Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
7ae1e1d0 GC |
13 | * Kernel Memory Controller |
14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
15 | * Authors: Glauber Costa and Suleiman Souhlal | |
16 | * | |
1575e68b JW |
17 | * Native page reclaim |
18 | * Charge lifetime sanitation | |
19 | * Lockless page tracking & accounting | |
20 | * Unified hierarchy configuration model | |
21 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
22 | * | |
8cdea7c0 BS |
23 | * This program is free software; you can redistribute it and/or modify |
24 | * it under the terms of the GNU General Public License as published by | |
25 | * the Free Software Foundation; either version 2 of the License, or | |
26 | * (at your option) any later version. | |
27 | * | |
28 | * This program is distributed in the hope that it will be useful, | |
29 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
30 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
31 | * GNU General Public License for more details. | |
32 | */ | |
33 | ||
3e32cb2e | 34 | #include <linux/page_counter.h> |
8cdea7c0 BS |
35 | #include <linux/memcontrol.h> |
36 | #include <linux/cgroup.h> | |
78fb7466 | 37 | #include <linux/mm.h> |
4ffef5fe | 38 | #include <linux/hugetlb.h> |
d13d1443 | 39 | #include <linux/pagemap.h> |
d52aa412 | 40 | #include <linux/smp.h> |
8a9f3ccd | 41 | #include <linux/page-flags.h> |
66e1707b | 42 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
43 | #include <linux/bit_spinlock.h> |
44 | #include <linux/rcupdate.h> | |
e222432b | 45 | #include <linux/limits.h> |
b9e15baf | 46 | #include <linux/export.h> |
8c7c6e34 | 47 | #include <linux/mutex.h> |
bb4cc1a8 | 48 | #include <linux/rbtree.h> |
b6ac57d5 | 49 | #include <linux/slab.h> |
66e1707b | 50 | #include <linux/swap.h> |
02491447 | 51 | #include <linux/swapops.h> |
66e1707b | 52 | #include <linux/spinlock.h> |
2e72b634 | 53 | #include <linux/eventfd.h> |
79bd9814 | 54 | #include <linux/poll.h> |
2e72b634 | 55 | #include <linux/sort.h> |
66e1707b | 56 | #include <linux/fs.h> |
d2ceb9b7 | 57 | #include <linux/seq_file.h> |
70ddf637 | 58 | #include <linux/vmpressure.h> |
b69408e8 | 59 | #include <linux/mm_inline.h> |
5d1ea48b | 60 | #include <linux/swap_cgroup.h> |
cdec2e42 | 61 | #include <linux/cpu.h> |
158e0a2d | 62 | #include <linux/oom.h> |
0056f4e6 | 63 | #include <linux/lockdep.h> |
79bd9814 | 64 | #include <linux/file.h> |
b23afb93 | 65 | #include <linux/tracehook.h> |
08e552c6 | 66 | #include "internal.h" |
d1a4c0b3 | 67 | #include <net/sock.h> |
4bd2c1ee | 68 | #include <net/ip.h> |
f35c3a8e | 69 | #include "slab.h" |
8cdea7c0 | 70 | |
8697d331 BS |
71 | #include <asm/uaccess.h> |
72 | ||
cc8e970c KM |
73 | #include <trace/events/vmscan.h> |
74 | ||
073219e9 TH |
75 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
76 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 77 | |
7d828602 JW |
78 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
79 | ||
a181b0e8 | 80 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 81 | |
f7e1cb6e JW |
82 | /* Socket memory accounting disabled? */ |
83 | static bool cgroup_memory_nosocket; | |
84 | ||
04823c83 VD |
85 | /* Kernel memory accounting disabled? */ |
86 | static bool cgroup_memory_nokmem; | |
87 | ||
21afa38e | 88 | /* Whether the swap controller is active */ |
c255a458 | 89 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 90 | int do_swap_account __read_mostly; |
c077719b | 91 | #else |
a0db00fc | 92 | #define do_swap_account 0 |
c077719b KH |
93 | #endif |
94 | ||
7941d214 JW |
95 | /* Whether legacy memory+swap accounting is active */ |
96 | static bool do_memsw_account(void) | |
97 | { | |
98 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; | |
99 | } | |
100 | ||
af7c4b0e JW |
101 | static const char * const mem_cgroup_stat_names[] = { |
102 | "cache", | |
103 | "rss", | |
b070e65c | 104 | "rss_huge", |
af7c4b0e | 105 | "mapped_file", |
c4843a75 | 106 | "dirty", |
3ea67d06 | 107 | "writeback", |
af7c4b0e JW |
108 | "swap", |
109 | }; | |
110 | ||
af7c4b0e JW |
111 | static const char * const mem_cgroup_events_names[] = { |
112 | "pgpgin", | |
113 | "pgpgout", | |
114 | "pgfault", | |
115 | "pgmajfault", | |
116 | }; | |
117 | ||
58cf188e SZ |
118 | static const char * const mem_cgroup_lru_names[] = { |
119 | "inactive_anon", | |
120 | "active_anon", | |
121 | "inactive_file", | |
122 | "active_file", | |
123 | "unevictable", | |
124 | }; | |
125 | ||
a0db00fc KS |
126 | #define THRESHOLDS_EVENTS_TARGET 128 |
127 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
128 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 129 | |
bb4cc1a8 AM |
130 | /* |
131 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
132 | * their hierarchy representation | |
133 | */ | |
134 | ||
135 | struct mem_cgroup_tree_per_zone { | |
136 | struct rb_root rb_root; | |
137 | spinlock_t lock; | |
138 | }; | |
139 | ||
140 | struct mem_cgroup_tree_per_node { | |
141 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
142 | }; | |
143 | ||
144 | struct mem_cgroup_tree { | |
145 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
146 | }; | |
147 | ||
148 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
149 | ||
9490ff27 KH |
150 | /* for OOM */ |
151 | struct mem_cgroup_eventfd_list { | |
152 | struct list_head list; | |
153 | struct eventfd_ctx *eventfd; | |
154 | }; | |
2e72b634 | 155 | |
79bd9814 TH |
156 | /* |
157 | * cgroup_event represents events which userspace want to receive. | |
158 | */ | |
3bc942f3 | 159 | struct mem_cgroup_event { |
79bd9814 | 160 | /* |
59b6f873 | 161 | * memcg which the event belongs to. |
79bd9814 | 162 | */ |
59b6f873 | 163 | struct mem_cgroup *memcg; |
79bd9814 TH |
164 | /* |
165 | * eventfd to signal userspace about the event. | |
166 | */ | |
167 | struct eventfd_ctx *eventfd; | |
168 | /* | |
169 | * Each of these stored in a list by the cgroup. | |
170 | */ | |
171 | struct list_head list; | |
fba94807 TH |
172 | /* |
173 | * register_event() callback will be used to add new userspace | |
174 | * waiter for changes related to this event. Use eventfd_signal() | |
175 | * on eventfd to send notification to userspace. | |
176 | */ | |
59b6f873 | 177 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 178 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
179 | /* |
180 | * unregister_event() callback will be called when userspace closes | |
181 | * the eventfd or on cgroup removing. This callback must be set, | |
182 | * if you want provide notification functionality. | |
183 | */ | |
59b6f873 | 184 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 185 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
186 | /* |
187 | * All fields below needed to unregister event when | |
188 | * userspace closes eventfd. | |
189 | */ | |
190 | poll_table pt; | |
191 | wait_queue_head_t *wqh; | |
192 | wait_queue_t wait; | |
193 | struct work_struct remove; | |
194 | }; | |
195 | ||
c0ff4b85 R |
196 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
197 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 198 | |
7dc74be0 DN |
199 | /* Stuffs for move charges at task migration. */ |
200 | /* | |
1dfab5ab | 201 | * Types of charges to be moved. |
7dc74be0 | 202 | */ |
1dfab5ab JW |
203 | #define MOVE_ANON 0x1U |
204 | #define MOVE_FILE 0x2U | |
205 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 206 | |
4ffef5fe DN |
207 | /* "mc" and its members are protected by cgroup_mutex */ |
208 | static struct move_charge_struct { | |
b1dd693e | 209 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
210 | struct mem_cgroup *from; |
211 | struct mem_cgroup *to; | |
1dfab5ab | 212 | unsigned long flags; |
4ffef5fe | 213 | unsigned long precharge; |
854ffa8d | 214 | unsigned long moved_charge; |
483c30b5 | 215 | unsigned long moved_swap; |
8033b97c DN |
216 | struct task_struct *moving_task; /* a task moving charges */ |
217 | wait_queue_head_t waitq; /* a waitq for other context */ | |
218 | } mc = { | |
2bd9bb20 | 219 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
220 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
221 | }; | |
4ffef5fe | 222 | |
4e416953 BS |
223 | /* |
224 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
225 | * limit reclaim to prevent infinite loops, if they ever occur. | |
226 | */ | |
a0db00fc | 227 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 228 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 229 | |
217bc319 KH |
230 | enum charge_type { |
231 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 232 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 233 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 234 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
235 | NR_CHARGE_TYPE, |
236 | }; | |
237 | ||
8c7c6e34 | 238 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
239 | enum res_type { |
240 | _MEM, | |
241 | _MEMSWAP, | |
242 | _OOM_TYPE, | |
510fc4e1 | 243 | _KMEM, |
d55f90bf | 244 | _TCP, |
86ae53e1 GC |
245 | }; |
246 | ||
a0db00fc KS |
247 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
248 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 249 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
250 | /* Used for OOM nofiier */ |
251 | #define OOM_CONTROL (0) | |
8c7c6e34 | 252 | |
0999821b GC |
253 | /* |
254 | * The memcg_create_mutex will be held whenever a new cgroup is created. | |
255 | * As a consequence, any change that needs to protect against new child cgroups | |
256 | * appearing has to hold it as well. | |
257 | */ | |
258 | static DEFINE_MUTEX(memcg_create_mutex); | |
259 | ||
70ddf637 AV |
260 | /* Some nice accessors for the vmpressure. */ |
261 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
262 | { | |
263 | if (!memcg) | |
264 | memcg = root_mem_cgroup; | |
265 | return &memcg->vmpressure; | |
266 | } | |
267 | ||
268 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
269 | { | |
270 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
271 | } | |
272 | ||
7ffc0edc MH |
273 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
274 | { | |
275 | return (memcg == root_mem_cgroup); | |
276 | } | |
277 | ||
4219b2da LZ |
278 | /* |
279 | * We restrict the id in the range of [1, 65535], so it can fit into | |
280 | * an unsigned short. | |
281 | */ | |
282 | #define MEM_CGROUP_ID_MAX USHRT_MAX | |
283 | ||
34c00c31 LZ |
284 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
285 | { | |
15a4c835 | 286 | return memcg->css.id; |
34c00c31 LZ |
287 | } |
288 | ||
adbe427b VD |
289 | /* |
290 | * A helper function to get mem_cgroup from ID. must be called under | |
291 | * rcu_read_lock(). The caller is responsible for calling | |
292 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping | |
293 | * refcnt from swap can be called against removed memcg.) | |
294 | */ | |
34c00c31 LZ |
295 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) |
296 | { | |
297 | struct cgroup_subsys_state *css; | |
298 | ||
7d699ddb | 299 | css = css_from_id(id, &memory_cgrp_subsys); |
34c00c31 LZ |
300 | return mem_cgroup_from_css(css); |
301 | } | |
302 | ||
127424c8 | 303 | #ifndef CONFIG_SLOB |
55007d84 | 304 | /* |
f7ce3190 | 305 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
306 | * The main reason for not using cgroup id for this: |
307 | * this works better in sparse environments, where we have a lot of memcgs, | |
308 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
309 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
310 | * 200 entry array for that. | |
55007d84 | 311 | * |
dbcf73e2 VD |
312 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
313 | * will double each time we have to increase it. | |
55007d84 | 314 | */ |
dbcf73e2 VD |
315 | static DEFINE_IDA(memcg_cache_ida); |
316 | int memcg_nr_cache_ids; | |
749c5415 | 317 | |
05257a1a VD |
318 | /* Protects memcg_nr_cache_ids */ |
319 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
320 | ||
321 | void memcg_get_cache_ids(void) | |
322 | { | |
323 | down_read(&memcg_cache_ids_sem); | |
324 | } | |
325 | ||
326 | void memcg_put_cache_ids(void) | |
327 | { | |
328 | up_read(&memcg_cache_ids_sem); | |
329 | } | |
330 | ||
55007d84 GC |
331 | /* |
332 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
333 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
334 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
335 | * tunable, but that is strictly not necessary. | |
336 | * | |
b8627835 | 337 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
338 | * this constant directly from cgroup, but it is understandable that this is |
339 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 340 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
341 | * increase ours as well if it increases. |
342 | */ | |
343 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 344 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 345 | |
d7f25f8a GC |
346 | /* |
347 | * A lot of the calls to the cache allocation functions are expected to be | |
348 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
349 | * conditional to this static branch, we'll have to allow modules that does | |
350 | * kmem_cache_alloc and the such to see this symbol as well | |
351 | */ | |
ef12947c | 352 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 353 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 354 | |
127424c8 | 355 | #endif /* !CONFIG_SLOB */ |
a8964b9b | 356 | |
f64c3f54 | 357 | static struct mem_cgroup_per_zone * |
e231875b | 358 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
f64c3f54 | 359 | { |
e231875b JZ |
360 | int nid = zone_to_nid(zone); |
361 | int zid = zone_idx(zone); | |
362 | ||
54f72fe0 | 363 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
364 | } |
365 | ||
ad7fa852 TH |
366 | /** |
367 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
368 | * @page: page of interest | |
369 | * | |
370 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
371 | * with @page is returned. The returned css remains associated with @page | |
372 | * until it is released. | |
373 | * | |
374 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
375 | * is returned. | |
376 | * | |
377 | * XXX: The above description of behavior on the default hierarchy isn't | |
378 | * strictly true yet as replace_page_cache_page() can modify the | |
379 | * association before @page is released even on the default hierarchy; | |
380 | * however, the current and planned usages don't mix the the two functions | |
381 | * and replace_page_cache_page() will soon be updated to make the invariant | |
382 | * actually true. | |
383 | */ | |
384 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
385 | { | |
386 | struct mem_cgroup *memcg; | |
387 | ||
ad7fa852 TH |
388 | memcg = page->mem_cgroup; |
389 | ||
9e10a130 | 390 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
391 | memcg = root_mem_cgroup; |
392 | ||
ad7fa852 TH |
393 | return &memcg->css; |
394 | } | |
395 | ||
2fc04524 VD |
396 | /** |
397 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
398 | * @page: the page | |
399 | * | |
400 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
401 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
402 | * is safe to call this function without holding a reference to @page. | |
403 | * | |
404 | * Note, this function is inherently racy, because there is nothing to prevent | |
405 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
406 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
407 | * do not care (such as procfs interfaces). | |
408 | */ | |
409 | ino_t page_cgroup_ino(struct page *page) | |
410 | { | |
411 | struct mem_cgroup *memcg; | |
412 | unsigned long ino = 0; | |
413 | ||
414 | rcu_read_lock(); | |
415 | memcg = READ_ONCE(page->mem_cgroup); | |
416 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
417 | memcg = parent_mem_cgroup(memcg); | |
418 | if (memcg) | |
419 | ino = cgroup_ino(memcg->css.cgroup); | |
420 | rcu_read_unlock(); | |
421 | return ino; | |
422 | } | |
423 | ||
f64c3f54 | 424 | static struct mem_cgroup_per_zone * |
e231875b | 425 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
f64c3f54 | 426 | { |
97a6c37b JW |
427 | int nid = page_to_nid(page); |
428 | int zid = page_zonenum(page); | |
f64c3f54 | 429 | |
e231875b | 430 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
f64c3f54 BS |
431 | } |
432 | ||
bb4cc1a8 AM |
433 | static struct mem_cgroup_tree_per_zone * |
434 | soft_limit_tree_node_zone(int nid, int zid) | |
435 | { | |
436 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
437 | } | |
438 | ||
439 | static struct mem_cgroup_tree_per_zone * | |
440 | soft_limit_tree_from_page(struct page *page) | |
441 | { | |
442 | int nid = page_to_nid(page); | |
443 | int zid = page_zonenum(page); | |
444 | ||
445 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
446 | } | |
447 | ||
cf2c8127 JW |
448 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
449 | struct mem_cgroup_tree_per_zone *mctz, | |
3e32cb2e | 450 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
451 | { |
452 | struct rb_node **p = &mctz->rb_root.rb_node; | |
453 | struct rb_node *parent = NULL; | |
454 | struct mem_cgroup_per_zone *mz_node; | |
455 | ||
456 | if (mz->on_tree) | |
457 | return; | |
458 | ||
459 | mz->usage_in_excess = new_usage_in_excess; | |
460 | if (!mz->usage_in_excess) | |
461 | return; | |
462 | while (*p) { | |
463 | parent = *p; | |
464 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
465 | tree_node); | |
466 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
467 | p = &(*p)->rb_left; | |
468 | /* | |
469 | * We can't avoid mem cgroups that are over their soft | |
470 | * limit by the same amount | |
471 | */ | |
472 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
473 | p = &(*p)->rb_right; | |
474 | } | |
475 | rb_link_node(&mz->tree_node, parent, p); | |
476 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
477 | mz->on_tree = true; | |
478 | } | |
479 | ||
cf2c8127 JW |
480 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
481 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 AM |
482 | { |
483 | if (!mz->on_tree) | |
484 | return; | |
485 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
486 | mz->on_tree = false; | |
487 | } | |
488 | ||
cf2c8127 JW |
489 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
490 | struct mem_cgroup_tree_per_zone *mctz) | |
bb4cc1a8 | 491 | { |
0a31bc97 JW |
492 | unsigned long flags; |
493 | ||
494 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 495 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 496 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
497 | } |
498 | ||
3e32cb2e JW |
499 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
500 | { | |
501 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 502 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
503 | unsigned long excess = 0; |
504 | ||
505 | if (nr_pages > soft_limit) | |
506 | excess = nr_pages - soft_limit; | |
507 | ||
508 | return excess; | |
509 | } | |
bb4cc1a8 AM |
510 | |
511 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
512 | { | |
3e32cb2e | 513 | unsigned long excess; |
bb4cc1a8 AM |
514 | struct mem_cgroup_per_zone *mz; |
515 | struct mem_cgroup_tree_per_zone *mctz; | |
bb4cc1a8 | 516 | |
e231875b | 517 | mctz = soft_limit_tree_from_page(page); |
bb4cc1a8 AM |
518 | /* |
519 | * Necessary to update all ancestors when hierarchy is used. | |
520 | * because their event counter is not touched. | |
521 | */ | |
522 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
e231875b | 523 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
3e32cb2e | 524 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
525 | /* |
526 | * We have to update the tree if mz is on RB-tree or | |
527 | * mem is over its softlimit. | |
528 | */ | |
529 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
530 | unsigned long flags; |
531 | ||
532 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
533 | /* if on-tree, remove it */ |
534 | if (mz->on_tree) | |
cf2c8127 | 535 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
536 | /* |
537 | * Insert again. mz->usage_in_excess will be updated. | |
538 | * If excess is 0, no tree ops. | |
539 | */ | |
cf2c8127 | 540 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 541 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
542 | } |
543 | } | |
544 | } | |
545 | ||
546 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
547 | { | |
bb4cc1a8 | 548 | struct mem_cgroup_tree_per_zone *mctz; |
e231875b JZ |
549 | struct mem_cgroup_per_zone *mz; |
550 | int nid, zid; | |
bb4cc1a8 | 551 | |
e231875b JZ |
552 | for_each_node(nid) { |
553 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
554 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
555 | mctz = soft_limit_tree_node_zone(nid, zid); | |
cf2c8127 | 556 | mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
557 | } |
558 | } | |
559 | } | |
560 | ||
561 | static struct mem_cgroup_per_zone * | |
562 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
563 | { | |
564 | struct rb_node *rightmost = NULL; | |
565 | struct mem_cgroup_per_zone *mz; | |
566 | ||
567 | retry: | |
568 | mz = NULL; | |
569 | rightmost = rb_last(&mctz->rb_root); | |
570 | if (!rightmost) | |
571 | goto done; /* Nothing to reclaim from */ | |
572 | ||
573 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
574 | /* | |
575 | * Remove the node now but someone else can add it back, | |
576 | * we will to add it back at the end of reclaim to its correct | |
577 | * position in the tree. | |
578 | */ | |
cf2c8127 | 579 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 580 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 581 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
582 | goto retry; |
583 | done: | |
584 | return mz; | |
585 | } | |
586 | ||
587 | static struct mem_cgroup_per_zone * | |
588 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
589 | { | |
590 | struct mem_cgroup_per_zone *mz; | |
591 | ||
0a31bc97 | 592 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 593 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 594 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
595 | return mz; |
596 | } | |
597 | ||
711d3d2c | 598 | /* |
484ebb3b GT |
599 | * Return page count for single (non recursive) @memcg. |
600 | * | |
711d3d2c KH |
601 | * Implementation Note: reading percpu statistics for memcg. |
602 | * | |
603 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
604 | * synchronization to implement "quick" read. There are trade-off between | |
605 | * reading cost and precision of value. Then, we may have a chance to implement | |
484ebb3b | 606 | * a periodic synchronization of counter in memcg's counter. |
711d3d2c KH |
607 | * |
608 | * But this _read() function is used for user interface now. The user accounts | |
609 | * memory usage by memory cgroup and he _always_ requires exact value because | |
610 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
611 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
612 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
613 | * | |
614 | * If there are kernel internal actions which can make use of some not-exact | |
615 | * value, and reading all cpu value can be performance bottleneck in some | |
484ebb3b | 616 | * common workload, threshold and synchronization as vmstat[] should be |
711d3d2c KH |
617 | * implemented. |
618 | */ | |
484ebb3b GT |
619 | static unsigned long |
620 | mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx) | |
c62b1a3b | 621 | { |
7a159cc9 | 622 | long val = 0; |
c62b1a3b | 623 | int cpu; |
c62b1a3b | 624 | |
484ebb3b | 625 | /* Per-cpu values can be negative, use a signed accumulator */ |
733a572e | 626 | for_each_possible_cpu(cpu) |
c0ff4b85 | 627 | val += per_cpu(memcg->stat->count[idx], cpu); |
484ebb3b GT |
628 | /* |
629 | * Summing races with updates, so val may be negative. Avoid exposing | |
630 | * transient negative values. | |
631 | */ | |
632 | if (val < 0) | |
633 | val = 0; | |
c62b1a3b KH |
634 | return val; |
635 | } | |
636 | ||
c0ff4b85 | 637 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
e9f8974f JW |
638 | enum mem_cgroup_events_index idx) |
639 | { | |
640 | unsigned long val = 0; | |
641 | int cpu; | |
642 | ||
733a572e | 643 | for_each_possible_cpu(cpu) |
c0ff4b85 | 644 | val += per_cpu(memcg->stat->events[idx], cpu); |
e9f8974f JW |
645 | return val; |
646 | } | |
647 | ||
c0ff4b85 | 648 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 649 | struct page *page, |
f627c2f5 | 650 | bool compound, int nr_pages) |
d52aa412 | 651 | { |
b2402857 KH |
652 | /* |
653 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
654 | * counted as CACHE even if it's on ANON LRU. | |
655 | */ | |
0a31bc97 | 656 | if (PageAnon(page)) |
b2402857 | 657 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
c0ff4b85 | 658 | nr_pages); |
d52aa412 | 659 | else |
b2402857 | 660 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
c0ff4b85 | 661 | nr_pages); |
55e462b0 | 662 | |
f627c2f5 KS |
663 | if (compound) { |
664 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
b070e65c DR |
665 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
666 | nr_pages); | |
f627c2f5 | 667 | } |
b070e65c | 668 | |
e401f176 KH |
669 | /* pagein of a big page is an event. So, ignore page size */ |
670 | if (nr_pages > 0) | |
c0ff4b85 | 671 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 672 | else { |
c0ff4b85 | 673 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
674 | nr_pages = -nr_pages; /* for event */ |
675 | } | |
e401f176 | 676 | |
13114716 | 677 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
678 | } |
679 | ||
e231875b JZ |
680 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
681 | int nid, | |
682 | unsigned int lru_mask) | |
bb2a0de9 | 683 | { |
e231875b | 684 | unsigned long nr = 0; |
889976db YH |
685 | int zid; |
686 | ||
e231875b | 687 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 688 | |
e231875b JZ |
689 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
690 | struct mem_cgroup_per_zone *mz; | |
691 | enum lru_list lru; | |
692 | ||
693 | for_each_lru(lru) { | |
694 | if (!(BIT(lru) & lru_mask)) | |
695 | continue; | |
696 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
697 | nr += mz->lru_size[lru]; | |
698 | } | |
699 | } | |
700 | return nr; | |
889976db | 701 | } |
bb2a0de9 | 702 | |
c0ff4b85 | 703 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 704 | unsigned int lru_mask) |
6d12e2d8 | 705 | { |
e231875b | 706 | unsigned long nr = 0; |
889976db | 707 | int nid; |
6d12e2d8 | 708 | |
31aaea4a | 709 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
710 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
711 | return nr; | |
d52aa412 KH |
712 | } |
713 | ||
f53d7ce3 JW |
714 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
715 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
716 | { |
717 | unsigned long val, next; | |
718 | ||
13114716 | 719 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 720 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 721 | /* from time_after() in jiffies.h */ |
f53d7ce3 JW |
722 | if ((long)next - (long)val < 0) { |
723 | switch (target) { | |
724 | case MEM_CGROUP_TARGET_THRESH: | |
725 | next = val + THRESHOLDS_EVENTS_TARGET; | |
726 | break; | |
bb4cc1a8 AM |
727 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
