Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
7ae1e1d0 GC |
13 | * Kernel Memory Controller |
14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
15 | * Authors: Glauber Costa and Suleiman Souhlal | |
16 | * | |
1575e68b JW |
17 | * Native page reclaim |
18 | * Charge lifetime sanitation | |
19 | * Lockless page tracking & accounting | |
20 | * Unified hierarchy configuration model | |
21 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
22 | * | |
8cdea7c0 BS |
23 | * This program is free software; you can redistribute it and/or modify |
24 | * it under the terms of the GNU General Public License as published by | |
25 | * the Free Software Foundation; either version 2 of the License, or | |
26 | * (at your option) any later version. | |
27 | * | |
28 | * This program is distributed in the hope that it will be useful, | |
29 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
30 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
31 | * GNU General Public License for more details. | |
32 | */ | |
33 | ||
3e32cb2e | 34 | #include <linux/page_counter.h> |
8cdea7c0 BS |
35 | #include <linux/memcontrol.h> |
36 | #include <linux/cgroup.h> | |
78fb7466 | 37 | #include <linux/mm.h> |
6e84f315 | 38 | #include <linux/sched/mm.h> |
3a4f8a0b | 39 | #include <linux/shmem_fs.h> |
4ffef5fe | 40 | #include <linux/hugetlb.h> |
d13d1443 | 41 | #include <linux/pagemap.h> |
1ff9e6e1 | 42 | #include <linux/vm_event_item.h> |
d52aa412 | 43 | #include <linux/smp.h> |
8a9f3ccd | 44 | #include <linux/page-flags.h> |
66e1707b | 45 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
46 | #include <linux/bit_spinlock.h> |
47 | #include <linux/rcupdate.h> | |
e222432b | 48 | #include <linux/limits.h> |
b9e15baf | 49 | #include <linux/export.h> |
8c7c6e34 | 50 | #include <linux/mutex.h> |
bb4cc1a8 | 51 | #include <linux/rbtree.h> |
b6ac57d5 | 52 | #include <linux/slab.h> |
66e1707b | 53 | #include <linux/swap.h> |
02491447 | 54 | #include <linux/swapops.h> |
66e1707b | 55 | #include <linux/spinlock.h> |
2e72b634 | 56 | #include <linux/eventfd.h> |
79bd9814 | 57 | #include <linux/poll.h> |
2e72b634 | 58 | #include <linux/sort.h> |
66e1707b | 59 | #include <linux/fs.h> |
d2ceb9b7 | 60 | #include <linux/seq_file.h> |
70ddf637 | 61 | #include <linux/vmpressure.h> |
b69408e8 | 62 | #include <linux/mm_inline.h> |
5d1ea48b | 63 | #include <linux/swap_cgroup.h> |
cdec2e42 | 64 | #include <linux/cpu.h> |
158e0a2d | 65 | #include <linux/oom.h> |
0056f4e6 | 66 | #include <linux/lockdep.h> |
79bd9814 | 67 | #include <linux/file.h> |
b23afb93 | 68 | #include <linux/tracehook.h> |
08e552c6 | 69 | #include "internal.h" |
d1a4c0b3 | 70 | #include <net/sock.h> |
4bd2c1ee | 71 | #include <net/ip.h> |
f35c3a8e | 72 | #include "slab.h" |
8cdea7c0 | 73 | |
7c0f6ba6 | 74 | #include <linux/uaccess.h> |
8697d331 | 75 | |
cc8e970c KM |
76 | #include <trace/events/vmscan.h> |
77 | ||
073219e9 TH |
78 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
79 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 80 | |
7d828602 JW |
81 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
82 | ||
a181b0e8 | 83 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 84 | |
f7e1cb6e JW |
85 | /* Socket memory accounting disabled? */ |
86 | static bool cgroup_memory_nosocket; | |
87 | ||
04823c83 VD |
88 | /* Kernel memory accounting disabled? */ |
89 | static bool cgroup_memory_nokmem; | |
90 | ||
21afa38e | 91 | /* Whether the swap controller is active */ |
c255a458 | 92 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 93 | int do_swap_account __read_mostly; |
c077719b | 94 | #else |
a0db00fc | 95 | #define do_swap_account 0 |
c077719b KH |
96 | #endif |
97 | ||
7941d214 JW |
98 | /* Whether legacy memory+swap accounting is active */ |
99 | static bool do_memsw_account(void) | |
100 | { | |
101 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; | |
102 | } | |
103 | ||
71cd3113 | 104 | static const char *const mem_cgroup_lru_names[] = { |
58cf188e SZ |
105 | "inactive_anon", |
106 | "active_anon", | |
107 | "inactive_file", | |
108 | "active_file", | |
109 | "unevictable", | |
110 | }; | |
111 | ||
a0db00fc KS |
112 | #define THRESHOLDS_EVENTS_TARGET 128 |
113 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
114 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 115 | |
bb4cc1a8 AM |
116 | /* |
117 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
118 | * their hierarchy representation | |
119 | */ | |
120 | ||
ef8f2327 | 121 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 122 | struct rb_root rb_root; |
fa90b2fd | 123 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
124 | spinlock_t lock; |
125 | }; | |
126 | ||
bb4cc1a8 AM |
127 | struct mem_cgroup_tree { |
128 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
129 | }; | |
130 | ||
131 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
132 | ||
9490ff27 KH |
133 | /* for OOM */ |
134 | struct mem_cgroup_eventfd_list { | |
135 | struct list_head list; | |
136 | struct eventfd_ctx *eventfd; | |
137 | }; | |
2e72b634 | 138 | |
79bd9814 TH |
139 | /* |
140 | * cgroup_event represents events which userspace want to receive. | |
141 | */ | |
3bc942f3 | 142 | struct mem_cgroup_event { |
79bd9814 | 143 | /* |
59b6f873 | 144 | * memcg which the event belongs to. |
79bd9814 | 145 | */ |
59b6f873 | 146 | struct mem_cgroup *memcg; |
79bd9814 TH |
147 | /* |
148 | * eventfd to signal userspace about the event. | |
149 | */ | |
150 | struct eventfd_ctx *eventfd; | |
151 | /* | |
152 | * Each of these stored in a list by the cgroup. | |
153 | */ | |
154 | struct list_head list; | |
fba94807 TH |
155 | /* |
156 | * register_event() callback will be used to add new userspace | |
157 | * waiter for changes related to this event. Use eventfd_signal() | |
158 | * on eventfd to send notification to userspace. | |
159 | */ | |
59b6f873 | 160 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 161 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
162 | /* |
163 | * unregister_event() callback will be called when userspace closes | |
164 | * the eventfd or on cgroup removing. This callback must be set, | |
165 | * if you want provide notification functionality. | |
166 | */ | |
59b6f873 | 167 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 168 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
169 | /* |
170 | * All fields below needed to unregister event when | |
171 | * userspace closes eventfd. | |
172 | */ | |
173 | poll_table pt; | |
174 | wait_queue_head_t *wqh; | |
ac6424b9 | 175 | wait_queue_entry_t wait; |
79bd9814 TH |
176 | struct work_struct remove; |
177 | }; | |
178 | ||
c0ff4b85 R |
179 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
180 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 181 | |
7dc74be0 DN |
182 | /* Stuffs for move charges at task migration. */ |
183 | /* | |
1dfab5ab | 184 | * Types of charges to be moved. |
7dc74be0 | 185 | */ |
1dfab5ab JW |
186 | #define MOVE_ANON 0x1U |
187 | #define MOVE_FILE 0x2U | |
188 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 189 | |
4ffef5fe DN |
190 | /* "mc" and its members are protected by cgroup_mutex */ |
191 | static struct move_charge_struct { | |
b1dd693e | 192 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 193 | struct mm_struct *mm; |
4ffef5fe DN |
194 | struct mem_cgroup *from; |
195 | struct mem_cgroup *to; | |
1dfab5ab | 196 | unsigned long flags; |
4ffef5fe | 197 | unsigned long precharge; |
854ffa8d | 198 | unsigned long moved_charge; |
483c30b5 | 199 | unsigned long moved_swap; |
8033b97c DN |
200 | struct task_struct *moving_task; /* a task moving charges */ |
201 | wait_queue_head_t waitq; /* a waitq for other context */ | |
202 | } mc = { | |
2bd9bb20 | 203 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
204 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
205 | }; | |
4ffef5fe | 206 | |
4e416953 BS |
207 | /* |
208 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
209 | * limit reclaim to prevent infinite loops, if they ever occur. | |
210 | */ | |
a0db00fc | 211 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 212 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 213 | |
217bc319 KH |
214 | enum charge_type { |
215 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 216 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 217 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 218 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
219 | NR_CHARGE_TYPE, |
220 | }; | |
221 | ||
8c7c6e34 | 222 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
223 | enum res_type { |
224 | _MEM, | |
225 | _MEMSWAP, | |
226 | _OOM_TYPE, | |
510fc4e1 | 227 | _KMEM, |
d55f90bf | 228 | _TCP, |
86ae53e1 GC |
229 | }; |
230 | ||
a0db00fc KS |
231 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
232 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 233 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
234 | /* Used for OOM nofiier */ |
235 | #define OOM_CONTROL (0) | |
8c7c6e34 | 236 | |
b05706f1 KT |
237 | /* |
238 | * Iteration constructs for visiting all cgroups (under a tree). If | |
239 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
240 | * be used for reference counting. | |
241 | */ | |
242 | #define for_each_mem_cgroup_tree(iter, root) \ | |
243 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
244 | iter != NULL; \ | |
245 | iter = mem_cgroup_iter(root, iter, NULL)) | |
246 | ||
247 | #define for_each_mem_cgroup(iter) \ | |
248 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
249 | iter != NULL; \ | |
250 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
251 | ||
7775face TH |
252 | static inline bool should_force_charge(void) |
253 | { | |
254 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
255 | (current->flags & PF_EXITING); | |
256 | } | |
257 | ||
70ddf637 AV |
258 | /* Some nice accessors for the vmpressure. */ |
259 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
260 | { | |
261 | if (!memcg) | |
262 | memcg = root_mem_cgroup; | |
263 | return &memcg->vmpressure; | |
264 | } | |
265 | ||
266 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
267 | { | |
268 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
269 | } | |
270 | ||
84c07d11 | 271 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 272 | /* |
f7ce3190 | 273 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
274 | * The main reason for not using cgroup id for this: |
275 | * this works better in sparse environments, where we have a lot of memcgs, | |
276 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
277 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
278 | * 200 entry array for that. | |
55007d84 | 279 | * |
dbcf73e2 VD |
280 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
281 | * will double each time we have to increase it. | |
55007d84 | 282 | */ |
dbcf73e2 VD |
283 | static DEFINE_IDA(memcg_cache_ida); |
284 | int memcg_nr_cache_ids; | |
749c5415 | 285 | |
05257a1a VD |
286 | /* Protects memcg_nr_cache_ids */ |
287 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
288 | ||
289 | void memcg_get_cache_ids(void) | |
290 | { | |
291 | down_read(&memcg_cache_ids_sem); | |
292 | } | |
293 | ||
294 | void memcg_put_cache_ids(void) | |
295 | { | |
296 | up_read(&memcg_cache_ids_sem); | |
297 | } | |
298 | ||
55007d84 GC |
299 | /* |
300 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
301 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
302 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
303 | * tunable, but that is strictly not necessary. | |
304 | * | |
b8627835 | 305 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
306 | * this constant directly from cgroup, but it is understandable that this is |
307 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 308 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
309 | * increase ours as well if it increases. |
310 | */ | |
311 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 312 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 313 | |
d7f25f8a GC |
314 | /* |
315 | * A lot of the calls to the cache allocation functions are expected to be | |
316 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
