3370 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3371 | if (ret) |
3372 | return ret; | |
3373 | ||
3374 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3375 | |
05b84301 | 3376 | if (type == _MEM) { |
2c488db2 | 3377 | thresholds = &memcg->thresholds; |
ce00a967 | 3378 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3379 | } else if (type == _MEMSWAP) { |
2c488db2 | 3380 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3381 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3382 | } else |
2e72b634 KS |
3383 | BUG(); |
3384 | ||
2e72b634 | 3385 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3386 | if (thresholds->primary) |
2e72b634 KS |
3387 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3388 | ||
2c488db2 | 3389 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3390 | |
3391 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3392 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3393 | GFP_KERNEL); |
2c488db2 | 3394 | if (!new) { |
2e72b634 KS |
3395 | ret = -ENOMEM; |
3396 | goto unlock; | |
3397 | } | |
2c488db2 | 3398 | new->size = size; |
2e72b634 KS |
3399 | |
3400 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3401 | if (thresholds->primary) { |
3402 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3403 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3404 | } |
3405 | ||
2e72b634 | 3406 | /* Add new threshold */ |
2c488db2 KS |
3407 | new->entries[size - 1].eventfd = eventfd; |
3408 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3409 | |
3410 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3411 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3412 | compare_thresholds, NULL); |
3413 | ||
3414 | /* Find current threshold */ | |
2c488db2 | 3415 | new->current_threshold = -1; |
2e72b634 | 3416 | for (i = 0; i < size; i++) { |
748dad36 | 3417 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3418 | /* |
2c488db2 KS |
3419 | * new->current_threshold will not be used until |
3420 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3421 | * it here. |
3422 | */ | |
2c488db2 | 3423 | ++new->current_threshold; |
748dad36 SZ |
3424 | } else |
3425 | break; | |
2e72b634 KS |
3426 | } |
3427 | ||
2c488db2 KS |
3428 | /* Free old spare buffer and save old primary buffer as spare */ |
3429 | kfree(thresholds->spare); | |
3430 | thresholds->spare = thresholds->primary; | |
3431 | ||
3432 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3433 | |
907860ed | 3434 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3435 | synchronize_rcu(); |
3436 | ||
2e72b634 KS |
3437 | unlock: |
3438 | mutex_unlock(&memcg->thresholds_lock); | |
3439 | ||
3440 | return ret; | |
3441 | } | |
3442 | ||
59b6f873 | 3443 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3444 | struct eventfd_ctx *eventfd, const char *args) |
3445 | { | |
59b6f873 | 3446 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3447 | } |
3448 | ||
59b6f873 | 3449 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3450 | struct eventfd_ctx *eventfd, const char *args) |
3451 | { | |
59b6f873 | 3452 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3453 | } |
3454 | ||
59b6f873 | 3455 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3456 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3457 | { |
2c488db2 KS |
3458 | struct mem_cgroup_thresholds *thresholds; |
3459 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3460 | unsigned long usage; |
2c488db2 | 3461 | int i, j, size; |
2e72b634 KS |
3462 | |
3463 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3464 | |
3465 | if (type == _MEM) { | |
2c488db2 | 3466 | thresholds = &memcg->thresholds; |
ce00a967 | 3467 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3468 | } else if (type == _MEMSWAP) { |
2c488db2 | 3469 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3470 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3471 | } else |
2e72b634 KS |
3472 | BUG(); |
3473 | ||
371528ca AV |
3474 | if (!thresholds->primary) |
3475 | goto unlock; | |
3476 | ||
2e72b634 KS |
3477 | /* Check if a threshold crossed before removing */ |
3478 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3479 | ||
3480 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3481 | size = 0; |
3482 | for (i = 0; i < thresholds->primary->size; i++) { | |
3483 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3484 | size++; |
3485 | } | |
3486 | ||
2c488db2 | 3487 | new = thresholds->spare; |
907860ed | 3488 | |
2e72b634 KS |
3489 | /* Set thresholds array to NULL if we don't have thresholds */ |
3490 | if (!size) { | |
2c488db2 KS |
3491 | kfree(new); |
3492 | new = NULL; | |
907860ed | 3493 | goto swap_buffers; |
2e72b634 KS |
3494 | } |
3495 | ||
2c488db2 | 3496 | new->size = size; |
2e72b634 KS |
3497 | |
3498 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3499 | new->current_threshold = -1; |
3500 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3501 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3502 | continue; |
3503 | ||
2c488db2 | 3504 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3505 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3506 | /* |
2c488db2 | 3507 | * new->current_threshold will not be used |
2e72b634 KS |
3508 | * until rcu_assign_pointer(), so it's safe to increment |
3509 | * it here. | |
3510 | */ | |
2c488db2 | 3511 | ++new->current_threshold; |
2e72b634 KS |
3512 | } |
3513 | j++; | |
3514 | } | |
3515 | ||
907860ed | 3516 | swap_buffers: |
2c488db2 KS |
3517 | /* Swap primary and spare array */ |
3518 | thresholds->spare = thresholds->primary; | |
8c757763 | 3519 | |
2c488db2 | 3520 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3521 | |
907860ed | 3522 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3523 | synchronize_rcu(); |
6611d8d7 MC |
3524 | |
3525 | /* If all events are unregistered, free the spare array */ | |
3526 | if (!new) { | |
3527 | kfree(thresholds->spare); | |
3528 | thresholds->spare = NULL; | |
3529 | } | |
371528ca | 3530 | unlock: |
2e72b634 | 3531 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3532 | } |
c1e862c1 | 3533 | |
59b6f873 | 3534 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3535 | struct eventfd_ctx *eventfd) |
3536 | { | |
59b6f873 | 3537 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3538 | } |
3539 | ||
59b6f873 | 3540 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3541 | struct eventfd_ctx *eventfd) |
3542 | { | |
59b6f873 | 3543 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3544 | } |
3545 | ||
59b6f873 | 3546 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3547 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3548 | { |
9490ff27 | 3549 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3550 | |
9490ff27 KH |
3551 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3552 | if (!event) | |
3553 | return -ENOMEM; | |
3554 | ||
1af8efe9 | 3555 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3556 | |
3557 | event->eventfd = eventfd; | |
3558 | list_add(&event->list, &memcg->oom_notify); | |
3559 | ||
3560 | /* already in OOM ? */ | |
c2b42d3c | 3561 | if (memcg->under_oom) |
9490ff27 | 3562 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3563 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3564 | |
3565 | return 0; | |
3566 | } | |
3567 | ||
59b6f873 | 3568 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3569 | struct eventfd_ctx *eventfd) |
9490ff27 | 3570 | { |
9490ff27 | 3571 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3572 | |
1af8efe9 | 3573 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3574 | |
c0ff4b85 | 3575 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3576 | if (ev->eventfd == eventfd) { |
3577 | list_del(&ev->list); | |
3578 | kfree(ev); | |
3579 | } | |
3580 | } | |
3581 | ||
1af8efe9 | 3582 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3583 | } |
3584 | ||
2da8ca82 | 3585 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3586 | { |
2da8ca82 | 3587 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3588 | |
791badbd | 3589 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3590 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
3c11ecf4 KH |
3591 | return 0; |
3592 | } | |
3593 | ||
182446d0 | 3594 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3595 | struct cftype *cft, u64 val) |
3596 | { | |
182446d0 | 3597 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3598 | |
3599 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3600 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3601 | return -EINVAL; |
3602 | ||
c0ff4b85 | 3603 | memcg->oom_kill_disable = val; |
4d845ebf | 3604 | if (!val) |
c0ff4b85 | 3605 | memcg_oom_recover(memcg); |
3dae7fec | 3606 | |
3c11ecf4 KH |
3607 | return 0; |
3608 | } | |
3609 | ||
52ebea74 TH |
3610 | #ifdef CONFIG_CGROUP_WRITEBACK |
3611 | ||
3612 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) | |
3613 | { | |
3614 | return &memcg->cgwb_list; | |
3615 | } | |
3616 | ||
841710aa TH |
3617 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3618 | { | |
3619 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3620 | } | |
3621 | ||
3622 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3623 | { | |
3624 | wb_domain_exit(&memcg->cgwb_domain); | |
3625 | } | |
3626 | ||
2529bb3a TH |
3627 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3628 | { | |
3629 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3630 | } | |
3631 | ||
841710aa TH |
3632 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3633 | { | |
3634 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3635 | ||
3636 | if (!memcg->css.parent) | |
3637 | return NULL; | |
3638 | ||
3639 | return &memcg->cgwb_domain; | |
3640 | } | |
3641 | ||
c2aa723a TH |
3642 | /** |
3643 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3644 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3645 | * @pfilepages: out parameter for number of file pages |
3646 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3647 | * @pdirty: out parameter for number of dirty pages |
3648 | * @pwriteback: out parameter for number of pages under writeback | |
3649 | * | |
c5edf9cd TH |
3650 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3651 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3652 | * is a bit more involved. | |
c2aa723a | 3653 | * |
c5edf9cd TH |
3654 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3655 | * headroom is calculated as the lowest headroom of itself and the | |
3656 | * ancestors. Note that this doesn't consider the actual amount of | |
3657 | * available memory in the system. The caller should further cap | |
3658 | * *@pheadroom accordingly. | |
c2aa723a | 3659 | */ |
c5edf9cd TH |
3660 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3661 | unsigned long *pheadroom, unsigned long *pdirty, | |
3662 | unsigned long *pwriteback) | |
c2aa723a TH |
3663 | { |
3664 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3665 | struct mem_cgroup *parent; | |
c2aa723a TH |
3666 | |
3667 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); | |
3668 | ||
3669 | /* this should eventually include NR_UNSTABLE_NFS */ | |
3670 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); | |
c5edf9cd TH |
3671 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3672 | (1 << LRU_ACTIVE_FILE)); | |
3673 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3674 | |
c2aa723a TH |
3675 | while ((parent = parent_mem_cgroup(memcg))) { |
3676 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); | |
3677 | unsigned long used = page_counter_read(&memcg->memory); | |
3678 | ||
c5edf9cd | 3679 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3680 | memcg = parent; |
3681 | } | |
c2aa723a TH |
3682 | } |
3683 | ||
841710aa TH |
3684 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3685 | ||
3686 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3687 | { | |
3688 | return 0; | |
3689 | } | |
3690 | ||
3691 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3692 | { | |
3693 | } | |
3694 | ||
2529bb3a TH |
3695 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3696 | { | |
3697 | } | |
3698 | ||
52ebea74 TH |
3699 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3700 | ||
3bc942f3 TH |
3701 | /* |
3702 | * DO NOT USE IN NEW FILES. | |
3703 | * | |
3704 | * "cgroup.event_control" implementation. | |
3705 | * | |
3706 | * This is way over-engineered. It tries to support fully configurable | |
3707 | * events for each user. Such level of flexibility is completely | |