728 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
729 | break; | |
f53d7ce3 JW |
730 | case MEM_CGROUP_TARGET_NUMAINFO: |
731 | next = val + NUMAINFO_EVENTS_TARGET; | |
732 | break; | |
733 | default: | |
734 | break; | |
735 | } | |
736 | __this_cpu_write(memcg->stat->targets[target], next); | |
737 | return true; | |
7a159cc9 | 738 | } |
f53d7ce3 | 739 | return false; |
d2265e6f KH |
740 | } |
741 | ||
742 | /* | |
743 | * Check events in order. | |
744 | * | |
745 | */ | |
c0ff4b85 | 746 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
747 | { |
748 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
749 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
750 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 751 | bool do_softlimit; |
82b3f2a7 | 752 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 753 | |
bb4cc1a8 AM |
754 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
755 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
756 | #if MAX_NUMNODES > 1 |
757 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
758 | MEM_CGROUP_TARGET_NUMAINFO); | |
759 | #endif | |
c0ff4b85 | 760 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
761 | if (unlikely(do_softlimit)) |
762 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 763 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 764 | if (unlikely(do_numainfo)) |
c0ff4b85 | 765 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 766 | #endif |
0a31bc97 | 767 | } |
d2265e6f KH |
768 | } |
769 | ||
cf475ad2 | 770 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 771 | { |
31a78f23 BS |
772 | /* |
773 | * mm_update_next_owner() may clear mm->owner to NULL | |
774 | * if it races with swapoff, page migration, etc. | |
775 | * So this can be called with p == NULL. | |
776 | */ | |
777 | if (unlikely(!p)) | |
778 | return NULL; | |
779 | ||
073219e9 | 780 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 781 | } |
33398cf2 | 782 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 783 | |
df381975 | 784 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 785 | { |
c0ff4b85 | 786 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 787 | |
54595fe2 KH |
788 | rcu_read_lock(); |
789 | do { | |
6f6acb00 MH |
790 | /* |
791 | * Page cache insertions can happen withou an | |
792 | * actual mm context, e.g. during disk probing | |
793 | * on boot, loopback IO, acct() writes etc. | |
794 | */ | |
795 | if (unlikely(!mm)) | |
df381975 | 796 | memcg = root_mem_cgroup; |
6f6acb00 MH |
797 | else { |
798 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
799 | if (unlikely(!memcg)) | |
800 | memcg = root_mem_cgroup; | |
801 | } | |
ec903c0c | 802 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 803 | rcu_read_unlock(); |
c0ff4b85 | 804 | return memcg; |
54595fe2 KH |
805 | } |
806 | ||
5660048c JW |
807 | /** |
808 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
809 | * @root: hierarchy root | |
810 | * @prev: previously returned memcg, NULL on first invocation | |
811 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
812 | * | |
813 | * Returns references to children of the hierarchy below @root, or | |
814 | * @root itself, or %NULL after a full round-trip. | |
815 | * | |
816 | * Caller must pass the return value in @prev on subsequent | |
817 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
818 | * to cancel a hierarchy walk before the round-trip is complete. | |
819 | * | |
820 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
821 | * divide up the memcgs in the hierarchy among all concurrent | |
822 | * reclaimers operating on the same zone and priority. | |
823 | */ | |
694fbc0f | 824 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 825 | struct mem_cgroup *prev, |
694fbc0f | 826 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 827 | { |
33398cf2 | 828 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 829 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 830 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 831 | struct mem_cgroup *pos = NULL; |
711d3d2c | 832 | |
694fbc0f AM |
833 | if (mem_cgroup_disabled()) |
834 | return NULL; | |
5660048c | 835 | |
9f3a0d09 JW |
836 | if (!root) |
837 | root = root_mem_cgroup; | |
7d74b06f | 838 | |
9f3a0d09 | 839 | if (prev && !reclaim) |
5ac8fb31 | 840 | pos = prev; |
14067bb3 | 841 | |
9f3a0d09 JW |
842 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
843 | if (prev) | |
5ac8fb31 | 844 | goto out; |
694fbc0f | 845 | return root; |
9f3a0d09 | 846 | } |
14067bb3 | 847 | |
542f85f9 | 848 | rcu_read_lock(); |
5f578161 | 849 | |
5ac8fb31 JW |
850 | if (reclaim) { |
851 | struct mem_cgroup_per_zone *mz; | |
852 | ||
853 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); | |
854 | iter = &mz->iter[reclaim->priority]; | |
855 | ||
856 | if (prev && reclaim->generation != iter->generation) | |
857 | goto out_unlock; | |
858 | ||
6df38689 | 859 | while (1) { |
4db0c3c2 | 860 | pos = READ_ONCE(iter->position); |
6df38689 VD |
861 | if (!pos || css_tryget(&pos->css)) |
862 | break; | |
5ac8fb31 | 863 | /* |
6df38689 VD |
864 | * css reference reached zero, so iter->position will |
865 | * be cleared by ->css_released. However, we should not | |
866 | * rely on this happening soon, because ->css_released | |
867 | * is called from a work queue, and by busy-waiting we | |
868 | * might block it. So we clear iter->position right | |
869 | * away. | |
5ac8fb31 | 870 | */ |
6df38689 VD |
871 | (void)cmpxchg(&iter->position, pos, NULL); |
872 | } | |
5ac8fb31 JW |
873 | } |
874 | ||
875 | if (pos) | |
876 | css = &pos->css; | |
877 | ||
878 | for (;;) { | |
879 | css = css_next_descendant_pre(css, &root->css); | |
880 | if (!css) { | |
881 | /* | |
882 | * Reclaimers share the hierarchy walk, and a | |
883 | * new one might jump in right at the end of | |
884 | * the hierarchy - make sure they see at least | |
885 | * one group and restart from the beginning. | |
886 | */ | |
887 | if (!prev) | |
888 | continue; | |
889 | break; | |
527a5ec9 | 890 | } |
7d74b06f | 891 | |
5ac8fb31 JW |
892 | /* |
893 | * Verify the css and acquire a reference. The root | |
894 | * is provided by the caller, so we know it's alive | |
895 | * and kicking, and don't take an extra reference. | |
896 | */ | |
897 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 898 | |
5ac8fb31 JW |
899 | if (css == &root->css) |
900 | break; | |
14067bb3 | 901 | |
b2052564 | 902 | if (css_tryget(css)) { |
5ac8fb31 JW |
903 | /* |
904 | * Make sure the memcg is initialized: | |
905 | * mem_cgroup_css_online() orders the the | |
906 | * initialization against setting the flag. | |
907 | */ | |
908 | if (smp_load_acquire(&memcg->initialized)) | |
909 | break; | |
542f85f9 | 910 | |
5ac8fb31 | 911 | css_put(css); |
527a5ec9 | 912 | } |
9f3a0d09 | 913 | |
5ac8fb31 | 914 | memcg = NULL; |
9f3a0d09 | 915 | } |
5ac8fb31 JW |
916 | |
917 | if (reclaim) { | |
5ac8fb31 | 918 | /* |
6df38689 VD |
919 | * The position could have already been updated by a competing |
920 | * thread, so check that the value hasn't changed since we read | |
921 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 922 | */ |
6df38689 VD |
923 | (void)cmpxchg(&iter->position, pos, memcg); |
924 | ||
5ac8fb31 JW |
925 | if (pos) |
926 | css_put(&pos->css); | |
927 | ||
928 | if (!memcg) | |
929 | iter->generation++; | |
930 | else if (!prev) | |
931 | reclaim->generation = iter->generation; | |
9f3a0d09 | 932 | } |
5ac8fb31 | 933 | |
542f85f9 MH |
934 | out_unlock: |
935 | rcu_read_unlock(); | |
5ac8fb31 | 936 | out: |
c40046f3 MH |
937 | if (prev && prev != root) |
938 | css_put(&prev->css); | |
939 | ||
9f3a0d09 | 940 | return memcg; |
14067bb3 | 941 | } |
7d74b06f | 942 | |
5660048c JW |
943 | /** |
944 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
945 | * @root: hierarchy root | |
946 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
947 | */ | |
948 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
949 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
950 | { |
951 | if (!root) | |
952 | root = root_mem_cgroup; | |
953 | if (prev && prev != root) | |
954 | css_put(&prev->css); | |
955 | } | |
7d74b06f | 956 | |
6df38689 VD |
957 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
958 | { | |
959 | struct mem_cgroup *memcg = dead_memcg; | |
960 | struct mem_cgroup_reclaim_iter *iter; | |
961 | struct mem_cgroup_per_zone *mz; | |
962 | int nid, zid; | |
963 | int i; | |
964 | ||
965 | while ((memcg = parent_mem_cgroup(memcg))) { | |
966 | for_each_node(nid) { | |
967 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
968 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; | |
969 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
970 | iter = &mz->iter[i]; | |
971 | cmpxchg(&iter->position, | |
972 | dead_memcg, NULL); | |
973 | } | |
974 | } | |
975 | } | |
976 | } | |
977 | } | |
978 | ||
9f3a0d09 JW |
979 | /* |
980 | * Iteration constructs for visiting all cgroups (under a tree). If | |
981 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
982 | * be used for reference counting. | |
983 | */ | |
984 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 985 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 986 | iter != NULL; \ |
527a5ec9 | 987 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 988 | |
9f3a0d09 | 989 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 990 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 991 | iter != NULL; \ |
527a5ec9 | 992 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 993 | |
925b7673 JW |
994 | /** |
995 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg | |
996 | * @zone: zone of the wanted lruvec | |
fa9add64 | 997 | * @memcg: memcg of the wanted lruvec |
925b7673 JW |
998 | * |
999 | * Returns the lru list vector holding pages for the given @zone and | |
1000 | * @mem. This can be the global zone lruvec, if the memory controller | |
1001 | * is disabled. | |
1002 | */ | |
1003 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, | |
1004 | struct mem_cgroup *memcg) | |
1005 | { | |
1006 | struct mem_cgroup_per_zone *mz; | |
bea8c150 | 1007 | struct lruvec *lruvec; |
925b7673 | 1008 | |
bea8c150 HD |
1009 | if (mem_cgroup_disabled()) { |
1010 | lruvec = &zone->lruvec; | |
1011 | goto out; | |
1012 | } | |
925b7673 | 1013 | |
e231875b | 1014 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
bea8c150 HD |
1015 | lruvec = &mz->lruvec; |
1016 | out: | |
1017 | /* | |
1018 | * Since a node can be onlined after the mem_cgroup was created, | |
1019 | * we have to be prepared to initialize lruvec->zone here; | |
1020 | * and if offlined then reonlined, we need to reinitialize it. | |
1021 | */ | |
1022 | if (unlikely(lruvec->zone != zone)) | |
1023 | lruvec->zone = zone; | |
1024 | return lruvec; | |
925b7673 JW |
1025 | } |
1026 | ||
925b7673 | 1027 | /** |
dfe0e773 | 1028 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1029 | * @page: the page |
fa9add64 | 1030 | * @zone: zone of the page |
dfe0e773 JW |
1031 | * |
1032 | * This function is only safe when following the LRU page isolation | |
1033 | * and putback protocol: the LRU lock must be held, and the page must | |
1034 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1035 | */ |
fa9add64 | 1036 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
08e552c6 | 1037 | { |
08e552c6 | 1038 | struct mem_cgroup_per_zone *mz; |
925b7673 | 1039 | struct mem_cgroup *memcg; |
bea8c150 | 1040 | struct lruvec *lruvec; |
6d12e2d8 | 1041 | |
bea8c150 HD |
1042 | if (mem_cgroup_disabled()) { |
1043 | lruvec = &zone->lruvec; | |
1044 | goto out; | |
1045 | } | |
925b7673 | 1046 | |
1306a85a | 1047 | memcg = page->mem_cgroup; |
7512102c | 1048 | /* |
dfe0e773 | 1049 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1050 | * possibly migrated - before they are charged. |
7512102c | 1051 | */ |
29833315 JW |
1052 | if (!memcg) |
1053 | memcg = root_mem_cgroup; | |
7512102c | 1054 | |
e231875b | 1055 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
bea8c150 HD |
1056 | lruvec = &mz->lruvec; |
1057 | out: | |
1058 | /* | |
1059 | * Since a node can be onlined after the mem_cgroup was created, | |
1060 | * we have to be prepared to initialize lruvec->zone here; | |
1061 | * and if offlined then reonlined, we need to reinitialize it. | |
1062 | */ | |
1063 | if (unlikely(lruvec->zone != zone)) | |
1064 | lruvec->zone = zone; | |
1065 | return lruvec; | |
08e552c6 | 1066 | } |
b69408e8 | 1067 | |
925b7673 | 1068 | /** |
fa9add64 HD |
1069 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1070 | * @lruvec: mem_cgroup per zone lru vector | |
1071 | * @lru: index of lru list the page is sitting on | |
1072 | * @nr_pages: positive when adding or negative when removing | |
925b7673 | 1073 | * |
fa9add64 HD |
1074 | * This function must be called when a page is added to or removed from an |
1075 | * lru list. | |
3f58a829 | 1076 | */ |
fa9add64 HD |
1077 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
1078 | int nr_pages) | |
3f58a829 MK |
1079 | { |
1080 | struct mem_cgroup_per_zone *mz; | |
fa9add64 | 1081 | unsigned long *lru_size; |
3f58a829 MK |
1082 | |
1083 | if (mem_cgroup_disabled()) | |
1084 | return; | |
1085 | ||
fa9add64 HD |
1086 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
1087 | lru_size = mz->lru_size + lru; | |
1088 | *lru_size += nr_pages; | |
1089 | VM_BUG_ON((long)(*lru_size) < 0); | |
08e552c6 | 1090 | } |
544122e5 | 1091 | |
2314b42d | 1092 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1093 | { |
2314b42d | 1094 | struct mem_cgroup *task_memcg; |
158e0a2d | 1095 | struct task_struct *p; |
ffbdccf5 | 1096 | bool ret; |
4c4a2214 | 1097 | |
158e0a2d | 1098 | p = find_lock_task_mm(task); |
de077d22 | 1099 | if (p) { |
2314b42d | 1100 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1101 | task_unlock(p); |
1102 | } else { | |
1103 | /* | |
1104 | * All threads may have already detached their mm's, but the oom | |
1105 | * killer still needs to detect if they have already been oom | |
1106 | * killed to prevent needlessly killing additional tasks. | |
1107 | */ | |
ffbdccf5 | 1108 | rcu_read_lock(); |
2314b42d JW |
1109 | task_memcg = mem_cgroup_from_task(task); |
1110 | css_get(&task_memcg->css); | |
ffbdccf5 | 1111 | rcu_read_unlock(); |
de077d22 | 1112 | } |
2314b42d JW |
1113 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1114 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1115 | return ret; |
1116 | } | |
1117 | ||
19942822 | 1118 | /** |
9d11ea9f | 1119 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1120 | * @memcg: the memory cgroup |
19942822 | 1121 | * |
9d11ea9f | 1122 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1123 | * pages. |
19942822 | 1124 | */ |
c0ff4b85 | 1125 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1126 | { |
3e32cb2e JW |
1127 | unsigned long margin = 0; |
1128 | unsigned long count; | |
1129 | unsigned long limit; | |
9d11ea9f | 1130 | |
3e32cb2e | 1131 | count = page_counter_read(&memcg->memory); |
4db0c3c2 | 1132 | limit = READ_ONCE(memcg->memory.limit); |
3e32cb2e JW |
1133 | if (count < limit) |
1134 | margin = limit - count; | |
1135 | ||
7941d214 | 1136 | if (do_memsw_account()) { |
3e32cb2e | 1137 | count = page_counter_read(&memcg->memsw); |
4db0c3c2 | 1138 | limit = READ_ONCE(memcg->memsw.limit); |
3e32cb2e JW |
1139 | if (count <= limit) |
1140 | margin = min(margin, limit - count); | |
1141 | } | |
1142 | ||
1143 | return margin; | |
19942822 JW |
1144 | } |
1145 | ||
32047e2a | 1146 | /* |
bdcbb659 | 1147 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1148 | * |
bdcbb659 QH |
1149 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1150 | * moving cgroups. This is for waiting at high-memory pressure | |
1151 | * caused by "move". | |
32047e2a | 1152 | */ |
c0ff4b85 | 1153 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1154 | { |
2bd9bb20 KH |
1155 | struct mem_cgroup *from; |
1156 | struct mem_cgroup *to; | |
4b534334 | 1157 | bool ret = false; |
2bd9bb20 KH |
1158 | /* |
1159 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1160 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1161 | */ | |
1162 | spin_lock(&mc.lock); | |
1163 | from = mc.from; | |
1164 | to = mc.to; | |
1165 | if (!from) | |
1166 | goto unlock; | |
3e92041d | 1167 | |
2314b42d JW |
1168 | ret = mem_cgroup_is_descendant(from, memcg) || |
1169 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1170 | unlock: |
1171 | spin_unlock(&mc.lock); | |
4b534334 KH |
1172 | return ret; |
1173 | } | |
1174 | ||
c0ff4b85 | 1175 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1176 | { |
1177 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1178 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1179 | DEFINE_WAIT(wait); |
1180 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1181 | /* moving charge context might have finished. */ | |
1182 | if (mc.moving_task) | |
1183 | schedule(); | |
1184 | finish_wait(&mc.waitq, &wait); | |
1185 | return true; | |
1186 | } | |
1187 | } | |
1188 | return false; | |
1189 | } | |
1190 | ||
58cf188e | 1191 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1192 | /** |
58cf188e | 1193 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1194 | * @memcg: The memory cgroup that went over limit |
1195 | * @p: Task that is going to be killed | |
1196 | * | |
1197 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1198 | * enabled | |
1199 | */ | |
1200 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1201 | { | |
e61734c5 | 1202 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
08088cb9 | 1203 | static DEFINE_MUTEX(oom_info_lock); |
58cf188e SZ |
1204 | struct mem_cgroup *iter; |
1205 | unsigned int i; | |
e222432b | 1206 | |
08088cb9 | 1207 | mutex_lock(&oom_info_lock); |
e222432b BS |
1208 | rcu_read_lock(); |
1209 | ||
2415b9f5 BV |
1210 | if (p) { |
1211 | pr_info("Task in "); | |
1212 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1213 | pr_cont(" killed as a result of limit of "); | |
1214 | } else { | |
1215 | pr_info("Memory limit reached of cgroup "); | |
1216 | } | |
1217 | ||
e61734c5 | 1218 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1219 | pr_cont("\n"); |
e222432b | 1220 | |
e222432b BS |
1221 | rcu_read_unlock(); |
1222 | ||
3e32cb2e JW |
1223 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1224 | K((u64)page_counter_read(&memcg->memory)), | |
1225 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1226 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1227 | K((u64)page_counter_read(&memcg->memsw)), | |
1228 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1229 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1230 | K((u64)page_counter_read(&memcg->kmem)), | |
1231 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1232 | |
1233 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1234 | pr_info("Memory cgroup stats for "); |
1235 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1236 | pr_cont(":"); |
1237 | ||
1238 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { | |
7941d214 | 1239 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
58cf188e | 1240 | continue; |
484ebb3b | 1241 | pr_cont(" %s:%luKB", mem_cgroup_stat_names[i], |
58cf188e SZ |
1242 | K(mem_cgroup_read_stat(iter, i))); |
1243 | } | |
1244 | ||
1245 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1246 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1247 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1248 | ||
1249 | pr_cont("\n"); | |
1250 | } | |
08088cb9 | 1251 | mutex_unlock(&oom_info_lock); |
e222432b BS |
1252 | } |
1253 | ||
81d39c20 KH |
1254 | /* |
1255 | * This function returns the number of memcg under hierarchy tree. Returns | |
1256 | * 1(self count) if no children. | |
1257 | */ | |
c0ff4b85 | 1258 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1259 | { |
1260 | int num = 0; | |
7d74b06f KH |
1261 | struct mem_cgroup *iter; |
1262 | ||
c0ff4b85 | 1263 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1264 | num++; |
81d39c20 KH |
1265 | return num; |
1266 | } | |
1267 | ||
a63d83f4 DR |
1268 | /* |
1269 | * Return the memory (and swap, if configured) limit for a memcg. | |
1270 | */ | |
3e32cb2e | 1271 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1272 | { |
3e32cb2e | 1273 | unsigned long limit; |
f3e8eb70 | 1274 | |
3e32cb2e | 1275 | limit = memcg->memory.limit; |
9a5a8f19 | 1276 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1277 | unsigned long memsw_limit; |
9a5a8f19 | 1278 | |
3e32cb2e JW |
1279 | memsw_limit = memcg->memsw.limit; |
1280 | limit = min(limit + total_swap_pages, memsw_limit); | |
9a5a8f19 | 1281 | } |
9a5a8f19 | 1282 | return limit; |
a63d83f4 DR |
1283 | } |
1284 | ||
19965460 DR |
1285 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1286 | int order) | |
9cbb78bb | 1287 | { |
6e0fc46d DR |
1288 | struct oom_control oc = { |
1289 | .zonelist = NULL, | |
1290 | .nodemask = NULL, | |
1291 | .gfp_mask = gfp_mask, | |
1292 | .order = order, | |
6e0fc46d | 1293 | }; |
9cbb78bb DR |
1294 | struct mem_cgroup *iter; |
1295 | unsigned long chosen_points = 0; | |
1296 | unsigned long totalpages; | |
1297 | unsigned int points = 0; | |
1298 | struct task_struct *chosen = NULL; | |
1299 | ||
dc56401f JW |
1300 | mutex_lock(&oom_lock); |
1301 | ||
876aafbf | 1302 | /* |
465adcf1 DR |
1303 | * If current has a pending SIGKILL or is exiting, then automatically |
1304 | * select it. The goal is to allow it to allocate so that it may | |
1305 | * quickly exit and free its memory. | |
876aafbf | 1306 | */ |
d003f371 | 1307 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
16e95196 | 1308 | mark_oom_victim(current); |
dc56401f | 1309 | goto unlock; |
876aafbf DR |
1310 | } |
1311 | ||
6e0fc46d | 1312 | check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg); |
3e32cb2e | 1313 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
9cbb78bb | 1314 | for_each_mem_cgroup_tree(iter, memcg) { |
72ec7029 | 1315 | struct css_task_iter it; |
9cbb78bb DR |
1316 | struct task_struct *task; |
1317 | ||
72ec7029 TH |
1318 | css_task_iter_start(&iter->css, &it); |
1319 | while ((task = css_task_iter_next(&it))) { | |
6e0fc46d | 1320 | switch (oom_scan_process_thread(&oc, task, totalpages)) { |
9cbb78bb DR |
1321 | case OOM_SCAN_SELECT: |
1322 | if (chosen) | |
1323 | put_task_struct(chosen); | |
1324 | chosen = task; | |
1325 | chosen_points = ULONG_MAX; | |
1326 | get_task_struct(chosen); | |
1327 | /* fall through */ | |
1328 | case OOM_SCAN_CONTINUE: | |
1329 | continue; | |
1330 | case OOM_SCAN_ABORT: | |
72ec7029 | 1331 | css_task_iter_end(&it); |
9cbb78bb DR |
1332 | mem_cgroup_iter_break(memcg, iter); |
1333 | if (chosen) | |
1334 | put_task_struct(chosen); | |
dc56401f | 1335 | goto unlock; |
9cbb78bb DR |
1336 | case OOM_SCAN_OK: |
1337 | break; | |
1338 | }; | |
1339 | points = oom_badness(task, memcg, NULL, totalpages); | |
d49ad935 DR |
1340 | if (!points || points < chosen_points) |
1341 | continue; | |
1342 | /* Prefer thread group leaders for display purposes */ | |
1343 | if (points == chosen_points && | |
1344 | thread_group_leader(chosen)) | |
1345 | continue; | |
1346 | ||
1347 | if (chosen) | |
1348 | put_task_struct(chosen); | |
1349 | chosen = task; | |
1350 | chosen_points = points; | |
1351 | get_task_struct(chosen); | |
9cbb78bb | 1352 | } |
72ec7029 | 1353 | css_task_iter_end(&it); |
9cbb78bb DR |
1354 | } |
1355 | ||
dc56401f JW |
1356 | if (chosen) { |
1357 | points = chosen_points * 1000 / totalpages; | |
6e0fc46d DR |
1358 | oom_kill_process(&oc, chosen, points, totalpages, memcg, |
1359 | "Memory cgroup out of memory"); | |
dc56401f JW |
1360 | } |
1361 | unlock: | |
1362 | mutex_unlock(&oom_lock); | |
9cbb78bb DR |
1363 | } |
1364 | ||
ae6e71d3 MC |
1365 | #if MAX_NUMNODES > 1 |
1366 | ||
4d0c066d KH |
1367 | /** |
1368 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1369 | * @memcg: the target memcg |
4d0c066d KH |
1370 | * @nid: the node ID to be checked. |
1371 | * @noswap : specify true here if the user wants flle only information. | |
1372 | * | |
1373 | * This function returns whether the specified memcg contains any | |
1374 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1375 | * pages in the node. | |
1376 | */ | |
c0ff4b85 | 1377 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1378 | int nid, bool noswap) |
1379 | { | |
c0ff4b85 | 1380 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1381 | return true; |
1382 | if (noswap || !total_swap_pages) | |
1383 | return false; | |
c0ff4b85 | 1384 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1385 | return true; |
1386 | return false; | |
1387 | ||
1388 | } | |
889976db YH |
1389 | |
1390 | /* | |
1391 | * Always updating the nodemask is not very good - even if we have an empty | |
1392 | * list or the wrong list here, we can start from some node and traverse all | |
1393 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1394 | * | |
1395 | */ | |
c0ff4b85 | 1396 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1397 | { |
1398 | int nid; | |
453a9bf3 KH |
1399 | /* |
1400 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1401 | * pagein/pageout changes since the last update. | |
1402 | */ | |
c0ff4b85 | 1403 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1404 | return; |
c0ff4b85 | 1405 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1406 | return; |
1407 | ||
889976db | 1408 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1409 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1410 | |
31aaea4a | 1411 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1412 | |
c0ff4b85 R |
1413 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1414 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1415 | } |
453a9bf3 | 1416 | |
c0ff4b85 R |
1417 | atomic_set(&memcg->numainfo_events, 0); |
1418 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1419 | } |
1420 | ||
1421 | /* | |
1422 | * Selecting a node where we start reclaim from. Because what we need is just | |
1423 | * reducing usage counter, start from anywhere is O,K. Considering | |
1424 | * memory reclaim from current node, there are pros. and cons. | |
1425 | * | |
1426 | * Freeing memory from current node means freeing memory from a node which | |
1427 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1428 | * hit limits, it will see a contention on a node. But freeing from remote | |
1429 | * node means more costs for memory reclaim because of memory latency. | |
1430 | * | |
1431 | * Now, we use round-robin. Better algorithm is welcomed. | |
1432 | */ | |
c0ff4b85 | 1433 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1434 | { |
1435 | int node; | |
1436 | ||
c0ff4b85 R |
1437 | mem_cgroup_may_update_nodemask(memcg); |
1438 | node = memcg->last_scanned_node; | |
889976db | 1439 | |
c0ff4b85 | 1440 | node = next_node(node, memcg->scan_nodes); |
889976db | 1441 | if (node == MAX_NUMNODES) |
c0ff4b85 | 1442 | node = first_node(memcg->scan_nodes); |
889976db YH |
1443 | /* |
1444 | * We call this when we hit limit, not when pages are added to LRU. | |
1445 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1446 | * memcg is too small and all pages are not on LRU. In that case, | |
1447 | * we use curret node. | |
1448 | */ | |
1449 | if (unlikely(node == MAX_NUMNODES)) | |
1450 | node = numa_node_id(); | |
1451 | ||
c0ff4b85 | 1452 | memcg->last_scanned_node = node; |
889976db YH |