317 | * conditional to this static branch, we'll have to allow modules that does | |
318 | * kmem_cache_alloc and the such to see this symbol as well | |
319 | */ | |
ef12947c | 320 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 321 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 322 | |
17cc4dfe TH |
323 | struct workqueue_struct *memcg_kmem_cache_wq; |
324 | ||
0a4465d3 KT |
325 | static int memcg_shrinker_map_size; |
326 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); | |
327 | ||
328 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) | |
329 | { | |
330 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); | |
331 | } | |
332 | ||
333 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, | |
334 | int size, int old_size) | |
335 | { | |
336 | struct memcg_shrinker_map *new, *old; | |
337 | int nid; | |
338 | ||
339 | lockdep_assert_held(&memcg_shrinker_map_mutex); | |
340 | ||
341 | for_each_node(nid) { | |
342 | old = rcu_dereference_protected( | |
343 | mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); | |
344 | /* Not yet online memcg */ | |
345 | if (!old) | |
346 | return 0; | |
347 | ||
348 | new = kvmalloc(sizeof(*new) + size, GFP_KERNEL); | |
349 | if (!new) | |
350 | return -ENOMEM; | |
351 | ||
352 | /* Set all old bits, clear all new bits */ | |
353 | memset(new->map, (int)0xff, old_size); | |
354 | memset((void *)new->map + old_size, 0, size - old_size); | |
355 | ||
356 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); | |
357 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); | |
358 | } | |
359 | ||
360 | return 0; | |
361 | } | |
362 | ||
363 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) | |
364 | { | |
365 | struct mem_cgroup_per_node *pn; | |
366 | struct memcg_shrinker_map *map; | |
367 | int nid; | |
368 | ||
369 | if (mem_cgroup_is_root(memcg)) | |
370 | return; | |
371 | ||
372 | for_each_node(nid) { | |
373 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
374 | map = rcu_dereference_protected(pn->shrinker_map, true); | |
375 | if (map) | |
376 | kvfree(map); | |
377 | rcu_assign_pointer(pn->shrinker_map, NULL); | |
378 | } | |
379 | } | |
380 | ||
381 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
382 | { | |
383 | struct memcg_shrinker_map *map; | |
384 | int nid, size, ret = 0; | |
385 | ||
386 | if (mem_cgroup_is_root(memcg)) | |
387 | return 0; | |
388 | ||
389 | mutex_lock(&memcg_shrinker_map_mutex); | |
390 | size = memcg_shrinker_map_size; | |
391 | for_each_node(nid) { | |
392 | map = kvzalloc(sizeof(*map) + size, GFP_KERNEL); | |
393 | if (!map) { | |
394 | memcg_free_shrinker_maps(memcg); | |
395 | ret = -ENOMEM; | |
396 | break; | |
397 | } | |
398 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); | |
399 | } | |
400 | mutex_unlock(&memcg_shrinker_map_mutex); | |
401 | ||
402 | return ret; | |
403 | } | |
404 | ||
405 | int memcg_expand_shrinker_maps(int new_id) | |
406 | { | |
407 | int size, old_size, ret = 0; | |
408 | struct mem_cgroup *memcg; | |
409 | ||
410 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); | |
411 | old_size = memcg_shrinker_map_size; | |
412 | if (size <= old_size) | |
413 | return 0; | |
414 | ||
415 | mutex_lock(&memcg_shrinker_map_mutex); | |
416 | if (!root_mem_cgroup) | |
417 | goto unlock; | |
418 | ||
419 | for_each_mem_cgroup(memcg) { | |
420 | if (mem_cgroup_is_root(memcg)) | |
421 | continue; | |
422 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); | |
423 | if (ret) | |
424 | goto unlock; | |
425 | } | |
426 | unlock: | |
427 | if (!ret) | |
428 | memcg_shrinker_map_size = size; | |
429 | mutex_unlock(&memcg_shrinker_map_mutex); | |
430 | return ret; | |
431 | } | |
fae91d6d KT |
432 | |
433 | void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id) | |
434 | { | |
435 | if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) { | |
436 | struct memcg_shrinker_map *map; | |
437 | ||
438 | rcu_read_lock(); | |
439 | map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map); | |
f90280d6 KT |
440 | /* Pairs with smp mb in shrink_slab() */ |
441 | smp_mb__before_atomic(); | |
fae91d6d KT |
442 | set_bit(shrinker_id, map->map); |
443 | rcu_read_unlock(); | |
444 | } | |
445 | } | |
446 | ||
0a4465d3 KT |
447 | #else /* CONFIG_MEMCG_KMEM */ |
448 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
449 | { | |
450 | return 0; | |
451 | } | |
452 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) { } | |
84c07d11 | 453 | #endif /* CONFIG_MEMCG_KMEM */ |
a8964b9b | 454 | |
ad7fa852 TH |
455 | /** |
456 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
457 | * @page: page of interest | |
458 | * | |
459 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
460 | * with @page is returned. The returned css remains associated with @page | |
461 | * until it is released. | |
462 | * | |
463 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
464 | * is returned. | |
ad7fa852 TH |
465 | */ |
466 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
467 | { | |
468 | struct mem_cgroup *memcg; | |
469 | ||
ad7fa852 TH |
470 | memcg = page->mem_cgroup; |
471 | ||
9e10a130 | 472 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
473 | memcg = root_mem_cgroup; |
474 | ||
ad7fa852 TH |
475 | return &memcg->css; |
476 | } | |
477 | ||
2fc04524 VD |
478 | /** |
479 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
480 | * @page: the page | |
481 | * | |
482 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
483 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
484 | * is safe to call this function without holding a reference to @page. | |
485 | * | |
486 | * Note, this function is inherently racy, because there is nothing to prevent | |
487 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
488 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
489 | * do not care (such as procfs interfaces). | |
490 | */ | |
491 | ino_t page_cgroup_ino(struct page *page) | |
492 | { | |
493 | struct mem_cgroup *memcg; | |
494 | unsigned long ino = 0; | |
495 | ||
496 | rcu_read_lock(); | |
497 | memcg = READ_ONCE(page->mem_cgroup); | |
498 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
499 | memcg = parent_mem_cgroup(memcg); | |
500 | if (memcg) | |
501 | ino = cgroup_ino(memcg->css.cgroup); | |
502 | rcu_read_unlock(); | |
503 | return ino; | |
504 | } | |
505 | ||
ef8f2327 MG |
506 | static struct mem_cgroup_per_node * |
507 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 508 | { |
97a6c37b | 509 | int nid = page_to_nid(page); |
f64c3f54 | 510 | |
ef8f2327 | 511 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
512 | } |
513 | ||
ef8f2327 MG |
514 | static struct mem_cgroup_tree_per_node * |
515 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 516 | { |
ef8f2327 | 517 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
518 | } |
519 | ||
ef8f2327 | 520 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
521 | soft_limit_tree_from_page(struct page *page) |
522 | { | |
523 | int nid = page_to_nid(page); | |
bb4cc1a8 | 524 | |
ef8f2327 | 525 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
526 | } |
527 | ||
ef8f2327 MG |
528 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
529 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 530 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
531 | { |
532 | struct rb_node **p = &mctz->rb_root.rb_node; | |
533 | struct rb_node *parent = NULL; | |
ef8f2327 | 534 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 535 | bool rightmost = true; |
bb4cc1a8 AM |
536 | |
537 | if (mz->on_tree) | |
538 | return; | |
539 | ||
540 | mz->usage_in_excess = new_usage_in_excess; | |
541 | if (!mz->usage_in_excess) | |
542 | return; | |
543 | while (*p) { | |
544 | parent = *p; | |
ef8f2327 | 545 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 546 | tree_node); |
fa90b2fd | 547 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 548 | p = &(*p)->rb_left; |
fa90b2fd DB |
549 | rightmost = false; |
550 | } | |
551 | ||
bb4cc1a8 AM |
552 | /* |
553 | * We can't avoid mem cgroups that are over their soft | |
554 | * limit by the same amount | |
555 | */ | |
556 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
557 | p = &(*p)->rb_right; | |
558 | } | |
fa90b2fd DB |
559 | |
560 | if (rightmost) | |
561 | mctz->rb_rightmost = &mz->tree_node; | |
562 | ||
bb4cc1a8 AM |
563 | rb_link_node(&mz->tree_node, parent, p); |
564 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
565 | mz->on_tree = true; | |
566 | } | |
567 | ||
ef8f2327 MG |
568 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
569 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
570 | { |
571 | if (!mz->on_tree) | |
572 | return; | |
fa90b2fd DB |
573 | |
574 | if (&mz->tree_node == mctz->rb_rightmost) | |
575 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
576 | ||
bb4cc1a8 AM |
577 | rb_erase(&mz->tree_node, &mctz->rb_root); |
578 | mz->on_tree = false; | |
579 | } | |
580 | ||
ef8f2327 MG |
581 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
582 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 583 | { |
0a31bc97 JW |
584 | unsigned long flags; |
585 | ||
586 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 587 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 588 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
589 | } |
590 | ||
3e32cb2e JW |
591 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
592 | { | |
593 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 594 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
595 | unsigned long excess = 0; |
596 | ||
597 | if (nr_pages > soft_limit) | |
598 | excess = nr_pages - soft_limit; | |
599 | ||
600 | return excess; | |
601 | } | |
bb4cc1a8 AM |
602 | |
603 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
604 | { | |
3e32cb2e | 605 | unsigned long excess; |
ef8f2327 MG |
606 | struct mem_cgroup_per_node *mz; |
607 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 608 | |
e231875b | 609 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
610 | if (!mctz) |
611 | return; | |
bb4cc1a8 AM |
612 | /* |
613 | * Necessary to update all ancestors when hierarchy is used. | |
614 | * because their event counter is not touched. | |
615 | */ | |
616 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 617 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 618 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
619 | /* |
620 | * We have to update the tree if mz is on RB-tree or | |
621 | * mem is over its softlimit. | |
622 | */ | |
623 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
624 | unsigned long flags; |
625 | ||
626 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
627 | /* if on-tree, remove it */ |
628 | if (mz->on_tree) | |
cf2c8127 | 629 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
630 | /* |
631 | * Insert again. mz->usage_in_excess will be updated. | |
632 | * If excess is 0, no tree ops. | |
633 | */ | |
cf2c8127 | 634 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 635 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
636 | } |
637 | } | |
638 | } | |
639 | ||
640 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
641 | { | |
ef8f2327 MG |
642 | struct mem_cgroup_tree_per_node *mctz; |
643 | struct mem_cgroup_per_node *mz; | |
644 | int nid; | |
bb4cc1a8 | 645 | |
e231875b | 646 | for_each_node(nid) { |
ef8f2327 MG |
647 | mz = mem_cgroup_nodeinfo(memcg, nid); |
648 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
649 | if (mctz) |
650 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
651 | } |
652 | } | |
653 | ||
ef8f2327 MG |
654 | static struct mem_cgroup_per_node * |
655 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 656 | { |
ef8f2327 | 657 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
658 | |
659 | retry: | |
660 | mz = NULL; | |
fa90b2fd | 661 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
662 | goto done; /* Nothing to reclaim from */ |
663 | ||
fa90b2fd DB |
664 | mz = rb_entry(mctz->rb_rightmost, |
665 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
666 | /* |
667 | * Remove the node now but someone else can add it back, | |
668 | * we will to add it back at the end of reclaim to its correct | |
669 | * position in the tree. | |
670 | */ | |