3708 | * unnecessary especially in the light of the planned unified hierarchy. | |
3709 | * | |
3710 | * Please deprecate this and replace with something simpler if at all | |
3711 | * possible. | |
3712 | */ | |
3713 | ||
79bd9814 TH |
3714 | /* |
3715 | * Unregister event and free resources. | |
3716 | * | |
3717 | * Gets called from workqueue. | |
3718 | */ | |
3bc942f3 | 3719 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3720 | { |
3bc942f3 TH |
3721 | struct mem_cgroup_event *event = |
3722 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3723 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3724 | |
3725 | remove_wait_queue(event->wqh, &event->wait); | |
3726 | ||
59b6f873 | 3727 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3728 | |
3729 | /* Notify userspace the event is going away. */ | |
3730 | eventfd_signal(event->eventfd, 1); | |
3731 | ||
3732 | eventfd_ctx_put(event->eventfd); | |
3733 | kfree(event); | |
59b6f873 | 3734 | css_put(&memcg->css); |
79bd9814 TH |
3735 | } |
3736 | ||
3737 | /* | |
3738 | * Gets called on POLLHUP on eventfd when user closes it. | |
3739 | * | |
3740 | * Called with wqh->lock held and interrupts disabled. | |
3741 | */ | |
3bc942f3 TH |
3742 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
3743 | int sync, void *key) | |
79bd9814 | 3744 | { |
3bc942f3 TH |
3745 | struct mem_cgroup_event *event = |
3746 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3747 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3748 | unsigned long flags = (unsigned long)key; |
3749 | ||
3750 | if (flags & POLLHUP) { | |
3751 | /* | |
3752 | * If the event has been detached at cgroup removal, we | |
3753 | * can simply return knowing the other side will cleanup | |
3754 | * for us. | |
3755 | * | |
3756 | * We can't race against event freeing since the other | |
3757 | * side will require wqh->lock via remove_wait_queue(), | |
3758 | * which we hold. | |
3759 | */ | |
fba94807 | 3760 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3761 | if (!list_empty(&event->list)) { |
3762 | list_del_init(&event->list); | |
3763 | /* | |
3764 | * We are in atomic context, but cgroup_event_remove() | |
3765 | * may sleep, so we have to call it in workqueue. | |
3766 | */ | |
3767 | schedule_work(&event->remove); | |
3768 | } | |
fba94807 | 3769 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3770 | } |
3771 | ||
3772 | return 0; | |
3773 | } | |
3774 | ||
3bc942f3 | 3775 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3776 | wait_queue_head_t *wqh, poll_table *pt) |
3777 | { | |
3bc942f3 TH |
3778 | struct mem_cgroup_event *event = |
3779 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3780 | |
3781 | event->wqh = wqh; | |
3782 | add_wait_queue(wqh, &event->wait); | |
3783 | } | |
3784 | ||
3785 | /* | |
3bc942f3 TH |
3786 | * DO NOT USE IN NEW FILES. |
3787 | * | |
79bd9814 TH |
3788 | * Parse input and register new cgroup event handler. |
3789 | * | |
3790 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3791 | * Interpretation of args is defined by control file implementation. | |
3792 | */ | |
451af504 TH |
3793 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
3794 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 3795 | { |
451af504 | 3796 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 3797 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 3798 | struct mem_cgroup_event *event; |
79bd9814 TH |
3799 | struct cgroup_subsys_state *cfile_css; |
3800 | unsigned int efd, cfd; | |
3801 | struct fd efile; | |
3802 | struct fd cfile; | |
fba94807 | 3803 | const char *name; |
79bd9814 TH |
3804 | char *endp; |
3805 | int ret; | |
3806 | ||
451af504 TH |
3807 | buf = strstrip(buf); |
3808 | ||
3809 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
3810 | if (*endp != ' ') |
3811 | return -EINVAL; | |
451af504 | 3812 | buf = endp + 1; |
79bd9814 | 3813 | |
451af504 | 3814 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
3815 | if ((*endp != ' ') && (*endp != '\0')) |
3816 | return -EINVAL; | |
451af504 | 3817 | buf = endp + 1; |
79bd9814 TH |
3818 | |
3819 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3820 | if (!event) | |
3821 | return -ENOMEM; | |
3822 | ||
59b6f873 | 3823 | event->memcg = memcg; |
79bd9814 | 3824 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
3825 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
3826 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
3827 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
3828 | |
3829 | efile = fdget(efd); | |
3830 | if (!efile.file) { | |
3831 | ret = -EBADF; | |
3832 | goto out_kfree; | |
3833 | } | |
3834 | ||
3835 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
3836 | if (IS_ERR(event->eventfd)) { | |
3837 | ret = PTR_ERR(event->eventfd); | |
3838 | goto out_put_efile; | |
3839 | } | |
3840 | ||
3841 | cfile = fdget(cfd); | |
3842 | if (!cfile.file) { | |
3843 | ret = -EBADF; | |
3844 | goto out_put_eventfd; | |
3845 | } | |
3846 | ||
3847 | /* the process need read permission on control file */ | |
3848 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
3849 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
3850 | if (ret < 0) | |
3851 | goto out_put_cfile; | |
3852 | ||
fba94807 TH |
3853 | /* |
3854 | * Determine the event callbacks and set them in @event. This used | |
3855 | * to be done via struct cftype but cgroup core no longer knows | |
3856 | * about these events. The following is crude but the whole thing | |
3857 | * is for compatibility anyway. | |
3bc942f3 TH |
3858 | * |
3859 | * DO NOT ADD NEW FILES. | |
fba94807 | 3860 | */ |
b583043e | 3861 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
3862 | |
3863 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
3864 | event->register_event = mem_cgroup_usage_register_event; | |
3865 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
3866 | } else if (!strcmp(name, "memory.oom_control")) { | |
3867 | event->register_event = mem_cgroup_oom_register_event; | |
3868 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
3869 | } else if (!strcmp(name, "memory.pressure_level")) { | |
3870 | event->register_event = vmpressure_register_event; | |
3871 | event->unregister_event = vmpressure_unregister_event; | |
3872 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
3873 | event->register_event = memsw_cgroup_usage_register_event; |
3874 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
3875 | } else { |
3876 | ret = -EINVAL; | |
3877 | goto out_put_cfile; | |
3878 | } | |
3879 | ||
79bd9814 | 3880 | /* |
b5557c4c TH |
3881 | * Verify @cfile should belong to @css. Also, remaining events are |
3882 | * automatically removed on cgroup destruction but the removal is | |
3883 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 3884 | */ |
b583043e | 3885 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 3886 | &memory_cgrp_subsys); |
79bd9814 | 3887 | ret = -EINVAL; |
5a17f543 | 3888 | if (IS_ERR(cfile_css)) |
79bd9814 | 3889 | goto out_put_cfile; |
5a17f543 TH |
3890 | if (cfile_css != css) { |
3891 | css_put(cfile_css); | |
79bd9814 | 3892 | goto out_put_cfile; |
5a17f543 | 3893 | } |
79bd9814 | 3894 | |
451af504 | 3895 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
3896 | if (ret) |
3897 | goto out_put_css; | |
3898 | ||
3899 | efile.file->f_op->poll(efile.file, &event->pt); | |
3900 | ||
fba94807 TH |
3901 | spin_lock(&memcg->event_list_lock); |
3902 | list_add(&event->list, &memcg->event_list); | |
3903 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
3904 | |
3905 | fdput(cfile); | |
3906 | fdput(efile); | |
3907 | ||
451af504 | 3908 | return nbytes; |
79bd9814 TH |
3909 | |
3910 | out_put_css: | |
b5557c4c | 3911 | css_put(css); |
79bd9814 TH |
3912 | out_put_cfile: |
3913 | fdput(cfile); | |
3914 | out_put_eventfd: | |
3915 | eventfd_ctx_put(event->eventfd); | |
3916 | out_put_efile: | |
3917 | fdput(efile); | |
3918 | out_kfree: | |
3919 | kfree(event); | |
3920 | ||
3921 | return ret; | |
3922 | } | |
3923 | ||
241994ed | 3924 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 3925 | { |
0eea1030 | 3926 | .name = "usage_in_bytes", |
8c7c6e34 | 3927 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 3928 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3929 | }, |
c84872e1 PE |
3930 | { |
3931 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 3932 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 3933 | .write = mem_cgroup_reset, |
791badbd | 3934 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 3935 | }, |
8cdea7c0 | 3936 | { |
0eea1030 | 3937 | .name = "limit_in_bytes", |
8c7c6e34 | 3938 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 3939 | .write = mem_cgroup_write, |
791badbd | 3940 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3941 | }, |
296c81d8 BS |
3942 | { |
3943 | .name = "soft_limit_in_bytes", | |
3944 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 3945 | .write = mem_cgroup_write, |
791badbd | 3946 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 3947 | }, |
8cdea7c0 BS |
3948 | { |
3949 | .name = "failcnt", | |
8c7c6e34 | 3950 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 3951 | .write = mem_cgroup_reset, |
791badbd | 3952 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3953 | }, |
d2ceb9b7 KH |
3954 | { |
3955 | .name = "stat", | |
2da8ca82 | 3956 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 3957 | }, |
c1e862c1 KH |
3958 | { |
3959 | .name = "force_empty", | |
6770c64e | 3960 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 3961 | }, |
18f59ea7 BS |
3962 | { |
3963 | .name = "use_hierarchy", | |
3964 | .write_u64 = mem_cgroup_hierarchy_write, | |
3965 | .read_u64 = mem_cgroup_hierarchy_read, | |
3966 | }, | |
79bd9814 | 3967 | { |
3bc942f3 | 3968 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 3969 | .write = memcg_write_event_control, |
7dbdb199 | 3970 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 3971 | }, |
a7885eb8 KM |
3972 | { |
3973 | .name = "swappiness", | |
3974 | .read_u64 = mem_cgroup_swappiness_read, | |
3975 | .write_u64 = mem_cgroup_swappiness_write, | |
3976 | }, | |
7dc74be0 DN |
3977 | { |
3978 | .name = "move_charge_at_immigrate", | |
3979 | .read_u64 = mem_cgroup_move_charge_read, | |
3980 | .write_u64 = mem_cgroup_move_charge_write, | |
3981 | }, | |
9490ff27 KH |
3982 | { |
3983 | .name = "oom_control", | |
2da8ca82 | 3984 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 3985 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
3986 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
3987 | }, | |
70ddf637 AV |
3988 | { |
3989 | .name = "pressure_level", | |
70ddf637 | 3990 | }, |
406eb0c9 YH |
3991 | #ifdef CONFIG_NUMA |
3992 | { | |
3993 | .name = "numa_stat", | |
2da8ca82 | 3994 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
3995 | }, |
3996 | #endif | |
510fc4e1 GC |
3997 | { |
3998 | .name = "kmem.limit_in_bytes", | |
3999 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4000 | .write = mem_cgroup_write, |
791badbd | 4001 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4002 | }, |
4003 | { | |
4004 | .name = "kmem.usage_in_bytes", | |
4005 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4006 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4007 | }, |
4008 | { | |
4009 | .name = "kmem.failcnt", | |
4010 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4011 | .write = mem_cgroup_reset, |
791badbd | 4012 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4013 | }, |
4014 | { | |
4015 | .name = "kmem.max_usage_in_bytes", | |
4016 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4017 | .write = mem_cgroup_reset, |
791badbd | 4018 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4019 | }, |
749c5415 GC |
4020 | #ifdef CONFIG_SLABINFO |
4021 | { | |
4022 | .name = "kmem.slabinfo", | |
b047501c VD |
4023 | .seq_start = slab_start, |
4024 | .seq_next = slab_next, | |
4025 | .seq_stop = slab_stop, | |
4026 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4027 | }, |
4028 | #endif | |
d55f90bf VD |
4029 | { |
4030 | .name = "kmem.tcp.limit_in_bytes", | |
4031 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4032 | .write = mem_cgroup_write, | |
4033 | .read_u64 = mem_cgroup_read_u64, | |
4034 | }, | |
4035 | { | |
4036 | .name = "kmem.tcp.usage_in_bytes", | |
4037 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4038 | .read_u64 = mem_cgroup_read_u64, | |
4039 | }, | |
4040 | { | |