1453 | return node; |
1454 | } | |
889976db | 1455 | #else |
c0ff4b85 | 1456 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1457 | { |
1458 | return 0; | |
1459 | } | |
1460 | #endif | |
1461 | ||
0608f43d AM |
1462 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
1463 | struct zone *zone, | |
1464 | gfp_t gfp_mask, | |
1465 | unsigned long *total_scanned) | |
1466 | { | |
1467 | struct mem_cgroup *victim = NULL; | |
1468 | int total = 0; | |
1469 | int loop = 0; | |
1470 | unsigned long excess; | |
1471 | unsigned long nr_scanned; | |
1472 | struct mem_cgroup_reclaim_cookie reclaim = { | |
1473 | .zone = zone, | |
1474 | .priority = 0, | |
1475 | }; | |
1476 | ||
3e32cb2e | 1477 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1478 | |
1479 | while (1) { | |
1480 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1481 | if (!victim) { | |
1482 | loop++; | |
1483 | if (loop >= 2) { | |
1484 | /* | |
1485 | * If we have not been able to reclaim | |
1486 | * anything, it might because there are | |
1487 | * no reclaimable pages under this hierarchy | |
1488 | */ | |
1489 | if (!total) | |
1490 | break; | |
1491 | /* | |
1492 | * We want to do more targeted reclaim. | |
1493 | * excess >> 2 is not to excessive so as to | |
1494 | * reclaim too much, nor too less that we keep | |
1495 | * coming back to reclaim from this cgroup | |
1496 | */ | |
1497 | if (total >= (excess >> 2) || | |
1498 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1499 | break; | |
1500 | } | |
1501 | continue; | |
1502 | } | |
0608f43d AM |
1503 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
1504 | zone, &nr_scanned); | |
1505 | *total_scanned += nr_scanned; | |
3e32cb2e | 1506 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1507 | break; |
6d61ef40 | 1508 | } |
0608f43d AM |
1509 | mem_cgroup_iter_break(root_memcg, victim); |
1510 | return total; | |
6d61ef40 BS |
1511 | } |
1512 | ||
0056f4e6 JW |
1513 | #ifdef CONFIG_LOCKDEP |
1514 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1515 | .name = "memcg_oom_lock", | |
1516 | }; | |
1517 | #endif | |
1518 | ||
fb2a6fc5 JW |
1519 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1520 | ||
867578cb KH |
1521 | /* |
1522 | * Check OOM-Killer is already running under our hierarchy. | |
1523 | * If someone is running, return false. | |
1524 | */ | |
fb2a6fc5 | 1525 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1526 | { |
79dfdacc | 1527 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1528 | |
fb2a6fc5 JW |
1529 | spin_lock(&memcg_oom_lock); |
1530 | ||
9f3a0d09 | 1531 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1532 | if (iter->oom_lock) { |
79dfdacc MH |
1533 | /* |
1534 | * this subtree of our hierarchy is already locked | |
1535 | * so we cannot give a lock. | |
1536 | */ | |
79dfdacc | 1537 | failed = iter; |
9f3a0d09 JW |
1538 | mem_cgroup_iter_break(memcg, iter); |
1539 | break; | |
23751be0 JW |
1540 | } else |
1541 | iter->oom_lock = true; | |
7d74b06f | 1542 | } |
867578cb | 1543 | |
fb2a6fc5 JW |
1544 | if (failed) { |
1545 | /* | |
1546 | * OK, we failed to lock the whole subtree so we have | |
1547 | * to clean up what we set up to the failing subtree | |
1548 | */ | |
1549 | for_each_mem_cgroup_tree(iter, memcg) { | |
1550 | if (iter == failed) { | |
1551 | mem_cgroup_iter_break(memcg, iter); | |
1552 | break; | |
1553 | } | |
1554 | iter->oom_lock = false; | |
79dfdacc | 1555 | } |
0056f4e6 JW |
1556 | } else |
1557 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1558 | |
1559 | spin_unlock(&memcg_oom_lock); | |
1560 | ||
1561 | return !failed; | |
a636b327 | 1562 | } |
0b7f569e | 1563 | |
fb2a6fc5 | 1564 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1565 | { |
7d74b06f KH |
1566 | struct mem_cgroup *iter; |
1567 | ||
fb2a6fc5 | 1568 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1569 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1570 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1571 | iter->oom_lock = false; |
fb2a6fc5 | 1572 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1573 | } |
1574 | ||
c0ff4b85 | 1575 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1576 | { |
1577 | struct mem_cgroup *iter; | |
1578 | ||
c2b42d3c | 1579 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1580 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1581 | iter->under_oom++; |
1582 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1583 | } |
1584 | ||
c0ff4b85 | 1585 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1586 | { |
1587 | struct mem_cgroup *iter; | |
1588 | ||
867578cb KH |
1589 | /* |
1590 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1591 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1592 | */ |
c2b42d3c | 1593 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1594 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1595 | if (iter->under_oom > 0) |
1596 | iter->under_oom--; | |
1597 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1598 | } |
1599 | ||
867578cb KH |
1600 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1601 | ||
dc98df5a | 1602 | struct oom_wait_info { |
d79154bb | 1603 | struct mem_cgroup *memcg; |
dc98df5a KH |
1604 | wait_queue_t wait; |
1605 | }; | |
1606 | ||
1607 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1608 | unsigned mode, int sync, void *arg) | |
1609 | { | |
d79154bb HD |
1610 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1611 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1612 | struct oom_wait_info *oom_wait_info; |
1613 | ||
1614 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1615 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1616 | |
2314b42d JW |
1617 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1618 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1619 | return 0; |
dc98df5a KH |
1620 | return autoremove_wake_function(wait, mode, sync, arg); |
1621 | } | |
1622 | ||
c0ff4b85 | 1623 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1624 | { |
c2b42d3c TH |
1625 | /* |
1626 | * For the following lockless ->under_oom test, the only required | |
1627 | * guarantee is that it must see the state asserted by an OOM when | |
1628 | * this function is called as a result of userland actions | |
1629 | * triggered by the notification of the OOM. This is trivially | |
1630 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1631 | * triggering notification. | |
1632 | */ | |
1633 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1634 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1635 | } |
1636 | ||
3812c8c8 | 1637 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1638 | { |
626ebc41 | 1639 | if (!current->memcg_may_oom) |
3812c8c8 | 1640 | return; |
867578cb | 1641 | /* |
49426420 JW |
1642 | * We are in the middle of the charge context here, so we |
1643 | * don't want to block when potentially sitting on a callstack | |
1644 | * that holds all kinds of filesystem and mm locks. | |
1645 | * | |
1646 | * Also, the caller may handle a failed allocation gracefully | |
1647 | * (like optional page cache readahead) and so an OOM killer | |
1648 | * invocation might not even be necessary. | |
1649 | * | |
1650 | * That's why we don't do anything here except remember the | |
1651 | * OOM context and then deal with it at the end of the page | |
1652 | * fault when the stack is unwound, the locks are released, | |
1653 | * and when we know whether the fault was overall successful. | |
867578cb | 1654 | */ |
49426420 | 1655 | css_get(&memcg->css); |
626ebc41 TH |
1656 | current->memcg_in_oom = memcg; |
1657 | current->memcg_oom_gfp_mask = mask; | |
1658 | current->memcg_oom_order = order; | |
3812c8c8 JW |
1659 | } |
1660 | ||
1661 | /** | |
1662 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1663 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1664 | * |
49426420 JW |
1665 | * This has to be called at the end of a page fault if the memcg OOM |
1666 | * handler was enabled. | |
3812c8c8 | 1667 | * |
49426420 | 1668 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1669 | * sleep on a waitqueue until the userspace task resolves the |
1670 | * situation. Sleeping directly in the charge context with all kinds | |
1671 | * of locks held is not a good idea, instead we remember an OOM state | |
1672 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1673 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1674 | * |
1675 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1676 | * completed, %false otherwise. |
3812c8c8 | 1677 | */ |
49426420 | 1678 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1679 | { |
626ebc41 | 1680 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1681 | struct oom_wait_info owait; |
49426420 | 1682 | bool locked; |
3812c8c8 JW |
1683 | |
1684 | /* OOM is global, do not handle */ | |
3812c8c8 | 1685 | if (!memcg) |
49426420 | 1686 | return false; |
3812c8c8 | 1687 | |
c32b3cbe | 1688 | if (!handle || oom_killer_disabled) |
49426420 | 1689 | goto cleanup; |
3812c8c8 JW |
1690 | |
1691 | owait.memcg = memcg; | |
1692 | owait.wait.flags = 0; | |
1693 | owait.wait.func = memcg_oom_wake_function; | |
1694 | owait.wait.private = current; | |
1695 | INIT_LIST_HEAD(&owait.wait.task_list); | |
867578cb | 1696 | |
3812c8c8 | 1697 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1698 | mem_cgroup_mark_under_oom(memcg); |
1699 | ||
1700 | locked = mem_cgroup_oom_trylock(memcg); | |
1701 | ||
1702 | if (locked) | |
1703 | mem_cgroup_oom_notify(memcg); | |
1704 | ||
1705 | if (locked && !memcg->oom_kill_disable) { | |
1706 | mem_cgroup_unmark_under_oom(memcg); | |
1707 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1708 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1709 | current->memcg_oom_order); | |
49426420 | 1710 | } else { |
3812c8c8 | 1711 | schedule(); |
49426420 JW |
1712 | mem_cgroup_unmark_under_oom(memcg); |
1713 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1714 | } | |
1715 | ||
1716 | if (locked) { | |
fb2a6fc5 JW |
1717 | mem_cgroup_oom_unlock(memcg); |
1718 | /* | |
1719 | * There is no guarantee that an OOM-lock contender | |
1720 | * sees the wakeups triggered by the OOM kill | |
1721 | * uncharges. Wake any sleepers explicitely. | |
1722 | */ | |
1723 | memcg_oom_recover(memcg); | |
1724 | } | |
49426420 | 1725 | cleanup: |
626ebc41 | 1726 | current->memcg_in_oom = NULL; |
3812c8c8 | 1727 | css_put(&memcg->css); |
867578cb | 1728 | return true; |
0b7f569e KH |
1729 | } |
1730 | ||
d7365e78 JW |
1731 | /** |
1732 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction | |
1733 | * @page: page that is going to change accounted state | |
32047e2a | 1734 | * |
d7365e78 JW |
1735 | * This function must mark the beginning of an accounted page state |
1736 | * change to prevent double accounting when the page is concurrently | |
1737 | * being moved to another memcg: | |
32047e2a | 1738 | * |
6de22619 | 1739 | * memcg = mem_cgroup_begin_page_stat(page); |
d7365e78 JW |
1740 | * if (TestClearPageState(page)) |
1741 | * mem_cgroup_update_page_stat(memcg, state, -1); | |
6de22619 | 1742 | * mem_cgroup_end_page_stat(memcg); |
d69b042f | 1743 | */ |
6de22619 | 1744 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page) |
89c06bd5 KH |
1745 | { |
1746 | struct mem_cgroup *memcg; | |
6de22619 | 1747 | unsigned long flags; |
89c06bd5 | 1748 | |
6de22619 JW |
1749 | /* |
1750 | * The RCU lock is held throughout the transaction. The fast | |
1751 | * path can get away without acquiring the memcg->move_lock | |
1752 | * because page moving starts with an RCU grace period. | |
1753 | * | |
1754 | * The RCU lock also protects the memcg from being freed when | |
1755 | * the page state that is going to change is the only thing | |
1756 | * preventing the page from being uncharged. | |
1757 | * E.g. end-writeback clearing PageWriteback(), which allows | |
1758 | * migration to go ahead and uncharge the page before the | |
1759 | * account transaction might be complete. | |
1760 | */ | |
d7365e78 JW |
1761 | rcu_read_lock(); |
1762 | ||
1763 | if (mem_cgroup_disabled()) | |
1764 | return NULL; | |
89c06bd5 | 1765 | again: |
1306a85a | 1766 | memcg = page->mem_cgroup; |
29833315 | 1767 | if (unlikely(!memcg)) |
d7365e78 JW |
1768 | return NULL; |
1769 | ||
bdcbb659 | 1770 | if (atomic_read(&memcg->moving_account) <= 0) |
d7365e78 | 1771 | return memcg; |
89c06bd5 | 1772 | |
6de22619 | 1773 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1774 | if (memcg != page->mem_cgroup) { |
6de22619 | 1775 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1776 | goto again; |
1777 | } | |
6de22619 JW |
1778 | |
1779 | /* | |
1780 | * When charge migration first begins, we can have locked and | |
1781 | * unlocked page stat updates happening concurrently. Track | |
1782 | * the task who has the lock for mem_cgroup_end_page_stat(). | |
1783 | */ | |
1784 | memcg->move_lock_task = current; | |
1785 | memcg->move_lock_flags = flags; | |
d7365e78 JW |
1786 | |
1787 | return memcg; | |
89c06bd5 | 1788 | } |
c4843a75 | 1789 | EXPORT_SYMBOL(mem_cgroup_begin_page_stat); |
89c06bd5 | 1790 | |
d7365e78 JW |
1791 | /** |
1792 | * mem_cgroup_end_page_stat - finish a page state statistics transaction | |
1793 | * @memcg: the memcg that was accounted against | |
d7365e78 | 1794 | */ |
6de22619 | 1795 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg) |
89c06bd5 | 1796 | { |
6de22619 JW |
1797 | if (memcg && memcg->move_lock_task == current) { |
1798 | unsigned long flags = memcg->move_lock_flags; | |
1799 | ||
1800 | memcg->move_lock_task = NULL; | |
1801 | memcg->move_lock_flags = 0; | |
1802 | ||
1803 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1804 | } | |
89c06bd5 | 1805 | |
d7365e78 | 1806 | rcu_read_unlock(); |
89c06bd5 | 1807 | } |
c4843a75 | 1808 | EXPORT_SYMBOL(mem_cgroup_end_page_stat); |
89c06bd5 | 1809 | |
cdec2e42 KH |
1810 | /* |
1811 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1812 | * TODO: maybe necessary to use big numbers in big irons. | |
1813 | */ | |
7ec99d62 | 1814 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1815 | struct memcg_stock_pcp { |
1816 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1817 | unsigned int nr_pages; |
cdec2e42 | 1818 | struct work_struct work; |
26fe6168 | 1819 | unsigned long flags; |
a0db00fc | 1820 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1821 | }; |
1822 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1823 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1824 | |
a0956d54 SS |
1825 | /** |
1826 | * consume_stock: Try to consume stocked charge on this cpu. | |
1827 | * @memcg: memcg to consume from. | |
1828 | * @nr_pages: how many pages to charge. | |
1829 | * | |
1830 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1831 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1832 | * service an allocation will refill the stock. | |
1833 | * | |
1834 | * returns true if successful, false otherwise. | |
cdec2e42 | 1835 | */ |
a0956d54 | 1836 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1837 | { |
1838 | struct memcg_stock_pcp *stock; | |
3e32cb2e | 1839 | bool ret = false; |
cdec2e42 | 1840 | |
a0956d54 | 1841 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 1842 | return ret; |
a0956d54 | 1843 | |
cdec2e42 | 1844 | stock = &get_cpu_var(memcg_stock); |
3e32cb2e | 1845 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1846 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1847 | ret = true; |
1848 | } | |
cdec2e42 KH |
1849 | put_cpu_var(memcg_stock); |
1850 | return ret; | |
1851 | } | |
1852 | ||
1853 | /* | |
3e32cb2e | 1854 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1855 | */ |
1856 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1857 | { | |
1858 | struct mem_cgroup *old = stock->cached; | |
1859 | ||
11c9ea4e | 1860 | if (stock->nr_pages) { |
3e32cb2e | 1861 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 1862 | if (do_memsw_account()) |
3e32cb2e | 1863 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1864 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1865 | stock->nr_pages = 0; |
cdec2e42 KH |
1866 | } |
1867 | stock->cached = NULL; | |
cdec2e42 KH |
1868 | } |
1869 | ||
1870 | /* | |
1871 | * This must be called under preempt disabled or must be called by | |
1872 | * a thread which is pinned to local cpu. | |
1873 | */ | |
1874 | static void drain_local_stock(struct work_struct *dummy) | |
1875 | { | |
7c8e0181 | 1876 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
cdec2e42 | 1877 | drain_stock(stock); |
26fe6168 | 1878 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
1879 | } |
1880 | ||
1881 | /* | |
3e32cb2e | 1882 | * Cache charges(val) to local per_cpu area. |
320cc51d | 1883 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1884 | */ |
c0ff4b85 | 1885 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1886 | { |
1887 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1888 | ||
c0ff4b85 | 1889 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1890 | drain_stock(stock); |
c0ff4b85 | 1891 | stock->cached = memcg; |
cdec2e42 | 1892 | } |
11c9ea4e | 1893 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
1894 | put_cpu_var(memcg_stock); |
1895 | } | |
1896 | ||
1897 | /* | |
c0ff4b85 | 1898 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 1899 | * of the hierarchy under it. |
cdec2e42 | 1900 | */ |
6d3d6aa2 | 1901 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 1902 | { |
26fe6168 | 1903 | int cpu, curcpu; |
d38144b7 | 1904 | |
6d3d6aa2 JW |
1905 | /* If someone's already draining, avoid adding running more workers. */ |
1906 | if (!mutex_trylock(&percpu_charge_mutex)) | |
1907 | return; | |
cdec2e42 | 1908 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 1909 | get_online_cpus(); |
5af12d0e | 1910 | curcpu = get_cpu(); |
cdec2e42 KH |
1911 | for_each_online_cpu(cpu) { |
1912 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1913 | struct mem_cgroup *memcg; |
26fe6168 | 1914 | |
c0ff4b85 R |
1915 | memcg = stock->cached; |
1916 | if (!memcg || !stock->nr_pages) | |
26fe6168 | 1917 | continue; |
2314b42d | 1918 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
3e92041d | 1919 | continue; |
d1a05b69 MH |
1920 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
1921 | if (cpu == curcpu) | |
1922 | drain_local_stock(&stock->work); | |
1923 | else | |
1924 | schedule_work_on(cpu, &stock->work); | |
1925 | } | |
cdec2e42 | 1926 | } |
5af12d0e | 1927 | put_cpu(); |
f894ffa8 | 1928 | put_online_cpus(); |
9f50fad6 | 1929 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
1930 | } |
1931 | ||
0db0628d | 1932 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
cdec2e42 KH |
1933 | unsigned long action, |
1934 | void *hcpu) | |
1935 | { | |
1936 | int cpu = (unsigned long)hcpu; | |
1937 | struct memcg_stock_pcp *stock; | |
1938 | ||
619d094b | 1939 | if (action == CPU_ONLINE) |
1489ebad | 1940 | return NOTIFY_OK; |
1489ebad | 1941 | |
d833049b | 1942 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
cdec2e42 | 1943 | return NOTIFY_OK; |
711d3d2c | 1944 | |
cdec2e42 KH |
1945 | stock = &per_cpu(memcg_stock, cpu); |
1946 | drain_stock(stock); | |
1947 | return NOTIFY_OK; | |
1948 | } | |
1949 | ||
f7e1cb6e JW |
1950 | static void reclaim_high(struct mem_cgroup *memcg, |
1951 | unsigned int nr_pages, | |
1952 | gfp_t gfp_mask) | |
1953 | { | |
1954 | do { | |
1955 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
1956 | continue; | |
1957 | mem_cgroup_events(memcg, MEMCG_HIGH, 1); | |
1958 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); | |
1959 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1960 | } | |
1961 | ||
1962 | static void high_work_func(struct work_struct *work) | |
1963 | { | |
1964 | struct mem_cgroup *memcg; | |
1965 | ||
1966 | memcg = container_of(work, struct mem_cgroup, high_work); | |
1967 | reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL); | |
1968 | } | |
1969 | ||
b23afb93 TH |
1970 | /* |
1971 | * Scheduled by try_charge() to be executed from the userland return path | |
1972 | * and reclaims memory over the high limit. | |
1973 | */ | |
1974 | void mem_cgroup_handle_over_high(void) | |
1975 | { | |
1976 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 1977 | struct mem_cgroup *memcg; |
b23afb93 TH |
1978 | |
1979 | if (likely(!nr_pages)) | |
1980 | return; | |
1981 | ||
f7e1cb6e JW |
1982 | memcg = get_mem_cgroup_from_mm(current->mm); |
1983 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
1984 | css_put(&memcg->css); |
1985 | current->memcg_nr_pages_over_high = 0; | |
1986 | } | |
1987 | ||
00501b53 JW |
1988 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1989 | unsigned int nr_pages) | |
8a9f3ccd | 1990 | { |
7ec99d62 | 1991 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 1992 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 1993 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 1994 | struct page_counter *counter; |
6539cc05 | 1995 | unsigned long nr_reclaimed; |
b70a2a21 JW |
1996 | bool may_swap = true; |
1997 | bool drained = false; | |
a636b327 | 1998 | |
ce00a967 | 1999 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2000 | return 0; |
6539cc05 | 2001 | retry: |
b6b6cc72 | 2002 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2003 | return 0; |
8a9f3ccd | 2004 | |
7941d214 | 2005 | if (!do_memsw_account() || |
6071ca52 JW |
2006 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2007 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2008 | goto done_restock; |
7941d214 | 2009 | if (do_memsw_account()) |
3e32cb2e JW |
2010 | page_counter_uncharge(&memcg->memsw, batch); |
2011 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2012 | } else { |
3e32cb2e | 2013 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2014 | may_swap = false; |
3fbe7244 | 2015 | } |
7a81b88c | 2016 | |
6539cc05 JW |
2017 | if (batch > nr_pages) { |
2018 | batch = nr_pages; | |
2019 | goto retry; | |
2020 | } | |
6d61ef40 | 2021 | |
06b078fc JW |
2022 | /* |
2023 | * Unlike in global OOM situations, memcg is not in a physical | |
2024 | * memory shortage. Allow dying and OOM-killed tasks to | |
2025 | * bypass the last charges so that they can exit quickly and | |
2026 | * free their memory. | |
2027 | */ | |
2028 | if (unlikely(test_thread_flag(TIF_MEMDIE) || | |
2029 | fatal_signal_pending(current) || | |
2030 | current->flags & PF_EXITING)) | |
10d53c74 | 2031 | goto force; |
06b078fc JW |
2032 | |
2033 | if (unlikely(task_in_memcg_oom(current))) | |
2034 | goto nomem; | |
2035 | ||
d0164adc | 2036 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2037 | goto nomem; |
4b534334 | 2038 | |
241994ed JW |
2039 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
2040 | ||
b70a2a21 JW |
2041 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2042 | gfp_mask, may_swap); | |
6539cc05 | 2043 | |
61e02c74 | 2044 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2045 | goto retry; |
28c34c29 | 2046 | |
b70a2a21 | 2047 | if (!drained) { |
6d3d6aa2 | 2048 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2049 | drained = true; |
2050 | goto retry; | |
2051 | } | |
2052 | ||
28c34c29 JW |
2053 | if (gfp_mask & __GFP_NORETRY) |
2054 | goto nomem; | |
6539cc05 JW |
2055 | /* |
2056 | * Even though the limit is exceeded at this point, reclaim | |
2057 | * may have been able to free some pages. Retry the charge | |
2058 | * before killing the task. | |
2059 | * | |
2060 | * Only for regular pages, though: huge pages are rather | |
2061 | * unlikely to succeed so close to the limit, and we fall back | |
2062 | * to regular pages anyway in case of failure. | |
2063 | */ | |
61e02c74 | 2064 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2065 | goto retry; |
2066 | /* | |
2067 | * At task move, charge accounts can be doubly counted. So, it's | |
2068 | * better to wait until the end of task_move if something is going on. | |
2069 | */ | |
2070 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2071 | goto retry; | |
2072 | ||
9b130619 JW |
2073 | if (nr_retries--) |
2074 | goto retry; | |
2075 | ||
06b078fc | 2076 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2077 | goto force; |
06b078fc | 2078 | |
6539cc05 | 2079 | if (fatal_signal_pending(current)) |
10d53c74 | 2080 | goto force; |
6539cc05 | 2081 | |
241994ed JW |
2082 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
2083 | ||
3608de07 JM |
2084 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
2085 | get_order(nr_pages * PAGE_SIZE)); | |
7a81b88c | 2086 | nomem: |
6d1fdc48 | 2087 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2088 | return -ENOMEM; |
10d53c74 TH |
2089 | force: |
2090 | /* | |
2091 | * The allocation either can't fail or will lead to more memory | |
2092 | * being freed very soon. Allow memory usage go over the limit | |
2093 | * temporarily by force charging it. | |
2094 | */ | |
2095 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2096 | if (do_memsw_account()) |
10d53c74 TH |
2097 | page_counter_charge(&memcg->memsw, nr_pages); |
2098 | css_get_many(&memcg->css, nr_pages); | |
2099 | ||
2100 | return 0; | |
6539cc05 JW |
2101 | |
2102 | done_restock: | |
e8ea14cc | 2103 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2104 | if (batch > nr_pages) |
2105 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2106 | |
241994ed | 2107 | /* |
b23afb93 TH |
2108 | * If the hierarchy is above the normal consumption range, schedule |
2109 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2110 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2111 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2112 | * not recorded as it most likely matches current's and won't | |
2113 | * change in the meantime. As high limit is checked again before | |
2114 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2115 | */ |
2116 | do { | |
b23afb93 | 2117 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2118 | /* Don't bother a random interrupted task */ |
2119 | if (in_interrupt()) { | |
2120 | schedule_work(&memcg->high_work); | |
2121 | break; | |
2122 | } | |
9516a18a | 2123 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2124 | set_notify_resume(current); |
2125 | break; | |
2126 | } | |
241994ed | 2127 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2128 | |
2129 | return 0; | |
7a81b88c | 2130 | } |
8a9f3ccd | 2131 | |
00501b53 | 2132 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2133 | { |
ce00a967 JW |
2134 | if (mem_cgroup_is_root(memcg)) |
2135 | return; | |
2136 | ||
3e32cb2e | 2137 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2138 | if (do_memsw_account()) |
3e32cb2e | 2139 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2140 | |
e8ea14cc | 2141 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2142 | } |
2143 | ||
0a31bc97 JW |
2144 | static void lock_page_lru(struct page *page, int *isolated) |
2145 | { | |
2146 | struct zone *zone = page_zone(page); | |
2147 | ||
2148 | spin_lock_irq(&zone->lru_lock); | |
2149 | if (PageLRU(page)) { | |
2150 | struct lruvec *lruvec; | |
2151 | ||
2152 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2153 | ClearPageLRU(page); | |
2154 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2155 | *isolated = 1; | |
2156 | } else | |
2157 | *isolated = 0; | |
2158 | } | |
2159 | ||
2160 | static void unlock_page_lru(struct page *page, int isolated) | |
2161 | { | |
2162 | struct zone *zone = page_zone(page); | |