cf2c8127 | 671 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 672 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 673 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
674 | goto retry; |
675 | done: | |
676 | return mz; | |
677 | } | |
678 | ||
ef8f2327 MG |
679 | static struct mem_cgroup_per_node * |
680 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 681 | { |
ef8f2327 | 682 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 683 | |
0a31bc97 | 684 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 685 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 686 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
687 | return mz; |
688 | } | |
689 | ||
db9adbcb JW |
690 | /** |
691 | * __mod_memcg_state - update cgroup memory statistics | |
692 | * @memcg: the memory cgroup | |
693 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
694 | * @val: delta to add to the counter, can be negative | |
695 | */ | |
696 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
697 | { | |
698 | long x; | |
699 | ||
700 | if (mem_cgroup_disabled()) | |
701 | return; | |
702 | ||
703 | x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); | |
704 | if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
705 | struct mem_cgroup *mi; |
706 | ||
707 | atomic_long_add(x, &memcg->vmstats_local[idx]); | |
708 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
709 | atomic_long_add(x, &mi->vmstats[idx]); | |
db9adbcb JW |
710 | x = 0; |
711 | } | |
712 | __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); | |
713 | } | |
714 | ||
42a30035 JW |
715 | static struct mem_cgroup_per_node * |
716 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
717 | { | |
718 | struct mem_cgroup *parent; | |
719 | ||
720 | parent = parent_mem_cgroup(pn->memcg); | |
721 | if (!parent) | |
722 | return NULL; | |
723 | return mem_cgroup_nodeinfo(parent, nid); | |
724 | } | |
725 | ||
db9adbcb JW |
726 | /** |
727 | * __mod_lruvec_state - update lruvec memory statistics | |
728 | * @lruvec: the lruvec | |
729 | * @idx: the stat item | |
730 | * @val: delta to add to the counter, can be negative | |
731 | * | |
732 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
733 | * function updates the all three counters that are affected by a | |
734 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
735 | */ | |
736 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
737 | int val) | |
738 | { | |
42a30035 | 739 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
db9adbcb | 740 | struct mem_cgroup_per_node *pn; |
42a30035 | 741 | struct mem_cgroup *memcg; |
db9adbcb JW |
742 | long x; |
743 | ||
744 | /* Update node */ | |
42a30035 | 745 | __mod_node_page_state(pgdat, idx, val); |
db9adbcb JW |
746 | |
747 | if (mem_cgroup_disabled()) | |
748 | return; | |
749 | ||
750 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); | |
42a30035 | 751 | memcg = pn->memcg; |
db9adbcb JW |
752 | |
753 | /* Update memcg */ | |
42a30035 | 754 | __mod_memcg_state(memcg, idx, val); |
db9adbcb JW |
755 | |
756 | /* Update lruvec */ | |
757 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); | |
758 | if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
759 | struct mem_cgroup_per_node *pi; |
760 | ||
761 | atomic_long_add(x, &pn->lruvec_stat_local[idx]); | |
762 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) | |
763 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
764 | x = 0; |
765 | } | |
766 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
767 | } | |
768 | ||
769 | /** | |
770 | * __count_memcg_events - account VM events in a cgroup | |
771 | * @memcg: the memory cgroup | |
772 | * @idx: the event item | |
773 | * @count: the number of events that occured | |
774 | */ | |
775 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
776 | unsigned long count) | |
777 | { | |
778 | unsigned long x; | |
779 | ||
780 | if (mem_cgroup_disabled()) | |
781 | return; | |
782 | ||
783 | x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); | |
784 | if (unlikely(x > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
785 | struct mem_cgroup *mi; |
786 | ||
787 | atomic_long_add(x, &memcg->vmevents_local[idx]); | |
788 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
789 | atomic_long_add(x, &mi->vmevents[idx]); | |
db9adbcb JW |
790 | x = 0; |
791 | } | |
792 | __this_cpu_write(memcg->vmstats_percpu->events[idx], x); | |
793 | } | |
794 | ||
42a30035 | 795 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 796 | { |
871789d4 | 797 | return atomic_long_read(&memcg->vmevents[event]); |
e9f8974f JW |
798 | } |
799 | ||
42a30035 JW |
800 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
801 | { | |
802 | return atomic_long_read(&memcg->vmevents_local[event]); | |
803 | } | |
804 | ||
c0ff4b85 | 805 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 806 | struct page *page, |
f627c2f5 | 807 | bool compound, int nr_pages) |
d52aa412 | 808 | { |
b2402857 KH |
809 | /* |
810 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
811 | * counted as CACHE even if it's on ANON LRU. | |
812 | */ | |
0a31bc97 | 813 | if (PageAnon(page)) |
c9019e9b | 814 | __mod_memcg_state(memcg, MEMCG_RSS, nr_pages); |
9a4caf1e | 815 | else { |
c9019e9b | 816 | __mod_memcg_state(memcg, MEMCG_CACHE, nr_pages); |
9a4caf1e | 817 | if (PageSwapBacked(page)) |
c9019e9b | 818 | __mod_memcg_state(memcg, NR_SHMEM, nr_pages); |
9a4caf1e | 819 | } |
55e462b0 | 820 | |
f627c2f5 KS |
821 | if (compound) { |
822 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
c9019e9b | 823 | __mod_memcg_state(memcg, MEMCG_RSS_HUGE, nr_pages); |
f627c2f5 | 824 | } |
b070e65c | 825 | |
e401f176 KH |
826 | /* pagein of a big page is an event. So, ignore page size */ |
827 | if (nr_pages > 0) | |
c9019e9b | 828 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 829 | else { |
c9019e9b | 830 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
831 | nr_pages = -nr_pages; /* for event */ |
832 | } | |
e401f176 | 833 | |
871789d4 | 834 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
835 | } |
836 | ||
f53d7ce3 JW |
837 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
838 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
839 | { |
840 | unsigned long val, next; | |
841 | ||
871789d4 CD |
842 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
843 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 844 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 845 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
846 | switch (target) { |
847 | case MEM_CGROUP_TARGET_THRESH: | |
848 | next = val + THRESHOLDS_EVENTS_TARGET; | |
849 | break; | |
bb4cc1a8 AM |
850 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
851 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
852 | break; | |
f53d7ce3 JW |
853 | case MEM_CGROUP_TARGET_NUMAINFO: |
854 | next = val + NUMAINFO_EVENTS_TARGET; | |
855 | break; | |
856 | default: | |
857 | break; | |
858 | } | |
871789d4 | 859 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 860 | return true; |
7a159cc9 | 861 | } |
f53d7ce3 | 862 | return false; |
d2265e6f KH |
863 | } |
864 | ||
865 | /* | |
866 | * Check events in order. | |
867 | * | |
868 | */ | |
c0ff4b85 | 869 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
870 | { |
871 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
872 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
873 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 874 | bool do_softlimit; |
82b3f2a7 | 875 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 876 | |
bb4cc1a8 AM |
877 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
878 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
879 | #if MAX_NUMNODES > 1 |
880 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
881 | MEM_CGROUP_TARGET_NUMAINFO); | |
882 | #endif | |
c0ff4b85 | 883 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
884 | if (unlikely(do_softlimit)) |
885 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 886 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 887 | if (unlikely(do_numainfo)) |
c0ff4b85 | 888 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 889 | #endif |
0a31bc97 | 890 | } |
d2265e6f KH |
891 | } |
892 | ||
cf475ad2 | 893 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 894 | { |
31a78f23 BS |
895 | /* |
896 | * mm_update_next_owner() may clear mm->owner to NULL | |
897 | * if it races with swapoff, page migration, etc. | |
898 | * So this can be called with p == NULL. | |
899 | */ | |
900 | if (unlikely(!p)) | |
901 | return NULL; | |
902 | ||
073219e9 | 903 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 904 | } |
33398cf2 | 905 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 906 | |
d46eb14b SB |
907 | /** |
908 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
909 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
910 | * | |
911 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
912 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
913 | * returned. | |
914 | */ | |
915 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 916 | { |
d46eb14b SB |
917 | struct mem_cgroup *memcg; |
918 | ||
919 | if (mem_cgroup_disabled()) | |
920 | return NULL; | |
0b7f569e | 921 | |
54595fe2 KH |
922 | rcu_read_lock(); |
923 | do { | |
6f6acb00 MH |
924 | /* |
925 | * Page cache insertions can happen withou an | |
926 | * actual mm context, e.g. during disk probing | |
927 | * on boot, loopback IO, acct() writes etc. | |
928 | */ | |
929 | if (unlikely(!mm)) | |
df381975 | 930 | memcg = root_mem_cgroup; |
6f6acb00 MH |
931 | else { |
932 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
933 | if (unlikely(!memcg)) | |
934 | memcg = root_mem_cgroup; | |
935 | } | |
ec903c0c | 936 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 937 | rcu_read_unlock(); |
c0ff4b85 | 938 | return memcg; |
54595fe2 | 939 | } |
d46eb14b SB |
940 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
941 | ||
f745c6f5 SB |
942 | /** |
943 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
944 | * @page: page from which memcg should be extracted. | |
945 | * | |
946 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
947 | * root_mem_cgroup is returned. | |
948 | */ | |
949 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
950 | { | |
951 | struct mem_cgroup *memcg = page->mem_cgroup; | |
952 | ||
953 | if (mem_cgroup_disabled()) | |
954 | return NULL; | |
955 | ||
956 | rcu_read_lock(); | |
957 | if (!memcg || !css_tryget_online(&memcg->css)) | |
958 | memcg = root_mem_cgroup; | |
959 | rcu_read_unlock(); | |
960 | return memcg; | |
961 | } | |
962 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
963 | ||
d46eb14b SB |
964 | /** |
965 | * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg. | |
966 | */ | |
967 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
968 | { | |
969 | if (unlikely(current->active_memcg)) { | |
970 | struct mem_cgroup *memcg = root_mem_cgroup; | |
971 | ||
972 | rcu_read_lock(); | |
973 | if (css_tryget_online(¤t->active_memcg->css)) | |
974 | memcg = current->active_memcg; | |
975 | rcu_read_unlock(); | |
976 | return memcg; | |
977 | } | |
978 | return get_mem_cgroup_from_mm(current->mm); | |
979 | } | |
54595fe2 | 980 | |
5660048c JW |
981 | /** |
982 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
983 | * @root: hierarchy root | |
984 | * @prev: previously returned memcg, NULL on first invocation | |
985 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
986 | * | |
987 | * Returns references to children of the hierarchy below @root, or | |
988 | * @root itself, or %NULL after a full round-trip. | |
989 | * | |
990 | * Caller must pass the return value in @prev on subsequent | |
991 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
992 | * to cancel a hierarchy walk before the round-trip is complete. | |
993 | * | |
b213b54f | 994 | * Reclaimers can specify a node and a priority level in @reclaim to |
5660048c | 995 | * divide up the memcgs in the hierarchy among all concurrent |
b213b54f | 996 | * reclaimers operating on the same node and priority. |
5660048c | 997 | */ |
694fbc0f | 998 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 999 | struct mem_cgroup *prev, |