4041 | .name = "kmem.tcp.failcnt", | |
4042 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4043 | .write = mem_cgroup_reset, | |
4044 | .read_u64 = mem_cgroup_read_u64, | |
4045 | }, | |
4046 | { | |
4047 | .name = "kmem.tcp.max_usage_in_bytes", | |
4048 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4049 | .write = mem_cgroup_reset, | |
4050 | .read_u64 = mem_cgroup_read_u64, | |
4051 | }, | |
6bc10349 | 4052 | { }, /* terminate */ |
af36f906 | 4053 | }; |
8c7c6e34 | 4054 | |
73f576c0 JW |
4055 | /* |
4056 | * Private memory cgroup IDR | |
4057 | * | |
4058 | * Swap-out records and page cache shadow entries need to store memcg | |
4059 | * references in constrained space, so we maintain an ID space that is | |
4060 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4061 | * memory-controlled cgroups to 64k. | |
4062 | * | |
4063 | * However, there usually are many references to the oflline CSS after | |
4064 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4065 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4066 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4067 | * relatively small ID space and prevent the creation of new cgroups | |
4068 | * even when there are much fewer than 64k cgroups - possibly none. | |
4069 | * | |
4070 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4071 | * be freed and recycled when it's no longer needed, which is usually | |
4072 | * when the CSS is offlined. | |
4073 | * | |
4074 | * The only exception to that are records of swapped out tmpfs/shmem | |
4075 | * pages that need to be attributed to live ancestors on swapin. But | |
4076 | * those references are manageable from userspace. | |
4077 | */ | |
4078 | ||
4079 | static DEFINE_IDR(mem_cgroup_idr); | |
4080 | ||
4081 | static void mem_cgroup_id_get(struct mem_cgroup *memcg) | |
4082 | { | |
4083 | atomic_inc(&memcg->id.ref); | |
4084 | } | |
4085 | ||
4086 | static void mem_cgroup_id_put(struct mem_cgroup *memcg) | |
4087 | { | |
4088 | if (atomic_dec_and_test(&memcg->id.ref)) { | |
4089 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
4090 | memcg->id.id = 0; | |
4091 | ||
4092 | /* Memcg ID pins CSS */ | |
4093 | css_put(&memcg->css); | |
4094 | } | |
4095 | } | |
4096 | ||
4097 | /** | |
4098 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4099 | * @id: the memcg id to look up | |
4100 | * | |
4101 | * Caller must hold rcu_read_lock(). | |
4102 | */ | |
4103 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4104 | { | |
4105 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4106 | return idr_find(&mem_cgroup_idr, id); | |
4107 | } | |
4108 | ||
c0ff4b85 | 4109 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4110 | { |
4111 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4112 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4113 | int zone, tmp = node; |
1ecaab2b KH |
4114 | /* |
4115 | * This routine is called against possible nodes. | |
4116 | * But it's BUG to call kmalloc() against offline node. | |
4117 | * | |
4118 | * TODO: this routine can waste much memory for nodes which will | |
4119 | * never be onlined. It's better to use memory hotplug callback | |
4120 | * function. | |
4121 | */ | |
41e3355d KH |
4122 | if (!node_state(node, N_NORMAL_MEMORY)) |
4123 | tmp = -1; | |
17295c88 | 4124 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4125 | if (!pn) |
4126 | return 1; | |
1ecaab2b | 4127 | |
1ecaab2b KH |
4128 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4129 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4130 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4131 | mz->usage_in_excess = 0; |
4132 | mz->on_tree = false; | |
d79154bb | 4133 | mz->memcg = memcg; |
1ecaab2b | 4134 | } |
54f72fe0 | 4135 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4136 | return 0; |
4137 | } | |
4138 | ||
c0ff4b85 | 4139 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4140 | { |
54f72fe0 | 4141 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4142 | } |
4143 | ||
0b8f73e1 | 4144 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4145 | { |
c8b2a36f | 4146 | int node; |
59927fb9 | 4147 | |
0b8f73e1 | 4148 | memcg_wb_domain_exit(memcg); |
c8b2a36f GC |
4149 | for_each_node(node) |
4150 | free_mem_cgroup_per_zone_info(memcg, node); | |
c8b2a36f | 4151 | free_percpu(memcg->stat); |
8ff69e2c | 4152 | kfree(memcg); |
59927fb9 | 4153 | } |
3afe36b1 | 4154 | |
0b8f73e1 | 4155 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4156 | { |
d142e3e6 | 4157 | struct mem_cgroup *memcg; |
0b8f73e1 | 4158 | size_t size; |
6d12e2d8 | 4159 | int node; |
8cdea7c0 | 4160 | |
0b8f73e1 JW |
4161 | size = sizeof(struct mem_cgroup); |
4162 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4163 | ||
4164 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4165 | if (!memcg) |
0b8f73e1 JW |
4166 | return NULL; |
4167 | ||
73f576c0 JW |
4168 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
4169 | 1, MEM_CGROUP_ID_MAX, | |
4170 | GFP_KERNEL); | |
4171 | if (memcg->id.id < 0) | |
4172 | goto fail; | |
4173 | ||
0b8f73e1 JW |
4174 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4175 | if (!memcg->stat) | |
4176 | goto fail; | |
78fb7466 | 4177 | |
3ed28fa1 | 4178 | for_each_node(node) |
c0ff4b85 | 4179 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
0b8f73e1 | 4180 | goto fail; |
f64c3f54 | 4181 | |
0b8f73e1 JW |
4182 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4183 | goto fail; | |
28dbc4b6 | 4184 | |
f7e1cb6e | 4185 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4186 | memcg->last_scanned_node = MAX_NUMNODES; |
4187 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4188 | mutex_init(&memcg->thresholds_lock); |
4189 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4190 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4191 | INIT_LIST_HEAD(&memcg->event_list); |
4192 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4193 | memcg->socket_pressure = jiffies; |
127424c8 | 4194 | #ifndef CONFIG_SLOB |
900a38f0 | 4195 | memcg->kmemcg_id = -1; |
900a38f0 | 4196 | #endif |
52ebea74 TH |
4197 | #ifdef CONFIG_CGROUP_WRITEBACK |
4198 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4199 | #endif | |
73f576c0 | 4200 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
4201 | return memcg; |
4202 | fail: | |
73f576c0 JW |
4203 | if (memcg->id.id > 0) |
4204 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
0b8f73e1 JW |
4205 | mem_cgroup_free(memcg); |
4206 | return NULL; | |
d142e3e6 GC |
4207 | } |
4208 | ||
0b8f73e1 JW |
4209 | static struct cgroup_subsys_state * __ref |
4210 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4211 | { |
0b8f73e1 JW |
4212 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4213 | struct mem_cgroup *memcg; | |
4214 | long error = -ENOMEM; | |
d142e3e6 | 4215 | |
0b8f73e1 JW |
4216 | memcg = mem_cgroup_alloc(); |
4217 | if (!memcg) | |
4218 | return ERR_PTR(error); | |
d142e3e6 | 4219 | |
0b8f73e1 JW |
4220 | memcg->high = PAGE_COUNTER_MAX; |
4221 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4222 | if (parent) { | |
4223 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4224 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4225 | } | |
4226 | if (parent && parent->use_hierarchy) { | |
4227 | memcg->use_hierarchy = true; | |
3e32cb2e | 4228 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4229 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4230 | page_counter_init(&memcg->memsw, &parent->memsw); |
4231 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4232 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4233 | } else { |
3e32cb2e | 4234 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4235 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4236 | page_counter_init(&memcg->memsw, NULL); |
4237 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4238 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4239 | /* |
4240 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4241 | * much sense so let cgroup subsystem know about this | |
4242 | * unfortunate state in our controller. | |
4243 | */ | |
d142e3e6 | 4244 | if (parent != root_mem_cgroup) |
073219e9 | 4245 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4246 | } |
d6441637 | 4247 | |
0b8f73e1 JW |
4248 | /* The following stuff does not apply to the root */ |
4249 | if (!parent) { | |
4250 | root_mem_cgroup = memcg; | |
4251 | return &memcg->css; | |
4252 | } | |
4253 | ||
b313aeee | 4254 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
4255 | if (error) |
4256 | goto fail; | |
127424c8 | 4257 | |
f7e1cb6e | 4258 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4259 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4260 | |
0b8f73e1 JW |
4261 | return &memcg->css; |
4262 | fail: | |
4263 | mem_cgroup_free(memcg); | |
ea3a9645 | 4264 | return ERR_PTR(-ENOMEM); |
0b8f73e1 JW |
4265 | } |
4266 | ||
73f576c0 | 4267 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 4268 | { |
73f576c0 JW |
4269 | /* Online state pins memcg ID, memcg ID pins CSS */ |
4270 | mem_cgroup_id_get(mem_cgroup_from_css(css)); | |
4271 | css_get(css); | |
2f7dd7a4 | 4272 | return 0; |
8cdea7c0 BS |
4273 | } |
4274 | ||
eb95419b | 4275 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4276 | { |
eb95419b | 4277 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4278 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4279 | |
4280 | /* | |
4281 | * Unregister events and notify userspace. | |
4282 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4283 | * directory to avoid race between userspace and kernelspace. | |
4284 | */ | |
fba94807 TH |
4285 | spin_lock(&memcg->event_list_lock); |
4286 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4287 | list_del_init(&event->list); |
4288 | schedule_work(&event->remove); | |
4289 | } | |
fba94807 | 4290 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4291 | |
567e9ab2 | 4292 | memcg_offline_kmem(memcg); |
52ebea74 | 4293 | wb_memcg_offline(memcg); |
73f576c0 JW |
4294 | |
4295 | mem_cgroup_id_put(memcg); | |
df878fb0 KH |
4296 | } |
4297 | ||
6df38689 VD |
4298 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4299 | { | |
4300 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4301 | ||
4302 | invalidate_reclaim_iterators(memcg); | |
4303 | } | |
4304 | ||
eb95419b | 4305 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4306 | { |
eb95419b | 4307 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4308 | |
f7e1cb6e | 4309 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4310 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4311 | |
0db15298 | 4312 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4313 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4314 | |
0b8f73e1 JW |
4315 | vmpressure_cleanup(&memcg->vmpressure); |
4316 | cancel_work_sync(&memcg->high_work); | |
4317 | mem_cgroup_remove_from_trees(memcg); | |
d886f4e4 | 4318 | memcg_free_kmem(memcg); |
0b8f73e1 | 4319 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4320 | } |
4321 | ||
1ced953b TH |
4322 | /** |
4323 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4324 | * @css: the target css | |
4325 | * | |
4326 | * Reset the states of the mem_cgroup associated with @css. This is | |
4327 | * invoked when the userland requests disabling on the default hierarchy | |
4328 | * but the memcg is pinned through dependency. The memcg should stop | |
4329 | * applying policies and should revert to the vanilla state as it may be | |
4330 | * made visible again. | |
4331 | * | |
4332 | * The current implementation only resets the essential configurations. | |
4333 | * This needs to be expanded to cover all the visible parts. | |
4334 | */ | |
4335 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4336 | { | |
4337 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4338 | ||
d334c9bc VD |
4339 | page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX); |
4340 | page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX); | |
4341 | page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX); | |
4342 | page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX); | |
4343 | page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
241994ed JW |
4344 | memcg->low = 0; |
4345 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4346 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4347 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4348 | } |
4349 | ||
02491447 | 4350 | #ifdef CONFIG_MMU |