2163 | ||
2164 | if (isolated) { | |
2165 | struct lruvec *lruvec; | |
2166 | ||
2167 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
2168 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
2169 | SetPageLRU(page); | |
2170 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2171 | } | |
2172 | spin_unlock_irq(&zone->lru_lock); | |
2173 | } | |
2174 | ||
00501b53 | 2175 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2176 | bool lrucare) |
7a81b88c | 2177 | { |
0a31bc97 | 2178 | int isolated; |
9ce70c02 | 2179 | |
1306a85a | 2180 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2181 | |
2182 | /* | |
2183 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2184 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2185 | */ | |
0a31bc97 JW |
2186 | if (lrucare) |
2187 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2188 | |
0a31bc97 JW |
2189 | /* |
2190 | * Nobody should be changing or seriously looking at | |
1306a85a | 2191 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2192 | * |
2193 | * - the page is uncharged | |
2194 | * | |
2195 | * - the page is off-LRU | |
2196 | * | |
2197 | * - an anonymous fault has exclusive page access, except for | |
2198 | * a locked page table | |
2199 | * | |
2200 | * - a page cache insertion, a swapin fault, or a migration | |
2201 | * have the page locked | |
2202 | */ | |
1306a85a | 2203 | page->mem_cgroup = memcg; |
9ce70c02 | 2204 | |
0a31bc97 JW |
2205 | if (lrucare) |
2206 | unlock_page_lru(page, isolated); | |
7a81b88c | 2207 | } |
66e1707b | 2208 | |
127424c8 | 2209 | #ifndef CONFIG_SLOB |
f3bb3043 | 2210 | static int memcg_alloc_cache_id(void) |
55007d84 | 2211 | { |
f3bb3043 VD |
2212 | int id, size; |
2213 | int err; | |
2214 | ||
dbcf73e2 | 2215 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2216 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2217 | if (id < 0) | |
2218 | return id; | |
55007d84 | 2219 | |
dbcf73e2 | 2220 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2221 | return id; |
2222 | ||
2223 | /* | |
2224 | * There's no space for the new id in memcg_caches arrays, | |
2225 | * so we have to grow them. | |
2226 | */ | |
05257a1a | 2227 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2228 | |
2229 | size = 2 * (id + 1); | |
55007d84 GC |
2230 | if (size < MEMCG_CACHES_MIN_SIZE) |
2231 | size = MEMCG_CACHES_MIN_SIZE; | |
2232 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2233 | size = MEMCG_CACHES_MAX_SIZE; | |
2234 | ||
f3bb3043 | 2235 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2236 | if (!err) |
2237 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2238 | if (!err) |
2239 | memcg_nr_cache_ids = size; | |
2240 | ||
2241 | up_write(&memcg_cache_ids_sem); | |
2242 | ||
f3bb3043 | 2243 | if (err) { |
dbcf73e2 | 2244 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2245 | return err; |
2246 | } | |
2247 | return id; | |
2248 | } | |
2249 | ||
2250 | static void memcg_free_cache_id(int id) | |
2251 | { | |
dbcf73e2 | 2252 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2253 | } |
2254 | ||
d5b3cf71 | 2255 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2256 | struct mem_cgroup *memcg; |
2257 | struct kmem_cache *cachep; | |
2258 | struct work_struct work; | |
2259 | }; | |
2260 | ||
d5b3cf71 | 2261 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2262 | { |
d5b3cf71 VD |
2263 | struct memcg_kmem_cache_create_work *cw = |
2264 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2265 | struct mem_cgroup *memcg = cw->memcg; |
2266 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2267 | |
d5b3cf71 | 2268 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2269 | |
5722d094 | 2270 | css_put(&memcg->css); |
d7f25f8a GC |
2271 | kfree(cw); |
2272 | } | |
2273 | ||
2274 | /* | |
2275 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2276 | */ |
d5b3cf71 VD |
2277 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2278 | struct kmem_cache *cachep) | |
d7f25f8a | 2279 | { |
d5b3cf71 | 2280 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2281 | |
776ed0f0 | 2282 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2283 | if (!cw) |
d7f25f8a | 2284 | return; |
8135be5a VD |
2285 | |
2286 | css_get(&memcg->css); | |
d7f25f8a GC |
2287 | |
2288 | cw->memcg = memcg; | |
2289 | cw->cachep = cachep; | |
d5b3cf71 | 2290 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2291 | |
d7f25f8a GC |
2292 | schedule_work(&cw->work); |
2293 | } | |
2294 | ||
d5b3cf71 VD |
2295 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2296 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2297 | { |
2298 | /* | |
2299 | * We need to stop accounting when we kmalloc, because if the | |
2300 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2301 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2302 | * |
2303 | * However, it is better to enclose the whole function. Depending on | |
2304 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2305 | * trigger an allocation. This too, will make us recurse. Because at | |
2306 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2307 | * the safest choice is to do it like this, wrapping the whole function. | |
2308 | */ | |
6f185c29 | 2309 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2310 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2311 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2312 | } |
c67a8a68 | 2313 | |
d7f25f8a GC |
2314 | /* |
2315 | * Return the kmem_cache we're supposed to use for a slab allocation. | |
2316 | * We try to use the current memcg's version of the cache. | |
2317 | * | |
2318 | * If the cache does not exist yet, if we are the first user of it, | |
2319 | * we either create it immediately, if possible, or create it asynchronously | |
2320 | * in a workqueue. | |
2321 | * In the latter case, we will let the current allocation go through with | |
2322 | * the original cache. | |
2323 | * | |
2324 | * Can't be called in interrupt context or from kernel threads. | |
2325 | * This function needs to be called with rcu_read_lock() held. | |
2326 | */ | |
230e9fc2 | 2327 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, gfp_t gfp) |
d7f25f8a GC |
2328 | { |
2329 | struct mem_cgroup *memcg; | |
959c8963 | 2330 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2331 | int kmemcg_id; |
d7f25f8a | 2332 | |
f7ce3190 | 2333 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2334 | |
230e9fc2 VD |
2335 | if (cachep->flags & SLAB_ACCOUNT) |
2336 | gfp |= __GFP_ACCOUNT; | |
2337 | ||
2338 | if (!(gfp & __GFP_ACCOUNT)) | |
2339 | return cachep; | |
2340 | ||
9d100c5e | 2341 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2342 | return cachep; |
2343 | ||
8135be5a | 2344 | memcg = get_mem_cgroup_from_mm(current->mm); |
4db0c3c2 | 2345 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2346 | if (kmemcg_id < 0) |
ca0dde97 | 2347 | goto out; |
d7f25f8a | 2348 | |
2a4db7eb | 2349 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2350 | if (likely(memcg_cachep)) |
2351 | return memcg_cachep; | |
ca0dde97 LZ |
2352 | |
2353 | /* | |
2354 | * If we are in a safe context (can wait, and not in interrupt | |
2355 | * context), we could be be predictable and return right away. | |
2356 | * This would guarantee that the allocation being performed | |
2357 | * already belongs in the new cache. | |
2358 | * | |
2359 | * However, there are some clashes that can arrive from locking. | |
2360 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2361 | * memcg_create_kmem_cache, this means no further allocation |
2362 | * could happen with the slab_mutex held. So it's better to | |
2363 | * defer everything. | |
ca0dde97 | 2364 | */ |
d5b3cf71 | 2365 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2366 | out: |
8135be5a | 2367 | css_put(&memcg->css); |
ca0dde97 | 2368 | return cachep; |
d7f25f8a | 2369 | } |
d7f25f8a | 2370 | |
8135be5a VD |
2371 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
2372 | { | |
2373 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2374 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2375 | } |
2376 | ||
f3ccb2c4 VD |
2377 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
2378 | struct mem_cgroup *memcg) | |
7ae1e1d0 | 2379 | { |
f3ccb2c4 VD |
2380 | unsigned int nr_pages = 1 << order; |
2381 | struct page_counter *counter; | |
7ae1e1d0 GC |
2382 | int ret; |
2383 | ||
567e9ab2 | 2384 | if (!memcg_kmem_online(memcg)) |
d05e83a6 | 2385 | return 0; |
6d42c232 | 2386 | |
f3ccb2c4 | 2387 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2388 | if (ret) |
f3ccb2c4 | 2389 | return ret; |
52c29b04 JW |
2390 | |
2391 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2392 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2393 | cancel_charge(memcg, nr_pages); | |
2394 | return -ENOMEM; | |
7ae1e1d0 GC |
2395 | } |
2396 | ||
f3ccb2c4 | 2397 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2398 | |
f3ccb2c4 | 2399 | return 0; |
7ae1e1d0 GC |
2400 | } |
2401 | ||
f3ccb2c4 | 2402 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2403 | { |
f3ccb2c4 VD |
2404 | struct mem_cgroup *memcg; |
2405 | int ret; | |
7ae1e1d0 | 2406 | |
f3ccb2c4 VD |
2407 | memcg = get_mem_cgroup_from_mm(current->mm); |
2408 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); | |
7ae1e1d0 | 2409 | css_put(&memcg->css); |
d05e83a6 | 2410 | return ret; |
7ae1e1d0 GC |
2411 | } |
2412 | ||
d05e83a6 | 2413 | void __memcg_kmem_uncharge(struct page *page, int order) |
7ae1e1d0 | 2414 | { |
1306a85a | 2415 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2416 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2417 | |
7ae1e1d0 GC |
2418 | if (!memcg) |
2419 | return; | |
2420 | ||
309381fe | 2421 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2422 | |
52c29b04 JW |
2423 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2424 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2425 | ||
f3ccb2c4 | 2426 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2427 | if (do_memsw_account()) |
f3ccb2c4 | 2428 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2429 | |
1306a85a | 2430 | page->mem_cgroup = NULL; |
f3ccb2c4 | 2431 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2432 | } |
127424c8 | 2433 | #endif /* !CONFIG_SLOB */ |
7ae1e1d0 | 2434 | |
ca3e0214 KH |
2435 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2436 | ||
ca3e0214 KH |
2437 | /* |
2438 | * Because tail pages are not marked as "used", set it. We're under | |
3ac808fd | 2439 | * zone->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2440 | */ |
e94c8a9c | 2441 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2442 | { |
e94c8a9c | 2443 | int i; |
ca3e0214 | 2444 | |
3d37c4a9 KH |
2445 | if (mem_cgroup_disabled()) |
2446 | return; | |
b070e65c | 2447 | |
29833315 | 2448 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2449 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2450 | |
1306a85a | 2451 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
b070e65c | 2452 | HPAGE_PMD_NR); |
ca3e0214 | 2453 | } |
12d27107 | 2454 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2455 | |
c255a458 | 2456 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 JW |
2457 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
2458 | bool charge) | |
d13d1443 | 2459 | { |
0a31bc97 JW |
2460 | int val = (charge) ? 1 : -1; |
2461 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); | |
d13d1443 | 2462 | } |
02491447 DN |
2463 | |
2464 | /** | |
2465 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2466 | * @entry: swap entry to be moved | |
2467 | * @from: mem_cgroup which the entry is moved from | |
2468 | * @to: mem_cgroup which the entry is moved to | |
2469 | * | |
2470 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2471 | * as the mem_cgroup's id of @from. | |
2472 | * | |
2473 | * Returns 0 on success, -EINVAL on failure. | |
2474 | * | |
3e32cb2e | 2475 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2476 | * both res and memsw, and called css_get(). |
2477 | */ | |
2478 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2479 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2480 | { |
2481 | unsigned short old_id, new_id; | |
2482 | ||
34c00c31 LZ |
2483 | old_id = mem_cgroup_id(from); |
2484 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2485 | |
2486 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2487 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2488 | mem_cgroup_swap_statistics(to, true); |
02491447 DN |
2489 | return 0; |
2490 | } | |
2491 | return -EINVAL; | |
2492 | } | |
2493 | #else | |
2494 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2495 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2496 | { |
2497 | return -EINVAL; | |
2498 | } | |
8c7c6e34 | 2499 | #endif |
d13d1443 | 2500 | |
3e32cb2e | 2501 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2502 | |
d38d2a75 | 2503 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2504 | unsigned long limit) |
628f4235 | 2505 | { |
3e32cb2e JW |
2506 | unsigned long curusage; |
2507 | unsigned long oldusage; | |
2508 | bool enlarge = false; | |
81d39c20 | 2509 | int retry_count; |
3e32cb2e | 2510 | int ret; |
81d39c20 KH |
2511 | |
2512 | /* | |
2513 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2514 | * is depends on callers. We set our retry-count to be function | |
2515 | * of # of children which we should visit in this loop. | |
2516 | */ | |
3e32cb2e JW |
2517 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2518 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2519 | |
3e32cb2e | 2520 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2521 | |
3e32cb2e | 2522 | do { |
628f4235 KH |
2523 | if (signal_pending(current)) { |
2524 | ret = -EINTR; | |
2525 | break; | |
2526 | } | |
3e32cb2e JW |
2527 | |
2528 | mutex_lock(&memcg_limit_mutex); | |
2529 | if (limit > memcg->memsw.limit) { | |
2530 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2531 | ret = -EINVAL; |
628f4235 KH |
2532 | break; |
2533 | } | |
3e32cb2e JW |
2534 | if (limit > memcg->memory.limit) |
2535 | enlarge = true; | |
2536 | ret = page_counter_limit(&memcg->memory, limit); | |
2537 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2538 | |
2539 | if (!ret) | |
2540 | break; | |
2541 | ||
b70a2a21 JW |
2542 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2543 | ||
3e32cb2e | 2544 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2545 | /* Usage is reduced ? */ |
f894ffa8 | 2546 | if (curusage >= oldusage) |
81d39c20 KH |
2547 | retry_count--; |
2548 | else | |
2549 | oldusage = curusage; | |
3e32cb2e JW |
2550 | } while (retry_count); |
2551 | ||
3c11ecf4 KH |
2552 | if (!ret && enlarge) |
2553 | memcg_oom_recover(memcg); | |
14797e23 | 2554 | |
8c7c6e34 KH |
2555 | return ret; |
2556 | } | |
2557 | ||
338c8431 | 2558 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2559 | unsigned long limit) |
8c7c6e34 | 2560 | { |
3e32cb2e JW |
2561 | unsigned long curusage; |
2562 | unsigned long oldusage; | |
2563 | bool enlarge = false; | |
81d39c20 | 2564 | int retry_count; |
3e32cb2e | 2565 | int ret; |
8c7c6e34 | 2566 | |
81d39c20 | 2567 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2568 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2569 | mem_cgroup_count_children(memcg); | |
2570 | ||
2571 | oldusage = page_counter_read(&memcg->memsw); | |
2572 | ||
2573 | do { | |
8c7c6e34 KH |
2574 | if (signal_pending(current)) { |
2575 | ret = -EINTR; | |
2576 | break; | |
2577 | } | |
3e32cb2e JW |
2578 | |
2579 | mutex_lock(&memcg_limit_mutex); | |
2580 | if (limit < memcg->memory.limit) { | |
2581 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2582 | ret = -EINVAL; |
8c7c6e34 KH |
2583 | break; |
2584 | } | |
3e32cb2e JW |
2585 | if (limit > memcg->memsw.limit) |
2586 | enlarge = true; | |
2587 | ret = page_counter_limit(&memcg->memsw, limit); | |
2588 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2589 | |
2590 | if (!ret) | |
2591 | break; | |
2592 | ||
b70a2a21 JW |
2593 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
2594 | ||
3e32cb2e | 2595 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 2596 | /* Usage is reduced ? */ |
8c7c6e34 | 2597 | if (curusage >= oldusage) |
628f4235 | 2598 | retry_count--; |
81d39c20 KH |
2599 | else |
2600 | oldusage = curusage; | |
3e32cb2e JW |
2601 | } while (retry_count); |
2602 | ||
3c11ecf4 KH |
2603 | if (!ret && enlarge) |
2604 | memcg_oom_recover(memcg); | |
3e32cb2e | 2605 | |
628f4235 KH |
2606 | return ret; |
2607 | } | |
2608 | ||
0608f43d AM |
2609 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2610 | gfp_t gfp_mask, | |
2611 | unsigned long *total_scanned) | |
2612 | { | |
2613 | unsigned long nr_reclaimed = 0; | |
2614 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2615 | unsigned long reclaimed; | |
2616 | int loop = 0; | |
2617 | struct mem_cgroup_tree_per_zone *mctz; | |
3e32cb2e | 2618 | unsigned long excess; |
0608f43d AM |
2619 | unsigned long nr_scanned; |
2620 | ||
2621 | if (order > 0) | |
2622 | return 0; | |
2623 | ||
2624 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); | |
2625 | /* | |
2626 | * This loop can run a while, specially if mem_cgroup's continuously | |
2627 | * keep exceeding their soft limit and putting the system under | |
2628 | * pressure | |
2629 | */ | |
2630 | do { | |
2631 | if (next_mz) | |
2632 | mz = next_mz; | |
2633 | else | |
2634 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2635 | if (!mz) | |
2636 | break; | |
2637 | ||
2638 | nr_scanned = 0; | |
2639 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, | |
2640 | gfp_mask, &nr_scanned); | |
2641 | nr_reclaimed += reclaimed; | |
2642 | *total_scanned += nr_scanned; | |
0a31bc97 | 2643 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2644 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2645 | |
2646 | /* | |
2647 | * If we failed to reclaim anything from this memory cgroup | |
2648 | * it is time to move on to the next cgroup | |
2649 | */ | |
2650 | next_mz = NULL; | |
bc2f2e7f VD |
2651 | if (!reclaimed) |
2652 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2653 | ||
3e32cb2e | 2654 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2655 | /* |
2656 | * One school of thought says that we should not add | |
2657 | * back the node to the tree if reclaim returns 0. | |
2658 | * But our reclaim could return 0, simply because due | |
2659 | * to priority we are exposing a smaller subset of | |
2660 | * memory to reclaim from. Consider this as a longer | |
2661 | * term TODO. | |
2662 | */ | |
2663 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2664 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2665 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2666 | css_put(&mz->memcg->css); |
2667 | loop++; | |
2668 | /* | |
2669 | * Could not reclaim anything and there are no more | |
2670 | * mem cgroups to try or we seem to be looping without | |
2671 | * reclaiming anything. | |
2672 | */ | |
2673 | if (!nr_reclaimed && | |
2674 | (next_mz == NULL || | |
2675 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2676 | break; | |
2677 | } while (!nr_reclaimed); | |
2678 | if (next_mz) | |
2679 | css_put(&next_mz->memcg->css); | |
2680 | return nr_reclaimed; | |
2681 | } | |
2682 | ||
ea280e7b TH |
2683 | /* |
2684 | * Test whether @memcg has children, dead or alive. Note that this | |
2685 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2686 | * returns %true if there are child csses according to the cgroup | |
2687 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2688 | */ | |
b5f99b53 GC |
2689 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2690 | { | |
ea280e7b TH |
2691 | bool ret; |
2692 | ||
696ac172 | 2693 | /* |
ea280e7b TH |
2694 | * The lock does not prevent addition or deletion of children, but |
2695 | * it prevents a new child from being initialized based on this | |
2696 | * parent in css_online(), so it's enough to decide whether | |
2697 | * hierarchically inherited attributes can still be changed or not. | |
696ac172 | 2698 | */ |
ea280e7b TH |
2699 | lockdep_assert_held(&memcg_create_mutex); |
2700 | ||
2701 | rcu_read_lock(); | |
2702 | ret = css_next_child(NULL, &memcg->css); | |
2703 | rcu_read_unlock(); | |
2704 | return ret; | |
b5f99b53 GC |
2705 | } |
2706 | ||
c26251f9 MH |
2707 | /* |
2708 | * Reclaims as many pages from the given memcg as possible and moves | |
2709 | * the rest to the parent. | |
2710 | * | |
2711 | * Caller is responsible for holding css reference for memcg. | |
2712 | */ | |
2713 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2714 | { | |
2715 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2716 | |
c1e862c1 KH |
2717 | /* we call try-to-free pages for make this cgroup empty */ |
2718 | lru_add_drain_all(); | |
f817ed48 | 2719 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2720 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2721 | int progress; |
c1e862c1 | 2722 | |
c26251f9 MH |
2723 | if (signal_pending(current)) |
2724 | return -EINTR; | |
2725 | ||
b70a2a21 JW |
2726 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2727 | GFP_KERNEL, true); | |
c1e862c1 | 2728 | if (!progress) { |
f817ed48 | 2729 | nr_retries--; |
c1e862c1 | 2730 | /* maybe some writeback is necessary */ |
8aa7e847 | 2731 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2732 | } |
f817ed48 KH |
2733 | |
2734 | } | |
ab5196c2 MH |
2735 | |
2736 | return 0; | |
cc847582 KH |
2737 | } |
2738 | ||
6770c64e TH |
2739 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2740 | char *buf, size_t nbytes, | |
2741 | loff_t off) | |
c1e862c1 | 2742 | { |
6770c64e | 2743 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2744 | |
d8423011 MH |
2745 | if (mem_cgroup_is_root(memcg)) |
2746 | return -EINVAL; | |
6770c64e | 2747 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2748 | } |
2749 | ||
182446d0 TH |
2750 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2751 | struct cftype *cft) | |
18f59ea7 | 2752 | { |
182446d0 | 2753 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2754 | } |
2755 | ||
182446d0 TH |
2756 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2757 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2758 | { |
2759 | int retval = 0; | |
182446d0 | 2760 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2761 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2762 | |
0999821b | 2763 | mutex_lock(&memcg_create_mutex); |
567fb435 GC |
2764 | |
2765 | if (memcg->use_hierarchy == val) | |
2766 | goto out; | |
2767 | ||
18f59ea7 | 2768 | /* |
af901ca1 | 2769 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2770 | * in the child subtrees. If it is unset, then the change can |
2771 | * occur, provided the current cgroup has no children. | |
2772 | * | |
2773 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2774 | * set if there are no children. | |
2775 | */ | |
c0ff4b85 | 2776 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2777 | (val == 1 || val == 0)) { |
ea280e7b | 2778 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2779 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2780 | else |
2781 | retval = -EBUSY; | |
2782 | } else | |
2783 | retval = -EINVAL; | |
567fb435 GC |
2784 | |
2785 | out: | |
0999821b | 2786 | mutex_unlock(&memcg_create_mutex); |
18f59ea7 BS |
2787 | |
2788 | return retval; | |
2789 | } | |
2790 | ||
3e32cb2e JW |
2791 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
2792 | enum mem_cgroup_stat_index idx) | |
ce00a967 JW |
2793 | { |
2794 | struct mem_cgroup *iter; | |
484ebb3b | 2795 | unsigned long val = 0; |
ce00a967 | 2796 | |
ce00a967 JW |
2797 | for_each_mem_cgroup_tree(iter, memcg) |
2798 | val += mem_cgroup_read_stat(iter, idx); | |
2799 | ||
ce00a967 JW |
2800 | return val; |
2801 | } | |
2802 | ||
6f646156 | 2803 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2804 | { |
c12176d3 | 2805 | unsigned long val; |
ce00a967 | 2806 | |
3e32cb2e JW |
2807 | if (mem_cgroup_is_root(memcg)) { |
2808 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); | |
2809 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); | |
2810 | if (swap) | |
2811 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); | |
2812 | } else { | |
ce00a967 | 2813 | if (!swap) |
3e32cb2e | 2814 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2815 | else |
3e32cb2e | 2816 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2817 | } |
c12176d3 | 2818 | return val; |
ce00a967 JW |
2819 | } |
2820 | ||
3e32cb2e JW |
2821 | enum { |
2822 | RES_USAGE, | |
2823 | RES_LIMIT, | |
2824 | RES_MAX_USAGE, | |
2825 | RES_FAILCNT, | |
2826 | RES_SOFT_LIMIT, | |
2827 | }; | |
ce00a967 | 2828 | |
791badbd | 2829 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2830 | struct cftype *cft) |
8cdea7c0 | 2831 | { |
182446d0 | 2832 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2833 | struct page_counter *counter; |
af36f906 | 2834 | |
3e32cb2e | 2835 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 2836 | case _MEM: |
3e32cb2e JW |
2837 | counter = &memcg->memory; |
2838 | break; | |
8c7c6e34 | 2839 | case _MEMSWAP: |
3e32cb2e JW |
2840 | counter = &memcg->memsw; |
2841 | break; | |
510fc4e1 | 2842 | case _KMEM: |
3e32cb2e | 2843 | counter = &memcg->kmem; |
510fc4e1 | 2844 | break; |
d55f90bf VD |
2845 | case _TCP: |
2846 | counter = &memcg->tcp_mem.memory_allocated; | |
2847 | break; | |
8c7c6e34 KH |
2848 | default: |
2849 | BUG(); | |
8c7c6e34 | 2850 | } |
3e32cb2e JW |
2851 | |
2852 | switch (MEMFILE_ATTR(cft->private)) { | |
2853 | case RES_USAGE: | |
2854 | if (counter == &memcg->memory) | |
c12176d3 | 2855 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 2856 | if (counter == &memcg->memsw) |
c12176d3 | 2857 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
2858 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
2859 | case RES_LIMIT: | |
2860 | return (u64)counter->limit * PAGE_SIZE; | |
2861 | case RES_MAX_USAGE: | |
2862 | return (u64)counter->watermark * PAGE_SIZE; | |
2863 | case RES_FAILCNT: | |
2864 | return counter->failcnt; | |
2865 | case RES_SOFT_LIMIT: | |
2866 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
2867 | default: | |
2868 | BUG(); | |
2869 | } | |
8cdea7c0 | 2870 | } |
510fc4e1 | 2871 | |
127424c8 | 2872 | #ifndef CONFIG_SLOB |
567e9ab2 | 2873 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 VD |
2874 | { |
2875 | int err = 0; | |
2876 | int memcg_id; | |
2877 | ||
2a4db7eb | 2878 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 2879 | BUG_ON(memcg->kmem_state); |
d6441637 | 2880 | |
510fc4e1 GC |
2881 | /* |
2882 | * For simplicity, we won't allow this to be disabled. It also can't | |
2883 | * be changed if the cgroup has children already, or if tasks had | |
2884 | * already joined. | |
2885 | * | |
2886 | * If tasks join before we set the limit, a person looking at | |
2887 | * kmem.usage_in_bytes will have no way to determine when it took | |
2888 | * place, which makes the value quite meaningless. | |
2889 | * | |
2890 | * After it first became limited, changes in the value of the limit are | |
2891 | * of course permitted. | |
510fc4e1 | 2892 | */ |
0999821b | 2893 | mutex_lock(&memcg_create_mutex); |
27bd4dbb | 2894 | if (cgroup_is_populated(memcg->css.cgroup) || |
ea280e7b | 2895 | (memcg->use_hierarchy && memcg_has_children(memcg))) |
d6441637 VD |
2896 | err = -EBUSY; |
2897 | mutex_unlock(&memcg_create_mutex); | |