694fbc0f | 1000 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1001 | { |
33398cf2 | 1002 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 1003 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1004 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1005 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1006 | |
694fbc0f AM |
1007 | if (mem_cgroup_disabled()) |
1008 | return NULL; | |
5660048c | 1009 | |
9f3a0d09 JW |
1010 | if (!root) |
1011 | root = root_mem_cgroup; | |
7d74b06f | 1012 | |
9f3a0d09 | 1013 | if (prev && !reclaim) |
5ac8fb31 | 1014 | pos = prev; |
14067bb3 | 1015 | |
9f3a0d09 JW |
1016 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1017 | if (prev) | |
5ac8fb31 | 1018 | goto out; |
694fbc0f | 1019 | return root; |
9f3a0d09 | 1020 | } |
14067bb3 | 1021 | |
542f85f9 | 1022 | rcu_read_lock(); |
5f578161 | 1023 | |
5ac8fb31 | 1024 | if (reclaim) { |
ef8f2327 | 1025 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1026 | |
ef8f2327 | 1027 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
5ac8fb31 JW |
1028 | iter = &mz->iter[reclaim->priority]; |
1029 | ||
1030 | if (prev && reclaim->generation != iter->generation) | |
1031 | goto out_unlock; | |
1032 | ||
6df38689 | 1033 | while (1) { |
4db0c3c2 | 1034 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1035 | if (!pos || css_tryget(&pos->css)) |
1036 | break; | |
5ac8fb31 | 1037 | /* |
6df38689 VD |
1038 | * css reference reached zero, so iter->position will |
1039 | * be cleared by ->css_released. However, we should not | |
1040 | * rely on this happening soon, because ->css_released | |
1041 | * is called from a work queue, and by busy-waiting we | |
1042 | * might block it. So we clear iter->position right | |
1043 | * away. | |
5ac8fb31 | 1044 | */ |
6df38689 VD |
1045 | (void)cmpxchg(&iter->position, pos, NULL); |
1046 | } | |
5ac8fb31 JW |
1047 | } |
1048 | ||
1049 | if (pos) | |
1050 | css = &pos->css; | |
1051 | ||
1052 | for (;;) { | |
1053 | css = css_next_descendant_pre(css, &root->css); | |
1054 | if (!css) { | |
1055 | /* | |
1056 | * Reclaimers share the hierarchy walk, and a | |
1057 | * new one might jump in right at the end of | |
1058 | * the hierarchy - make sure they see at least | |
1059 | * one group and restart from the beginning. | |
1060 | */ | |
1061 | if (!prev) | |
1062 | continue; | |
1063 | break; | |
527a5ec9 | 1064 | } |
7d74b06f | 1065 | |
5ac8fb31 JW |
1066 | /* |
1067 | * Verify the css and acquire a reference. The root | |
1068 | * is provided by the caller, so we know it's alive | |
1069 | * and kicking, and don't take an extra reference. | |
1070 | */ | |
1071 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1072 | |
5ac8fb31 JW |
1073 | if (css == &root->css) |
1074 | break; | |
14067bb3 | 1075 | |
0b8f73e1 JW |
1076 | if (css_tryget(css)) |
1077 | break; | |
9f3a0d09 | 1078 | |
5ac8fb31 | 1079 | memcg = NULL; |
9f3a0d09 | 1080 | } |
5ac8fb31 JW |
1081 | |
1082 | if (reclaim) { | |
5ac8fb31 | 1083 | /* |
6df38689 VD |
1084 | * The position could have already been updated by a competing |
1085 | * thread, so check that the value hasn't changed since we read | |
1086 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1087 | */ |
6df38689 VD |
1088 | (void)cmpxchg(&iter->position, pos, memcg); |
1089 | ||
5ac8fb31 JW |
1090 | if (pos) |
1091 | css_put(&pos->css); | |
1092 | ||
1093 | if (!memcg) | |
1094 | iter->generation++; | |
1095 | else if (!prev) | |
1096 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1097 | } |
5ac8fb31 | 1098 | |
542f85f9 MH |
1099 | out_unlock: |
1100 | rcu_read_unlock(); | |
5ac8fb31 | 1101 | out: |
c40046f3 MH |
1102 | if (prev && prev != root) |
1103 | css_put(&prev->css); | |
1104 | ||
9f3a0d09 | 1105 | return memcg; |
14067bb3 | 1106 | } |
7d74b06f | 1107 | |
5660048c JW |
1108 | /** |
1109 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1110 | * @root: hierarchy root | |
1111 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1112 | */ | |
1113 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1114 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1115 | { |
1116 | if (!root) | |
1117 | root = root_mem_cgroup; | |
1118 | if (prev && prev != root) | |
1119 | css_put(&prev->css); | |
1120 | } | |
7d74b06f | 1121 | |
6df38689 VD |
1122 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1123 | { | |
1124 | struct mem_cgroup *memcg = dead_memcg; | |
1125 | struct mem_cgroup_reclaim_iter *iter; | |
ef8f2327 MG |
1126 | struct mem_cgroup_per_node *mz; |
1127 | int nid; | |
6df38689 VD |
1128 | int i; |
1129 | ||
9f15bde6 | 1130 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
6df38689 | 1131 | for_each_node(nid) { |
ef8f2327 MG |
1132 | mz = mem_cgroup_nodeinfo(memcg, nid); |
1133 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
1134 | iter = &mz->iter[i]; | |
1135 | cmpxchg(&iter->position, | |
1136 | dead_memcg, NULL); | |
6df38689 VD |
1137 | } |
1138 | } | |
1139 | } | |
1140 | } | |
1141 | ||
7c5f64f8 VD |
1142 | /** |
1143 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1144 | * @memcg: hierarchy root | |
1145 | * @fn: function to call for each task | |
1146 | * @arg: argument passed to @fn | |
1147 | * | |
1148 | * This function iterates over tasks attached to @memcg or to any of its | |
1149 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1150 | * value, the function breaks the iteration loop and returns the value. | |
1151 | * Otherwise, it will iterate over all tasks and return 0. | |
1152 | * | |
1153 | * This function must not be called for the root memory cgroup. | |
1154 | */ | |
1155 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1156 | int (*fn)(struct task_struct *, void *), void *arg) | |
1157 | { | |
1158 | struct mem_cgroup *iter; | |
1159 | int ret = 0; | |
1160 | ||
1161 | BUG_ON(memcg == root_mem_cgroup); | |
1162 | ||
1163 | for_each_mem_cgroup_tree(iter, memcg) { | |
1164 | struct css_task_iter it; | |
1165 | struct task_struct *task; | |
1166 | ||
bc2fb7ed | 1167 | css_task_iter_start(&iter->css, 0, &it); |
7c5f64f8 VD |
1168 | while (!ret && (task = css_task_iter_next(&it))) |
1169 | ret = fn(task, arg); | |
1170 | css_task_iter_end(&it); | |
1171 | if (ret) { | |
1172 | mem_cgroup_iter_break(memcg, iter); | |
1173 | break; | |
1174 | } | |
1175 | } | |
1176 | return ret; | |
1177 | } | |
1178 | ||
925b7673 | 1179 | /** |
dfe0e773 | 1180 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1181 | * @page: the page |
f144c390 | 1182 | * @pgdat: pgdat of the page |
dfe0e773 JW |
1183 | * |
1184 | * This function is only safe when following the LRU page isolation | |
1185 | * and putback protocol: the LRU lock must be held, and the page must | |
1186 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1187 | */ |
599d0c95 | 1188 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1189 | { |
ef8f2327 | 1190 | struct mem_cgroup_per_node *mz; |
925b7673 | 1191 | struct mem_cgroup *memcg; |
bea8c150 | 1192 | struct lruvec *lruvec; |
6d12e2d8 | 1193 | |
bea8c150 | 1194 | if (mem_cgroup_disabled()) { |
599d0c95 | 1195 | lruvec = &pgdat->lruvec; |
bea8c150 HD |
1196 | goto out; |
1197 | } | |
925b7673 | 1198 | |
1306a85a | 1199 | memcg = page->mem_cgroup; |
7512102c | 1200 | /* |
dfe0e773 | 1201 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1202 | * possibly migrated - before they are charged. |
7512102c | 1203 | */ |
29833315 JW |
1204 | if (!memcg) |
1205 | memcg = root_mem_cgroup; | |
7512102c | 1206 | |
ef8f2327 | 1207 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1208 | lruvec = &mz->lruvec; |
1209 | out: | |
1210 | /* | |
1211 | * Since a node can be onlined after the mem_cgroup was created, | |
1212 | * we have to be prepared to initialize lruvec->zone here; | |
1213 | * and if offlined then reonlined, we need to reinitialize it. | |
1214 | */ | |
599d0c95 MG |
1215 | if (unlikely(lruvec->pgdat != pgdat)) |
1216 | lruvec->pgdat = pgdat; | |
bea8c150 | 1217 | return lruvec; |
08e552c6 | 1218 | } |
b69408e8 | 1219 | |
925b7673 | 1220 | /** |
fa9add64 HD |
1221 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1222 | * @lruvec: mem_cgroup per zone lru vector | |
1223 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1224 | * @zid: zone id of the accounted pages |
fa9add64 | 1225 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1226 | * |
ca707239 HD |
1227 | * This function must be called under lru_lock, just before a page is added |
1228 | * to or just after a page is removed from an lru list (that ordering being | |
1229 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1230 | */ |
fa9add64 | 1231 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1232 | int zid, int nr_pages) |
3f58a829 | 1233 | { |
ef8f2327 | 1234 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1235 | unsigned long *lru_size; |
ca707239 | 1236 | long size; |
3f58a829 MK |
1237 | |
1238 | if (mem_cgroup_disabled()) | |
1239 | return; | |
1240 | ||
ef8f2327 | 1241 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1242 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1243 | |
1244 | if (nr_pages < 0) | |
1245 | *lru_size += nr_pages; | |
1246 | ||
1247 | size = *lru_size; | |
b4536f0c MH |
1248 | if (WARN_ONCE(size < 0, |
1249 | "%s(%p, %d, %d): lru_size %ld\n", | |
1250 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1251 | VM_BUG_ON(1); |
1252 | *lru_size = 0; | |
1253 | } | |
1254 | ||
1255 | if (nr_pages > 0) | |
1256 | *lru_size += nr_pages; | |
08e552c6 | 1257 | } |
544122e5 | 1258 | |
2314b42d | 1259 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1260 | { |
2314b42d | 1261 | struct mem_cgroup *task_memcg; |
158e0a2d | 1262 | struct task_struct *p; |
ffbdccf5 | 1263 | bool ret; |
4c4a2214 | 1264 | |
158e0a2d | 1265 | p = find_lock_task_mm(task); |
de077d22 | 1266 | if (p) { |
2314b42d | 1267 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1268 | task_unlock(p); |
1269 | } else { | |
1270 | /* | |
1271 | * All threads may have already detached their mm's, but the oom | |
1272 | * killer still needs to detect if they have already been oom | |
1273 | * killed to prevent needlessly killing additional tasks. | |
1274 | */ | |
ffbdccf5 | 1275 | rcu_read_lock(); |
2314b42d JW |
1276 | task_memcg = mem_cgroup_from_task(task); |
1277 | css_get(&task_memcg->css); | |
ffbdccf5 | 1278 | rcu_read_unlock(); |
de077d22 | 1279 | } |
2314b42d JW |
1280 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1281 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1282 | return ret; |
1283 | } | |
1284 | ||
19942822 | 1285 | /** |
9d11ea9f | 1286 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1287 | * @memcg: the memory cgroup |
19942822 | 1288 | * |
9d11ea9f | 1289 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1290 | * pages. |
19942822 | 1291 | */ |
c0ff4b85 | 1292 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1293 | { |
3e32cb2e JW |
1294 | unsigned long margin = 0; |
1295 | unsigned long count; | |
1296 | unsigned long limit; | |
9d11ea9f | 1297 | |
3e32cb2e | 1298 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1299 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1300 | if (count < limit) |
1301 | margin = limit - count; | |
1302 | ||
7941d214 | 1303 | if (do_memsw_account()) { |
3e32cb2e | 1304 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1305 | limit = READ_ONCE(memcg->memsw.max); |
3e32cb2e JW |
1306 | if (count <= limit) |
1307 | margin = min(margin, limit - count); | |
cbedbac3 LR |
1308 | else |
1309 | margin = 0; | |
3e32cb2e JW |
1310 | } |
1311 | ||
1312 | return margin; | |
19942822 JW |
1313 | } |
1314 | ||
32047e2a | 1315 | /* |
bdcbb659 | 1316 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1317 | * |
bdcbb659 QH |
1318 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1319 | * moving cgroups. This is for waiting at high-memory pressure | |
1320 | * caused by "move". | |
32047e2a | 1321 | */ |
c0ff4b85 | 1322 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1323 | { |
2bd9bb20 KH |
1324 | struct mem_cgroup *from; |
1325 | struct mem_cgroup *to; | |
4b534334 | 1326 | bool ret = false; |
2bd9bb20 KH |
1327 | /* |
1328 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1329 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1330 | */ | |