7dc74be0 | 4351 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4352 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4353 | { |
05b84301 | 4354 | int ret; |
9476db97 | 4355 | |
d0164adc MG |
4356 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4357 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4358 | if (!ret) { |
854ffa8d | 4359 | mc.precharge += count; |
854ffa8d DN |
4360 | return ret; |
4361 | } | |
9476db97 JW |
4362 | |
4363 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4364 | while (count--) { |
00501b53 | 4365 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
38c5d72f | 4366 | if (ret) |
38c5d72f | 4367 | return ret; |
854ffa8d | 4368 | mc.precharge++; |
9476db97 | 4369 | cond_resched(); |
854ffa8d | 4370 | } |
9476db97 | 4371 | return 0; |
4ffef5fe DN |
4372 | } |
4373 | ||
4ffef5fe DN |
4374 | union mc_target { |
4375 | struct page *page; | |
02491447 | 4376 | swp_entry_t ent; |
4ffef5fe DN |
4377 | }; |
4378 | ||
4ffef5fe | 4379 | enum mc_target_type { |
8d32ff84 | 4380 | MC_TARGET_NONE = 0, |
4ffef5fe | 4381 | MC_TARGET_PAGE, |
02491447 | 4382 | MC_TARGET_SWAP, |
4ffef5fe DN |
4383 | }; |
4384 | ||
90254a65 DN |
4385 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4386 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4387 | { |
90254a65 | 4388 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4389 | |
90254a65 DN |
4390 | if (!page || !page_mapped(page)) |
4391 | return NULL; | |
4392 | if (PageAnon(page)) { | |
1dfab5ab | 4393 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4394 | return NULL; |
1dfab5ab JW |
4395 | } else { |
4396 | if (!(mc.flags & MOVE_FILE)) | |
4397 | return NULL; | |
4398 | } | |
90254a65 DN |
4399 | if (!get_page_unless_zero(page)) |
4400 | return NULL; | |
4401 | ||
4402 | return page; | |
4403 | } | |
4404 | ||
4b91355e | 4405 | #ifdef CONFIG_SWAP |
90254a65 | 4406 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 4407 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 4408 | { |
90254a65 DN |
4409 | struct page *page = NULL; |
4410 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4411 | ||
1dfab5ab | 4412 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4413 | return NULL; |
4b91355e KH |
4414 | /* |
4415 | * Because lookup_swap_cache() updates some statistics counter, | |
4416 | * we call find_get_page() with swapper_space directly. | |
4417 | */ | |
33806f06 | 4418 | page = find_get_page(swap_address_space(ent), ent.val); |
7941d214 | 4419 | if (do_memsw_account()) |
90254a65 DN |
4420 | entry->val = ent.val; |
4421 | ||
4422 | return page; | |
4423 | } | |
4b91355e KH |
4424 | #else |
4425 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 4426 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
4427 | { |
4428 | return NULL; | |
4429 | } | |
4430 | #endif | |
90254a65 | 4431 | |
87946a72 DN |
4432 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4433 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4434 | { | |
4435 | struct page *page = NULL; | |
87946a72 DN |
4436 | struct address_space *mapping; |
4437 | pgoff_t pgoff; | |
4438 | ||
4439 | if (!vma->vm_file) /* anonymous vma */ | |
4440 | return NULL; | |
1dfab5ab | 4441 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4442 | return NULL; |
4443 | ||
87946a72 | 4444 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4445 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4446 | |
4447 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4448 | #ifdef CONFIG_SWAP |
4449 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4450 | if (shmem_mapping(mapping)) { |
4451 | page = find_get_entry(mapping, pgoff); | |
4452 | if (radix_tree_exceptional_entry(page)) { | |
4453 | swp_entry_t swp = radix_to_swp_entry(page); | |
7941d214 | 4454 | if (do_memsw_account()) |
139b6a6f JW |
4455 | *entry = swp; |
4456 | page = find_get_page(swap_address_space(swp), swp.val); | |
4457 | } | |
4458 | } else | |
4459 | page = find_get_page(mapping, pgoff); | |
4460 | #else | |
4461 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4462 | #endif |
87946a72 DN |
4463 | return page; |
4464 | } | |
4465 | ||
b1b0deab CG |
4466 | /** |
4467 | * mem_cgroup_move_account - move account of the page | |
4468 | * @page: the page | |
25843c2b | 4469 | * @compound: charge the page as compound or small page |
b1b0deab CG |
4470 | * @from: mem_cgroup which the page is moved from. |
4471 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4472 | * | |
3ac808fd | 4473 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
4474 | * |
4475 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4476 | * from old cgroup. | |
4477 | */ | |
4478 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 4479 | bool compound, |
b1b0deab CG |
4480 | struct mem_cgroup *from, |
4481 | struct mem_cgroup *to) | |
4482 | { | |
4483 | unsigned long flags; | |
f627c2f5 | 4484 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 4485 | int ret; |
c4843a75 | 4486 | bool anon; |
b1b0deab CG |
4487 | |
4488 | VM_BUG_ON(from == to); | |
4489 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 4490 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
4491 | |
4492 | /* | |
6a93ca8f | 4493 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 4494 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 4495 | */ |
f627c2f5 | 4496 | ret = -EBUSY; |
b1b0deab CG |
4497 | if (!trylock_page(page)) |
4498 | goto out; | |
4499 | ||
4500 | ret = -EINVAL; | |
4501 | if (page->mem_cgroup != from) | |
4502 | goto out_unlock; | |
4503 | ||
c4843a75 GT |
4504 | anon = PageAnon(page); |
4505 | ||
b1b0deab CG |
4506 | spin_lock_irqsave(&from->move_lock, flags); |
4507 | ||
c4843a75 | 4508 | if (!anon && page_mapped(page)) { |
b1b0deab CG |
4509 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
4510 | nr_pages); | |
4511 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
4512 | nr_pages); | |
4513 | } | |
4514 | ||
c4843a75 GT |
4515 | /* |
4516 | * move_lock grabbed above and caller set from->moving_account, so | |
4517 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. | |
4518 | * So mapping should be stable for dirty pages. | |
4519 | */ | |
4520 | if (!anon && PageDirty(page)) { | |
4521 | struct address_space *mapping = page_mapping(page); | |
4522 | ||
4523 | if (mapping_cap_account_dirty(mapping)) { | |
4524 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4525 | nr_pages); | |
4526 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4527 | nr_pages); | |
4528 | } | |
4529 | } | |
4530 | ||
b1b0deab CG |
4531 | if (PageWriteback(page)) { |
4532 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4533 | nr_pages); | |
4534 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4535 | nr_pages); | |
4536 | } | |
4537 | ||
4538 | /* | |
4539 | * It is safe to change page->mem_cgroup here because the page | |
4540 | * is referenced, charged, and isolated - we can't race with | |
4541 | * uncharging, charging, migration, or LRU putback. | |
4542 | */ | |
4543 | ||
4544 | /* caller should have done css_get */ | |
4545 | page->mem_cgroup = to; | |
4546 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4547 | ||
4548 | ret = 0; | |
4549 | ||
4550 | local_irq_disable(); | |
f627c2f5 | 4551 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 4552 | memcg_check_events(to, page); |
f627c2f5 | 4553 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
4554 | memcg_check_events(from, page); |
4555 | local_irq_enable(); | |
4556 | out_unlock: | |
4557 | unlock_page(page); | |
4558 | out: | |
4559 | return ret; | |
4560 | } | |
4561 | ||
7cf7806c LR |
4562 | /** |
4563 | * get_mctgt_type - get target type of moving charge | |
4564 | * @vma: the vma the pte to be checked belongs | |
4565 | * @addr: the address corresponding to the pte to be checked | |
4566 | * @ptent: the pte to be checked | |
4567 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
4568 | * | |
4569 | * Returns | |
4570 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4571 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4572 | * move charge. if @target is not NULL, the page is stored in target->page | |
4573 | * with extra refcnt got(Callers should handle it). | |
4574 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
4575 | * target for charge migration. if @target is not NULL, the entry is stored | |
4576 | * in target->ent. | |
4577 | * | |
4578 | * Called with pte lock held. | |
4579 | */ | |
4580 | ||
8d32ff84 | 4581 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4582 | unsigned long addr, pte_t ptent, union mc_target *target) |
4583 | { | |
4584 | struct page *page = NULL; | |
8d32ff84 | 4585 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4586 | swp_entry_t ent = { .val = 0 }; |
4587 | ||
4588 | if (pte_present(ptent)) | |
4589 | page = mc_handle_present_pte(vma, addr, ptent); | |
4590 | else if (is_swap_pte(ptent)) | |
48406ef8 | 4591 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 4592 | else if (pte_none(ptent)) |
87946a72 | 4593 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4594 | |
4595 | if (!page && !ent.val) | |
8d32ff84 | 4596 | return ret; |
02491447 | 4597 | if (page) { |
02491447 | 4598 | /* |
0a31bc97 | 4599 | * Do only loose check w/o serialization. |
1306a85a | 4600 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4601 | * not under LRU exclusion. |
02491447 | 4602 | */ |
1306a85a | 4603 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4604 | ret = MC_TARGET_PAGE; |
4605 | if (target) | |
4606 | target->page = page; | |
4607 | } | |
4608 | if (!ret || !target) | |
4609 | put_page(page); | |
4610 | } | |
90254a65 DN |
4611 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4612 | if (ent.val && !ret && | |
34c00c31 | 4613 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4614 | ret = MC_TARGET_SWAP; |
4615 | if (target) | |
4616 | target->ent = ent; | |
4ffef5fe | 4617 | } |
4ffef5fe DN |
4618 | return ret; |
4619 | } | |
4620 | ||
12724850 NH |
4621 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4622 | /* | |
4623 | * We don't consider swapping or file mapped pages because THP does not | |
4624 | * support them for now. | |
4625 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4626 | */ | |
4627 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4628 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4629 | { | |
4630 | struct page *page = NULL; | |
12724850 NH |
4631 | enum mc_target_type ret = MC_TARGET_NONE; |
4632 | ||
4633 | page = pmd_page(pmd); | |
309381fe | 4634 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4635 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4636 | return ret; |
1306a85a | 4637 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4638 | ret = MC_TARGET_PAGE; |
4639 | if (target) { | |
4640 | get_page(page); | |
4641 | target->page = page; | |
4642 | } | |
4643 | } | |
4644 | return ret; | |
4645 | } | |
4646 | #else | |
4647 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4648 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4649 | { | |
4650 | return MC_TARGET_NONE; | |
4651 | } | |
4652 | #endif | |
4653 | ||
4ffef5fe DN |
4654 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4655 | unsigned long addr, unsigned long end, | |
4656 | struct mm_walk *walk) | |
4657 | { | |
26bcd64a | 4658 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4659 | pte_t *pte; |
4660 | spinlock_t *ptl; | |
4661 | ||
b6ec57f4 KS |
4662 | ptl = pmd_trans_huge_lock(pmd, vma); |
4663 | if (ptl) { | |
12724850 NH |
4664 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4665 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4666 | spin_unlock(ptl); |
1a5a9906 | 4667 | return 0; |
12724850 | 4668 | } |
03319327 | 4669 | |
45f83cef AA |
4670 | if (pmd_trans_unstable(pmd)) |
4671 | return 0; | |
4ffef5fe DN |
4672 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4673 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4674 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4675 | mc.precharge++; /* increment precharge temporarily */ |
4676 | pte_unmap_unlock(pte - 1, ptl); | |
4677 | cond_resched(); | |
4678 | ||
7dc74be0 DN |
4679 | return 0; |
4680 | } | |
4681 | ||
4ffef5fe DN |
4682 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4683 | { | |
4684 | unsigned long precharge; | |
4ffef5fe | 4685 | |
26bcd64a NH |
4686 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4687 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4688 | .mm = mm, | |
4689 | }; | |