2898 | if (err) | |
2899 | goto out; | |
510fc4e1 | 2900 | |
f3bb3043 | 2901 | memcg_id = memcg_alloc_cache_id(); |
d6441637 VD |
2902 | if (memcg_id < 0) { |
2903 | err = memcg_id; | |
2904 | goto out; | |
2905 | } | |
2906 | ||
ef12947c | 2907 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 2908 | /* |
567e9ab2 | 2909 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 2910 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
2911 | * guarantee no one starts accounting before all call sites are |
2912 | * patched. | |
2913 | */ | |
900a38f0 | 2914 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 2915 | memcg->kmem_state = KMEM_ONLINE; |
510fc4e1 | 2916 | out: |
d6441637 | 2917 | return err; |
d6441637 VD |
2918 | } |
2919 | ||
55007d84 | 2920 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
510fc4e1 | 2921 | { |
55007d84 | 2922 | int ret = 0; |
510fc4e1 | 2923 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
55007d84 | 2924 | |
d6441637 VD |
2925 | if (!parent) |
2926 | return 0; | |
55007d84 | 2927 | |
8c0145b6 | 2928 | mutex_lock(&memcg_limit_mutex); |
55007d84 | 2929 | /* |
567e9ab2 JW |
2930 | * If the parent cgroup is not kmem-online now, it cannot be |
2931 | * onlined after this point, because it has at least one child | |
2932 | * already. | |
55007d84 | 2933 | */ |
04823c83 VD |
2934 | if (memcg_kmem_online(parent) || |
2935 | (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nokmem)) | |
567e9ab2 | 2936 | ret = memcg_online_kmem(memcg); |
8c0145b6 | 2937 | mutex_unlock(&memcg_limit_mutex); |
55007d84 | 2938 | return ret; |
510fc4e1 | 2939 | } |
8e0a8912 | 2940 | |
8e0a8912 JW |
2941 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
2942 | { | |
2943 | struct cgroup_subsys_state *css; | |
2944 | struct mem_cgroup *parent, *child; | |
2945 | int kmemcg_id; | |
2946 | ||
2947 | if (memcg->kmem_state != KMEM_ONLINE) | |
2948 | return; | |
2949 | /* | |
2950 | * Clear the online state before clearing memcg_caches array | |
2951 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
2952 | * guarantees that no cache will be created for this cgroup | |
2953 | * after we are done (see memcg_create_kmem_cache()). | |
2954 | */ | |
2955 | memcg->kmem_state = KMEM_ALLOCATED; | |
2956 | ||
2957 | memcg_deactivate_kmem_caches(memcg); | |
2958 | ||
2959 | kmemcg_id = memcg->kmemcg_id; | |
2960 | BUG_ON(kmemcg_id < 0); | |
2961 | ||
2962 | parent = parent_mem_cgroup(memcg); | |
2963 | if (!parent) | |
2964 | parent = root_mem_cgroup; | |
2965 | ||
2966 | /* | |
2967 | * Change kmemcg_id of this cgroup and all its descendants to the | |
2968 | * parent's id, and then move all entries from this cgroup's list_lrus | |
2969 | * to ones of the parent. After we have finished, all list_lrus | |
2970 | * corresponding to this cgroup are guaranteed to remain empty. The | |
2971 | * ordering is imposed by list_lru_node->lock taken by | |
2972 | * memcg_drain_all_list_lrus(). | |
2973 | */ | |
2974 | css_for_each_descendant_pre(css, &memcg->css) { | |
2975 | child = mem_cgroup_from_css(css); | |
2976 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
2977 | child->kmemcg_id = parent->kmemcg_id; | |
2978 | if (!memcg->use_hierarchy) | |
2979 | break; | |
2980 | } | |
2981 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); | |
2982 | ||
2983 | memcg_free_cache_id(kmemcg_id); | |
2984 | } | |
2985 | ||
2986 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2987 | { | |
2988 | if (memcg->kmem_state == KMEM_ALLOCATED) { | |
2989 | memcg_destroy_kmem_caches(memcg); | |
2990 | static_branch_dec(&memcg_kmem_enabled_key); | |
2991 | WARN_ON(page_counter_read(&memcg->kmem)); | |
2992 | } | |
8e0a8912 | 2993 | } |
d6441637 | 2994 | #else |
127424c8 JW |
2995 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
2996 | { | |
2997 | return 0; | |
2998 | } | |
2999 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3000 | { | |
3001 | } | |
3002 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3003 | { | |
3004 | } | |
3005 | #endif /* !CONFIG_SLOB */ | |
3006 | ||
d6441637 | 3007 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3008 | unsigned long limit) |
d6441637 | 3009 | { |
127424c8 JW |
3010 | int ret; |
3011 | ||
3012 | mutex_lock(&memcg_limit_mutex); | |
3013 | /* Top-level cgroup doesn't propagate from root */ | |
3014 | if (!memcg_kmem_online(memcg)) { | |
3015 | ret = memcg_online_kmem(memcg); | |
3016 | if (ret) | |
3017 | goto out; | |
3018 | } | |
3019 | ret = page_counter_limit(&memcg->kmem, limit); | |
3020 | out: | |
3021 | mutex_unlock(&memcg_limit_mutex); | |
3022 | return ret; | |
d6441637 | 3023 | } |
510fc4e1 | 3024 | |
d55f90bf VD |
3025 | static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit) |
3026 | { | |
3027 | int ret; | |
3028 | ||
3029 | mutex_lock(&memcg_limit_mutex); | |
3030 | ||
3031 | ret = page_counter_limit(&memcg->tcp_mem.memory_allocated, limit); | |
3032 | if (ret) | |
3033 | goto out; | |
3034 | ||
3035 | if (!memcg->tcp_mem.active) { | |
3036 | /* | |
3037 | * The active flag needs to be written after the static_key | |
3038 | * update. This is what guarantees that the socket activation | |
3039 | * function is the last one to run. See sock_update_memcg() for | |
3040 | * details, and note that we don't mark any socket as belonging | |
3041 | * to this memcg until that flag is up. | |
3042 | * | |
3043 | * We need to do this, because static_keys will span multiple | |
3044 | * sites, but we can't control their order. If we mark a socket | |
3045 | * as accounted, but the accounting functions are not patched in | |
3046 | * yet, we'll lose accounting. | |
3047 | * | |
3048 | * We never race with the readers in sock_update_memcg(), | |
3049 | * because when this value change, the code to process it is not | |
3050 | * patched in yet. | |
3051 | */ | |
3052 | static_branch_inc(&memcg_sockets_enabled_key); | |
3053 | memcg->tcp_mem.active = true; | |
3054 | } | |
3055 | out: | |
3056 | mutex_unlock(&memcg_limit_mutex); | |
3057 | return ret; | |
3058 | } | |
d55f90bf | 3059 | |
628f4235 KH |
3060 | /* |
3061 | * The user of this function is... | |
3062 | * RES_LIMIT. | |
3063 | */ | |
451af504 TH |
3064 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3065 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3066 | { |
451af504 | 3067 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3068 | unsigned long nr_pages; |
628f4235 KH |
3069 | int ret; |
3070 | ||
451af504 | 3071 | buf = strstrip(buf); |
650c5e56 | 3072 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3073 | if (ret) |
3074 | return ret; | |
af36f906 | 3075 | |
3e32cb2e | 3076 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3077 | case RES_LIMIT: |
4b3bde4c BS |
3078 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3079 | ret = -EINVAL; | |
3080 | break; | |
3081 | } | |
3e32cb2e JW |
3082 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3083 | case _MEM: | |
3084 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3085 | break; |
3e32cb2e JW |
3086 | case _MEMSWAP: |
3087 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3088 | break; |
3e32cb2e JW |
3089 | case _KMEM: |
3090 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3091 | break; | |
d55f90bf VD |
3092 | case _TCP: |
3093 | ret = memcg_update_tcp_limit(memcg, nr_pages); | |
3094 | break; | |
3e32cb2e | 3095 | } |
296c81d8 | 3096 | break; |
3e32cb2e JW |
3097 | case RES_SOFT_LIMIT: |
3098 | memcg->soft_limit = nr_pages; | |
3099 | ret = 0; | |
628f4235 KH |
3100 | break; |
3101 | } | |
451af504 | 3102 | return ret ?: nbytes; |
8cdea7c0 BS |
3103 | } |
3104 | ||
6770c64e TH |
3105 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3106 | size_t nbytes, loff_t off) | |
c84872e1 | 3107 | { |
6770c64e | 3108 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3109 | struct page_counter *counter; |
c84872e1 | 3110 | |
3e32cb2e JW |
3111 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3112 | case _MEM: | |
3113 | counter = &memcg->memory; | |
3114 | break; | |
3115 | case _MEMSWAP: | |
3116 | counter = &memcg->memsw; | |
3117 | break; | |
3118 | case _KMEM: | |
3119 | counter = &memcg->kmem; | |
3120 | break; | |
d55f90bf VD |
3121 | case _TCP: |
3122 | counter = &memcg->tcp_mem.memory_allocated; | |
3123 | break; | |
3e32cb2e JW |
3124 | default: |
3125 | BUG(); | |
3126 | } | |
af36f906 | 3127 | |
3e32cb2e | 3128 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3129 | case RES_MAX_USAGE: |
3e32cb2e | 3130 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3131 | break; |
3132 | case RES_FAILCNT: | |
3e32cb2e | 3133 | counter->failcnt = 0; |
29f2a4da | 3134 | break; |
3e32cb2e JW |
3135 | default: |
3136 | BUG(); | |
29f2a4da | 3137 | } |
f64c3f54 | 3138 | |
6770c64e | 3139 | return nbytes; |
c84872e1 PE |
3140 | } |
3141 | ||
182446d0 | 3142 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3143 | struct cftype *cft) |
3144 | { | |
182446d0 | 3145 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3146 | } |
3147 | ||
02491447 | 3148 | #ifdef CONFIG_MMU |
182446d0 | 3149 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3150 | struct cftype *cft, u64 val) |
3151 | { | |
182446d0 | 3152 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3153 | |
1dfab5ab | 3154 | if (val & ~MOVE_MASK) |
7dc74be0 | 3155 | return -EINVAL; |
ee5e8472 | 3156 | |
7dc74be0 | 3157 | /* |
ee5e8472 GC |
3158 | * No kind of locking is needed in here, because ->can_attach() will |
3159 | * check this value once in the beginning of the process, and then carry | |
3160 | * on with stale data. This means that changes to this value will only | |
3161 | * affect task migrations starting after the change. | |
7dc74be0 | 3162 | */ |
c0ff4b85 | 3163 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3164 | return 0; |
3165 | } | |
02491447 | 3166 | #else |
182446d0 | 3167 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3168 | struct cftype *cft, u64 val) |
3169 | { | |
3170 | return -ENOSYS; | |
3171 | } | |
3172 | #endif | |
7dc74be0 | 3173 | |
406eb0c9 | 3174 | #ifdef CONFIG_NUMA |
2da8ca82 | 3175 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3176 | { |
25485de6 GT |
3177 | struct numa_stat { |
3178 | const char *name; | |
3179 | unsigned int lru_mask; | |
3180 | }; | |
3181 | ||
3182 | static const struct numa_stat stats[] = { | |
3183 | { "total", LRU_ALL }, | |
3184 | { "file", LRU_ALL_FILE }, | |
3185 | { "anon", LRU_ALL_ANON }, | |
3186 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3187 | }; | |
3188 | const struct numa_stat *stat; | |
406eb0c9 | 3189 | int nid; |
25485de6 | 3190 | unsigned long nr; |
2da8ca82 | 3191 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3192 | |
25485de6 GT |
3193 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3194 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3195 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3196 | for_each_node_state(nid, N_MEMORY) { | |
3197 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3198 | stat->lru_mask); | |
3199 | seq_printf(m, " N%d=%lu", nid, nr); | |
3200 | } | |
3201 | seq_putc(m, '\n'); | |
406eb0c9 | 3202 | } |
406eb0c9 | 3203 | |
071aee13 YH |
3204 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3205 | struct mem_cgroup *iter; | |
3206 | ||
3207 | nr = 0; | |
3208 | for_each_mem_cgroup_tree(iter, memcg) | |
3209 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3210 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3211 | for_each_node_state(nid, N_MEMORY) { | |
3212 | nr = 0; | |
3213 | for_each_mem_cgroup_tree(iter, memcg) | |
3214 | nr += mem_cgroup_node_nr_lru_pages( | |
3215 | iter, nid, stat->lru_mask); | |
3216 | seq_printf(m, " N%d=%lu", nid, nr); | |
3217 | } | |
3218 | seq_putc(m, '\n'); | |
406eb0c9 | 3219 | } |
406eb0c9 | 3220 | |
406eb0c9 YH |
3221 | return 0; |
3222 | } | |
3223 | #endif /* CONFIG_NUMA */ | |
3224 | ||
2da8ca82 | 3225 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3226 | { |
2da8ca82 | 3227 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3228 | unsigned long memory, memsw; |
af7c4b0e JW |
3229 | struct mem_cgroup *mi; |
3230 | unsigned int i; | |
406eb0c9 | 3231 | |
0ca44b14 GT |
3232 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
3233 | MEM_CGROUP_STAT_NSTATS); | |
3234 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != | |
3235 | MEM_CGROUP_EVENTS_NSTATS); | |
70bc068c RS |
3236 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3237 | ||
af7c4b0e | 3238 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
7941d214 | 3239 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3240 | continue; |
484ebb3b | 3241 | seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i], |
af7c4b0e | 3242 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
1dd3a273 | 3243 | } |
7b854121 | 3244 | |
af7c4b0e JW |
3245 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
3246 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], | |
3247 | mem_cgroup_read_events(memcg, i)); | |
3248 | ||
3249 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3250 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3251 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3252 | ||
14067bb3 | 3253 | /* Hierarchical information */ |
3e32cb2e JW |
3254 | memory = memsw = PAGE_COUNTER_MAX; |
3255 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3256 | memory = min(memory, mi->memory.limit); | |
3257 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3258 | } |
3e32cb2e JW |
3259 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3260 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3261 | if (do_memsw_account()) |
3e32cb2e JW |
3262 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3263 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3264 | |
af7c4b0e | 3265 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
484ebb3b | 3266 | unsigned long long val = 0; |
af7c4b0e | 3267 | |
7941d214 | 3268 | if (i == MEM_CGROUP_STAT_SWAP && !do_memsw_account()) |
1dd3a273 | 3269 | continue; |
af7c4b0e JW |
3270 | for_each_mem_cgroup_tree(mi, memcg) |
3271 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; | |
484ebb3b | 3272 | seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val); |
af7c4b0e JW |
3273 | } |
3274 | ||
3275 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { | |
3276 | unsigned long long val = 0; | |
3277 | ||
3278 | for_each_mem_cgroup_tree(mi, memcg) | |
3279 | val += mem_cgroup_read_events(mi, i); | |
3280 | seq_printf(m, "total_%s %llu\n", | |
3281 | mem_cgroup_events_names[i], val); | |
3282 | } | |
3283 | ||
3284 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3285 | unsigned long long val = 0; | |
3286 | ||
3287 | for_each_mem_cgroup_tree(mi, memcg) | |
3288 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3289 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3290 | } |
14067bb3 | 3291 | |
7f016ee8 | 3292 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 KM |
3293 | { |
3294 | int nid, zid; | |
3295 | struct mem_cgroup_per_zone *mz; | |
89abfab1 | 3296 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3297 | unsigned long recent_rotated[2] = {0, 0}; |
3298 | unsigned long recent_scanned[2] = {0, 0}; | |
3299 | ||
3300 | for_each_online_node(nid) | |
3301 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
e231875b | 3302 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
89abfab1 | 3303 | rstat = &mz->lruvec.reclaim_stat; |
7f016ee8 | 3304 | |
89abfab1 HD |
3305 | recent_rotated[0] += rstat->recent_rotated[0]; |
3306 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3307 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3308 | recent_scanned[1] += rstat->recent_scanned[1]; | |
7f016ee8 | 3309 | } |
78ccf5b5 JW |
3310 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3311 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3312 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3313 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3314 | } |
3315 | #endif | |
3316 | ||
d2ceb9b7 KH |
3317 | return 0; |
3318 | } | |
3319 | ||
182446d0 TH |
3320 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3321 | struct cftype *cft) | |
a7885eb8 | 3322 | { |
182446d0 | 3323 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3324 | |
1f4c025b | 3325 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3326 | } |
3327 | ||
182446d0 TH |
3328 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3329 | struct cftype *cft, u64 val) | |
a7885eb8 | 3330 | { |
182446d0 | 3331 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3332 | |
3dae7fec | 3333 | if (val > 100) |
a7885eb8 KM |
3334 | return -EINVAL; |
3335 | ||
14208b0e | 3336 | if (css->parent) |
3dae7fec JW |
3337 | memcg->swappiness = val; |
3338 | else | |
3339 | vm_swappiness = val; | |
068b38c1 | 3340 | |
a7885eb8 KM |
3341 | return 0; |
3342 | } | |
3343 | ||
2e72b634 KS |
3344 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3345 | { | |
3346 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3347 | unsigned long usage; |
2e72b634 KS |
3348 | int i; |
3349 | ||
3350 | rcu_read_lock(); | |
3351 | if (!swap) | |
2c488db2 | 3352 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3353 | else |
2c488db2 | 3354 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3355 | |
3356 | if (!t) | |
3357 | goto unlock; | |
3358 | ||
ce00a967 | 3359 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3360 | |
3361 | /* | |
748dad36 | 3362 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3363 | * If it's not true, a threshold was crossed after last |
3364 | * call of __mem_cgroup_threshold(). | |
3365 | */ | |
5407a562 | 3366 | i = t->current_threshold; |
2e72b634 KS |
3367 | |
3368 | /* | |
3369 | * Iterate backward over array of thresholds starting from | |
3370 | * current_threshold and check if a threshold is crossed. | |
3371 | * If none of thresholds below usage is crossed, we read | |
3372 | * only one element of the array here. | |
3373 | */ | |
3374 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3375 | eventfd_signal(t->entries[i].eventfd, 1); | |
3376 | ||
3377 | /* i = current_threshold + 1 */ | |
3378 | i++; | |
3379 | ||
3380 | /* | |
3381 | * Iterate forward over array of thresholds starting from | |
3382 | * current_threshold+1 and check if a threshold is crossed. | |
3383 | * If none of thresholds above usage is crossed, we read | |
3384 | * only one element of the array here. | |
3385 | */ | |
3386 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3387 | eventfd_signal(t->entries[i].eventfd, 1); | |
3388 | ||
3389 | /* Update current_threshold */ | |
5407a562 | 3390 | t->current_threshold = i - 1; |
2e72b634 KS |
3391 | unlock: |
3392 | rcu_read_unlock(); | |
3393 | } | |
3394 | ||
3395 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3396 | { | |
ad4ca5f4 KS |
3397 | while (memcg) { |
3398 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3399 | if (do_memsw_account()) |
ad4ca5f4 KS |
3400 | __mem_cgroup_threshold(memcg, true); |
3401 | ||
3402 | memcg = parent_mem_cgroup(memcg); | |
3403 | } | |
2e72b634 KS |
3404 | } |
3405 | ||
3406 | static int compare_thresholds(const void *a, const void *b) | |
3407 | { | |
3408 | const struct mem_cgroup_threshold *_a = a; | |
3409 | const struct mem_cgroup_threshold *_b = b; | |
3410 | ||
2bff24a3 GT |
3411 | if (_a->threshold > _b->threshold) |
3412 | return 1; | |
3413 | ||
3414 | if (_a->threshold < _b->threshold) | |
3415 | return -1; | |
3416 | ||
3417 | return 0; | |
2e72b634 KS |
3418 | } |
3419 | ||
c0ff4b85 | 3420 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3421 | { |
3422 | struct mem_cgroup_eventfd_list *ev; | |
3423 | ||
2bcf2e92 MH |
3424 | spin_lock(&memcg_oom_lock); |
3425 | ||
c0ff4b85 | 3426 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3427 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3428 | |
3429 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3430 | return 0; |
3431 | } | |
3432 | ||
c0ff4b85 | 3433 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3434 | { |
7d74b06f KH |
3435 | struct mem_cgroup *iter; |
3436 | ||
c0ff4b85 | 3437 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3438 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3439 | } |
3440 | ||
59b6f873 | 3441 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3442 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3443 | { |
2c488db2 KS |
3444 | struct mem_cgroup_thresholds *thresholds; |
3445 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3446 | unsigned long threshold; |
3447 | unsigned long usage; | |
2c488db2 | 3448 | int i, size, ret; |
2e72b634 | 3449 | |
650c5e56 | 3450 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3451 | if (ret) |
3452 | return ret; | |
3453 | ||
3454 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3455 | |
05b84301 | 3456 | if (type == _MEM) { |
2c488db2 | 3457 | thresholds = &memcg->thresholds; |
ce00a967 | 3458 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3459 | } else if (type == _MEMSWAP) { |
2c488db2 | 3460 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3461 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3462 | } else |
2e72b634 KS |
3463 | BUG(); |
3464 | ||
2e72b634 | 3465 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3466 | if (thresholds->primary) |
2e72b634 KS |
3467 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3468 | ||
2c488db2 | 3469 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3470 | |
3471 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3472 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3473 | GFP_KERNEL); |
2c488db2 | 3474 | if (!new) { |
2e72b634 KS |
3475 | ret = -ENOMEM; |
3476 | goto unlock; | |
3477 | } | |
2c488db2 | 3478 | new->size = size; |
2e72b634 KS |
3479 | |
3480 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3481 | if (thresholds->primary) { |
3482 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3483 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3484 | } |
3485 | ||
2e72b634 | 3486 | /* Add new threshold */ |
2c488db2 KS |
3487 | new->entries[size - 1].eventfd = eventfd; |
3488 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3489 | |
3490 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3491 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3492 | compare_thresholds, NULL); |
3493 | ||
3494 | /* Find current threshold */ | |
2c488db2 | 3495 | new->current_threshold = -1; |
2e72b634 | 3496 | for (i = 0; i < size; i++) { |
748dad36 | 3497 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3498 | /* |
2c488db2 KS |
3499 | * new->current_threshold will not be used until |
3500 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3501 | * it here. |
3502 | */ | |
2c488db2 | 3503 | ++new->current_threshold; |
748dad36 SZ |
3504 | } else |
3505 | break; | |
2e72b634 KS |
3506 | } |
3507 | ||
2c488db2 KS |
3508 | /* Free old spare buffer and save old primary buffer as spare */ |
3509 | kfree(thresholds->spare); | |
3510 | thresholds->spare = thresholds->primary; | |
3511 | ||
3512 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3513 | |
907860ed | 3514 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3515 | synchronize_rcu(); |
3516 | ||
2e72b634 KS |
3517 | unlock: |
3518 | mutex_unlock(&memcg->thresholds_lock); | |
3519 | ||
3520 | return ret; | |
3521 | } | |
3522 | ||
59b6f873 | 3523 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3524 | struct eventfd_ctx *eventfd, const char *args) |
3525 | { | |
59b6f873 | 3526 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3527 | } |
3528 | ||
59b6f873 | 3529 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3530 | struct eventfd_ctx *eventfd, const char *args) |
3531 | { | |
59b6f873 | 3532 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3533 | } |
3534 | ||
59b6f873 | 3535 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3536 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3537 | { |
2c488db2 KS |
3538 | struct mem_cgroup_thresholds *thresholds; |
3539 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3540 | unsigned long usage; |
2c488db2 | 3541 | int i, j, size; |
2e72b634 KS |
3542 | |
3543 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3544 | |
3545 | if (type == _MEM) { | |
2c488db2 | 3546 | thresholds = &memcg->thresholds; |
ce00a967 | 3547 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3548 | } else if (type == _MEMSWAP) { |
2c488db2 | 3549 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3550 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3551 | } else |
2e72b634 KS |
3552 | BUG(); |
3553 | ||
371528ca AV |
3554 | if (!thresholds->primary) |
3555 | goto unlock; | |
3556 | ||
2e72b634 KS |
3557 | /* Check if a threshold crossed before removing */ |
3558 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3559 | ||
3560 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3561 | size = 0; |
3562 | for (i = 0; i < thresholds->primary->size; i++) { | |
3563 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3564 | size++; |
3565 | } | |
3566 | ||
2c488db2 | 3567 | new = thresholds->spare; |
907860ed | 3568 | |
2e72b634 KS |
3569 | /* Set thresholds array to NULL if we don't have thresholds */ |
3570 | if (!size) { | |
2c488db2 KS |
3571 | kfree(new); |
3572 | new = NULL; | |
907860ed | 3573 | goto swap_buffers; |
2e72b634 KS |
3574 | } |
3575 | ||
2c488db2 | 3576 | new->size = size; |
2e72b634 KS |
3577 | |
3578 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3579 | new->current_threshold = -1; |
3580 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3581 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3582 | continue; |
3583 | ||
2c488db2 | 3584 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3585 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3586 | /* |
2c488db2 | 3587 | * new->current_threshold will not be used |
2e72b634 KS |
3588 | * until rcu_assign_pointer(), so it's safe to increment |
3589 | * it here. | |
3590 | */ | |
2c488db2 | 3591 | ++new->current_threshold; |
2e72b634 KS |
3592 | } |
3593 | j++; | |
3594 | } | |
3595 | ||
907860ed | 3596 | swap_buffers: |
2c488db2 KS |
3597 | /* Swap primary and spare array */ |
3598 | thresholds->spare = thresholds->primary; | |
8c757763 | 3599 | |
2c488db2 | 3600 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3601 | |
907860ed | 3602 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3603 | synchronize_rcu(); |
6611d8d7 MC |
3604 | |
3605 | /* If all events are unregistered, free the spare array */ | |
3606 | if (!new) { | |
3607 | kfree(thresholds->spare); | |
3608 | thresholds->spare = NULL; | |
3609 | } | |
371528ca | 3610 | unlock: |
2e72b634 | 3611 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3612 | } |
c1e862c1 | 3613 | |
59b6f873 | 3614 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3615 | struct eventfd_ctx *eventfd) |
3616 | { | |
59b6f873 | 3617 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3618 | } |
3619 | ||
59b6f873 | 3620 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3621 | struct eventfd_ctx *eventfd) |
3622 | { | |
59b6f873 | 3623 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3624 | } |
3625 | ||
59b6f873 | 3626 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3627 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3628 | { |
9490ff27 | 3629 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3630 | |
9490ff27 KH |
3631 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3632 | if (!event) | |