1331 | spin_lock(&mc.lock); | |
1332 | from = mc.from; | |
1333 | to = mc.to; | |
1334 | if (!from) | |
1335 | goto unlock; | |
3e92041d | 1336 | |
2314b42d JW |
1337 | ret = mem_cgroup_is_descendant(from, memcg) || |
1338 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1339 | unlock: |
1340 | spin_unlock(&mc.lock); | |
4b534334 KH |
1341 | return ret; |
1342 | } | |
1343 | ||
c0ff4b85 | 1344 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1345 | { |
1346 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1347 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1348 | DEFINE_WAIT(wait); |
1349 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1350 | /* moving charge context might have finished. */ | |
1351 | if (mc.moving_task) | |
1352 | schedule(); | |
1353 | finish_wait(&mc.waitq, &wait); | |
1354 | return true; | |
1355 | } | |
1356 | } | |
1357 | return false; | |
1358 | } | |
1359 | ||
8ad6e404 | 1360 | static const unsigned int memcg1_stats[] = { |
71cd3113 JW |
1361 | MEMCG_CACHE, |
1362 | MEMCG_RSS, | |
1363 | MEMCG_RSS_HUGE, | |
1364 | NR_SHMEM, | |
1365 | NR_FILE_MAPPED, | |
1366 | NR_FILE_DIRTY, | |
1367 | NR_WRITEBACK, | |
1368 | MEMCG_SWAP, | |
1369 | }; | |
1370 | ||
1371 | static const char *const memcg1_stat_names[] = { | |
1372 | "cache", | |
1373 | "rss", | |
1374 | "rss_huge", | |
1375 | "shmem", | |
1376 | "mapped_file", | |
1377 | "dirty", | |
1378 | "writeback", | |
1379 | "swap", | |
1380 | }; | |
1381 | ||
58cf188e | 1382 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1383 | /** |
f0c867d9 | 1384 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1385 | * memory controller. | |
e222432b BS |
1386 | * @memcg: The memory cgroup that went over limit |
1387 | * @p: Task that is going to be killed | |
1388 | * | |
1389 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1390 | * enabled | |
1391 | */ | |
f0c867d9 | 1392 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1393 | { |
e222432b BS |
1394 | rcu_read_lock(); |
1395 | ||
f0c867d9 | 1396 | if (memcg) { |
1397 | pr_cont(",oom_memcg="); | |
1398 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1399 | } else | |
1400 | pr_cont(",global_oom"); | |
2415b9f5 | 1401 | if (p) { |
f0c867d9 | 1402 | pr_cont(",task_memcg="); |
2415b9f5 | 1403 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1404 | } |
e222432b | 1405 | rcu_read_unlock(); |
f0c867d9 | 1406 | } |
1407 | ||
1408 | /** | |
1409 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1410 | * memory controller. | |
1411 | * @memcg: The memory cgroup that went over limit | |
1412 | */ | |
1413 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1414 | { | |
1415 | struct mem_cgroup *iter; | |
1416 | unsigned int i; | |
e222432b | 1417 | |
3e32cb2e JW |
1418 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1419 | K((u64)page_counter_read(&memcg->memory)), | |
bbec2e15 | 1420 | K((u64)memcg->memory.max), memcg->memory.failcnt); |
3e32cb2e JW |
1421 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", |
1422 | K((u64)page_counter_read(&memcg->memsw)), | |
bbec2e15 | 1423 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); |
3e32cb2e JW |
1424 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", |
1425 | K((u64)page_counter_read(&memcg->kmem)), | |
bbec2e15 | 1426 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); |
58cf188e SZ |
1427 | |
1428 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1429 | pr_info("Memory cgroup stats for "); |
1430 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1431 | pr_cont(":"); |
1432 | ||
71cd3113 JW |
1433 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
1434 | if (memcg1_stats[i] == MEMCG_SWAP && !do_swap_account) | |
58cf188e | 1435 | continue; |
71cd3113 | 1436 | pr_cont(" %s:%luKB", memcg1_stat_names[i], |
205b20cc JW |
1437 | K(memcg_page_state_local(iter, |
1438 | memcg1_stats[i]))); | |
58cf188e SZ |
1439 | } |
1440 | ||
1441 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1442 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
205b20cc JW |
1443 | K(memcg_page_state_local(iter, |
1444 | NR_LRU_BASE + i))); | |
58cf188e SZ |
1445 | |
1446 | pr_cont("\n"); | |
1447 | } | |
e222432b BS |
1448 | } |
1449 | ||
a63d83f4 DR |
1450 | /* |
1451 | * Return the memory (and swap, if configured) limit for a memcg. | |
1452 | */ | |
bbec2e15 | 1453 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1454 | { |
bbec2e15 | 1455 | unsigned long max; |
f3e8eb70 | 1456 | |
bbec2e15 | 1457 | max = memcg->memory.max; |
9a5a8f19 | 1458 | if (mem_cgroup_swappiness(memcg)) { |
bbec2e15 RG |
1459 | unsigned long memsw_max; |
1460 | unsigned long swap_max; | |
9a5a8f19 | 1461 | |
bbec2e15 RG |
1462 | memsw_max = memcg->memsw.max; |
1463 | swap_max = memcg->swap.max; | |
1464 | swap_max = min(swap_max, (unsigned long)total_swap_pages); | |
1465 | max = min(max + swap_max, memsw_max); | |
9a5a8f19 | 1466 | } |
bbec2e15 | 1467 | return max; |
a63d83f4 DR |
1468 | } |
1469 | ||
b6e6edcf | 1470 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1471 | int order) |
9cbb78bb | 1472 | { |
6e0fc46d DR |
1473 | struct oom_control oc = { |
1474 | .zonelist = NULL, | |
1475 | .nodemask = NULL, | |
2a966b77 | 1476 | .memcg = memcg, |
6e0fc46d DR |
1477 | .gfp_mask = gfp_mask, |
1478 | .order = order, | |
6e0fc46d | 1479 | }; |
7c5f64f8 | 1480 | bool ret; |
9cbb78bb | 1481 | |
7775face TH |
1482 | if (mutex_lock_killable(&oom_lock)) |
1483 | return true; | |
1484 | /* | |
1485 | * A few threads which were not waiting at mutex_lock_killable() can | |
1486 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1487 | */ | |
1488 | ret = should_force_charge() || out_of_memory(&oc); | |
dc56401f | 1489 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1490 | return ret; |
9cbb78bb DR |
1491 | } |
1492 | ||
ae6e71d3 MC |
1493 | #if MAX_NUMNODES > 1 |
1494 | ||
4d0c066d KH |
1495 | /** |
1496 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1497 | * @memcg: the target memcg |
4d0c066d KH |
1498 | * @nid: the node ID to be checked. |
1499 | * @noswap : specify true here if the user wants flle only information. | |
1500 | * | |
1501 | * This function returns whether the specified memcg contains any | |
1502 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1503 | * pages in the node. | |
1504 | */ | |
c0ff4b85 | 1505 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1506 | int nid, bool noswap) |
1507 | { | |
2b487e59 JW |
1508 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); |
1509 | ||
205b20cc JW |
1510 | if (lruvec_page_state_local(lruvec, NR_INACTIVE_FILE) || |
1511 | lruvec_page_state_local(lruvec, NR_ACTIVE_FILE)) | |
4d0c066d KH |
1512 | return true; |
1513 | if (noswap || !total_swap_pages) | |
1514 | return false; | |
205b20cc JW |
1515 | if (lruvec_page_state_local(lruvec, NR_INACTIVE_ANON) || |
1516 | lruvec_page_state_local(lruvec, NR_ACTIVE_ANON)) | |
4d0c066d KH |
1517 | return true; |
1518 | return false; | |
1519 | ||
1520 | } | |
889976db YH |
1521 | |
1522 | /* | |
1523 | * Always updating the nodemask is not very good - even if we have an empty | |
1524 | * list or the wrong list here, we can start from some node and traverse all | |
1525 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1526 | * | |
1527 | */ | |
c0ff4b85 | 1528 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1529 | { |
1530 | int nid; | |
453a9bf3 KH |
1531 | /* |
1532 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1533 | * pagein/pageout changes since the last update. | |
1534 | */ | |
c0ff4b85 | 1535 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1536 | return; |
c0ff4b85 | 1537 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1538 | return; |
1539 | ||
889976db | 1540 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1541 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1542 | |
31aaea4a | 1543 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1544 | |
c0ff4b85 R |
1545 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1546 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1547 | } |
453a9bf3 | 1548 | |
c0ff4b85 R |
1549 | atomic_set(&memcg->numainfo_events, 0); |
1550 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1551 | } |
1552 | ||
1553 | /* | |
1554 | * Selecting a node where we start reclaim from. Because what we need is just | |
1555 | * reducing usage counter, start from anywhere is O,K. Considering | |
1556 | * memory reclaim from current node, there are pros. and cons. | |
1557 | * | |
1558 | * Freeing memory from current node means freeing memory from a node which | |
1559 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1560 | * hit limits, it will see a contention on a node. But freeing from remote | |
1561 | * node means more costs for memory reclaim because of memory latency. | |
1562 | * | |
1563 | * Now, we use round-robin. Better algorithm is welcomed. | |
1564 | */ | |
c0ff4b85 | 1565 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1566 | { |
1567 | int node; | |
1568 | ||
c0ff4b85 R |
1569 | mem_cgroup_may_update_nodemask(memcg); |
1570 | node = memcg->last_scanned_node; | |
889976db | 1571 | |
0edaf86c | 1572 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1573 | /* |
fda3d69b MH |
1574 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1575 | * last time it really checked all the LRUs due to rate limiting. | |
1576 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1577 | */ |
1578 | if (unlikely(node == MAX_NUMNODES)) | |
1579 | node = numa_node_id(); | |
1580 | ||
c0ff4b85 | 1581 | memcg->last_scanned_node = node; |
889976db YH |
1582 | return node; |
1583 | } | |
889976db | 1584 | #else |
c0ff4b85 | 1585 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1586 | { |
1587 | return 0; | |
1588 | } | |
1589 | #endif | |
1590 | ||
0608f43d | 1591 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1592 | pg_data_t *pgdat, |
0608f43d AM |
1593 | gfp_t gfp_mask, |
1594 | unsigned long *total_scanned) | |
1595 | { | |
1596 | struct mem_cgroup *victim = NULL; | |
1597 | int total = 0; | |
1598 | int loop = 0; | |
1599 | unsigned long excess; | |
1600 | unsigned long nr_scanned; | |
1601 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1602 | .pgdat = pgdat, |
0608f43d AM |
1603 | .priority = 0, |
1604 | }; | |
1605 | ||
3e32cb2e | 1606 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1607 | |
1608 | while (1) { | |
1609 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1610 | if (!victim) { | |
1611 | loop++; | |
1612 | if (loop >= 2) { | |
1613 | /* | |
1614 | * If we have not been able to reclaim | |
1615 | * anything, it might because there are | |
1616 | * no reclaimable pages under this hierarchy | |
1617 | */ | |
1618 | if (!total) | |
1619 | break; | |
1620 | /* | |
1621 | * We want to do more targeted reclaim. | |
1622 | * excess >> 2 is not to excessive so as to | |
1623 | * reclaim too much, nor too less that we keep | |
1624 | * coming back to reclaim from this cgroup | |
1625 | */ | |
1626 | if (total >= (excess >> 2) || | |
1627 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1628 | break; | |
1629 | } | |
1630 | continue; | |
1631 | } | |
a9dd0a83 | 1632 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1633 | pgdat, &nr_scanned); |
0608f43d | 1634 | *total_scanned += nr_scanned; |
3e32cb2e | 1635 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1636 | break; |
6d61ef40 | 1637 | } |
0608f43d AM |
1638 | mem_cgroup_iter_break(root_memcg, victim); |
1639 | return total; | |
6d61ef40 BS |
1640 | } |
1641 | ||
0056f4e6 JW |
1642 | #ifdef CONFIG_LOCKDEP |
1643 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1644 | .name = "memcg_oom_lock", | |
1645 | }; | |
1646 | #endif | |
1647 | ||
fb2a6fc5 JW |
1648 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1649 | ||
867578cb KH |
1650 | /* |
1651 | * Check OOM-Killer is already running under our hierarchy. | |
1652 | * If someone is running, return false. | |
1653 | */ | |
fb2a6fc5 | 1654 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1655 | { |
79dfdacc | 1656 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1657 | |
fb2a6fc5 JW |
1658 | spin_lock(&memcg_oom_lock); |
1659 | ||
