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