3633 | return -ENOMEM; | |
3634 | ||
1af8efe9 | 3635 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3636 | |
3637 | event->eventfd = eventfd; | |
3638 | list_add(&event->list, &memcg->oom_notify); | |
3639 | ||
3640 | /* already in OOM ? */ | |
c2b42d3c | 3641 | if (memcg->under_oom) |
9490ff27 | 3642 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3643 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3644 | |
3645 | return 0; | |
3646 | } | |
3647 | ||
59b6f873 | 3648 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3649 | struct eventfd_ctx *eventfd) |
9490ff27 | 3650 | { |
9490ff27 | 3651 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3652 | |
1af8efe9 | 3653 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3654 | |
c0ff4b85 | 3655 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3656 | if (ev->eventfd == eventfd) { |
3657 | list_del(&ev->list); | |
3658 | kfree(ev); | |
3659 | } | |
3660 | } | |
3661 | ||
1af8efe9 | 3662 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3663 | } |
3664 | ||
2da8ca82 | 3665 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3666 | { |
2da8ca82 | 3667 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3668 | |
791badbd | 3669 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3670 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
3c11ecf4 KH |
3671 | return 0; |
3672 | } | |
3673 | ||
182446d0 | 3674 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3675 | struct cftype *cft, u64 val) |
3676 | { | |
182446d0 | 3677 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3678 | |
3679 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3680 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3681 | return -EINVAL; |
3682 | ||
c0ff4b85 | 3683 | memcg->oom_kill_disable = val; |
4d845ebf | 3684 | if (!val) |
c0ff4b85 | 3685 | memcg_oom_recover(memcg); |
3dae7fec | 3686 | |
3c11ecf4 KH |
3687 | return 0; |
3688 | } | |
3689 | ||
52ebea74 TH |
3690 | #ifdef CONFIG_CGROUP_WRITEBACK |
3691 | ||
3692 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) | |
3693 | { | |
3694 | return &memcg->cgwb_list; | |
3695 | } | |
3696 | ||
841710aa TH |
3697 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3698 | { | |
3699 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3700 | } | |
3701 | ||
3702 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3703 | { | |
3704 | wb_domain_exit(&memcg->cgwb_domain); | |
3705 | } | |
3706 | ||
2529bb3a TH |
3707 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3708 | { | |
3709 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3710 | } | |
3711 | ||
841710aa TH |
3712 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3713 | { | |
3714 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3715 | ||
3716 | if (!memcg->css.parent) | |
3717 | return NULL; | |
3718 | ||
3719 | return &memcg->cgwb_domain; | |
3720 | } | |
3721 | ||
c2aa723a TH |
3722 | /** |
3723 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3724 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3725 | * @pfilepages: out parameter for number of file pages |
3726 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3727 | * @pdirty: out parameter for number of dirty pages |
3728 | * @pwriteback: out parameter for number of pages under writeback | |
3729 | * | |
c5edf9cd TH |
3730 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3731 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3732 | * is a bit more involved. | |
c2aa723a | 3733 | * |
c5edf9cd TH |
3734 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3735 | * headroom is calculated as the lowest headroom of itself and the | |
3736 | * ancestors. Note that this doesn't consider the actual amount of | |
3737 | * available memory in the system. The caller should further cap | |
3738 | * *@pheadroom accordingly. | |
c2aa723a | 3739 | */ |
c5edf9cd TH |
3740 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3741 | unsigned long *pheadroom, unsigned long *pdirty, | |
3742 | unsigned long *pwriteback) | |
c2aa723a TH |
3743 | { |
3744 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3745 | struct mem_cgroup *parent; | |
c2aa723a TH |
3746 | |
3747 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); | |
3748 | ||
3749 | /* this should eventually include NR_UNSTABLE_NFS */ | |
3750 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); | |
c5edf9cd TH |
3751 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3752 | (1 << LRU_ACTIVE_FILE)); | |
3753 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3754 | |
c2aa723a TH |
3755 | while ((parent = parent_mem_cgroup(memcg))) { |
3756 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); | |
3757 | unsigned long used = page_counter_read(&memcg->memory); | |
3758 | ||
c5edf9cd | 3759 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3760 | memcg = parent; |
3761 | } | |
c2aa723a TH |
3762 | } |
3763 | ||
841710aa TH |
3764 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3765 | ||
3766 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3767 | { | |
3768 | return 0; | |
3769 | } | |
3770 | ||
3771 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3772 | { | |
3773 | } | |
3774 | ||
2529bb3a TH |
3775 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3776 | { | |
3777 | } | |
3778 | ||
52ebea74 TH |
3779 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3780 | ||
3bc942f3 TH |
3781 | /* |
3782 | * DO NOT USE IN NEW FILES. | |
3783 | * | |
3784 | * "cgroup.event_control" implementation. | |
3785 | * | |
3786 | * This is way over-engineered. It tries to support fully configurable | |
3787 | * events for each user. Such level of flexibility is completely | |
3788 | * unnecessary especially in the light of the planned unified hierarchy. | |
3789 | * | |
3790 | * Please deprecate this and replace with something simpler if at all | |
3791 | * possible. | |
3792 | */ | |
3793 | ||
79bd9814 TH |
3794 | /* |
3795 | * Unregister event and free resources. | |
3796 | * | |
3797 | * Gets called from workqueue. | |
3798 | */ | |
3bc942f3 | 3799 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3800 | { |
3bc942f3 TH |
3801 | struct mem_cgroup_event *event = |
3802 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3803 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3804 | |
3805 | remove_wait_queue(event->wqh, &event->wait); | |
3806 | ||
59b6f873 | 3807 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3808 | |
3809 | /* Notify userspace the event is going away. */ | |
3810 | eventfd_signal(event->eventfd, 1); | |
3811 | ||
3812 | eventfd_ctx_put(event->eventfd); | |
3813 | kfree(event); | |
59b6f873 | 3814 | css_put(&memcg->css); |
79bd9814 TH |
3815 | } |
3816 | ||
3817 | /* | |
3818 | * Gets called on POLLHUP on eventfd when user closes it. | |
3819 | * | |
3820 | * Called with wqh->lock held and interrupts disabled. | |
3821 | */ | |
3bc942f3 TH |
3822 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
3823 | int sync, void *key) | |
79bd9814 | 3824 | { |
3bc942f3 TH |
3825 | struct mem_cgroup_event *event = |
3826 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3827 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3828 | unsigned long flags = (unsigned long)key; |
3829 | ||
3830 | if (flags & POLLHUP) { | |
3831 | /* | |
3832 | * If the event has been detached at cgroup removal, we | |
3833 | * can simply return knowing the other side will cleanup | |
3834 | * for us. | |
3835 | * | |
3836 | * We can't race against event freeing since the other | |
3837 | * side will require wqh->lock via remove_wait_queue(), | |
3838 | * which we hold. | |
3839 | */ | |
fba94807 | 3840 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3841 | if (!list_empty(&event->list)) { |
3842 | list_del_init(&event->list); | |
3843 | /* | |
3844 | * We are in atomic context, but cgroup_event_remove() | |
3845 | * may sleep, so we have to call it in workqueue. | |
3846 | */ | |
3847 | schedule_work(&event->remove); | |
3848 | } | |
fba94807 | 3849 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3850 | } |
3851 | ||
3852 | return 0; | |
3853 | } | |
3854 | ||
3bc942f3 | 3855 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3856 | wait_queue_head_t *wqh, poll_table *pt) |
3857 | { | |
3bc942f3 TH |
3858 | struct mem_cgroup_event *event = |
3859 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3860 | |
3861 | event->wqh = wqh; | |
3862 | add_wait_queue(wqh, &event->wait); | |
3863 | } | |
3864 | ||
3865 | /* | |
3bc942f3 TH |
3866 | * DO NOT USE IN NEW FILES. |
3867 | * | |
79bd9814 TH |
3868 | * Parse input and register new cgroup event handler. |
3869 | * | |
3870 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3871 | * Interpretation of args is defined by control file implementation. | |
3872 | */ | |
451af504 TH |
3873 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
3874 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 3875 | { |
451af504 | 3876 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 3877 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 3878 | struct mem_cgroup_event *event; |
79bd9814 TH |
3879 | struct cgroup_subsys_state *cfile_css; |
3880 | unsigned int efd, cfd; | |
3881 | struct fd efile; | |
3882 | struct fd cfile; | |
fba94807 | 3883 | const char *name; |
79bd9814 TH |
3884 | char *endp; |
3885 | int ret; | |
3886 | ||
451af504 TH |
3887 | buf = strstrip(buf); |
3888 | ||
3889 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
3890 | if (*endp != ' ') |
3891 | return -EINVAL; | |
451af504 | 3892 | buf = endp + 1; |
79bd9814 | 3893 | |
451af504 | 3894 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
3895 | if ((*endp != ' ') && (*endp != '\0')) |
3896 | return -EINVAL; | |
451af504 | 3897 | buf = endp + 1; |
79bd9814 TH |
3898 | |
3899 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3900 | if (!event) | |
3901 | return -ENOMEM; | |
3902 | ||
59b6f873 | 3903 | event->memcg = memcg; |
79bd9814 | 3904 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
3905 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
3906 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
3907 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
3908 | |
3909 | efile = fdget(efd); | |
3910 | if (!efile.file) { | |
3911 | ret = -EBADF; | |
3912 | goto out_kfree; | |
3913 | } | |
3914 | ||
3915 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
3916 | if (IS_ERR(event->eventfd)) { | |
3917 | ret = PTR_ERR(event->eventfd); | |
3918 | goto out_put_efile; | |
3919 | } | |
3920 | ||
3921 | cfile = fdget(cfd); | |
3922 | if (!cfile.file) { | |
3923 | ret = -EBADF; | |
3924 | goto out_put_eventfd; | |
3925 | } | |
3926 | ||
3927 | /* the process need read permission on control file */ | |
3928 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
3929 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
3930 | if (ret < 0) | |
3931 | goto out_put_cfile; | |
3932 | ||
fba94807 TH |
3933 | /* |
3934 | * Determine the event callbacks and set them in @event. This used | |
3935 | * to be done via struct cftype but cgroup core no longer knows | |
3936 | * about these events. The following is crude but the whole thing | |
3937 | * is for compatibility anyway. | |
3bc942f3 TH |
3938 | * |
3939 | * DO NOT ADD NEW FILES. | |
fba94807 | 3940 | */ |
b583043e | 3941 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
3942 | |
3943 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
3944 | event->register_event = mem_cgroup_usage_register_event; | |
3945 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
3946 | } else if (!strcmp(name, "memory.oom_control")) { | |
3947 | event->register_event = mem_cgroup_oom_register_event; | |
3948 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
3949 | } else if (!strcmp(name, "memory.pressure_level")) { | |
3950 | event->register_event = vmpressure_register_event; | |
3951 | event->unregister_event = vmpressure_unregister_event; | |
3952 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
3953 | event->register_event = memsw_cgroup_usage_register_event; |
3954 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
3955 | } else { |
3956 | ret = -EINVAL; | |
3957 | goto out_put_cfile; | |
3958 | } | |
3959 | ||
79bd9814 | 3960 | /* |
b5557c4c TH |
3961 | * Verify @cfile should belong to @css. Also, remaining events are |
3962 | * automatically removed on cgroup destruction but the removal is | |
3963 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 3964 | */ |
b583043e | 3965 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 3966 | &memory_cgrp_subsys); |
79bd9814 | 3967 | ret = -EINVAL; |
5a17f543 | 3968 | if (IS_ERR(cfile_css)) |
79bd9814 | 3969 | goto out_put_cfile; |
5a17f543 TH |
3970 | if (cfile_css != css) { |
3971 | css_put(cfile_css); | |
79bd9814 | 3972 | goto out_put_cfile; |
5a17f543 | 3973 | } |
79bd9814 | 3974 | |
451af504 | 3975 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
3976 | if (ret) |
3977 | goto out_put_css; | |
3978 | ||
3979 | efile.file->f_op->poll(efile.file, &event->pt); | |
3980 | ||
fba94807 TH |
3981 | spin_lock(&memcg->event_list_lock); |
3982 | list_add(&event->list, &memcg->event_list); | |
3983 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
3984 | |
3985 | fdput(cfile); | |
3986 | fdput(efile); | |
3987 | ||
451af504 | 3988 | return nbytes; |
79bd9814 TH |
3989 | |
3990 | out_put_css: | |
b5557c4c | 3991 | css_put(css); |
79bd9814 TH |
3992 | out_put_cfile: |
3993 | fdput(cfile); | |
3994 | out_put_eventfd: | |
3995 | eventfd_ctx_put(event->eventfd); | |
3996 | out_put_efile: | |
3997 | fdput(efile); | |
3998 | out_kfree: | |
3999 | kfree(event); | |
4000 | ||
4001 | return ret; | |
4002 | } | |
4003 | ||
241994ed | 4004 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4005 | { |
0eea1030 | 4006 | .name = "usage_in_bytes", |
8c7c6e34 | 4007 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4008 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4009 | }, |
c84872e1 PE |
4010 | { |
4011 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4012 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4013 | .write = mem_cgroup_reset, |
791badbd | 4014 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4015 | }, |
8cdea7c0 | 4016 | { |
0eea1030 | 4017 | .name = "limit_in_bytes", |
8c7c6e34 | 4018 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4019 | .write = mem_cgroup_write, |
791badbd | 4020 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4021 | }, |
296c81d8 BS |
4022 | { |
4023 | .name = "soft_limit_in_bytes", | |
4024 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4025 | .write = mem_cgroup_write, |
791badbd | 4026 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4027 | }, |
8cdea7c0 BS |
4028 | { |
4029 | .name = "failcnt", | |
8c7c6e34 | 4030 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4031 | .write = mem_cgroup_reset, |
791badbd | 4032 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4033 | }, |
d2ceb9b7 KH |
4034 | { |
4035 | .name = "stat", | |
2da8ca82 | 4036 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4037 | }, |
c1e862c1 KH |
4038 | { |
4039 | .name = "force_empty", | |
6770c64e | 4040 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4041 | }, |
18f59ea7 BS |
4042 | { |
4043 | .name = "use_hierarchy", | |
4044 | .write_u64 = mem_cgroup_hierarchy_write, | |
4045 | .read_u64 = mem_cgroup_hierarchy_read, | |
4046 | }, | |
79bd9814 | 4047 | { |
3bc942f3 | 4048 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4049 | .write = memcg_write_event_control, |
7dbdb199 | 4050 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4051 | }, |
a7885eb8 KM |
4052 | { |
4053 | .name = "swappiness", | |
4054 | .read_u64 = mem_cgroup_swappiness_read, | |
4055 | .write_u64 = mem_cgroup_swappiness_write, | |
4056 | }, | |
7dc74be0 DN |
4057 | { |
4058 | .name = "move_charge_at_immigrate", | |
4059 | .read_u64 = mem_cgroup_move_charge_read, | |
4060 | .write_u64 = mem_cgroup_move_charge_write, | |
4061 | }, | |
9490ff27 KH |
4062 | { |
4063 | .name = "oom_control", | |
2da8ca82 | 4064 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4065 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4066 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4067 | }, | |
70ddf637 AV |
4068 | { |
4069 | .name = "pressure_level", | |
70ddf637 | 4070 | }, |
406eb0c9 YH |
4071 | #ifdef CONFIG_NUMA |
4072 | { | |
4073 | .name = "numa_stat", | |
2da8ca82 | 4074 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4075 | }, |
4076 | #endif | |
510fc4e1 GC |
4077 | { |
4078 | .name = "kmem.limit_in_bytes", | |
4079 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4080 | .write = mem_cgroup_write, |
791badbd | 4081 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4082 | }, |
4083 | { | |
4084 | .name = "kmem.usage_in_bytes", | |
4085 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4086 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4087 | }, |
4088 | { | |
4089 | .name = "kmem.failcnt", | |
4090 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4091 | .write = mem_cgroup_reset, |
791badbd | 4092 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4093 | }, |
4094 | { | |
4095 | .name = "kmem.max_usage_in_bytes", | |
4096 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4097 | .write = mem_cgroup_reset, |
791badbd | 4098 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4099 | }, |
749c5415 GC |
4100 | #ifdef CONFIG_SLABINFO |
4101 | { | |
4102 | .name = "kmem.slabinfo", | |
b047501c VD |
4103 | .seq_start = slab_start, |
4104 | .seq_next = slab_next, | |
4105 | .seq_stop = slab_stop, | |
4106 | .seq_show = memcg_slab_show, | |
749c5415 GC |
4107 | }, |
4108 | #endif | |
d55f90bf VD |
4109 | { |
4110 | .name = "kmem.tcp.limit_in_bytes", | |
4111 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4112 | .write = mem_cgroup_write, | |
4113 | .read_u64 = mem_cgroup_read_u64, | |
4114 | }, | |
4115 | { | |
4116 | .name = "kmem.tcp.usage_in_bytes", | |
4117 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4118 | .read_u64 = mem_cgroup_read_u64, | |
4119 | }, | |
4120 | { | |
4121 | .name = "kmem.tcp.failcnt", | |
4122 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4123 | .write = mem_cgroup_reset, | |
4124 | .read_u64 = mem_cgroup_read_u64, | |
4125 | }, | |
4126 | { | |
4127 | .name = "kmem.tcp.max_usage_in_bytes", | |
4128 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4129 | .write = mem_cgroup_reset, | |
4130 | .read_u64 = mem_cgroup_read_u64, | |
4131 | }, | |
6bc10349 | 4132 | { }, /* terminate */ |
af36f906 | 4133 | }; |
8c7c6e34 | 4134 | |
c0ff4b85 | 4135 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4136 | { |
4137 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4138 | struct mem_cgroup_per_zone *mz; |
41e3355d | 4139 | int zone, tmp = node; |
1ecaab2b KH |
4140 | /* |
4141 | * This routine is called against possible nodes. | |
4142 | * But it's BUG to call kmalloc() against offline node. | |
4143 | * | |
4144 | * TODO: this routine can waste much memory for nodes which will | |
4145 | * never be onlined. It's better to use memory hotplug callback | |
4146 | * function. | |
4147 | */ | |
41e3355d KH |
4148 | if (!node_state(node, N_NORMAL_MEMORY)) |
4149 | tmp = -1; | |
17295c88 | 4150 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4151 | if (!pn) |
4152 | return 1; | |
1ecaab2b | 4153 | |
1ecaab2b KH |
4154 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4155 | mz = &pn->zoneinfo[zone]; | |
bea8c150 | 4156 | lruvec_init(&mz->lruvec); |
bb4cc1a8 AM |
4157 | mz->usage_in_excess = 0; |
4158 | mz->on_tree = false; | |
d79154bb | 4159 | mz->memcg = memcg; |
1ecaab2b | 4160 | } |
54f72fe0 | 4161 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4162 | return 0; |
4163 | } | |
4164 | ||
c0ff4b85 | 4165 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4166 | { |
54f72fe0 | 4167 | kfree(memcg->nodeinfo[node]); |
1ecaab2b KH |
4168 | } |
4169 | ||
33327948 KH |
4170 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4171 | { | |
d79154bb | 4172 | struct mem_cgroup *memcg; |
8ff69e2c | 4173 | size_t size; |
33327948 | 4174 | |
8ff69e2c VD |
4175 | size = sizeof(struct mem_cgroup); |
4176 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
33327948 | 4177 | |
8ff69e2c | 4178 | memcg = kzalloc(size, GFP_KERNEL); |
d79154bb | 4179 | if (!memcg) |
e7bbcdf3 DC |
4180 | return NULL; |
4181 | ||
d79154bb HD |
4182 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4183 | if (!memcg->stat) | |
d2e61b8d | 4184 | goto out_free; |
841710aa TH |
4185 | |
4186 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) | |
4187 | goto out_free_stat; | |
4188 | ||
d79154bb | 4189 | return memcg; |
d2e61b8d | 4190 | |
841710aa TH |
4191 | out_free_stat: |
4192 | free_percpu(memcg->stat); | |
d2e61b8d | 4193 | out_free: |
8ff69e2c | 4194 | kfree(memcg); |
d2e61b8d | 4195 | return NULL; |
33327948 KH |
4196 | } |
4197 | ||
59927fb9 | 4198 | /* |
c8b2a36f GC |
4199 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
4200 | * (scanning all at force_empty is too costly...) | |
4201 | * | |
4202 | * Instead of clearing all references at force_empty, we remember | |
4203 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4204 | * it goes down to 0. | |
4205 | * | |
4206 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
59927fb9 | 4207 | */ |
c8b2a36f GC |
4208 | |
4209 | static void __mem_cgroup_free(struct mem_cgroup *memcg) | |
59927fb9 | 4210 | { |
c8b2a36f | 4211 | int node; |
59927fb9 | 4212 | |
f7e1cb6e JW |
4213 | cancel_work_sync(&memcg->high_work); |
4214 | ||
bb4cc1a8 | 4215 | mem_cgroup_remove_from_trees(memcg); |
c8b2a36f GC |
4216 | |
4217 | for_each_node(node) | |
4218 | free_mem_cgroup_per_zone_info(memcg, node); | |
4219 | ||
4220 | free_percpu(memcg->stat); | |
841710aa | 4221 | memcg_wb_domain_exit(memcg); |
8ff69e2c | 4222 | kfree(memcg); |
59927fb9 | 4223 | } |
3afe36b1 | 4224 | |
0eb253e2 | 4225 | static struct cgroup_subsys_state * __ref |
eb95419b | 4226 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
8cdea7c0 | 4227 | { |
d142e3e6 | 4228 | struct mem_cgroup *memcg; |
04046e1a | 4229 | long error = -ENOMEM; |
6d12e2d8 | 4230 | int node; |
8cdea7c0 | 4231 | |
c0ff4b85 R |
4232 | memcg = mem_cgroup_alloc(); |
4233 | if (!memcg) | |
04046e1a | 4234 | return ERR_PTR(error); |
78fb7466 | 4235 | |
3ed28fa1 | 4236 | for_each_node(node) |
c0ff4b85 | 4237 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
6d12e2d8 | 4238 | goto free_out; |
f64c3f54 | 4239 | |
c077719b | 4240 | /* root ? */ |
eb95419b | 4241 | if (parent_css == NULL) { |
a41c58a6 | 4242 | root_mem_cgroup = memcg; |
3e32cb2e | 4243 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4244 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4245 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4246 | page_counter_init(&memcg->memsw, NULL); |
4247 | page_counter_init(&memcg->kmem, NULL); | |
18f59ea7 | 4248 | } |
28dbc4b6 | 4249 | |
f7e1cb6e | 4250 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4251 | memcg->last_scanned_node = MAX_NUMNODES; |
4252 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4253 | memcg->move_charge_at_immigrate = 0; |
4254 | mutex_init(&memcg->thresholds_lock); | |
4255 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4256 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4257 | INIT_LIST_HEAD(&memcg->event_list); |
4258 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4259 | memcg->socket_pressure = jiffies; |
127424c8 | 4260 | #ifndef CONFIG_SLOB |
900a38f0 | 4261 | memcg->kmemcg_id = -1; |
900a38f0 | 4262 | #endif |
52ebea74 TH |
4263 | #ifdef CONFIG_CGROUP_WRITEBACK |
4264 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4265 | #endif | |
d142e3e6 GC |
4266 | return &memcg->css; |
4267 | ||
4268 | free_out: | |
4269 | __mem_cgroup_free(memcg); | |
4270 | return ERR_PTR(error); | |
4271 | } | |
4272 | ||
4273 | static int | |
eb95419b | 4274 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
d142e3e6 | 4275 | { |
eb95419b | 4276 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 4277 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
2f7dd7a4 | 4278 | int ret; |
d142e3e6 | 4279 | |
15a4c835 | 4280 | if (css->id > MEM_CGROUP_ID_MAX) |
4219b2da LZ |
4281 | return -ENOSPC; |
4282 | ||
63876986 | 4283 | if (!parent) |
d142e3e6 GC |
4284 | return 0; |
4285 | ||
0999821b | 4286 | mutex_lock(&memcg_create_mutex); |
d142e3e6 GC |
4287 | |
4288 | memcg->use_hierarchy = parent->use_hierarchy; | |
4289 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4290 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4291 | ||
4292 | if (parent->use_hierarchy) { | |
3e32cb2e | 4293 | page_counter_init(&memcg->memory, &parent->memory); |
241994ed | 4294 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4295 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4296 | page_counter_init(&memcg->memsw, &parent->memsw); |
4297 | page_counter_init(&memcg->kmem, &parent->kmem); | |
d55f90bf VD |
4298 | page_counter_init(&memcg->tcp_mem.memory_allocated, |
4299 | &parent->tcp_mem.memory_allocated); | |
55007d84 | 4300 | |
7bcc1bb1 | 4301 | /* |
8d76a979 LZ |
4302 | * No need to take a reference to the parent because cgroup |
4303 | * core guarantees its existence. | |
7bcc1bb1 | 4304 | */ |
18f59ea7 | 4305 | } else { |
3e32cb2e | 4306 | page_counter_init(&memcg->memory, NULL); |
241994ed | 4307 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4308 | memcg->soft_limit = PAGE_COUNTER_MAX; |
3e32cb2e JW |
4309 | page_counter_init(&memcg->memsw, NULL); |
4310 | page_counter_init(&memcg->kmem, NULL); | |
d55f90bf | 4311 | page_counter_init(&memcg->tcp_mem.memory_allocated, NULL); |
8c7f6edb TH |
4312 | /* |
4313 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4314 | * much sense so let cgroup subsystem know about this | |
4315 | * unfortunate state in our controller. | |
4316 | */ | |
d142e3e6 | 4317 | if (parent != root_mem_cgroup) |
073219e9 | 4318 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4319 | } |
0999821b | 4320 | mutex_unlock(&memcg_create_mutex); |
d6441637 | 4321 | |
3893e302 | 4322 | ret = memcg_propagate_kmem(memcg); |
2f7dd7a4 JW |
4323 | if (ret) |
4324 | return ret; | |
127424c8 | 4325 | |
f7e1cb6e | 4326 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4327 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4328 | |
2f7dd7a4 JW |
4329 | /* |
4330 | * Make sure the memcg is initialized: mem_cgroup_iter() | |
4331 | * orders reading memcg->initialized against its callers | |
4332 | * reading the memcg members. | |
4333 | */ | |
4334 | smp_store_release(&memcg->initialized, 1); | |
4335 | ||
4336 | return 0; | |
8cdea7c0 BS |
4337 | } |
4338 | ||
eb95419b | 4339 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4340 | { |
eb95419b | 4341 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4342 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4343 | |
4344 | /* | |
4345 | * Unregister events and notify userspace. | |
4346 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4347 | * directory to avoid race between userspace and kernelspace. | |
4348 | */ | |
fba94807 TH |
4349 | spin_lock(&memcg->event_list_lock); |
4350 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4351 | list_del_init(&event->list); |
4352 | schedule_work(&event->remove); | |
4353 | } | |
fba94807 | 4354 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4355 | |