9f3a0d09 | 1660 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1661 | if (iter->oom_lock) { |
79dfdacc MH |
1662 | /* |
1663 | * this subtree of our hierarchy is already locked | |
1664 | * so we cannot give a lock. | |
1665 | */ | |
79dfdacc | 1666 | failed = iter; |
9f3a0d09 JW |
1667 | mem_cgroup_iter_break(memcg, iter); |
1668 | break; | |
23751be0 JW |
1669 | } else |
1670 | iter->oom_lock = true; | |
7d74b06f | 1671 | } |
867578cb | 1672 | |
fb2a6fc5 JW |
1673 | if (failed) { |
1674 | /* | |
1675 | * OK, we failed to lock the whole subtree so we have | |
1676 | * to clean up what we set up to the failing subtree | |
1677 | */ | |
1678 | for_each_mem_cgroup_tree(iter, memcg) { | |
1679 | if (iter == failed) { | |
1680 | mem_cgroup_iter_break(memcg, iter); | |
1681 | break; | |
1682 | } | |
1683 | iter->oom_lock = false; | |
79dfdacc | 1684 | } |
0056f4e6 JW |
1685 | } else |
1686 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1687 | |
1688 | spin_unlock(&memcg_oom_lock); | |
1689 | ||
1690 | return !failed; | |
a636b327 | 1691 | } |
0b7f569e | 1692 | |
fb2a6fc5 | 1693 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1694 | { |
7d74b06f KH |
1695 | struct mem_cgroup *iter; |
1696 | ||
fb2a6fc5 | 1697 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1698 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1699 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1700 | iter->oom_lock = false; |
fb2a6fc5 | 1701 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1702 | } |
1703 | ||
c0ff4b85 | 1704 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1705 | { |
1706 | struct mem_cgroup *iter; | |
1707 | ||
c2b42d3c | 1708 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1709 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1710 | iter->under_oom++; |
1711 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1712 | } |
1713 | ||
c0ff4b85 | 1714 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1715 | { |
1716 | struct mem_cgroup *iter; | |
1717 | ||
867578cb KH |
1718 | /* |
1719 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1720 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1721 | */ |
c2b42d3c | 1722 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1723 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1724 | if (iter->under_oom > 0) |
1725 | iter->under_oom--; | |
1726 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1727 | } |
1728 | ||
867578cb KH |
1729 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1730 | ||
dc98df5a | 1731 | struct oom_wait_info { |
d79154bb | 1732 | struct mem_cgroup *memcg; |
ac6424b9 | 1733 | wait_queue_entry_t wait; |
dc98df5a KH |
1734 | }; |
1735 | ||
ac6424b9 | 1736 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1737 | unsigned mode, int sync, void *arg) |
1738 | { | |
d79154bb HD |
1739 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1740 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1741 | struct oom_wait_info *oom_wait_info; |
1742 | ||
1743 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1744 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1745 | |
2314b42d JW |
1746 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1747 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1748 | return 0; |
dc98df5a KH |
1749 | return autoremove_wake_function(wait, mode, sync, arg); |
1750 | } | |
1751 | ||
c0ff4b85 | 1752 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1753 | { |
c2b42d3c TH |
1754 | /* |
1755 | * For the following lockless ->under_oom test, the only required | |
1756 | * guarantee is that it must see the state asserted by an OOM when | |
1757 | * this function is called as a result of userland actions | |
1758 | * triggered by the notification of the OOM. This is trivially | |
1759 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1760 | * triggering notification. | |
1761 | */ | |
1762 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1763 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1764 | } |
1765 | ||
29ef680a MH |
1766 | enum oom_status { |
1767 | OOM_SUCCESS, | |
1768 | OOM_FAILED, | |
1769 | OOM_ASYNC, | |
1770 | OOM_SKIPPED | |
1771 | }; | |
1772 | ||
1773 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1774 | { |
7056d3a3 MH |
1775 | enum oom_status ret; |
1776 | bool locked; | |
1777 | ||
29ef680a MH |
1778 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1779 | return OOM_SKIPPED; | |
1780 | ||
7a1adfdd RG |
1781 | memcg_memory_event(memcg, MEMCG_OOM); |
1782 | ||
867578cb | 1783 | /* |
49426420 JW |
1784 | * We are in the middle of the charge context here, so we |
1785 | * don't want to block when potentially sitting on a callstack | |
1786 | * that holds all kinds of filesystem and mm locks. | |
1787 | * | |
29ef680a MH |
1788 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1789 | * handling until the charge can succeed; remember the context and put | |
1790 | * the task to sleep at the end of the page fault when all locks are | |
1791 | * released. | |
49426420 | 1792 | * |
29ef680a MH |
1793 | * On the other hand, in-kernel OOM killer allows for an async victim |
1794 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1795 | * relying on the oom victim to make a forward progress and we can | |
1796 | * invoke the oom killer here. | |
1797 | * | |
1798 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1799 | * victim and then we have to bail out from the charge path. | |
867578cb | 1800 | */ |
29ef680a MH |
1801 | if (memcg->oom_kill_disable) { |
1802 | if (!current->in_user_fault) | |
1803 | return OOM_SKIPPED; | |
1804 | css_get(&memcg->css); | |
1805 | current->memcg_in_oom = memcg; | |
1806 | current->memcg_oom_gfp_mask = mask; | |
1807 | current->memcg_oom_order = order; | |
1808 | ||
1809 | return OOM_ASYNC; | |
1810 | } | |
1811 | ||
7056d3a3 MH |
1812 | mem_cgroup_mark_under_oom(memcg); |
1813 | ||
1814 | locked = mem_cgroup_oom_trylock(memcg); | |
1815 | ||
1816 | if (locked) | |
1817 | mem_cgroup_oom_notify(memcg); | |
1818 | ||
1819 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1820 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1821 | ret = OOM_SUCCESS; |
1822 | else | |
1823 | ret = OOM_FAILED; | |
1824 | ||
1825 | if (locked) | |
1826 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1827 | |
7056d3a3 | 1828 | return ret; |
3812c8c8 JW |
1829 | } |
1830 | ||
1831 | /** | |
1832 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1833 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1834 | * |
49426420 JW |
1835 | * This has to be called at the end of a page fault if the memcg OOM |
1836 | * handler was enabled. | |
3812c8c8 | 1837 | * |
49426420 | 1838 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1839 | * sleep on a waitqueue until the userspace task resolves the |
1840 | * situation. Sleeping directly in the charge context with all kinds | |
1841 | * of locks held is not a good idea, instead we remember an OOM state | |
1842 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1843 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1844 | * |
1845 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1846 | * completed, %false otherwise. |
3812c8c8 | 1847 | */ |
49426420 | 1848 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1849 | { |
626ebc41 | 1850 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1851 | struct oom_wait_info owait; |
49426420 | 1852 | bool locked; |
3812c8c8 JW |
1853 | |
1854 | /* OOM is global, do not handle */ | |
3812c8c8 | 1855 | if (!memcg) |
49426420 | 1856 | return false; |
3812c8c8 | 1857 | |
7c5f64f8 | 1858 | if (!handle) |
49426420 | 1859 | goto cleanup; |
3812c8c8 JW |
1860 | |
1861 | owait.memcg = memcg; | |
1862 | owait.wait.flags = 0; | |
1863 | owait.wait.func = memcg_oom_wake_function; | |
1864 | owait.wait.private = current; | |
2055da97 | 1865 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1866 | |
3812c8c8 | 1867 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1868 | mem_cgroup_mark_under_oom(memcg); |
1869 | ||
1870 | locked = mem_cgroup_oom_trylock(memcg); | |
1871 | ||
1872 | if (locked) | |
1873 | mem_cgroup_oom_notify(memcg); | |
1874 | ||
1875 | if (locked && !memcg->oom_kill_disable) { | |
1876 | mem_cgroup_unmark_under_oom(memcg); | |
1877 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1878 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1879 | current->memcg_oom_order); | |
49426420 | 1880 | } else { |
3812c8c8 | 1881 | schedule(); |
49426420 JW |
1882 | mem_cgroup_unmark_under_oom(memcg); |
1883 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1884 | } | |
1885 | ||
1886 | if (locked) { | |
fb2a6fc5 JW |
1887 | mem_cgroup_oom_unlock(memcg); |
1888 | /* | |
1889 | * There is no guarantee that an OOM-lock contender | |
1890 | * sees the wakeups triggered by the OOM kill | |
1891 | * uncharges. Wake any sleepers explicitely. | |
1892 | */ | |
1893 | memcg_oom_recover(memcg); | |
1894 | } | |
49426420 | 1895 | cleanup: |
626ebc41 | 1896 | current->memcg_in_oom = NULL; |
3812c8c8 | 1897 | css_put(&memcg->css); |
867578cb | 1898 | return true; |
0b7f569e KH |
1899 | } |
1900 | ||
3d8b38eb RG |
1901 | /** |
1902 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
1903 | * @victim: task to be killed by the OOM killer | |
1904 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
1905 | * | |
1906 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
1907 | * by killing all belonging OOM-killable tasks. | |
1908 | * | |
1909 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
1910 | */ | |
1911 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
1912 | struct mem_cgroup *oom_domain) | |
1913 | { | |
1914 | struct mem_cgroup *oom_group = NULL; | |
1915 | struct mem_cgroup *memcg; | |
1916 | ||
1917 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
1918 | return NULL; | |
1919 | ||
1920 | if (!oom_domain) | |
1921 | oom_domain = root_mem_cgroup; | |
1922 | ||
1923 | rcu_read_lock(); | |
1924 | ||
1925 | memcg = mem_cgroup_from_task(victim); | |
1926 | if (memcg == root_mem_cgroup) | |
1927 | goto out; | |
1928 | ||
1929 | /* | |
1930 | * Traverse the memory cgroup hierarchy from the victim task's | |
1931 | * cgroup up to the OOMing cgroup (or root) to find the | |
1932 | * highest-level memory cgroup with oom.group set. | |
1933 | */ | |
1934 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
1935 | if (memcg->oom_group) | |
1936 | oom_group = memcg; | |
1937 | ||
1938 | if (memcg == oom_domain) | |
1939 | break; | |
1940 | } | |
1941 | ||
1942 | if (oom_group) | |
1943 | css_get(&oom_group->css); | |
1944 | out: | |
1945 | rcu_read_unlock(); | |
1946 | ||
1947 | return oom_group; | |
1948 | } | |
1949 | ||
1950 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
1951 | { | |
1952 | pr_info("Tasks in "); | |
1953 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1954 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
1955 | } | |
1956 | ||
d7365e78 | 1957 | /** |
81f8c3a4 JW |
1958 | * lock_page_memcg - lock a page->mem_cgroup binding |
1959 | * @page: the page | |
32047e2a | 1960 | * |
81f8c3a4 | 1961 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
1962 | * another cgroup. |
1963 | * | |
1964 | * It ensures lifetime of the returned memcg. Caller is responsible | |
1965 | * for the lifetime of the page; __unlock_page_memcg() is available | |
1966 | * when @page might get freed inside the locked section. | |
d69b042f | 1967 | */ |
739f79fc | 1968 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 KH |
1969 | { |
1970 | struct mem_cgroup *memcg; | |
6de22619 | 1971 | unsigned long flags; |
89c06bd5 | 1972 | |
6de22619 JW |
1973 | /* |
1974 | * The RCU lock is held throughout the transaction. The fast | |
1975 | * path can get away without acquiring the memcg->move_lock | |
1976 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
1977 | * |
1978 | * The RCU lock also protects the memcg from being freed when | |
1979 | * the page state that is going to change is the only thing | |
1980 | * preventing the page itself from being freed. E.g. writeback | |
1981 | * doesn't hold a page reference and relies on PG_writeback to | |
1982 | * keep off truncation, migration and so forth. | |
1983 | */ | |
d7365e78 JW |