dfe076b0 | 4690 | down_read(&mm->mmap_sem); |
26bcd64a | 4691 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
dfe076b0 | 4692 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4693 | |
4694 | precharge = mc.precharge; | |
4695 | mc.precharge = 0; | |
4696 | ||
4697 | return precharge; | |
4698 | } | |
4699 | ||
4ffef5fe DN |
4700 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4701 | { | |
dfe076b0 DN |
4702 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4703 | ||
4704 | VM_BUG_ON(mc.moving_task); | |
4705 | mc.moving_task = current; | |
4706 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4707 | } |
4708 | ||
dfe076b0 DN |
4709 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4710 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4711 | { |
2bd9bb20 KH |
4712 | struct mem_cgroup *from = mc.from; |
4713 | struct mem_cgroup *to = mc.to; | |
4714 | ||
4ffef5fe | 4715 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4716 | if (mc.precharge) { |
00501b53 | 4717 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4718 | mc.precharge = 0; |
4719 | } | |
4720 | /* | |
4721 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4722 | * we must uncharge here. | |
4723 | */ | |
4724 | if (mc.moved_charge) { | |
00501b53 | 4725 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4726 | mc.moved_charge = 0; |
4ffef5fe | 4727 | } |
483c30b5 DN |
4728 | /* we must fixup refcnts and charges */ |
4729 | if (mc.moved_swap) { | |
483c30b5 | 4730 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4731 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4732 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4733 | |
05b84301 | 4734 | /* |
3e32cb2e JW |
4735 | * we charged both to->memory and to->memsw, so we |
4736 | * should uncharge to->memory. | |
05b84301 | 4737 | */ |
ce00a967 | 4738 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4739 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4740 | ||
e8ea14cc | 4741 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 4742 | |
4050377b | 4743 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
4744 | mc.moved_swap = 0; |
4745 | } | |
dfe076b0 DN |
4746 | memcg_oom_recover(from); |
4747 | memcg_oom_recover(to); | |
4748 | wake_up_all(&mc.waitq); | |
4749 | } | |
4750 | ||
4751 | static void mem_cgroup_clear_mc(void) | |
4752 | { | |
264a0ae1 TH |
4753 | struct mm_struct *mm = mc.mm; |
4754 | ||
dfe076b0 DN |
4755 | /* |
4756 | * we must clear moving_task before waking up waiters at the end of | |
4757 | * task migration. | |
4758 | */ | |
4759 | mc.moving_task = NULL; | |
4760 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4761 | spin_lock(&mc.lock); |
4ffef5fe DN |
4762 | mc.from = NULL; |
4763 | mc.to = NULL; | |
264a0ae1 | 4764 | mc.mm = NULL; |
2bd9bb20 | 4765 | spin_unlock(&mc.lock); |
264a0ae1 TH |
4766 | |
4767 | mmput(mm); | |
4ffef5fe DN |
4768 | } |
4769 | ||
1f7dd3e5 | 4770 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4771 | { |
1f7dd3e5 | 4772 | struct cgroup_subsys_state *css; |
eed67d75 | 4773 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 4774 | struct mem_cgroup *from; |
4530eddb | 4775 | struct task_struct *leader, *p; |
9f2115f9 | 4776 | struct mm_struct *mm; |
1dfab5ab | 4777 | unsigned long move_flags; |
9f2115f9 | 4778 | int ret = 0; |
7dc74be0 | 4779 | |
1f7dd3e5 TH |
4780 | /* charge immigration isn't supported on the default hierarchy */ |
4781 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
4782 | return 0; |
4783 | ||
4530eddb TH |
4784 | /* |
4785 | * Multi-process migrations only happen on the default hierarchy | |
4786 | * where charge immigration is not used. Perform charge | |
4787 | * immigration if @tset contains a leader and whine if there are | |
4788 | * multiple. | |
4789 | */ | |
4790 | p = NULL; | |
1f7dd3e5 | 4791 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
4792 | WARN_ON_ONCE(p); |
4793 | p = leader; | |
1f7dd3e5 | 4794 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
4795 | } |
4796 | if (!p) | |
4797 | return 0; | |
4798 | ||
1f7dd3e5 TH |
4799 | /* |
4800 | * We are now commited to this value whatever it is. Changes in this | |
4801 | * tunable will only affect upcoming migrations, not the current one. | |
4802 | * So we need to save it, and keep it going. | |
4803 | */ | |
4804 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
4805 | if (!move_flags) | |
4806 | return 0; | |
4807 | ||
9f2115f9 TH |
4808 | from = mem_cgroup_from_task(p); |
4809 | ||
4810 | VM_BUG_ON(from == memcg); | |
4811 | ||
4812 | mm = get_task_mm(p); | |
4813 | if (!mm) | |
4814 | return 0; | |
4815 | /* We move charges only when we move a owner of the mm */ | |
4816 | if (mm->owner == p) { | |
4817 | VM_BUG_ON(mc.from); | |
4818 | VM_BUG_ON(mc.to); | |
4819 | VM_BUG_ON(mc.precharge); | |
4820 | VM_BUG_ON(mc.moved_charge); | |
4821 | VM_BUG_ON(mc.moved_swap); | |
4822 | ||
4823 | spin_lock(&mc.lock); | |
264a0ae1 | 4824 | mc.mm = mm; |
9f2115f9 TH |
4825 | mc.from = from; |
4826 | mc.to = memcg; | |
4827 | mc.flags = move_flags; | |
4828 | spin_unlock(&mc.lock); | |
4829 | /* We set mc.moving_task later */ | |
4830 | ||
4831 | ret = mem_cgroup_precharge_mc(mm); | |
4832 | if (ret) | |
4833 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
4834 | } else { |
4835 | mmput(mm); | |
7dc74be0 DN |
4836 | } |
4837 | return ret; | |
4838 | } | |
4839 | ||
1f7dd3e5 | 4840 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4841 | { |
4e2f245d JW |
4842 | if (mc.to) |
4843 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
4844 | } |
4845 | ||
4ffef5fe DN |
4846 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4847 | unsigned long addr, unsigned long end, | |
4848 | struct mm_walk *walk) | |
7dc74be0 | 4849 | { |
4ffef5fe | 4850 | int ret = 0; |
26bcd64a | 4851 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4852 | pte_t *pte; |
4853 | spinlock_t *ptl; | |
12724850 NH |
4854 | enum mc_target_type target_type; |
4855 | union mc_target target; | |
4856 | struct page *page; | |
4ffef5fe | 4857 | |
b6ec57f4 KS |
4858 | ptl = pmd_trans_huge_lock(pmd, vma); |
4859 | if (ptl) { | |
62ade86a | 4860 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 4861 | spin_unlock(ptl); |
12724850 NH |
4862 | return 0; |
4863 | } | |
4864 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
4865 | if (target_type == MC_TARGET_PAGE) { | |
4866 | page = target.page; | |
4867 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 4868 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 4869 | mc.from, mc.to)) { |
12724850 NH |
4870 | mc.precharge -= HPAGE_PMD_NR; |
4871 | mc.moved_charge += HPAGE_PMD_NR; | |
4872 | } | |
4873 | putback_lru_page(page); | |
4874 | } | |
4875 | put_page(page); | |
4876 | } | |
bf929152 | 4877 | spin_unlock(ptl); |
1a5a9906 | 4878 | return 0; |
12724850 NH |
4879 | } |
4880 | ||
45f83cef AA |
4881 | if (pmd_trans_unstable(pmd)) |
4882 | return 0; | |
4ffef5fe DN |
4883 | retry: |
4884 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4885 | for (; addr != end; addr += PAGE_SIZE) { | |
4886 | pte_t ptent = *(pte++); | |
02491447 | 4887 | swp_entry_t ent; |
4ffef5fe DN |
4888 | |
4889 | if (!mc.precharge) | |
4890 | break; | |
4891 | ||
8d32ff84 | 4892 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
4893 | case MC_TARGET_PAGE: |
4894 | page = target.page; | |
53f9263b KS |
4895 | /* |
4896 | * We can have a part of the split pmd here. Moving it | |
4897 | * can be done but it would be too convoluted so simply | |
4898 | * ignore such a partial THP and keep it in original | |
4899 | * memcg. There should be somebody mapping the head. | |
4900 | */ | |
4901 | if (PageTransCompound(page)) | |
4902 | goto put; | |
4ffef5fe DN |
4903 | if (isolate_lru_page(page)) |
4904 | goto put; | |
f627c2f5 KS |
4905 | if (!mem_cgroup_move_account(page, false, |
4906 | mc.from, mc.to)) { | |
4ffef5fe | 4907 | mc.precharge--; |
854ffa8d DN |
4908 | /* we uncharge from mc.from later. */ |
4909 | mc.moved_charge++; | |
4ffef5fe DN |
4910 | } |
4911 | putback_lru_page(page); | |
8d32ff84 | 4912 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
4913 | put_page(page); |
4914 | break; | |
02491447 DN |
4915 | case MC_TARGET_SWAP: |
4916 | ent = target.ent; | |
e91cbb42 | 4917 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 4918 | mc.precharge--; |
483c30b5 DN |
4919 | /* we fixup refcnts and charges later. */ |
4920 | mc.moved_swap++; | |
4921 | } | |
02491447 | 4922 | break; |
4ffef5fe DN |
4923 | default: |
4924 | break; | |
4925 | } | |
4926 | } | |
4927 | pte_unmap_unlock(pte - 1, ptl); | |
4928 | cond_resched(); | |
4929 | ||
4930 | if (addr != end) { | |
4931 | /* | |
4932 | * We have consumed all precharges we got in can_attach(). | |
4933 | * We try charge one by one, but don't do any additional | |
4934 | * charges to mc.to if we have failed in charge once in attach() | |
4935 | * phase. | |
4936 | */ | |
854ffa8d | 4937 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4938 | if (!ret) |
4939 | goto retry; | |
4940 | } | |
4941 | ||
4942 | return ret; | |
4943 | } | |
4944 | ||
264a0ae1 | 4945 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 4946 | { |
26bcd64a NH |
4947 | struct mm_walk mem_cgroup_move_charge_walk = { |
4948 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
264a0ae1 | 4949 | .mm = mc.mm, |
26bcd64a | 4950 | }; |
4ffef5fe DN |
4951 | |
4952 | lru_add_drain_all(); | |
312722cb | 4953 | /* |
81f8c3a4 JW |
4954 | * Signal lock_page_memcg() to take the memcg's move_lock |
4955 | * while we're moving its pages to another memcg. Then wait | |
4956 | * for already started RCU-only updates to finish. | |
312722cb JW |
4957 | */ |
4958 | atomic_inc(&mc.from->moving_account); | |
4959 | synchronize_rcu(); | |
dfe076b0 | 4960 | retry: |
264a0ae1 | 4961 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
4962 | /* |
4963 | * Someone who are holding the mmap_sem might be waiting in | |
4964 | * waitq. So we cancel all extra charges, wake up all waiters, | |
4965 | * and retry. Because we cancel precharges, we might not be able | |
4966 | * to move enough charges, but moving charge is a best-effort | |
4967 | * feature anyway, so it wouldn't be a big problem. | |
4968 | */ | |
4969 | __mem_cgroup_clear_mc(); | |
4970 | cond_resched(); | |
4971 | goto retry; | |
4972 | } | |
26bcd64a NH |
4973 | /* |
4974 | * When we have consumed all precharges and failed in doing | |
4975 | * additional charge, the page walk just aborts. | |
4976 | */ | |
4977 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); | |
264a0ae1 | 4978 | up_read(&mc.mm->mmap_sem); |
312722cb | 4979 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
4980 | } |
4981 | ||
264a0ae1 | 4982 | static void mem_cgroup_move_task(void) |
67e465a7 | 4983 | { |
264a0ae1 TH |
4984 | if (mc.to) { |
4985 | mem_cgroup_move_charge(); | |
a433658c | 4986 | mem_cgroup_clear_mc(); |
264a0ae1 | 4987 | } |
67e465a7 | 4988 | } |
5cfb80a7 | 4989 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 4990 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
4991 | { |
4992 | return 0; | |
4993 | } | |
1f7dd3e5 | 4994 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
4995 | { |
4996 | } | |
264a0ae1 | 4997 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
4998 | { |
4999 | } | |
5000 | #endif | |
67e465a7 | 5001 | |
f00baae7 TH |
5002 | /* |
5003 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5004 | * to verify whether we're attached to the default hierarchy on each mount |
5005 | * attempt. | |
f00baae7 | 5006 | */ |
eb95419b | 5007 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5008 | { |
5009 | /* | |
aa6ec29b | 5010 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5011 | * guarantees that @root doesn't have any children, so turning it |
5012 | * on for the root memcg is enough. | |
5013 | */ | |
9e10a130 | 5014 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5015 | root_mem_cgroup->use_hierarchy = true; |
5016 | else | |
5017 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5018 | } |
5019 | ||
241994ed JW |
5020 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5021 | struct cftype *cft) | |
5022 | { | |
f5fc3c5d JW |
5023 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5024 | ||
5025 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5026 | } |
5027 | ||
5028 | static int memory_low_show(struct seq_file *m, void *v) | |
5029 | { | |
5030 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5031 | unsigned long low = READ_ONCE(memcg->low); |