33cb876e | 4356 | vmpressure_cleanup(&memcg->vmpressure); |
2a4db7eb | 4357 | |
567e9ab2 | 4358 | memcg_offline_kmem(memcg); |
52ebea74 TH |
4359 | |
4360 | wb_memcg_offline(memcg); | |
df878fb0 KH |
4361 | } |
4362 | ||
6df38689 VD |
4363 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4364 | { | |
4365 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4366 | ||
4367 | invalidate_reclaim_iterators(memcg); | |
4368 | } | |
4369 | ||
eb95419b | 4370 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4371 | { |
eb95419b | 4372 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4373 | |
f7e1cb6e | 4374 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4375 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4376 | |
d55f90bf VD |
4377 | if (memcg->tcp_mem.active) |
4378 | static_branch_dec(&memcg_sockets_enabled_key); | |
3893e302 | 4379 | |
d886f4e4 | 4380 | memcg_free_kmem(memcg); |
465939a1 | 4381 | __mem_cgroup_free(memcg); |
8cdea7c0 BS |
4382 | } |
4383 | ||
1ced953b TH |
4384 | /** |
4385 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4386 | * @css: the target css | |
4387 | * | |
4388 | * Reset the states of the mem_cgroup associated with @css. This is | |
4389 | * invoked when the userland requests disabling on the default hierarchy | |
4390 | * but the memcg is pinned through dependency. The memcg should stop | |
4391 | * applying policies and should revert to the vanilla state as it may be | |
4392 | * made visible again. | |
4393 | * | |
4394 | * The current implementation only resets the essential configurations. | |
4395 | * This needs to be expanded to cover all the visible parts. | |
4396 | */ | |
4397 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4398 | { | |
4399 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4400 | ||
3e32cb2e JW |
4401 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
4402 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); | |
4403 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); | |
241994ed JW |
4404 | memcg->low = 0; |
4405 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4406 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4407 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4408 | } |
4409 | ||
02491447 | 4410 | #ifdef CONFIG_MMU |
7dc74be0 | 4411 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4412 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4413 | { |
05b84301 | 4414 | int ret; |
9476db97 | 4415 | |
d0164adc MG |
4416 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4417 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4418 | if (!ret) { |
854ffa8d | 4419 | mc.precharge += count; |
854ffa8d DN |
4420 | return ret; |
4421 | } | |
9476db97 JW |
4422 | |
4423 | /* Try charges one by one with reclaim */ | |
854ffa8d | 4424 | while (count--) { |
00501b53 | 4425 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
38c5d72f | 4426 | if (ret) |
38c5d72f | 4427 | return ret; |
854ffa8d | 4428 | mc.precharge++; |
9476db97 | 4429 | cond_resched(); |
854ffa8d | 4430 | } |
9476db97 | 4431 | return 0; |
4ffef5fe DN |
4432 | } |
4433 | ||
4434 | /** | |
8d32ff84 | 4435 | * get_mctgt_type - get target type of moving charge |
4ffef5fe DN |
4436 | * @vma: the vma the pte to be checked belongs |
4437 | * @addr: the address corresponding to the pte to be checked | |
4438 | * @ptent: the pte to be checked | |
02491447 | 4439 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4440 | * |
4441 | * Returns | |
4442 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4443 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4444 | * move charge. if @target is not NULL, the page is stored in target->page | |
4445 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4446 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4447 | * target for charge migration. if @target is not NULL, the entry is stored | |
4448 | * in target->ent. | |
4ffef5fe DN |
4449 | * |
4450 | * Called with pte lock held. | |
4451 | */ | |
4ffef5fe DN |
4452 | union mc_target { |
4453 | struct page *page; | |
02491447 | 4454 | swp_entry_t ent; |
4ffef5fe DN |
4455 | }; |
4456 | ||
4ffef5fe | 4457 | enum mc_target_type { |
8d32ff84 | 4458 | MC_TARGET_NONE = 0, |
4ffef5fe | 4459 | MC_TARGET_PAGE, |
02491447 | 4460 | MC_TARGET_SWAP, |
4ffef5fe DN |
4461 | }; |
4462 | ||
90254a65 DN |
4463 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4464 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4465 | { |
90254a65 | 4466 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4467 | |
90254a65 DN |
4468 | if (!page || !page_mapped(page)) |
4469 | return NULL; | |
4470 | if (PageAnon(page)) { | |
1dfab5ab | 4471 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4472 | return NULL; |
1dfab5ab JW |
4473 | } else { |
4474 | if (!(mc.flags & MOVE_FILE)) | |
4475 | return NULL; | |
4476 | } | |
90254a65 DN |
4477 | if (!get_page_unless_zero(page)) |
4478 | return NULL; | |
4479 | ||
4480 | return page; | |
4481 | } | |
4482 | ||
4b91355e | 4483 | #ifdef CONFIG_SWAP |
90254a65 DN |
4484 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
4485 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4486 | { | |
90254a65 DN |
4487 | struct page *page = NULL; |
4488 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4489 | ||
1dfab5ab | 4490 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4491 | return NULL; |
4b91355e KH |
4492 | /* |
4493 | * Because lookup_swap_cache() updates some statistics counter, | |
4494 | * we call find_get_page() with swapper_space directly. | |
4495 | */ | |
33806f06 | 4496 | page = find_get_page(swap_address_space(ent), ent.val); |
7941d214 | 4497 | if (do_memsw_account()) |
90254a65 DN |
4498 | entry->val = ent.val; |
4499 | ||
4500 | return page; | |
4501 | } | |
4b91355e KH |
4502 | #else |
4503 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4504 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4505 | { | |
4506 | return NULL; | |
4507 | } | |
4508 | #endif | |
90254a65 | 4509 | |
87946a72 DN |
4510 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4511 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4512 | { | |
4513 | struct page *page = NULL; | |
87946a72 DN |
4514 | struct address_space *mapping; |
4515 | pgoff_t pgoff; | |
4516 | ||
4517 | if (!vma->vm_file) /* anonymous vma */ | |
4518 | return NULL; | |
1dfab5ab | 4519 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4520 | return NULL; |
4521 | ||
87946a72 | 4522 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4523 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4524 | |
4525 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4526 | #ifdef CONFIG_SWAP |
4527 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4528 | if (shmem_mapping(mapping)) { |
4529 | page = find_get_entry(mapping, pgoff); | |
4530 | if (radix_tree_exceptional_entry(page)) { | |
4531 | swp_entry_t swp = radix_to_swp_entry(page); | |
7941d214 | 4532 | if (do_memsw_account()) |
139b6a6f JW |
4533 | *entry = swp; |
4534 | page = find_get_page(swap_address_space(swp), swp.val); | |
4535 | } | |
4536 | } else | |
4537 | page = find_get_page(mapping, pgoff); | |
4538 | #else | |
4539 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4540 | #endif |
87946a72 DN |
4541 | return page; |
4542 | } | |
4543 | ||
b1b0deab CG |
4544 | /** |
4545 | * mem_cgroup_move_account - move account of the page | |
4546 | * @page: the page | |
4547 | * @nr_pages: number of regular pages (>1 for huge pages) | |
4548 | * @from: mem_cgroup which the page is moved from. | |
4549 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4550 | * | |
3ac808fd | 4551 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
4552 | * |
4553 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4554 | * from old cgroup. | |
4555 | */ | |
4556 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 4557 | bool compound, |
b1b0deab CG |
4558 | struct mem_cgroup *from, |
4559 | struct mem_cgroup *to) | |
4560 | { | |
4561 | unsigned long flags; | |
f627c2f5 | 4562 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 4563 | int ret; |
c4843a75 | 4564 | bool anon; |
b1b0deab CG |
4565 | |
4566 | VM_BUG_ON(from == to); | |
4567 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 4568 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
4569 | |
4570 | /* | |
45637bab HD |
4571 | * Prevent mem_cgroup_replace_page() from looking at |
4572 | * page->mem_cgroup of its source page while we change it. | |
b1b0deab | 4573 | */ |
f627c2f5 | 4574 | ret = -EBUSY; |
b1b0deab CG |
4575 | if (!trylock_page(page)) |
4576 | goto out; | |
4577 | ||
4578 | ret = -EINVAL; | |
4579 | if (page->mem_cgroup != from) | |
4580 | goto out_unlock; | |
4581 | ||
c4843a75 GT |
4582 | anon = PageAnon(page); |
4583 | ||
b1b0deab CG |
4584 | spin_lock_irqsave(&from->move_lock, flags); |
4585 | ||
c4843a75 | 4586 | if (!anon && page_mapped(page)) { |
b1b0deab CG |
4587 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
4588 | nr_pages); | |
4589 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], | |
4590 | nr_pages); | |
4591 | } | |
4592 | ||
c4843a75 GT |
4593 | /* |
4594 | * move_lock grabbed above and caller set from->moving_account, so | |
4595 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. | |
4596 | * So mapping should be stable for dirty pages. | |
4597 | */ | |
4598 | if (!anon && PageDirty(page)) { | |
4599 | struct address_space *mapping = page_mapping(page); | |
4600 | ||
4601 | if (mapping_cap_account_dirty(mapping)) { | |
4602 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4603 | nr_pages); | |
4604 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], | |
4605 | nr_pages); | |
4606 | } | |
4607 | } | |
4608 | ||
b1b0deab CG |
4609 | if (PageWriteback(page)) { |
4610 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4611 | nr_pages); | |
4612 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], | |
4613 | nr_pages); | |
4614 | } | |
4615 | ||
4616 | /* | |
4617 | * It is safe to change page->mem_cgroup here because the page | |
4618 | * is referenced, charged, and isolated - we can't race with | |
4619 | * uncharging, charging, migration, or LRU putback. | |
4620 | */ | |
4621 | ||
4622 | /* caller should have done css_get */ | |
4623 | page->mem_cgroup = to; | |
4624 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4625 | ||
4626 | ret = 0; | |
4627 | ||
4628 | local_irq_disable(); | |
f627c2f5 | 4629 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 4630 | memcg_check_events(to, page); |
f627c2f5 | 4631 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
4632 | memcg_check_events(from, page); |
4633 | local_irq_enable(); | |
4634 | out_unlock: | |
4635 | unlock_page(page); | |
4636 | out: | |
4637 | return ret; | |
4638 | } | |
4639 | ||
8d32ff84 | 4640 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4641 | unsigned long addr, pte_t ptent, union mc_target *target) |
4642 | { | |
4643 | struct page *page = NULL; | |
8d32ff84 | 4644 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4645 | swp_entry_t ent = { .val = 0 }; |
4646 | ||
4647 | if (pte_present(ptent)) | |
4648 | page = mc_handle_present_pte(vma, addr, ptent); | |
4649 | else if (is_swap_pte(ptent)) | |
4650 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
0661a336 | 4651 | else if (pte_none(ptent)) |
87946a72 | 4652 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4653 | |
4654 | if (!page && !ent.val) | |
8d32ff84 | 4655 | return ret; |
02491447 | 4656 | if (page) { |
02491447 | 4657 | /* |
0a31bc97 | 4658 | * Do only loose check w/o serialization. |
1306a85a | 4659 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4660 | * not under LRU exclusion. |
02491447 | 4661 | */ |
1306a85a | 4662 | if (page->mem_cgroup == mc.from) { |
02491447 DN |
4663 | ret = MC_TARGET_PAGE; |
4664 | if (target) | |
4665 | target->page = page; | |
4666 | } | |
4667 | if (!ret || !target) | |
4668 | put_page(page); | |
4669 | } | |
90254a65 DN |
4670 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4671 | if (ent.val && !ret && | |
34c00c31 | 4672 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4673 | ret = MC_TARGET_SWAP; |
4674 | if (target) | |
4675 | target->ent = ent; | |
4ffef5fe | 4676 | } |
4ffef5fe DN |
4677 | return ret; |
4678 | } | |
4679 | ||
12724850 NH |
4680 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4681 | /* | |
4682 | * We don't consider swapping or file mapped pages because THP does not | |
4683 | * support them for now. | |
4684 | * Caller should make sure that pmd_trans_huge(pmd) is true. | |
4685 | */ | |
4686 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4687 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4688 | { | |
4689 | struct page *page = NULL; | |
12724850 NH |
4690 | enum mc_target_type ret = MC_TARGET_NONE; |
4691 | ||
4692 | page = pmd_page(pmd); | |
309381fe | 4693 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4694 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4695 | return ret; |
1306a85a | 4696 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4697 | ret = MC_TARGET_PAGE; |
4698 | if (target) { | |
4699 | get_page(page); | |
4700 | target->page = page; | |
4701 | } | |
4702 | } | |
4703 | return ret; | |
4704 | } | |
4705 | #else | |
4706 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4707 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4708 | { | |
4709 | return MC_TARGET_NONE; | |
4710 | } | |
4711 | #endif | |
4712 | ||
4ffef5fe DN |
4713 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4714 | unsigned long addr, unsigned long end, | |
4715 | struct mm_walk *walk) | |
4716 | { | |
26bcd64a | 4717 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4718 | pte_t *pte; |
4719 | spinlock_t *ptl; | |
4720 | ||
4b471e88 | 4721 | if (pmd_trans_huge_lock(pmd, vma, &ptl)) { |
12724850 NH |
4722 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4723 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4724 | spin_unlock(ptl); |
1a5a9906 | 4725 | return 0; |
12724850 | 4726 | } |
03319327 | 4727 | |
45f83cef AA |
4728 | if (pmd_trans_unstable(pmd)) |
4729 | return 0; | |
4ffef5fe DN |
4730 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4731 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4732 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4733 | mc.precharge++; /* increment precharge temporarily */ |
4734 | pte_unmap_unlock(pte - 1, ptl); | |
4735 | cond_resched(); | |
4736 | ||
7dc74be0 DN |
4737 | return 0; |
4738 | } | |
4739 | ||
4ffef5fe DN |
4740 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4741 | { | |
4742 | unsigned long precharge; | |
4ffef5fe | 4743 | |
26bcd64a NH |
4744 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4745 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4746 | .mm = mm, | |
4747 | }; | |
dfe076b0 | 4748 | down_read(&mm->mmap_sem); |
26bcd64a | 4749 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
dfe076b0 | 4750 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4751 | |
4752 | precharge = mc.precharge; | |
4753 | mc.precharge = 0; | |
4754 | ||
4755 | return precharge; | |
4756 | } | |
4757 | ||
4ffef5fe DN |
4758 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4759 | { | |
dfe076b0 DN |
4760 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4761 | ||
4762 | VM_BUG_ON(mc.moving_task); | |
4763 | mc.moving_task = current; | |
4764 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4765 | } |
4766 | ||
dfe076b0 DN |
4767 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4768 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4769 | { |
2bd9bb20 KH |
4770 | struct mem_cgroup *from = mc.from; |
4771 | struct mem_cgroup *to = mc.to; | |
4772 | ||
4ffef5fe | 4773 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4774 | if (mc.precharge) { |
00501b53 | 4775 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4776 | mc.precharge = 0; |
4777 | } | |
4778 | /* | |
4779 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4780 | * we must uncharge here. | |
4781 | */ | |
4782 | if (mc.moved_charge) { | |
00501b53 | 4783 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4784 | mc.moved_charge = 0; |
4ffef5fe | 4785 | } |
483c30b5 DN |
4786 | /* we must fixup refcnts and charges */ |
4787 | if (mc.moved_swap) { | |
483c30b5 | 4788 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4789 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4790 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4791 | |
05b84301 | 4792 | /* |
3e32cb2e JW |
4793 | * we charged both to->memory and to->memsw, so we |
4794 | * should uncharge to->memory. | |
05b84301 | 4795 | */ |
ce00a967 | 4796 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4797 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4798 | ||
e8ea14cc | 4799 | css_put_many(&mc.from->css, mc.moved_swap); |
3e32cb2e | 4800 | |
4050377b | 4801 | /* we've already done css_get(mc.to) */ |
483c30b5 DN |
4802 | mc.moved_swap = 0; |
4803 | } | |
dfe076b0 DN |
4804 | memcg_oom_recover(from); |
4805 | memcg_oom_recover(to); | |
4806 | wake_up_all(&mc.waitq); | |
4807 | } | |
4808 | ||
4809 | static void mem_cgroup_clear_mc(void) | |
4810 | { | |
dfe076b0 DN |
4811 | /* |
4812 | * we must clear moving_task before waking up waiters at the end of | |
4813 | * task migration. | |
4814 | */ | |
4815 | mc.moving_task = NULL; | |
4816 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4817 | spin_lock(&mc.lock); |
4ffef5fe DN |
4818 | mc.from = NULL; |
4819 | mc.to = NULL; | |
2bd9bb20 | 4820 | spin_unlock(&mc.lock); |
4ffef5fe DN |
4821 | } |
4822 | ||
1f7dd3e5 | 4823 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4824 | { |
1f7dd3e5 | 4825 | struct cgroup_subsys_state *css; |
eed67d75 | 4826 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 4827 | struct mem_cgroup *from; |
4530eddb | 4828 | struct task_struct *leader, *p; |
9f2115f9 | 4829 | struct mm_struct *mm; |
1dfab5ab | 4830 | unsigned long move_flags; |
9f2115f9 | 4831 | int ret = 0; |
7dc74be0 | 4832 | |
1f7dd3e5 TH |
4833 | /* charge immigration isn't supported on the default hierarchy */ |
4834 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
4835 | return 0; |
4836 | ||
4530eddb TH |
4837 | /* |
4838 | * Multi-process migrations only happen on the default hierarchy | |
4839 | * where charge immigration is not used. Perform charge | |
4840 | * immigration if @tset contains a leader and whine if there are | |
4841 | * multiple. | |
4842 | */ | |
4843 | p = NULL; | |
1f7dd3e5 | 4844 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
4845 | WARN_ON_ONCE(p); |
4846 | p = leader; | |
1f7dd3e5 | 4847 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
4848 | } |
4849 | if (!p) | |
4850 | return 0; | |
4851 | ||
1f7dd3e5 TH |
4852 | /* |
4853 | * We are now commited to this value whatever it is. Changes in this | |
4854 | * tunable will only affect upcoming migrations, not the current one. | |
4855 | * So we need to save it, and keep it going. | |
4856 | */ | |
4857 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
4858 | if (!move_flags) | |
4859 | return 0; | |
4860 | ||
9f2115f9 TH |
4861 | from = mem_cgroup_from_task(p); |
4862 | ||
4863 | VM_BUG_ON(from == memcg); | |
4864 | ||
4865 | mm = get_task_mm(p); | |
4866 | if (!mm) | |
4867 | return 0; | |
4868 | /* We move charges only when we move a owner of the mm */ | |
4869 | if (mm->owner == p) { | |
4870 | VM_BUG_ON(mc.from); | |
4871 | VM_BUG_ON(mc.to); | |
4872 | VM_BUG_ON(mc.precharge); | |
4873 | VM_BUG_ON(mc.moved_charge); | |
4874 | VM_BUG_ON(mc.moved_swap); | |
4875 | ||
4876 | spin_lock(&mc.lock); | |
4877 | mc.from = from; | |
4878 | mc.to = memcg; | |
4879 | mc.flags = move_flags; | |
4880 | spin_unlock(&mc.lock); | |
4881 | /* We set mc.moving_task later */ | |
4882 | ||
4883 | ret = mem_cgroup_precharge_mc(mm); | |
4884 | if (ret) | |
4885 | mem_cgroup_clear_mc(); | |
7dc74be0 | 4886 | } |
9f2115f9 | 4887 | mmput(mm); |
7dc74be0 DN |
4888 | return ret; |
4889 | } | |
4890 | ||
1f7dd3e5 | 4891 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4892 | { |
4e2f245d JW |
4893 | if (mc.to) |
4894 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
4895 | } |
4896 | ||
4ffef5fe DN |
4897 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4898 | unsigned long addr, unsigned long end, | |
4899 | struct mm_walk *walk) | |
7dc74be0 | 4900 | { |
4ffef5fe | 4901 | int ret = 0; |
26bcd64a | 4902 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4903 | pte_t *pte; |
4904 | spinlock_t *ptl; | |
12724850 NH |
4905 | enum mc_target_type target_type; |
4906 | union mc_target target; | |
4907 | struct page *page; | |
4ffef5fe | 4908 | |
4b471e88 | 4909 | if (pmd_trans_huge_lock(pmd, vma, &ptl)) { |
62ade86a | 4910 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 4911 | spin_unlock(ptl); |
12724850 NH |
4912 | return 0; |
4913 | } | |
4914 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
4915 | if (target_type == MC_TARGET_PAGE) { | |
4916 | page = target.page; | |
4917 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 4918 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 4919 | mc.from, mc.to)) { |
12724850 NH |
4920 | mc.precharge -= HPAGE_PMD_NR; |
4921 | mc.moved_charge += HPAGE_PMD_NR; | |
4922 | } | |
4923 | putback_lru_page(page); | |
4924 | } | |
4925 | put_page(page); | |
4926 | } | |
bf929152 | 4927 | spin_unlock(ptl); |
1a5a9906 | 4928 | return 0; |
12724850 NH |
4929 | } |
4930 | ||
45f83cef AA |
4931 | if (pmd_trans_unstable(pmd)) |
4932 | return 0; | |
4ffef5fe DN |
4933 | retry: |
4934 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4935 | for (; addr != end; addr += PAGE_SIZE) { | |
4936 | pte_t ptent = *(pte++); | |
02491447 | 4937 | swp_entry_t ent; |
4ffef5fe DN |
4938 | |
4939 | if (!mc.precharge) | |
4940 | break; | |
4941 | ||
8d32ff84 | 4942 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
4ffef5fe DN |
4943 | case MC_TARGET_PAGE: |
4944 | page = target.page; | |
53f9263b KS |
4945 | /* |
4946 | * We can have a part of the split pmd here. Moving it | |
4947 | * can be done but it would be too convoluted so simply | |
4948 | * ignore such a partial THP and keep it in original | |
4949 | * memcg. There should be somebody mapping the head. | |
4950 | */ | |
4951 | if (PageTransCompound(page)) | |
4952 | goto put; | |
4ffef5fe DN |
4953 | if (isolate_lru_page(page)) |
4954 | goto put; | |
f627c2f5 KS |
4955 | if (!mem_cgroup_move_account(page, false, |
4956 | mc.from, mc.to)) { | |
4ffef5fe | 4957 | mc.precharge--; |
854ffa8d DN |
4958 | /* we uncharge from mc.from later. */ |
4959 | mc.moved_charge++; | |
4ffef5fe DN |
4960 | } |
4961 | putback_lru_page(page); | |
8d32ff84 | 4962 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
4963 | put_page(page); |
4964 | break; | |
02491447 DN |
4965 | case MC_TARGET_SWAP: |
4966 | ent = target.ent; | |
e91cbb42 | 4967 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 4968 | mc.precharge--; |
483c30b5 DN |
4969 | /* we fixup refcnts and charges later. */ |
4970 | mc.moved_swap++; | |
4971 | } | |
02491447 | 4972 | break; |
4ffef5fe DN |
4973 | default: |
4974 | break; | |
4975 | } | |
4976 | } | |
4977 | pte_unmap_unlock(pte - 1, ptl); | |
4978 | cond_resched(); | |
4979 | ||
4980 | if (addr != end) { | |
4981 | /* | |
4982 | * We have consumed all precharges we got in can_attach(). | |
4983 | * We try charge one by one, but don't do any additional | |
4984 | * charges to mc.to if we have failed in charge once in attach() | |
4985 | * phase. | |
4986 | */ | |
854ffa8d | 4987 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4988 | if (!ret) |
4989 | goto retry; | |
4990 | } | |
4991 | ||
4992 | return ret; | |
4993 | } | |
4994 | ||
4995 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
4996 | { | |
26bcd64a NH |
4997 | struct mm_walk mem_cgroup_move_charge_walk = { |
4998 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
4999 | .mm = mm, | |
5000 | }; | |
4ffef5fe DN |
5001 | |
5002 | lru_add_drain_all(); | |
312722cb JW |
5003 | /* |
5004 | * Signal mem_cgroup_begin_page_stat() to take the memcg's | |
5005 | * move_lock while we're moving its pages to another memcg. | |
5006 | * Then wait for already started RCU-only updates to finish. | |
5007 | */ | |
5008 | atomic_inc(&mc.from->moving_account); | |
5009 | synchronize_rcu(); | |
dfe076b0 DN |
5010 | retry: |
5011 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5012 | /* | |
5013 | * Someone who are holding the mmap_sem might be waiting in | |
5014 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5015 | * and retry. Because we cancel precharges, we might not be able | |
5016 | * to move enough charges, but moving charge is a best-effort | |
5017 | * feature anyway, so it wouldn't be a big problem. | |
5018 | */ | |
5019 | __mem_cgroup_clear_mc(); | |
5020 | cond_resched(); | |
5021 | goto retry; | |
5022 | } | |
26bcd64a NH |
5023 | /* |
5024 | * When we have consumed all precharges and failed in doing | |
5025 | * additional charge, the page walk just aborts. | |
5026 | */ | |
5027 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); | |
dfe076b0 | 5028 | up_read(&mm->mmap_sem); |
312722cb | 5029 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5030 | } |
5031 | ||
1f7dd3e5 | 5032 | static void mem_cgroup_move_task(struct cgroup_taskset *tset) |
67e465a7 | 5033 | { |
1f7dd3e5 TH |
5034 | struct cgroup_subsys_state *css; |
5035 | struct task_struct *p = cgroup_taskset_first(tset, &css); | |
a433658c | 5036 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5037 | |
dfe076b0 | 5038 | if (mm) { |
a433658c KM |
5039 | if (mc.to) |
5040 | mem_cgroup_move_charge(mm); | |
dfe076b0 DN |
5041 | mmput(mm); |
5042 | } | |
a433658c KM |
5043 | if (mc.to) |
5044 | mem_cgroup_clear_mc(); | |
67e465a7 | 5045 | } |
5cfb80a7 | 5046 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5047 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5048 | { |
5049 | return 0; | |
5050 | } | |
1f7dd3e5 | 5051 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5052 | { |
5053 | } | |
1f7dd3e5 | 5054 | static void mem_cgroup_move_task(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5055 | { |
5056 | } | |
5057 | #endif | |
67e465a7 | 5058 | |
f00baae7 TH |
5059 | /* |
5060 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5061 | * to verify whether we're attached to the default hierarchy on each mount |
5062 | * attempt. | |
f00baae7 | 5063 | */ |
eb95419b | 5064 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5065 | { |
5066 | /* | |
aa6ec29b | 5067 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5068 | * guarantees that @root doesn't have any children, so turning it |
5069 | * on for the root memcg is enough. | |
5070 | */ | |
9e10a130 | 5071 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5072 | root_mem_cgroup->use_hierarchy = true; |
5073 | else | |
5074 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5075 | } |
5076 | ||
241994ed JW |
5077 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5078 | struct cftype *cft) | |
5079 | { | |
f5fc3c5d JW |
5080 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5081 | ||
5082 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5083 | } |
5084 | ||
5085 | static int memory_low_show(struct seq_file *m, void *v) | |
5086 | { | |
5087 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5088 | unsigned long low = READ_ONCE(memcg->low); |
241994ed JW |
5089 | |