1984 | rcu_read_lock(); |
1985 | ||
1986 | if (mem_cgroup_disabled()) | |
739f79fc | 1987 | return NULL; |
89c06bd5 | 1988 | again: |
1306a85a | 1989 | memcg = page->mem_cgroup; |
29833315 | 1990 | if (unlikely(!memcg)) |
739f79fc | 1991 | return NULL; |
d7365e78 | 1992 | |
bdcbb659 | 1993 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 1994 | return memcg; |
89c06bd5 | 1995 | |
6de22619 | 1996 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1997 | if (memcg != page->mem_cgroup) { |
6de22619 | 1998 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1999 | goto again; |
2000 | } | |
6de22619 JW |
2001 | |
2002 | /* | |
2003 | * When charge migration first begins, we can have locked and | |
2004 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 2005 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
2006 | */ |
2007 | memcg->move_lock_task = current; | |
2008 | memcg->move_lock_flags = flags; | |
d7365e78 | 2009 | |
739f79fc | 2010 | return memcg; |
89c06bd5 | 2011 | } |
81f8c3a4 | 2012 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2013 | |
d7365e78 | 2014 | /** |
739f79fc JW |
2015 | * __unlock_page_memcg - unlock and unpin a memcg |
2016 | * @memcg: the memcg | |
2017 | * | |
2018 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2019 | */ |
739f79fc | 2020 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2021 | { |
6de22619 JW |
2022 | if (memcg && memcg->move_lock_task == current) { |
2023 | unsigned long flags = memcg->move_lock_flags; | |
2024 | ||
2025 | memcg->move_lock_task = NULL; | |
2026 | memcg->move_lock_flags = 0; | |
2027 | ||
2028 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2029 | } | |
89c06bd5 | 2030 | |
d7365e78 | 2031 | rcu_read_unlock(); |
89c06bd5 | 2032 | } |
739f79fc JW |
2033 | |
2034 | /** | |
2035 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
2036 | * @page: the page | |
2037 | */ | |
2038 | void unlock_page_memcg(struct page *page) | |
2039 | { | |
2040 | __unlock_page_memcg(page->mem_cgroup); | |
2041 | } | |
81f8c3a4 | 2042 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2043 | |
cdec2e42 KH |
2044 | struct memcg_stock_pcp { |
2045 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2046 | unsigned int nr_pages; |
cdec2e42 | 2047 | struct work_struct work; |
26fe6168 | 2048 | unsigned long flags; |
a0db00fc | 2049 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2050 | }; |
2051 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2052 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2053 | |
a0956d54 SS |
2054 | /** |
2055 | * consume_stock: Try to consume stocked charge on this cpu. | |
2056 | * @memcg: memcg to consume from. | |
2057 | * @nr_pages: how many pages to charge. | |
2058 | * | |
2059 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2060 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2061 | * service an allocation will refill the stock. | |
2062 | * | |
2063 | * returns true if successful, false otherwise. | |
cdec2e42 | 2064 | */ |
a0956d54 | 2065 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2066 | { |
2067 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2068 | unsigned long flags; |
3e32cb2e | 2069 | bool ret = false; |
cdec2e42 | 2070 | |
a983b5eb | 2071 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2072 | return ret; |
a0956d54 | 2073 | |
db2ba40c JW |
2074 | local_irq_save(flags); |
2075 | ||
2076 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2077 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2078 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2079 | ret = true; |
2080 | } | |
db2ba40c JW |
2081 | |
2082 | local_irq_restore(flags); | |
2083 | ||
cdec2e42 KH |
2084 | return ret; |
2085 | } | |
2086 | ||
2087 | /* | |
3e32cb2e | 2088 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2089 | */ |
2090 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2091 | { | |
2092 | struct mem_cgroup *old = stock->cached; | |
2093 | ||
11c9ea4e | 2094 | if (stock->nr_pages) { |
3e32cb2e | 2095 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2096 | if (do_memsw_account()) |
3e32cb2e | 2097 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2098 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2099 | stock->nr_pages = 0; |
cdec2e42 KH |
2100 | } |
2101 | stock->cached = NULL; | |
cdec2e42 KH |
2102 | } |
2103 | ||
cdec2e42 KH |
2104 | static void drain_local_stock(struct work_struct *dummy) |
2105 | { | |
db2ba40c JW |
2106 | struct memcg_stock_pcp *stock; |
2107 | unsigned long flags; | |
2108 | ||
72f0184c MH |
2109 | /* |
2110 | * The only protection from memory hotplug vs. drain_stock races is | |
2111 | * that we always operate on local CPU stock here with IRQ disabled | |
2112 | */ | |
db2ba40c JW |
2113 | local_irq_save(flags); |
2114 | ||
2115 | stock = this_cpu_ptr(&memcg_stock); | |
cdec2e42 | 2116 | drain_stock(stock); |
26fe6168 | 2117 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2118 | |
2119 | local_irq_restore(flags); | |
cdec2e42 KH |
2120 | } |
2121 | ||
2122 | /* | |
3e32cb2e | 2123 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2124 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2125 | */ |
c0ff4b85 | 2126 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2127 | { |
db2ba40c JW |
2128 | struct memcg_stock_pcp *stock; |
2129 | unsigned long flags; | |
2130 | ||
2131 | local_irq_save(flags); | |
cdec2e42 | 2132 | |
db2ba40c | 2133 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2134 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2135 | drain_stock(stock); |
c0ff4b85 | 2136 | stock->cached = memcg; |
cdec2e42 | 2137 | } |
11c9ea4e | 2138 | stock->nr_pages += nr_pages; |
db2ba40c | 2139 | |
a983b5eb | 2140 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2141 | drain_stock(stock); |
2142 | ||
db2ba40c | 2143 | local_irq_restore(flags); |
cdec2e42 KH |
2144 | } |
2145 | ||
2146 | /* | |
c0ff4b85 | 2147 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2148 | * of the hierarchy under it. |
cdec2e42 | 2149 | */ |
6d3d6aa2 | 2150 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2151 | { |
26fe6168 | 2152 | int cpu, curcpu; |
d38144b7 | 2153 | |
6d3d6aa2 JW |
2154 | /* If someone's already draining, avoid adding running more workers. */ |
2155 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2156 | return; | |
72f0184c MH |
2157 | /* |
2158 | * Notify other cpus that system-wide "drain" is running | |
2159 | * We do not care about races with the cpu hotplug because cpu down | |
2160 | * as well as workers from this path always operate on the local | |
2161 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2162 | */ | |
5af12d0e | 2163 | curcpu = get_cpu(); |
cdec2e42 KH |
2164 | for_each_online_cpu(cpu) { |
2165 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2166 | struct mem_cgroup *memcg; |
26fe6168 | 2167 | |
c0ff4b85 | 2168 | memcg = stock->cached; |
72f0184c | 2169 | if (!memcg || !stock->nr_pages || !css_tryget(&memcg->css)) |
26fe6168 | 2170 | continue; |
72f0184c MH |
2171 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) { |
2172 | css_put(&memcg->css); | |
3e92041d | 2173 | continue; |
72f0184c | 2174 | } |
d1a05b69 MH |
2175 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2176 | if (cpu == curcpu) | |
2177 | drain_local_stock(&stock->work); | |
2178 | else | |
2179 | schedule_work_on(cpu, &stock->work); | |
2180 | } | |
72f0184c | 2181 | css_put(&memcg->css); |
cdec2e42 | 2182 | } |
5af12d0e | 2183 | put_cpu(); |
9f50fad6 | 2184 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2185 | } |
2186 | ||
308167fc | 2187 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2188 | { |
cdec2e42 | 2189 | struct memcg_stock_pcp *stock; |
42a30035 | 2190 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2191 | |
cdec2e42 KH |
2192 | stock = &per_cpu(memcg_stock, cpu); |
2193 | drain_stock(stock); | |
a983b5eb JW |
2194 | |
2195 | for_each_mem_cgroup(memcg) { | |
2196 | int i; | |
2197 | ||
2198 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2199 | int nid; | |
2200 | long x; | |
2201 | ||
871789d4 | 2202 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
42a30035 JW |
2203 | if (x) { |
2204 | atomic_long_add(x, &memcg->vmstats_local[i]); | |
2205 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
2206 | atomic_long_add(x, &memcg->vmstats[i]); | |
2207 | } | |
a983b5eb JW |
2208 | |
2209 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2210 | continue; | |
2211 | ||
2212 | for_each_node(nid) { | |
2213 | struct mem_cgroup_per_node *pn; | |
2214 | ||
2215 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2216 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
42a30035 JW |
2217 | if (x) { |
2218 | atomic_long_add(x, &pn->lruvec_stat_local[i]); | |
2219 | do { | |
2220 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2221 | } while ((pn = parent_nodeinfo(pn, nid))); | |
2222 | } | |
a983b5eb JW |
2223 | } |
2224 | } | |
2225 | ||
e27be240 | 2226 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2227 | long x; |
2228 | ||
871789d4 | 2229 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
42a30035 JW |
2230 | if (x) { |
2231 | atomic_long_add(x, &memcg->vmevents_local[i]); | |
2232 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
2233 | atomic_long_add(x, &memcg->vmevents[i]); | |
2234 | } | |
a983b5eb JW |
2235 | } |
2236 | } | |
2237 | ||
308167fc | 2238 | return 0; |
cdec2e42 KH |
2239 | } |
2240 | ||
f7e1cb6e JW |
2241 | static void reclaim_high(struct mem_cgroup *memcg, |
2242 | unsigned int nr_pages, | |
2243 | gfp_t gfp_mask) | |
2244 | { | |
2245 | do { | |
2246 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
2247 | continue; | |
e27be240 | 2248 | memcg_memory_event(memcg, MEMCG_HIGH); |
f7e1cb6e JW |
2249 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
2250 | } while ((memcg = parent_mem_cgroup(memcg))); | |
2251 | } | |
2252 | ||
2253 | static void high_work_func(struct work_struct *work) | |
2254 | { | |
2255 | struct mem_cgroup *memcg; | |
2256 | ||
2257 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2258 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2259 | } |
2260 | ||
b23afb93 TH |
2261 | /* |
2262 | * Scheduled by try_charge() to be executed from the userland return path | |
2263 | * and reclaims memory over the high limit. | |
2264 | */ | |
2265 | void mem_cgroup_handle_over_high(void) | |
2266 | { | |
2267 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 2268 | struct mem_cgroup *memcg; |
b23afb93 TH |
2269 | |
2270 | if (likely(!nr_pages)) | |
2271 | return; | |
2272 | ||
f7e1cb6e JW |
2273 | memcg = get_mem_cgroup_from_mm(current->mm); |
2274 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
2275 | css_put(&memcg->css); |
2276 | current->memcg_nr_pages_over_high = 0; | |
2277 | } | |
2278 | ||
00501b53 JW |
2279 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2280 | unsigned int nr_pages) | |
8a9f3ccd | 2281 | { |
a983b5eb | 2282 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
9b130619 | 2283 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2284 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2285 | struct page_counter *counter; |
6539cc05 | 2286 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2287 | bool may_swap = true; |
2288 | bool drained = false; | |
29ef680a MH |
2289 | bool oomed = false; |
2290 | enum oom_status oom_status; | |
a636b327 | 2291 | |
ce00a967 | 2292 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2293 | return 0; |
6539cc05 | 2294 | retry: |
b6b6cc72 | 2295 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2296 | return 0; |
8a9f3ccd | 2297 | |
7941d214 | 2298 | if (!do_memsw_account() || |
6071ca52 JW |
2299 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2300 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2301 | goto done_restock; |
7941d214 | 2302 | if (do_memsw_account()) |
3e32cb2e JW |
2303 | page_counter_uncharge(&memcg->memsw, batch); |
2304 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2305 | } else { |
3e32cb2e | 2306 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2307 | may_swap = false; |
3fbe7244 | 2308 | } |
7a81b88c | 2309 | |
6539cc05 JW |
2310 | if (batch > nr_pages) { |
2311 | batch = nr_pages; | |