241994ed JW |
5032 | |
5033 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 5034 | seq_puts(m, "max\n"); |
241994ed JW |
5035 | else |
5036 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5037 | ||
5038 | return 0; | |
5039 | } | |
5040 | ||
5041 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5042 | char *buf, size_t nbytes, loff_t off) | |
5043 | { | |
5044 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5045 | unsigned long low; | |
5046 | int err; | |
5047 | ||
5048 | buf = strstrip(buf); | |
d2973697 | 5049 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5050 | if (err) |
5051 | return err; | |
5052 | ||
5053 | memcg->low = low; | |
5054 | ||
5055 | return nbytes; | |
5056 | } | |
5057 | ||
5058 | static int memory_high_show(struct seq_file *m, void *v) | |
5059 | { | |
5060 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5061 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
5062 | |
5063 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5064 | seq_puts(m, "max\n"); |
241994ed JW |
5065 | else |
5066 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5067 | ||
5068 | return 0; | |
5069 | } | |
5070 | ||
5071 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5072 | char *buf, size_t nbytes, loff_t off) | |
5073 | { | |
5074 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 5075 | unsigned long nr_pages; |
241994ed JW |
5076 | unsigned long high; |
5077 | int err; | |
5078 | ||
5079 | buf = strstrip(buf); | |
d2973697 | 5080 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5081 | if (err) |
5082 | return err; | |
5083 | ||
5084 | memcg->high = high; | |
5085 | ||
588083bb JW |
5086 | nr_pages = page_counter_read(&memcg->memory); |
5087 | if (nr_pages > high) | |
5088 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5089 | GFP_KERNEL, true); | |
5090 | ||
2529bb3a | 5091 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5092 | return nbytes; |
5093 | } | |
5094 | ||
5095 | static int memory_max_show(struct seq_file *m, void *v) | |
5096 | { | |
5097 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5098 | unsigned long max = READ_ONCE(memcg->memory.limit); |
241994ed JW |
5099 | |
5100 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5101 | seq_puts(m, "max\n"); |
241994ed JW |
5102 | else |
5103 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5104 | ||
5105 | return 0; | |
5106 | } | |
5107 | ||
5108 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5109 | char *buf, size_t nbytes, loff_t off) | |
5110 | { | |
5111 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
5112 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5113 | bool drained = false; | |
241994ed JW |
5114 | unsigned long max; |
5115 | int err; | |
5116 | ||
5117 | buf = strstrip(buf); | |
d2973697 | 5118 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5119 | if (err) |
5120 | return err; | |
5121 | ||
b6e6edcf JW |
5122 | xchg(&memcg->memory.limit, max); |
5123 | ||
5124 | for (;;) { | |
5125 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5126 | ||
5127 | if (nr_pages <= max) | |
5128 | break; | |
5129 | ||
5130 | if (signal_pending(current)) { | |
5131 | err = -EINTR; | |
5132 | break; | |
5133 | } | |
5134 | ||
5135 | if (!drained) { | |
5136 | drain_all_stock(memcg); | |
5137 | drained = true; | |
5138 | continue; | |
5139 | } | |
5140 | ||
5141 | if (nr_reclaims) { | |
5142 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5143 | GFP_KERNEL, true)) | |
5144 | nr_reclaims--; | |
5145 | continue; | |
5146 | } | |
5147 | ||
5148 | mem_cgroup_events(memcg, MEMCG_OOM, 1); | |
5149 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) | |
5150 | break; | |
5151 | } | |
241994ed | 5152 | |
2529bb3a | 5153 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5154 | return nbytes; |
5155 | } | |
5156 | ||
5157 | static int memory_events_show(struct seq_file *m, void *v) | |
5158 | { | |
5159 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5160 | ||
5161 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5162 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5163 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5164 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5165 | ||
5166 | return 0; | |
5167 | } | |
5168 | ||
587d9f72 JW |
5169 | static int memory_stat_show(struct seq_file *m, void *v) |
5170 | { | |
5171 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
72b54e73 VD |
5172 | unsigned long stat[MEMCG_NR_STAT]; |
5173 | unsigned long events[MEMCG_NR_EVENTS]; | |
587d9f72 JW |
5174 | int i; |
5175 | ||
5176 | /* | |
5177 | * Provide statistics on the state of the memory subsystem as | |
5178 | * well as cumulative event counters that show past behavior. | |
5179 | * | |
5180 | * This list is ordered following a combination of these gradients: | |
5181 | * 1) generic big picture -> specifics and details | |
5182 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
5183 | * | |
5184 | * Current memory state: | |
5185 | */ | |
5186 | ||
72b54e73 VD |
5187 | tree_stat(memcg, stat); |
5188 | tree_events(memcg, events); | |
5189 | ||
587d9f72 | 5190 | seq_printf(m, "anon %llu\n", |
72b54e73 | 5191 | (u64)stat[MEM_CGROUP_STAT_RSS] * PAGE_SIZE); |
587d9f72 | 5192 | seq_printf(m, "file %llu\n", |
72b54e73 | 5193 | (u64)stat[MEM_CGROUP_STAT_CACHE] * PAGE_SIZE); |
12580e4b VD |
5194 | seq_printf(m, "kernel_stack %llu\n", |
5195 | (u64)stat[MEMCG_KERNEL_STACK] * PAGE_SIZE); | |
27ee57c9 VD |
5196 | seq_printf(m, "slab %llu\n", |
5197 | (u64)(stat[MEMCG_SLAB_RECLAIMABLE] + | |
5198 | stat[MEMCG_SLAB_UNRECLAIMABLE]) * PAGE_SIZE); | |
b2807f07 | 5199 | seq_printf(m, "sock %llu\n", |
72b54e73 | 5200 | (u64)stat[MEMCG_SOCK] * PAGE_SIZE); |
587d9f72 JW |
5201 | |
5202 | seq_printf(m, "file_mapped %llu\n", | |
72b54e73 | 5203 | (u64)stat[MEM_CGROUP_STAT_FILE_MAPPED] * PAGE_SIZE); |
587d9f72 | 5204 | seq_printf(m, "file_dirty %llu\n", |
72b54e73 | 5205 | (u64)stat[MEM_CGROUP_STAT_DIRTY] * PAGE_SIZE); |
587d9f72 | 5206 | seq_printf(m, "file_writeback %llu\n", |
72b54e73 | 5207 | (u64)stat[MEM_CGROUP_STAT_WRITEBACK] * PAGE_SIZE); |
587d9f72 JW |
5208 | |
5209 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5210 | struct mem_cgroup *mi; | |
5211 | unsigned long val = 0; | |
5212 | ||
5213 | for_each_mem_cgroup_tree(mi, memcg) | |
5214 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)); | |
5215 | seq_printf(m, "%s %llu\n", | |
5216 | mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE); | |
5217 | } | |
5218 | ||
27ee57c9 VD |
5219 | seq_printf(m, "slab_reclaimable %llu\n", |
5220 | (u64)stat[MEMCG_SLAB_RECLAIMABLE] * PAGE_SIZE); | |
5221 | seq_printf(m, "slab_unreclaimable %llu\n", | |
5222 | (u64)stat[MEMCG_SLAB_UNRECLAIMABLE] * PAGE_SIZE); | |
5223 | ||
587d9f72 JW |
5224 | /* Accumulated memory events */ |
5225 | ||
5226 | seq_printf(m, "pgfault %lu\n", | |
72b54e73 | 5227 | events[MEM_CGROUP_EVENTS_PGFAULT]); |
587d9f72 | 5228 | seq_printf(m, "pgmajfault %lu\n", |
72b54e73 | 5229 | events[MEM_CGROUP_EVENTS_PGMAJFAULT]); |
587d9f72 JW |
5230 | |
5231 | return 0; | |
5232 | } | |
5233 | ||
241994ed JW |
5234 | static struct cftype memory_files[] = { |
5235 | { | |
5236 | .name = "current", | |
f5fc3c5d | 5237 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5238 | .read_u64 = memory_current_read, |
5239 | }, | |
5240 | { | |
5241 | .name = "low", | |
5242 | .flags = CFTYPE_NOT_ON_ROOT, | |
5243 | .seq_show = memory_low_show, | |
5244 | .write = memory_low_write, | |
5245 | }, | |
5246 | { | |
5247 | .name = "high", | |
5248 | .flags = CFTYPE_NOT_ON_ROOT, | |
5249 | .seq_show = memory_high_show, | |
5250 | .write = memory_high_write, | |
5251 | }, | |
5252 | { | |
5253 | .name = "max", | |
5254 | .flags = CFTYPE_NOT_ON_ROOT, | |
5255 | .seq_show = memory_max_show, | |
5256 | .write = memory_max_write, | |
5257 | }, | |
5258 | { | |
5259 | .name = "events", | |
5260 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5261 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5262 | .seq_show = memory_events_show, |
5263 | }, | |
587d9f72 JW |
5264 | { |
5265 | .name = "stat", | |
5266 | .flags = CFTYPE_NOT_ON_ROOT, | |
5267 | .seq_show = memory_stat_show, | |
5268 | }, | |
241994ed JW |
5269 | { } /* terminate */ |
5270 | }; | |
5271 | ||
073219e9 | 5272 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5273 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5274 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5275 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5276 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5277 | .css_free = mem_cgroup_css_free, |
1ced953b | 5278 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5279 | .can_attach = mem_cgroup_can_attach, |
5280 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 5281 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5282 | .bind = mem_cgroup_bind, |
241994ed JW |
5283 | .dfl_cftypes = memory_files, |
5284 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5285 | .early_init = 0, |
8cdea7c0 | 5286 | }; |
c077719b | 5287 | |
241994ed JW |
5288 | /** |
5289 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5290 | * @root: the highest ancestor to consider | |
5291 | * @memcg: the memory cgroup to check | |
5292 | * | |
5293 | * Returns %true if memory consumption of @memcg, and that of all | |
5294 | * configurable ancestors up to @root, is below the normal range. | |
5295 | */ | |
5296 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5297 | { | |
5298 | if (mem_cgroup_disabled()) | |
5299 | return false; | |
5300 | ||
5301 | /* | |
5302 | * The toplevel group doesn't have a configurable range, so | |
5303 | * it's never low when looked at directly, and it is not | |
5304 | * considered an ancestor when assessing the hierarchy. | |
5305 | */ | |
5306 | ||
5307 | if (memcg == root_mem_cgroup) | |
5308 | return false; | |
5309 | ||
4e54dede | 5310 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5311 | return false; |
5312 | ||
5313 | while (memcg != root) { | |
5314 | memcg = parent_mem_cgroup(memcg); | |
5315 | ||
5316 | if (memcg == root_mem_cgroup) | |
5317 | break; | |
5318 | ||
4e54dede | 5319 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5320 | return false; |
5321 | } | |
5322 | return true; | |
5323 | } | |
5324 | ||
00501b53 JW |
5325 | /** |
5326 | * mem_cgroup_try_charge - try charging a page | |
5327 | * @page: page to charge | |
5328 | * @mm: mm context of the victim | |
5329 | * @gfp_mask: reclaim mode | |
5330 | * @memcgp: charged memcg return | |
25843c2b | 5331 | * @compound: charge the page as compound or small page |
00501b53 JW |
5332 | * |
5333 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5334 | * pages according to @gfp_mask if necessary. | |
5335 | * | |
5336 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5337 | * Otherwise, an error code is returned. | |
5338 | * | |
5339 | * After page->mapping has been set up, the caller must finalize the | |
5340 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5341 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5342 | */ | |
5343 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
5344 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
5345 | bool compound) | |
00501b53 JW |
5346 | { |
5347 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 5348 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5349 | int ret = 0; |
5350 | ||
5351 | if (mem_cgroup_disabled()) | |
5352 | goto out; | |
5353 | ||
5354 | if (PageSwapCache(page)) { | |
00501b53 JW |
5355 | /* |
5356 | * Every swap fault against a single page tries to charge the | |
5357 | * page, bail as early as possible. shmem_unuse() encounters | |
5358 | * already charged pages, too. The USED bit is protected by | |
5359 | * the page lock, which serializes swap cache removal, which | |
5360 | * in turn serializes uncharging. | |
5361 | */ | |
e993d905 | 5362 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1306a85a | 5363 | if (page->mem_cgroup) |
00501b53 | 5364 | goto out; |
e993d905 | 5365 | |
37e84351 | 5366 | if (do_swap_account) { |
e993d905 VD |
5367 | swp_entry_t ent = { .val = page_private(page), }; |
5368 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5369 | ||
5370 | rcu_read_lock(); | |
5371 | memcg = mem_cgroup_from_id(id); | |
5372 | if (memcg && !css_tryget_online(&memcg->css)) | |
5373 | memcg = NULL; | |
5374 | rcu_read_unlock(); | |
5375 | } | |
00501b53 JW |
5376 | } |
5377 | ||
00501b53 JW |