5090 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 5091 | seq_puts(m, "max\n"); |
241994ed JW |
5092 | else |
5093 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5094 | ||
5095 | return 0; | |
5096 | } | |
5097 | ||
5098 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5099 | char *buf, size_t nbytes, loff_t off) | |
5100 | { | |
5101 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5102 | unsigned long low; | |
5103 | int err; | |
5104 | ||
5105 | buf = strstrip(buf); | |
d2973697 | 5106 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5107 | if (err) |
5108 | return err; | |
5109 | ||
5110 | memcg->low = low; | |
5111 | ||
5112 | return nbytes; | |
5113 | } | |
5114 | ||
5115 | static int memory_high_show(struct seq_file *m, void *v) | |
5116 | { | |
5117 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5118 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
5119 | |
5120 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5121 | seq_puts(m, "max\n"); |
241994ed JW |
5122 | else |
5123 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5124 | ||
5125 | return 0; | |
5126 | } | |
5127 | ||
5128 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5129 | char *buf, size_t nbytes, loff_t off) | |
5130 | { | |
5131 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5132 | unsigned long high; | |
5133 | int err; | |
5134 | ||
5135 | buf = strstrip(buf); | |
d2973697 | 5136 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5137 | if (err) |
5138 | return err; | |
5139 | ||
5140 | memcg->high = high; | |
5141 | ||
2529bb3a | 5142 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5143 | return nbytes; |
5144 | } | |
5145 | ||
5146 | static int memory_max_show(struct seq_file *m, void *v) | |
5147 | { | |
5148 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5149 | unsigned long max = READ_ONCE(memcg->memory.limit); |
241994ed JW |
5150 | |
5151 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5152 | seq_puts(m, "max\n"); |
241994ed JW |
5153 | else |
5154 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5155 | ||
5156 | return 0; | |
5157 | } | |
5158 | ||
5159 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5160 | char *buf, size_t nbytes, loff_t off) | |
5161 | { | |
5162 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5163 | unsigned long max; | |
5164 | int err; | |
5165 | ||
5166 | buf = strstrip(buf); | |
d2973697 | 5167 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5168 | if (err) |
5169 | return err; | |
5170 | ||
5171 | err = mem_cgroup_resize_limit(memcg, max); | |
5172 | if (err) | |
5173 | return err; | |
5174 | ||
2529bb3a | 5175 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5176 | return nbytes; |
5177 | } | |
5178 | ||
5179 | static int memory_events_show(struct seq_file *m, void *v) | |
5180 | { | |
5181 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5182 | ||
5183 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); | |
5184 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); | |
5185 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); | |
5186 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); | |
5187 | ||
5188 | return 0; | |
5189 | } | |
5190 | ||
5191 | static struct cftype memory_files[] = { | |
5192 | { | |
5193 | .name = "current", | |
f5fc3c5d | 5194 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5195 | .read_u64 = memory_current_read, |
5196 | }, | |
5197 | { | |
5198 | .name = "low", | |
5199 | .flags = CFTYPE_NOT_ON_ROOT, | |
5200 | .seq_show = memory_low_show, | |
5201 | .write = memory_low_write, | |
5202 | }, | |
5203 | { | |
5204 | .name = "high", | |
5205 | .flags = CFTYPE_NOT_ON_ROOT, | |
5206 | .seq_show = memory_high_show, | |
5207 | .write = memory_high_write, | |
5208 | }, | |
5209 | { | |
5210 | .name = "max", | |
5211 | .flags = CFTYPE_NOT_ON_ROOT, | |
5212 | .seq_show = memory_max_show, | |
5213 | .write = memory_max_write, | |
5214 | }, | |
5215 | { | |
5216 | .name = "events", | |
5217 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5218 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5219 | .seq_show = memory_events_show, |
5220 | }, | |
5221 | { } /* terminate */ | |
5222 | }; | |
5223 | ||
073219e9 | 5224 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5225 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5226 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5227 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5228 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5229 | .css_free = mem_cgroup_css_free, |
1ced953b | 5230 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5231 | .can_attach = mem_cgroup_can_attach, |
5232 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5233 | .attach = mem_cgroup_move_task, |
f00baae7 | 5234 | .bind = mem_cgroup_bind, |
241994ed JW |
5235 | .dfl_cftypes = memory_files, |
5236 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5237 | .early_init = 0, |
8cdea7c0 | 5238 | }; |
c077719b | 5239 | |
241994ed JW |
5240 | /** |
5241 | * mem_cgroup_low - check if memory consumption is below the normal range | |
5242 | * @root: the highest ancestor to consider | |
5243 | * @memcg: the memory cgroup to check | |
5244 | * | |
5245 | * Returns %true if memory consumption of @memcg, and that of all | |
5246 | * configurable ancestors up to @root, is below the normal range. | |
5247 | */ | |
5248 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5249 | { | |
5250 | if (mem_cgroup_disabled()) | |
5251 | return false; | |
5252 | ||
5253 | /* | |
5254 | * The toplevel group doesn't have a configurable range, so | |
5255 | * it's never low when looked at directly, and it is not | |
5256 | * considered an ancestor when assessing the hierarchy. | |
5257 | */ | |
5258 | ||
5259 | if (memcg == root_mem_cgroup) | |
5260 | return false; | |
5261 | ||
4e54dede | 5262 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5263 | return false; |
5264 | ||
5265 | while (memcg != root) { | |
5266 | memcg = parent_mem_cgroup(memcg); | |
5267 | ||
5268 | if (memcg == root_mem_cgroup) | |
5269 | break; | |
5270 | ||
4e54dede | 5271 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5272 | return false; |
5273 | } | |
5274 | return true; | |
5275 | } | |
5276 | ||
00501b53 JW |
5277 | /** |
5278 | * mem_cgroup_try_charge - try charging a page | |
5279 | * @page: page to charge | |
5280 | * @mm: mm context of the victim | |
5281 | * @gfp_mask: reclaim mode | |
5282 | * @memcgp: charged memcg return | |
5283 | * | |
5284 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5285 | * pages according to @gfp_mask if necessary. | |
5286 | * | |
5287 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5288 | * Otherwise, an error code is returned. | |
5289 | * | |
5290 | * After page->mapping has been set up, the caller must finalize the | |
5291 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5292 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5293 | */ | |
5294 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
5295 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
5296 | bool compound) | |
00501b53 JW |
5297 | { |
5298 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 5299 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5300 | int ret = 0; |
5301 | ||
5302 | if (mem_cgroup_disabled()) | |
5303 | goto out; | |
5304 | ||
5305 | if (PageSwapCache(page)) { | |
00501b53 JW |
5306 | /* |
5307 | * Every swap fault against a single page tries to charge the | |
5308 | * page, bail as early as possible. shmem_unuse() encounters | |
5309 | * already charged pages, too. The USED bit is protected by | |
5310 | * the page lock, which serializes swap cache removal, which | |
5311 | * in turn serializes uncharging. | |
5312 | */ | |
e993d905 | 5313 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
1306a85a | 5314 | if (page->mem_cgroup) |
00501b53 | 5315 | goto out; |
e993d905 | 5316 | |
7941d214 | 5317 | if (do_memsw_account()) { |
e993d905 VD |
5318 | swp_entry_t ent = { .val = page_private(page), }; |
5319 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5320 | ||
5321 | rcu_read_lock(); | |
5322 | memcg = mem_cgroup_from_id(id); | |
5323 | if (memcg && !css_tryget_online(&memcg->css)) | |
5324 | memcg = NULL; | |
5325 | rcu_read_unlock(); | |
5326 | } | |
00501b53 JW |
5327 | } |
5328 | ||
00501b53 JW |
5329 | if (!memcg) |
5330 | memcg = get_mem_cgroup_from_mm(mm); | |
5331 | ||
5332 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5333 | ||
5334 | css_put(&memcg->css); | |
00501b53 JW |
5335 | out: |
5336 | *memcgp = memcg; | |
5337 | return ret; | |
5338 | } | |
5339 | ||
5340 | /** | |
5341 | * mem_cgroup_commit_charge - commit a page charge | |
5342 | * @page: page to charge | |
5343 | * @memcg: memcg to charge the page to | |
5344 | * @lrucare: page might be on LRU already | |
5345 | * | |
5346 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5347 | * after page->mapping has been set up. This must happen atomically | |
5348 | * as part of the page instantiation, i.e. under the page table lock | |
5349 | * for anonymous pages, under the page lock for page and swap cache. | |
5350 | * | |
5351 | * In addition, the page must not be on the LRU during the commit, to | |
5352 | * prevent racing with task migration. If it might be, use @lrucare. | |
5353 | * | |
5354 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5355 | */ | |
5356 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 5357 | bool lrucare, bool compound) |
00501b53 | 5358 | { |
f627c2f5 | 5359 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5360 | |
5361 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5362 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5363 | ||
5364 | if (mem_cgroup_disabled()) | |
5365 | return; | |
5366 | /* | |
5367 | * Swap faults will attempt to charge the same page multiple | |
5368 | * times. But reuse_swap_page() might have removed the page | |
5369 | * from swapcache already, so we can't check PageSwapCache(). | |
5370 | */ | |
5371 | if (!memcg) | |
5372 | return; | |
5373 | ||
6abb5a86 JW |
5374 | commit_charge(page, memcg, lrucare); |
5375 | ||
6abb5a86 | 5376 | local_irq_disable(); |
f627c2f5 | 5377 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
5378 | memcg_check_events(memcg, page); |
5379 | local_irq_enable(); | |
00501b53 | 5380 | |
7941d214 | 5381 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
5382 | swp_entry_t entry = { .val = page_private(page) }; |
5383 | /* | |
5384 | * The swap entry might not get freed for a long time, | |
5385 | * let's not wait for it. The page already received a | |
5386 | * memory+swap charge, drop the swap entry duplicate. | |
5387 | */ | |
5388 | mem_cgroup_uncharge_swap(entry); | |
5389 | } | |
5390 | } | |
5391 | ||
5392 | /** | |
5393 | * mem_cgroup_cancel_charge - cancel a page charge | |
5394 | * @page: page to charge | |
5395 | * @memcg: memcg to charge the page to | |
5396 | * | |
5397 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5398 | */ | |
f627c2f5 KS |
5399 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
5400 | bool compound) | |
00501b53 | 5401 | { |
f627c2f5 | 5402 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5403 | |
5404 | if (mem_cgroup_disabled()) | |
5405 | return; | |
5406 | /* | |
5407 | * Swap faults will attempt to charge the same page multiple | |
5408 | * times. But reuse_swap_page() might have removed the page | |
5409 | * from swapcache already, so we can't check PageSwapCache(). | |
5410 | */ | |
5411 | if (!memcg) | |
5412 | return; | |
5413 | ||
00501b53 JW |
5414 | cancel_charge(memcg, nr_pages); |
5415 | } | |
5416 | ||
747db954 | 5417 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
747db954 JW |
5418 | unsigned long nr_anon, unsigned long nr_file, |
5419 | unsigned long nr_huge, struct page *dummy_page) | |
5420 | { | |
18eca2e6 | 5421 | unsigned long nr_pages = nr_anon + nr_file; |
747db954 JW |
5422 | unsigned long flags; |
5423 | ||
ce00a967 | 5424 | if (!mem_cgroup_is_root(memcg)) { |
18eca2e6 | 5425 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 5426 | if (do_memsw_account()) |
18eca2e6 | 5427 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 JW |
5428 | memcg_oom_recover(memcg); |
5429 | } | |
747db954 JW |
5430 | |
5431 | local_irq_save(flags); | |
5432 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); | |
5433 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); | |
5434 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); | |
5435 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); | |
18eca2e6 | 5436 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
747db954 JW |
5437 | memcg_check_events(memcg, dummy_page); |
5438 | local_irq_restore(flags); | |
e8ea14cc JW |
5439 | |
5440 | if (!mem_cgroup_is_root(memcg)) | |
18eca2e6 | 5441 | css_put_many(&memcg->css, nr_pages); |
747db954 JW |
5442 | } |
5443 | ||
5444 | static void uncharge_list(struct list_head *page_list) | |
5445 | { | |
5446 | struct mem_cgroup *memcg = NULL; | |
747db954 JW |
5447 | unsigned long nr_anon = 0; |
5448 | unsigned long nr_file = 0; | |
5449 | unsigned long nr_huge = 0; | |
5450 | unsigned long pgpgout = 0; | |
747db954 JW |
5451 | struct list_head *next; |
5452 | struct page *page; | |
5453 | ||
5454 | next = page_list->next; | |
5455 | do { | |
5456 | unsigned int nr_pages = 1; | |
747db954 JW |
5457 | |
5458 | page = list_entry(next, struct page, lru); | |
5459 | next = page->lru.next; | |
5460 | ||
5461 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5462 | VM_BUG_ON_PAGE(page_count(page), page); | |
5463 | ||
1306a85a | 5464 | if (!page->mem_cgroup) |
747db954 JW |
5465 | continue; |
5466 | ||
5467 | /* | |
5468 | * Nobody should be changing or seriously looking at | |
1306a85a | 5469 | * page->mem_cgroup at this point, we have fully |
29833315 | 5470 | * exclusive access to the page. |
747db954 JW |
5471 | */ |
5472 | ||
1306a85a | 5473 | if (memcg != page->mem_cgroup) { |
747db954 | 5474 | if (memcg) { |
18eca2e6 JW |
5475 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5476 | nr_huge, page); | |
5477 | pgpgout = nr_anon = nr_file = nr_huge = 0; | |
747db954 | 5478 | } |
1306a85a | 5479 | memcg = page->mem_cgroup; |
747db954 JW |
5480 | } |
5481 | ||
5482 | if (PageTransHuge(page)) { | |
5483 | nr_pages <<= compound_order(page); | |
5484 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
5485 | nr_huge += nr_pages; | |
5486 | } | |
5487 | ||
5488 | if (PageAnon(page)) | |
5489 | nr_anon += nr_pages; | |
5490 | else | |
5491 | nr_file += nr_pages; | |
5492 | ||
1306a85a | 5493 | page->mem_cgroup = NULL; |
747db954 JW |
5494 | |
5495 | pgpgout++; | |
5496 | } while (next != page_list); | |
5497 | ||
5498 | if (memcg) | |
18eca2e6 JW |
5499 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
5500 | nr_huge, page); | |
747db954 JW |
5501 | } |
5502 | ||
0a31bc97 JW |
5503 | /** |
5504 | * mem_cgroup_uncharge - uncharge a page | |
5505 | * @page: page to uncharge | |
5506 | * | |
5507 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5508 | * mem_cgroup_commit_charge(). | |
5509 | */ | |
5510 | void mem_cgroup_uncharge(struct page *page) | |
5511 | { | |
0a31bc97 JW |
5512 | if (mem_cgroup_disabled()) |
5513 | return; | |
5514 | ||
747db954 | 5515 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5516 | if (!page->mem_cgroup) |
0a31bc97 JW |
5517 | return; |
5518 | ||
747db954 JW |
5519 | INIT_LIST_HEAD(&page->lru); |
5520 | uncharge_list(&page->lru); | |
5521 | } | |
0a31bc97 | 5522 | |
747db954 JW |
5523 | /** |
5524 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5525 | * @page_list: list of pages to uncharge | |
5526 | * | |
5527 | * Uncharge a list of pages previously charged with | |
5528 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5529 | */ | |
5530 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5531 | { | |
5532 | if (mem_cgroup_disabled()) | |
5533 | return; | |
0a31bc97 | 5534 | |
747db954 JW |
5535 | if (!list_empty(page_list)) |
5536 | uncharge_list(page_list); | |
0a31bc97 JW |
5537 | } |
5538 | ||
5539 | /** | |
45637bab | 5540 | * mem_cgroup_replace_page - migrate a charge to another page |
0a31bc97 JW |
5541 | * @oldpage: currently charged page |
5542 | * @newpage: page to transfer the charge to | |
0a31bc97 JW |
5543 | * |
5544 | * Migrate the charge from @oldpage to @newpage. | |
5545 | * | |
5546 | * Both pages must be locked, @newpage->mapping must be set up. | |
25be6a65 | 5547 | * Either or both pages might be on the LRU already. |
0a31bc97 | 5548 | */ |
45637bab | 5549 | void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage) |
0a31bc97 | 5550 | { |
29833315 | 5551 | struct mem_cgroup *memcg; |
0a31bc97 JW |
5552 | int isolated; |
5553 | ||
5554 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5555 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 5556 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
5557 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5558 | newpage); | |
0a31bc97 JW |
5559 | |
5560 | if (mem_cgroup_disabled()) | |
5561 | return; | |
5562 | ||
5563 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5564 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5565 | return; |
5566 | ||
45637bab | 5567 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 5568 | memcg = oldpage->mem_cgroup; |
29833315 | 5569 | if (!memcg) |
0a31bc97 JW |
5570 | return; |
5571 | ||
45637bab | 5572 | lock_page_lru(oldpage, &isolated); |
1306a85a | 5573 | oldpage->mem_cgroup = NULL; |
45637bab | 5574 | unlock_page_lru(oldpage, isolated); |
0a31bc97 | 5575 | |
45637bab | 5576 | commit_charge(newpage, memcg, true); |
0a31bc97 JW |
5577 | } |
5578 | ||
ef12947c | 5579 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
5580 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
5581 | ||
5582 | void sock_update_memcg(struct sock *sk) | |
5583 | { | |
5584 | struct mem_cgroup *memcg; | |
5585 | ||
5586 | /* Socket cloning can throw us here with sk_cgrp already | |
5587 | * filled. It won't however, necessarily happen from | |
5588 | * process context. So the test for root memcg given | |
5589 | * the current task's memcg won't help us in this case. | |
5590 | * | |
5591 | * Respecting the original socket's memcg is a better | |
5592 | * decision in this case. | |
5593 | */ | |
5594 | if (sk->sk_memcg) { | |
5595 | BUG_ON(mem_cgroup_is_root(sk->sk_memcg)); | |
5596 | css_get(&sk->sk_memcg->css); | |
5597 | return; | |
5598 | } | |
5599 | ||
5600 | rcu_read_lock(); | |
5601 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
5602 | if (memcg == root_mem_cgroup) |
5603 | goto out; | |
f7e1cb6e JW |
5604 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcp_mem.active) |
5605 | goto out; | |
f7e1cb6e | 5606 | if (css_tryget_online(&memcg->css)) |
11092087 | 5607 | sk->sk_memcg = memcg; |
f7e1cb6e | 5608 | out: |
11092087 JW |
5609 | rcu_read_unlock(); |
5610 | } | |
5611 | EXPORT_SYMBOL(sock_update_memcg); | |
5612 | ||
5613 | void sock_release_memcg(struct sock *sk) | |
5614 | { | |
5615 | WARN_ON(!sk->sk_memcg); | |
5616 | css_put(&sk->sk_memcg->css); | |
5617 | } | |
5618 | ||
5619 | /** | |
5620 | * mem_cgroup_charge_skmem - charge socket memory | |
5621 | * @memcg: memcg to charge | |
5622 | * @nr_pages: number of pages to charge | |
5623 | * | |
5624 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
5625 | * @memcg's configured limit, %false if the charge had to be forced. | |
5626 | */ | |
5627 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5628 | { | |
f7e1cb6e | 5629 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 5630 | |
f7e1cb6e JW |
5631 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
5632 | struct page_counter *counter; | |
5633 | ||
5634 | if (page_counter_try_charge(&memcg->tcp_mem.memory_allocated, | |
5635 | nr_pages, &counter)) { | |
5636 | memcg->tcp_mem.memory_pressure = 0; | |
5637 | return true; | |
5638 | } | |
5639 | page_counter_charge(&memcg->tcp_mem.memory_allocated, nr_pages); | |
5640 | memcg->tcp_mem.memory_pressure = 1; | |
5641 | return false; | |
11092087 | 5642 | } |
d886f4e4 | 5643 | |
f7e1cb6e JW |
5644 | /* Don't block in the packet receive path */ |
5645 | if (in_softirq()) | |
5646 | gfp_mask = GFP_NOWAIT; | |
5647 | ||
5648 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) | |
5649 | return true; | |
5650 | ||
5651 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
5652 | return false; |
5653 | } | |
5654 | ||
5655 | /** | |
5656 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
5657 | * @memcg - memcg to uncharge | |
5658 | * @nr_pages - number of pages to uncharge | |
5659 | */ | |
5660 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5661 | { | |
f7e1cb6e JW |
5662 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
5663 | page_counter_uncharge(&memcg->tcp_mem.memory_allocated, | |
5664 | nr_pages); | |
5665 | return; | |
5666 | } | |
d886f4e4 | 5667 | |
f7e1cb6e JW |
5668 | page_counter_uncharge(&memcg->memory, nr_pages); |
5669 | css_put_many(&memcg->css, nr_pages); | |
11092087 JW |
5670 | } |
5671 | ||
f7e1cb6e JW |
5672 | static int __init cgroup_memory(char *s) |
5673 | { | |
5674 | char *token; | |
5675 | ||
5676 | while ((token = strsep(&s, ",")) != NULL) { | |
5677 | if (!*token) | |
5678 | continue; | |
5679 | if (!strcmp(token, "nosocket")) | |
5680 | cgroup_memory_nosocket = true; | |
04823c83 VD |
5681 | if (!strcmp(token, "nokmem")) |
5682 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
5683 | } |
5684 | return 0; | |
5685 | } | |
5686 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 5687 | |
2d11085e | 5688 | /* |
1081312f MH |
5689 | * subsys_initcall() for memory controller. |
5690 | * | |
5691 | * Some parts like hotcpu_notifier() have to be initialized from this context | |
5692 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically | |
5693 | * everything that doesn't depend on a specific mem_cgroup structure should | |
5694 | * be initialized from here. | |
2d11085e MH |
5695 | */ |
5696 | static int __init mem_cgroup_init(void) | |
5697 | { | |
95a045f6 JW |
5698 | int cpu, node; |
5699 | ||
2d11085e | 5700 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
95a045f6 JW |
5701 | |
5702 | for_each_possible_cpu(cpu) | |
5703 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5704 | drain_local_stock); | |
5705 | ||
5706 | for_each_node(node) { | |
5707 | struct mem_cgroup_tree_per_node *rtpn; | |
5708 | int zone; | |
5709 | ||
5710 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5711 | node_online(node) ? node : NUMA_NO_NODE); | |
5712 | ||
5713 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
5714 | struct mem_cgroup_tree_per_zone *rtpz; | |
5715 | ||
5716 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
5717 | rtpz->rb_root = RB_ROOT; | |
5718 | spin_lock_init(&rtpz->lock); | |
5719 | } | |
5720 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
5721 | } | |
5722 | ||
2d11085e MH |
5723 | return 0; |
5724 | } | |
5725 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5726 | |
5727 | #ifdef CONFIG_MEMCG_SWAP | |
5728 | /** | |
5729 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5730 | * @page: page whose memsw charge to transfer | |
5731 | * @entry: swap entry to move the charge to | |
5732 | * | |
5733 | * Transfer the memsw charge of @page to @entry. | |
5734 | */ | |
5735 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5736 | { | |
5737 | struct mem_cgroup *memcg; | |
5738 | unsigned short oldid; | |
5739 | ||
5740 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5741 | VM_BUG_ON_PAGE(page_count(page), page); | |
5742 | ||
7941d214 | 5743 | if (!do_memsw_account()) |
21afa38e JW |
5744 | return; |
5745 | ||
5746 | memcg = page->mem_cgroup; | |
5747 | ||
5748 | /* Readahead page, never charged */ | |
5749 | if (!memcg) | |
5750 | return; | |
5751 | ||
5752 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); | |
5753 | VM_BUG_ON_PAGE(oldid, page); | |
5754 | mem_cgroup_swap_statistics(memcg, true); | |
5755 | ||
5756 | page->mem_cgroup = NULL; | |
5757 | ||
5758 | if (!mem_cgroup_is_root(memcg)) | |
5759 | page_counter_uncharge(&memcg->memory, 1); | |
5760 | ||
ce9ce665 SAS |
5761 | /* |
5762 | * Interrupts should be disabled here because the caller holds the | |
5763 | * mapping->tree_lock lock which is taken with interrupts-off. It is | |
5764 | * important here to have the interrupts disabled because it is the | |
5765 | * only synchronisation we have for udpating the per-CPU variables. | |
5766 | */ | |
5767 | VM_BUG_ON(!irqs_disabled()); | |
f627c2f5 | 5768 | mem_cgroup_charge_statistics(memcg, page, false, -1); |
21afa38e JW |
5769 | memcg_check_events(memcg, page); |
5770 | } | |
5771 | ||
5772 | /** | |
5773 | * mem_cgroup_uncharge_swap - uncharge a swap entry | |
5774 | * @entry: swap entry to uncharge | |
5775 | * | |
5776 | * Drop the memsw charge associated with @entry. | |
5777 | */ | |
5778 | void mem_cgroup_uncharge_swap(swp_entry_t entry) | |
5779 | { | |
5780 | struct mem_cgroup *memcg; | |
5781 | unsigned short id; | |
5782 | ||
7941d214 | 5783 | if (!do_memsw_account()) |
21afa38e JW |
5784 | return; |
5785 | ||
5786 | id = swap_cgroup_record(entry, 0); | |
5787 | rcu_read_lock(); | |
adbe427b | 5788 | memcg = mem_cgroup_from_id(id); |
21afa38e JW |
5789 | if (memcg) { |
5790 | if (!mem_cgroup_is_root(memcg)) | |
5791 | page_counter_uncharge(&memcg->memsw, 1); | |
5792 | mem_cgroup_swap_statistics(memcg, false); | |
5793 | css_put(&memcg->css); | |
5794 | } | |
5795 | rcu_read_unlock(); | |
5796 | } | |
5797 | ||
5798 | /* for remember boot option*/ | |
5799 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
5800 | static int really_do_swap_account __initdata = 1; | |
5801 | #else | |
5802 | static int really_do_swap_account __initdata; | |
5803 | #endif | |
5804 | ||
5805 | static int __init enable_swap_account(char *s) | |
5806 | { | |
5807 | if (!strcmp(s, "1")) | |
5808 | really_do_swap_account = 1; | |
5809 | else if (!strcmp(s, "0")) | |
5810 | really_do_swap_account = 0; | |
5811 | return 1; | |
5812 | } | |
5813 | __setup("swapaccount=", enable_swap_account); | |
5814 | ||
5815 | static struct cftype memsw_cgroup_files[] = { | |
5816 | { | |
5817 | .name = "memsw.usage_in_bytes", | |
5818 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
5819 | .read_u64 = mem_cgroup_read_u64, | |
5820 | }, | |
5821 | { | |
5822 | .name = "memsw.max_usage_in_bytes", | |
5823 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
5824 | .write = mem_cgroup_reset, | |
5825 | .read_u64 = mem_cgroup_read_u64, | |
5826 | }, | |
5827 | { | |
5828 | .name = "memsw.limit_in_bytes", | |
5829 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
5830 | .write = mem_cgroup_write, | |
5831 | .read_u64 = mem_cgroup_read_u64, | |
5832 | }, | |
5833 | { | |
5834 | .name = "memsw.failcnt", | |
5835 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
5836 | .write = mem_cgroup_reset, | |
5837 | .read_u64 = mem_cgroup_read_u64, | |
5838 | }, | |
5839 | { }, /* terminate */ | |
5840 | }; | |
5841 | ||
5842 | static int __init mem_cgroup_swap_init(void) | |
5843 | { | |
5844 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
5845 | do_swap_account = 1; | |
5846 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, | |
5847 | memsw_cgroup_files)); | |
5848 | } | |
5849 | return 0; | |
5850 | } | |
5851 | subsys_initcall(mem_cgroup_swap_init); | |
5852 | ||
5853 | #endif /* CONFIG_MEMCG_SWAP */ |