5378 | if (!memcg) |
5379 | memcg = get_mem_cgroup_from_mm(mm); | |
5380 | ||
5381 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5382 | ||
5383 | css_put(&memcg->css); | |
00501b53 JW |
5384 | out: |
5385 | *memcgp = memcg; | |
5386 | return ret; | |
5387 | } | |
5388 | ||
5389 | /** | |
5390 | * mem_cgroup_commit_charge - commit a page charge | |
5391 | * @page: page to charge | |
5392 | * @memcg: memcg to charge the page to | |
5393 | * @lrucare: page might be on LRU already | |
25843c2b | 5394 | * @compound: charge the page as compound or small page |
00501b53 JW |
5395 | * |
5396 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5397 | * after page->mapping has been set up. This must happen atomically | |
5398 | * as part of the page instantiation, i.e. under the page table lock | |
5399 | * for anonymous pages, under the page lock for page and swap cache. | |
5400 | * | |
5401 | * In addition, the page must not be on the LRU during the commit, to | |
5402 | * prevent racing with task migration. If it might be, use @lrucare. | |
5403 | * | |
5404 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5405 | */ | |
5406 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 5407 | bool lrucare, bool compound) |
00501b53 | 5408 | { |
f627c2f5 | 5409 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5410 | |
5411 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5412 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5413 | ||
5414 | if (mem_cgroup_disabled()) | |
5415 | return; | |
5416 | /* | |
5417 | * Swap faults will attempt to charge the same page multiple | |
5418 | * times. But reuse_swap_page() might have removed the page | |
5419 | * from swapcache already, so we can't check PageSwapCache(). | |
5420 | */ | |
5421 | if (!memcg) | |
5422 | return; | |
5423 | ||
6abb5a86 JW |
5424 | commit_charge(page, memcg, lrucare); |
5425 | ||
6abb5a86 | 5426 | local_irq_disable(); |
f627c2f5 | 5427 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
5428 | memcg_check_events(memcg, page); |
5429 | local_irq_enable(); | |
00501b53 | 5430 | |
7941d214 | 5431 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
5432 | swp_entry_t entry = { .val = page_private(page) }; |
5433 | /* | |
5434 | * The swap entry might not get freed for a long time, | |
5435 | * let's not wait for it. The page already received a | |
5436 | * memory+swap charge, drop the swap entry duplicate. | |
5437 | */ | |
5438 | mem_cgroup_uncharge_swap(entry); | |
5439 | } | |
5440 | } | |
5441 | ||
5442 | /** | |
5443 | * mem_cgroup_cancel_charge - cancel a page charge | |
5444 | * @page: page to charge | |
5445 | * @memcg: memcg to charge the page to | |
25843c2b | 5446 | * @compound: charge the page as compound or small page |
00501b53 JW |
5447 | * |
5448 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5449 | */ | |
f627c2f5 KS |
5450 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
5451 | bool compound) | |
00501b53 | 5452 | { |
f627c2f5 | 5453 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5454 | |
5455 | if (mem_cgroup_disabled()) | |
5456 | return; | |
5457 | /* | |
5458 | * Swap faults will attempt to charge the same page multiple | |
5459 | * times. But reuse_swap_page() might have removed the page | |
5460 | * from swapcache already, so we can't check PageSwapCache(). | |
5461 | */ | |
5462 | if (!memcg) | |
5463 | return; | |
5464 | ||
00501b53 JW |
5465 | cancel_charge(memcg, nr_pages); |
5466 | } | |
5467 | ||
747db954 | 5468 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 | 5469 | unsigned long nr_anon, unsigned long nr_file, |
5e8d35f8 VD |
5470 | unsigned long nr_huge, unsigned long nr_kmem, |
5471 | struct page *dummy_page) | |
747db954 | 5472 | { |
5e8d35f8 | 5473 | unsigned long nr_pages = nr_anon + nr_file + nr_kmem; |
747db954 JW |
5474 | unsigned long flags; |
5475 | ||
ce00a967 | 5476 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 | 5477 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 5478 | if (do_memsw_account()) |
18eca2e6 | 5479 | page_counter_uncharge(&memcg->memsw, nr_pages); |
5e8d35f8 VD |
5480 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && nr_kmem) |
5481 | page_counter_uncharge(&memcg->kmem, nr_kmem); | |
ce00a967 JW |
5482 | memcg_oom_recover(memcg); |
5483 | } | |
747db954 JW |
5484 | |
5485 | local_irq_save(flags); | |
5486 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5487 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5488 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5489 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5490 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5491 | memcg_check_events(memcg, dummy_page); |
5492 | local_irq_restore(flags); | |
e8ea14cc JW |
5493 | |
5494 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5495 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5496 | } |
5497 | ||
5498 | static void uncharge_list(struct list_head *page_list) | |
5499 | { | |
5500 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5501 | unsigned long nr_anon = 0; |
5502 | unsigned long nr_file = 0; | |
5503 | unsigned long nr_huge = 0; | |
5e8d35f8 | 5504 | unsigned long nr_kmem = 0; |
747db954 | 5505 | unsigned long pgpgout = 0; |
747db954 JW |
5506 | struct list_head *next; |
5507 | struct page *page; | |
5508 | ||
8b592656 JW |
5509 | /* |
5510 | * Note that the list can be a single page->lru; hence the | |
5511 | * do-while loop instead of a simple list_for_each_entry(). | |
5512 | */ | |
747db954 JW |
5513 | next = page_list->next; |
5514 | do { | |
747db954 JW |
5515 | page = list_entry(next, struct page, lru); |
5516 | next = page->lru.next; | |
5517 | ||
5518 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5519 | VM_BUG_ON_PAGE(page_count(page), page); | |
5520 | ||
1306a85a | 5521 | if (!page->mem_cgroup) |
747db954 JW |
5522 | continue; |
5523 | ||
5524 | /* | |
5525 | * Nobody should be changing or seriously looking at | |
1306a85a | 5526 | * page->mem_cgroup at this point, we have fully |
29833315 | 5527 | * exclusive access to the page. |
747db954 JW |
5528 | */ |
5529 | ||
1306a85a | 5530 | if (memcg != page->mem_cgroup) { |
747db954 | 5531 | if (memcg) { |
18eca2e6 | 5532 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5e8d35f8 VD |
5533 | nr_huge, nr_kmem, page); |
5534 | pgpgout = nr_anon = nr_file = | |
5535 | nr_huge = nr_kmem = 0; | |
747db954 | 5536 | } |
1306a85a | 5537 | memcg = page->mem_cgroup; |
747db954 JW |
5538 | } |
5539 | ||
5e8d35f8 VD |
5540 | if (!PageKmemcg(page)) { |
5541 | unsigned int nr_pages = 1; | |
747db954 | 5542 | |
5e8d35f8 VD |
5543 | if (PageTransHuge(page)) { |
5544 | nr_pages <<= compound_order(page); | |
5e8d35f8 VD |
5545 | nr_huge += nr_pages; |
5546 | } | |
5547 | if (PageAnon(page)) | |
5548 | nr_anon += nr_pages; | |
5549 | else | |
5550 | nr_file += nr_pages; | |
5551 | pgpgout++; | |
5552 | } else | |
5553 | nr_kmem += 1 << compound_order(page); | |
747db954 | 5554 | |
1306a85a | 5555 | page->mem_cgroup = NULL; |
747db954 JW |
5556 | } while (next != page_list); |
5557 | ||
5558 | if (memcg) | |
18eca2e6 | 5559 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5e8d35f8 | 5560 | nr_huge, nr_kmem, page); |
747db954 JW |
5561 | } |
5562 | ||
0a31bc97 JW |
5563 | /** |
5564 | * mem_cgroup_uncharge - uncharge a page | |
5565 | * @page: page to uncharge | |
5566 | * | |
5567 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5568 | * mem_cgroup_commit_charge(). | |
5569 | */ | |
5570 | void mem_cgroup_uncharge(struct page *page) | |
5571 | { | |
0a31bc97 JW |
5572 | if (mem_cgroup_disabled()) |
5573 | return; | |
5574 | ||
747db954 | 5575 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5576 | if (!page->mem_cgroup) |
0a31bc97 JW |
5577 | return; |
5578 | ||
747db954 JW |
5579 | INIT_LIST_HEAD(&page->lru); |
5580 | uncharge_list(&page->lru); | |
5581 | } | |
0a31bc97 | 5582 | |
747db954 JW |
5583 | /** |
5584 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5585 | * @page_list: list of pages to uncharge | |
5586 | * | |
5587 | * Uncharge a list of pages previously charged with | |
5588 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5589 | */ | |
5590 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5591 | { | |
5592 | if (mem_cgroup_disabled()) | |
5593 | return; | |
0a31bc97 | 5594 | |
747db954 JW |
5595 | if (!list_empty(page_list)) |
5596 | uncharge_list(page_list); | |
0a31bc97 JW |
5597 | } |
5598 | ||
5599 | /** | |
6a93ca8f JW |
5600 | * mem_cgroup_migrate - charge a page's replacement |
5601 | * @oldpage: currently circulating page | |
5602 | * @newpage: replacement page | |
0a31bc97 | 5603 | * |
6a93ca8f JW |
5604 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
5605 | * be uncharged upon free. | |
0a31bc97 JW |
5606 | * |
5607 | * Both pages must be locked, @newpage->mapping must be set up. | |
5608 | */ | |
6a93ca8f | 5609 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 5610 | { |
29833315 | 5611 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
5612 | unsigned int nr_pages; |
5613 | bool compound; | |
d93c4130 | 5614 | unsigned long flags; |
0a31bc97 JW |
5615 | |
5616 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5617 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 5618 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
5619 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5620 | newpage); | |
0a31bc97 JW |
5621 | |
5622 | if (mem_cgroup_disabled()) | |
5623 | return; | |
5624 | ||
5625 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5626 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5627 | return; |
5628 | ||
45637bab | 5629 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 5630 | memcg = oldpage->mem_cgroup; |
29833315 | 5631 | if (!memcg) |
0a31bc97 JW |
5632 | return; |
5633 | ||
44b7a8d3 JW |
5634 | /* Force-charge the new page. The old one will be freed soon */ |
5635 | compound = PageTransHuge(newpage); | |
5636 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
5637 | ||
5638 | page_counter_charge(&memcg->memory, nr_pages); | |
5639 | if (do_memsw_account()) | |
5640 | page_counter_charge(&memcg->memsw, nr_pages); | |
5641 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 5642 | |
9cf7666a | 5643 | commit_charge(newpage, memcg, false); |
44b7a8d3 | 5644 | |
d93c4130 | 5645 | local_irq_save(flags); |
44b7a8d3 JW |
5646 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
5647 | memcg_check_events(memcg, newpage); | |
d93c4130 | 5648 | local_irq_restore(flags); |
0a31bc97 JW |
5649 | } |
5650 | ||
ef12947c | 5651 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
5652 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
5653 | ||
5654 | void sock_update_memcg(struct sock *sk) | |
5655 | { | |
5656 | struct mem_cgroup *memcg; | |
5657 | ||
5658 | /* Socket cloning can throw us here with sk_cgrp already | |
5659 | * filled. It won't however, necessarily happen from | |
5660 | * process context. So the test for root memcg given | |
5661 | * the current task's memcg won't help us in this case. | |
5662 | * | |
5663 | * Respecting the original socket's memcg is a better | |
5664 | * decision in this case. | |
5665 | */ | |
5666 | if (sk->sk_memcg) { | |
5667 | BUG_ON(mem_cgroup_is_root(sk->sk_memcg)); | |
5668 | css_get(&sk->sk_memcg->css); | |
5669 | return; | |
5670 | } | |
5671 | ||
5672 | rcu_read_lock(); | |
5673 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
5674 | if (memcg == root_mem_cgroup) |
5675 | goto out; | |
0db15298 | 5676 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 5677 | goto out; |
f7e1cb6e | 5678 | if (css_tryget_online(&memcg->css)) |
11092087 | 5679 | sk->sk_memcg = memcg; |
f7e1cb6e | 5680 | out: |
11092087 JW |
5681 | rcu_read_unlock(); |
5682 | } | |
5683 | EXPORT_SYMBOL(sock_update_memcg); | |
5684 | ||
5685 | void sock_release_memcg(struct sock *sk) | |
5686 | { | |
5687 | WARN_ON(!sk->sk_memcg); | |
5688 | css_put(&sk->sk_memcg->css); | |
5689 | } | |
5690 | ||
5691 | /** | |
5692 | * mem_cgroup_charge_skmem - charge socket memory | |
5693 | * @memcg: memcg to charge | |
5694 | * @nr_pages: number of pages to charge | |
5695 | * | |
5696 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
5697 | * @memcg's configured limit, %false if the charge had to be forced. | |
5698 | */ | |
5699 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5700 | { | |
f7e1cb6e | 5701 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 5702 | |
f7e1cb6e | 5703 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5704 | struct page_counter *fail; |
f7e1cb6e | 5705 | |
0db15298 JW |
5706 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
5707 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
5708 | return true; |
5709 | } | |
0db15298 JW |
5710 | page_counter_charge(&memcg->tcpmem, nr_pages); |
5711 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 5712 | return false; |
11092087 | 5713 | } |
d886f4e4 | 5714 | |
f7e1cb6e JW |
5715 | /* Don't block in the packet receive path */ |
5716 | if (in_softirq()) | |
5717 | gfp_mask = GFP_NOWAIT; | |
5718 | ||
b2807f07 JW |
5719 | this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages); |
5720 | ||
f7e1cb6e JW |
5721 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
5722 | return true; | |
5723 | ||
5724 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
5725 | return false; |
5726 | } | |
5727 | ||
5728 | /** | |
5729 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
5730 | * @memcg - memcg to uncharge | |
5731 | * @nr_pages - number of pages to uncharge | |
5732 | */ | |
5733 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5734 | { | |
f7e1cb6e | 5735 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5736 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
5737 | return; |
5738 | } | |
d886f4e4 | 5739 | |
b2807f07 JW |
5740 | this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages); |
5741 | ||
f7e1cb6e JW |
5742 | page_counter_uncharge(&memcg->memory, nr_pages); |
5743 | css_put_many(&memcg->css, nr_pages); | |
11092087 JW |
5744 | } |
5745 | ||
f7e1cb6e JW |
5746 | static int __init cgroup_memory(char *s) |
5747 | { | |
5748 | char *token; | |
5749 | ||
5750 | while ((token = strsep(&s, ",")) != NULL) { | |
5751 | if (!*token) | |
5752 | continue; | |
5753 | if (!strcmp(token, "nosocket")) | |
5754 | cgroup_memory_nosocket = true; | |
04823c83 VD |
5755 | if (!strcmp(token, "nokmem")) |
5756 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
5757 | } |
5758 | return 0; | |
5759 | } | |
5760 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 5761 | |
2d11085e | 5762 | /* |
1081312f MH |
5763 | * subsys_initcall() for memory controller. |
5764 | * | |
5765 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
5766 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
5767 | * everything that doesn't depend on a specific mem_cgroup structure should | |
5768 | * be initialized from here. | |
2d11085e MH |
5769 | */ |
5770 | static int __init mem_cgroup_init(void) | |
5771 | { | |
95a045f6 JW |
5772 | int cpu, node; |
5773 | ||
2d11085e | 5774 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
95a045f6 JW |
5775 | |
5776 | for_each_possible_cpu(cpu) | |
5777 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5778 | drain_local_stock); | |
5779 | ||
5780 | for_each_node(node) { | |
5781 | struct mem_cgroup_tree_per_node *rtpn; | |
5782 | int zone; | |
5783 | ||
5784 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5785 | node_online(node) ? node : NUMA_NO_NODE); | |
5786 | ||
5787 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5788 | struct mem_cgroup_tree_per_zone *rtpz; | |
5789 | ||
5790 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5791 | rtpz->rb_root = RB_ROOT; | |
5792 | spin_lock_init(&rtpz->lock); | |
5793 | } | |
5794 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5795 | } | |
5796 | ||
2d11085e MH |
5797 | return 0; |
5798 | } | |
5799 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5800 | |
5801 | #ifdef CONFIG_MEMCG_SWAP | |
5802 | /** | |
5803 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5804 | * @page: page whose memsw charge to transfer | |
5805 | * @entry: swap entry to move the charge to | |
5806 | * | |
5807 | * Transfer the memsw charge of @page to @entry. | |
5808 | */ | |
5809 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5810 | { | |
5811 | struct mem_cgroup *memcg; | |
5812 | unsigned short oldid; | |
5813 | ||
5814 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5815 | VM_BUG_ON_PAGE(page_count(page), page); | |
5816 | ||
7941d214 | 5817 | if (!do_memsw_account()) |
21afa38e JW |
5818 | return; |
5819 | ||
5820 | memcg = page->mem_cgroup; | |
5821 | ||
5822 | /* Readahead page, never charged */ | |
5823 | if (!memcg) | |
5824 | return; | |
5825 | ||
73f576c0 | 5826 | mem_cgroup_id_get(memcg); |
21afa38e JW |
5827 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); |
5828 | VM_BUG_ON_PAGE(oldid, page); | |
5829 | mem_cgroup_swap_statistics(memcg, true); | |
5830 | ||
5831 | page->mem_cgroup = NULL; | |
5832 | ||
5833 | if (!mem_cgroup_is_root(memcg)) | |
5834 | page_counter_uncharge(&memcg->memory, 1); | |
5835 | ||
ce9ce665 SAS |
5836 | /* |
5837 | * Interrupts should be disabled here because the caller holds the | |
5838 | * mapping->tree_lock lock which is taken with interrupts-off. It is | |
5839 | * important here to have the interrupts disabled because it is the | |
5840 | * only synchronisation we have for udpating the per-CPU variables. | |
5841 | */ | |
5842 | VM_BUG_ON(!irqs_disabled()); | |
f627c2f5 | 5843 | mem_cgroup_charge_statistics(memcg, page, false, -1); |
21afa38e | 5844 | memcg_check_events(memcg, page); |
73f576c0 JW |
5845 | |
5846 | if (!mem_cgroup_is_root(memcg)) | |
5847 | css_put(&memcg->css); | |
21afa38e JW |
5848 | } |
5849 | ||
37e84351 VD |
5850 | /* |
5851 | * mem_cgroup_try_charge_swap - try charging a swap entry | |
5852 | * @page: page being added to swap | |
5853 | * @entry: swap entry to charge | |
5854 | * | |
5855 | * Try to charge @entry to the memcg that @page belongs to. | |
5856 | * | |
5857 | * Returns 0 on success, -ENOMEM on failure. | |
5858 | */ | |
5859 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
5860 | { | |
5861 | struct mem_cgroup *memcg; | |
5862 | struct page_counter *counter; | |
5863 | unsigned short oldid; | |
5864 | ||
5865 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
5866 | return 0; | |
5867 | ||
5868 | memcg = page->mem_cgroup; | |
5869 | ||
5870 | /* Readahead page, never charged */ | |
5871 | if (!memcg) | |
5872 | return 0; | |
5873 | ||
5874 | if (!mem_cgroup_is_root(memcg) && | |
5875 | !page_counter_try_charge(&memcg->swap, 1, &counter)) | |
5876 | return -ENOMEM; | |
5877 | ||
73f576c0 | 5878 | mem_cgroup_id_get(memcg); |
37e84351 VD |
5879 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); |
5880 | VM_BUG_ON_PAGE(oldid, page); | |
5881 | mem_cgroup_swap_statistics(memcg, true); | |
5882 | ||
37e84351 VD |
5883 | return 0; |
5884 | } | |
5885 | ||
21afa38e JW |
5886 | /** |
5887 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5888 | * @entry: swap entry to uncharge | |
5889 | * | |
37e84351 | 5890 | * Drop the swap charge associated with @entry. |
21afa38e JW |
5891 | */ |
5892 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5893 | { | |
5894 | struct mem_cgroup *memcg; | |
5895 | unsigned short id; | |
5896 | ||
37e84351 | 5897 | if (!do_swap_account) |
21afa38e JW |
5898 | return; |
5899 | ||
5900 | id = swap_cgroup_record(entry, 0); | |
5901 | rcu_read_lock(); | |
adbe427b | 5902 | memcg = mem_cgroup_from_id(id); |
21afa38e | 5903 | if (memcg) { |
37e84351 VD |
5904 | if (!mem_cgroup_is_root(memcg)) { |
5905 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5906 | page_counter_uncharge(&memcg->swap, 1); | |
5907 | else | |
5908 | page_counter_uncharge(&memcg->memsw, 1); | |
5909 | } | |
21afa38e | 5910 | mem_cgroup_swap_statistics(memcg, false); |
73f576c0 | 5911 | mem_cgroup_id_put(memcg); |
21afa38e JW |
5912 | } |
5913 | rcu_read_unlock(); | |
5914 | } | |
5915 | ||
d8b38438 VD |
5916 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
5917 | { | |
5918 | long nr_swap_pages = get_nr_swap_pages(); | |
5919 | ||
5920 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5921 | return nr_swap_pages; | |
5922 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
5923 | nr_swap_pages = min_t(long, nr_swap_pages, | |
5924 | READ_ONCE(memcg->swap.limit) - | |
5925 | page_counter_read(&memcg->swap)); | |
5926 | return nr_swap_pages; | |
5927 | } | |
5928 | ||
5ccc5aba VD |
5929 | bool mem_cgroup_swap_full(struct page *page) |
5930 | { | |
5931 | struct mem_cgroup *memcg; | |
5932 | ||
5933 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
5934 | ||
5935 | if (vm_swap_full()) | |
5936 | return true; | |
5937 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
5938 | return false; | |
5939 | ||
5940 | memcg = page->mem_cgroup; | |
5941 | if (!memcg) | |
5942 | return false; | |
5943 | ||
5944 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
5945 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit) | |
5946 | return true; | |
5947 | ||
5948 | return false; | |
5949 | } | |
5950 | ||
21afa38e JW |
5951 | /* for remember boot option*/ |
5952 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
5953 | static int really_do_swap_account __initdata = 1; | |
5954 | #else | |
5955 | static int really_do_swap_account __initdata; | |
5956 | #endif | |
5957 | ||
5958 | static int __init enable_swap_account(char *s) | |
5959 | { | |
5960 | if (!strcmp(s, "1")) | |
5961 | really_do_swap_account = 1; | |
5962 | else if (!strcmp(s, "0")) | |
5963 | really_do_swap_account = 0; | |
5964 | return 1; | |
5965 | } | |
5966 | __setup("swapaccount=", enable_swap_account); | |
5967 | ||
37e84351 VD |
5968 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
5969 | struct cftype *cft) | |
5970 | { | |
5971 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5972 | ||
5973 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
5974 | } | |
5975 | ||
5976 | static int swap_max_show(struct seq_file *m, void *v) | |
5977 | { | |
5978 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5979 | unsigned long max = READ_ONCE(memcg->swap.limit); | |
5980 | ||
5981 | if (max == PAGE_COUNTER_MAX) | |
5982 | seq_puts(m, "max\n"); | |
5983 | else | |
5984 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5985 | ||
5986 | return 0; | |
5987 | } | |
5988 | ||
5989 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
5990 | char *buf, size_t nbytes, loff_t off) | |
5991 | { | |
5992 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5993 | unsigned long max; | |
5994 | int err; | |
5995 | ||
5996 | buf = strstrip(buf); | |
5997 | err = page_counter_memparse(buf, "max", &max); | |
5998 | if (err) | |
5999 | return err; | |
6000 | ||
6001 | mutex_lock(&memcg_limit_mutex); | |
6002 | err = page_counter_limit(&memcg->swap, max); | |
6003 | mutex_unlock(&memcg_limit_mutex); | |
6004 | if (err) | |
6005 | return err; | |
6006 | ||
6007 | return nbytes; | |
6008 | } | |
6009 | ||
6010 | static struct cftype swap_files[] = { | |
6011 | { | |
6012 | .name = "swap.current", | |
6013 | .flags = CFTYPE_NOT_ON_ROOT, | |
6014 | .read_u64 = swap_current_read, | |
6015 | }, | |
6016 | { | |
6017 | .name = "swap.max", | |
6018 | .flags = CFTYPE_NOT_ON_ROOT, | |
6019 | .seq_show = swap_max_show, | |
6020 | .write = swap_max_write, | |
6021 | }, | |
6022 | { } /* terminate */ | |
6023 | }; | |
6024 | ||
21afa38e JW |
6025 | static struct cftype memsw_cgroup_files[] = { |
6026 | { | |
6027 | .name = "memsw.usage_in_bytes", | |
6028 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6029 | .read_u64 = mem_cgroup_read_u64, | |
6030 | }, | |
6031 | { | |
6032 | .name = "memsw.max_usage_in_bytes", | |
6033 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6034 | .write = mem_cgroup_reset, | |
6035 | .read_u64 = mem_cgroup_read_u64, | |
6036 | }, | |
6037 | { | |
6038 | .name = "memsw.limit_in_bytes", | |
6039 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6040 | .write = mem_cgroup_write, | |
6041 | .read_u64 = mem_cgroup_read_u64, | |
6042 | }, | |
6043 | { | |
6044 | .name = "memsw.failcnt", | |
6045 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6046 | .write = mem_cgroup_reset, | |
6047 | .read_u64 = mem_cgroup_read_u64, | |
6048 | }, | |
6049 | { }, /* terminate */ | |
6050 | }; | |
6051 | ||
6052 | static int __init mem_cgroup_swap_init(void) | |
6053 | { | |
6054 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6055 | do_swap_account = 1; | |
37e84351 VD |
6056 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
6057 | swap_files)); | |
21afa38e JW |
6058 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6059 | memsw_cgroup_files)); | |
6060 | } | |
6061 | return 0; | |
6062 | } | |
6063 | subsys_initcall(mem_cgroup_swap_init); | |
6064 | ||
6065 | #endif /* CONFIG_MEMCG_SWAP */ |