2312 | goto retry; | |
2313 | } | |
6d61ef40 | 2314 | |
06b078fc JW |
2315 | /* |
2316 | * Unlike in global OOM situations, memcg is not in a physical | |
2317 | * memory shortage. Allow dying and OOM-killed tasks to | |
2318 | * bypass the last charges so that they can exit quickly and | |
2319 | * free their memory. | |
2320 | */ | |
7775face | 2321 | if (unlikely(should_force_charge())) |
10d53c74 | 2322 | goto force; |
06b078fc | 2323 | |
89a28483 JW |
2324 | /* |
2325 | * Prevent unbounded recursion when reclaim operations need to | |
2326 | * allocate memory. This might exceed the limits temporarily, | |
2327 | * but we prefer facilitating memory reclaim and getting back | |
2328 | * under the limit over triggering OOM kills in these cases. | |
2329 | */ | |
2330 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2331 | goto force; | |
2332 | ||
06b078fc JW |
2333 | if (unlikely(task_in_memcg_oom(current))) |
2334 | goto nomem; | |
2335 | ||
d0164adc | 2336 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2337 | goto nomem; |
4b534334 | 2338 | |
e27be240 | 2339 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2340 | |
b70a2a21 JW |
2341 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2342 | gfp_mask, may_swap); | |
6539cc05 | 2343 | |
61e02c74 | 2344 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2345 | goto retry; |
28c34c29 | 2346 | |
b70a2a21 | 2347 | if (!drained) { |
6d3d6aa2 | 2348 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2349 | drained = true; |
2350 | goto retry; | |
2351 | } | |
2352 | ||
28c34c29 JW |
2353 | if (gfp_mask & __GFP_NORETRY) |
2354 | goto nomem; | |
6539cc05 JW |
2355 | /* |
2356 | * Even though the limit is exceeded at this point, reclaim | |
2357 | * may have been able to free some pages. Retry the charge | |
2358 | * before killing the task. | |
2359 | * | |
2360 | * Only for regular pages, though: huge pages are rather | |
2361 | * unlikely to succeed so close to the limit, and we fall back | |
2362 | * to regular pages anyway in case of failure. | |
2363 | */ | |
61e02c74 | 2364 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2365 | goto retry; |
2366 | /* | |
2367 | * At task move, charge accounts can be doubly counted. So, it's | |
2368 | * better to wait until the end of task_move if something is going on. | |
2369 | */ | |
2370 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2371 | goto retry; | |
2372 | ||
9b130619 JW |
2373 | if (nr_retries--) |
2374 | goto retry; | |
2375 | ||
29ef680a MH |
2376 | if (gfp_mask & __GFP_RETRY_MAYFAIL && oomed) |
2377 | goto nomem; | |
2378 | ||
06b078fc | 2379 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2380 | goto force; |
06b078fc | 2381 | |
6539cc05 | 2382 | if (fatal_signal_pending(current)) |
10d53c74 | 2383 | goto force; |
6539cc05 | 2384 | |
29ef680a MH |
2385 | /* |
2386 | * keep retrying as long as the memcg oom killer is able to make | |
2387 | * a forward progress or bypass the charge if the oom killer | |
2388 | * couldn't make any progress. | |
2389 | */ | |
2390 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2391 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2392 | switch (oom_status) { |
2393 | case OOM_SUCCESS: | |
2394 | nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
2395 | oomed = true; | |
2396 | goto retry; | |
2397 | case OOM_FAILED: | |
2398 | goto force; | |
2399 | default: | |
2400 | goto nomem; | |
2401 | } | |
7a81b88c | 2402 | nomem: |
6d1fdc48 | 2403 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2404 | return -ENOMEM; |
10d53c74 TH |
2405 | force: |
2406 | /* | |
2407 | * The allocation either can't fail or will lead to more memory | |
2408 | * being freed very soon. Allow memory usage go over the limit | |
2409 | * temporarily by force charging it. | |
2410 | */ | |
2411 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2412 | if (do_memsw_account()) |
10d53c74 TH |
2413 | page_counter_charge(&memcg->memsw, nr_pages); |
2414 | css_get_many(&memcg->css, nr_pages); | |
2415 | ||
2416 | return 0; | |
6539cc05 JW |
2417 | |
2418 | done_restock: | |
e8ea14cc | 2419 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2420 | if (batch > nr_pages) |
2421 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2422 | |
241994ed | 2423 | /* |
b23afb93 TH |
2424 | * If the hierarchy is above the normal consumption range, schedule |
2425 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2426 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2427 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2428 | * not recorded as it most likely matches current's and won't | |
2429 | * change in the meantime. As high limit is checked again before | |
2430 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2431 | */ |
2432 | do { | |
b23afb93 | 2433 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2434 | /* Don't bother a random interrupted task */ |
2435 | if (in_interrupt()) { | |
2436 | schedule_work(&memcg->high_work); | |
2437 | break; | |
2438 | } | |
9516a18a | 2439 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2440 | set_notify_resume(current); |
2441 | break; | |
2442 | } | |
241994ed | 2443 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2444 | |
2445 | return 0; | |
7a81b88c | 2446 | } |
8a9f3ccd | 2447 | |
00501b53 | 2448 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2449 | { |
ce00a967 JW |
2450 | if (mem_cgroup_is_root(memcg)) |
2451 | return; | |
2452 | ||
3e32cb2e | 2453 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2454 | if (do_memsw_account()) |
3e32cb2e | 2455 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2456 | |
e8ea14cc | 2457 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2458 | } |
2459 | ||
0a31bc97 JW |
2460 | static void lock_page_lru(struct page *page, int *isolated) |
2461 | { | |
f4b7e272 | 2462 | pg_data_t *pgdat = page_pgdat(page); |
0a31bc97 | 2463 | |
f4b7e272 | 2464 | spin_lock_irq(&pgdat->lru_lock); |
0a31bc97 JW |
2465 | if (PageLRU(page)) { |
2466 | struct lruvec *lruvec; | |
2467 | ||
f4b7e272 | 2468 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
0a31bc97 JW |
2469 | ClearPageLRU(page); |
2470 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2471 | *isolated = 1; | |
2472 | } else | |
2473 | *isolated = 0; | |
2474 | } | |
2475 | ||
2476 | static void unlock_page_lru(struct page *page, int isolated) | |
2477 | { | |
f4b7e272 | 2478 | pg_data_t *pgdat = page_pgdat(page); |
0a31bc97 JW |
2479 | |
2480 | if (isolated) { | |
2481 | struct lruvec *lruvec; | |
2482 | ||
f4b7e272 | 2483 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
0a31bc97 JW |
2484 | VM_BUG_ON_PAGE(PageLRU(page), page); |
2485 | SetPageLRU(page); | |
2486 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2487 | } | |
f4b7e272 | 2488 | spin_unlock_irq(&pgdat->lru_lock); |
0a31bc97 JW |
2489 | } |
2490 | ||
00501b53 | 2491 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2492 | bool lrucare) |
7a81b88c | 2493 | { |
0a31bc97 | 2494 | int isolated; |
9ce70c02 | 2495 | |
1306a85a | 2496 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2497 | |
2498 | /* | |
2499 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2500 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2501 | */ | |
0a31bc97 JW |
2502 | if (lrucare) |
2503 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2504 | |
0a31bc97 JW |
2505 | /* |
2506 | * Nobody should be changing or seriously looking at | |
1306a85a | 2507 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2508 | * |
2509 | * - the page is uncharged | |
2510 | * | |
2511 | * - the page is off-LRU | |
2512 | * | |
2513 | * - an anonymous fault has exclusive page access, except for | |
2514 | * a locked page table | |
2515 | * | |
2516 | * - a page cache insertion, a swapin fault, or a migration | |
2517 | * have the page locked | |
2518 | */ | |
1306a85a | 2519 | page->mem_cgroup = memcg; |
9ce70c02 | 2520 | |
0a31bc97 JW |
2521 | if (lrucare) |
2522 | unlock_page_lru(page, isolated); | |
7a81b88c | 2523 | } |
66e1707b | 2524 | |
84c07d11 | 2525 | #ifdef CONFIG_MEMCG_KMEM |
f3bb3043 | 2526 | static int memcg_alloc_cache_id(void) |
55007d84 | 2527 | { |
f3bb3043 VD |
2528 | int id, size; |
2529 | int err; | |
2530 | ||
dbcf73e2 | 2531 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2532 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2533 | if (id < 0) | |
2534 | return id; | |
55007d84 | 2535 | |
dbcf73e2 | 2536 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2537 | return id; |
2538 | ||
2539 | /* | |
2540 | * There's no space for the new id in memcg_caches arrays, | |
2541 | * so we have to grow them. | |
2542 | */ | |
05257a1a | 2543 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2544 | |
2545 | size = 2 * (id + 1); | |
55007d84 GC |
2546 | if (size < MEMCG_CACHES_MIN_SIZE) |
2547 | size = MEMCG_CACHES_MIN_SIZE; | |
2548 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2549 | size = MEMCG_CACHES_MAX_SIZE; | |
2550 | ||
f3bb3043 | 2551 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2552 | if (!err) |
2553 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2554 | if (!err) |
2555 | memcg_nr_cache_ids = size; | |
2556 | ||
2557 | up_write(&memcg_cache_ids_sem); | |
2558 | ||
f3bb3043 | 2559 | if (err) { |
dbcf73e2 | 2560 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2561 | return err; |
2562 | } | |
2563 | return id; | |
2564 | } | |
2565 | ||
2566 | static void memcg_free_cache_id(int id) | |
2567 | { | |
dbcf73e2 | 2568 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2569 | } |
2570 | ||
d5b3cf71 | 2571 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2572 | struct mem_cgroup *memcg; |
2573 | struct kmem_cache *cachep; | |
2574 | struct work_struct work; | |
2575 | }; | |
2576 | ||
d5b3cf71 | 2577 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2578 | { |
d5b3cf71 VD |
2579 | struct memcg_kmem_cache_create_work *cw = |
2580 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2581 | struct mem_cgroup *memcg = cw->memcg; |
2582 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2583 | |
d5b3cf71 | 2584 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2585 | |
5722d094 | 2586 | css_put(&memcg->css); |
d7f25f8a GC |
2587 | kfree(cw); |
2588 | } | |
2589 | ||
2590 | /* | |
2591 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2592 | */ |
85cfb245 | 2593 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
d5b3cf71 | 2594 | struct kmem_cache *cachep) |
d7f25f8a | 2595 | { |
d5b3cf71 | 2596 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2597 | |
c892fd82 | 2598 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT | __GFP_NOWARN); |
8135be5a | 2599 | if (!cw) |
d7f25f8a | 2600 | return; |
8135be5a VD |
2601 | |
2602 | css_get(&memcg->css); | |
d7f25f8a GC |
2603 | |
2604 | cw->memcg = memcg; | |
2605 | cw->cachep = cachep; | |
d5b3cf71 | 2606 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2607 | |
17cc4dfe | 2608 | queue_work(memcg_kmem_cache_wq, &cw->work); |
d7f25f8a GC |
2609 | } |
2610 | ||
45264778 VD |
2611 | static inline bool memcg_kmem_bypass(void) |
2612 | { | |
2613 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
2614 | return true; | |
2615 | return false; | |
2616 | } | |
2617 | ||
2618 | /** | |
2619 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation | |
2620 | * @cachep: the original global kmem cache | |
2621 | * | |
d7f25f8a GC |
2622 | * Return the kmem_cache we're supposed to use for a slab allocation. |
2623 | * We try to use the current memcg's version of the cache. | |
2624 | * | |
45264778 VD |
2625 | * If the cache does not exist yet, if we are the first user of it, we |
2626 | * create it asynchronously in a workqueue and let the current allocation | |
2627 | * go through with the original cache. | |
d7f25f8a | 2628 | * |
45264778 VD |
2629 | * This function takes a reference to the cache it returns to assure it |
2630 | * won't get destroyed while we are working with it. Once the caller is | |
2631 | * done with it, memcg_kmem_put_cache() must be called to release the | |
2632 | * reference. | |
d7f25f8a | 2633 | */ |
45264778 | 2634 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2635 | { |
2636 | struct mem_cgroup *memcg; | |
959c8963 | 2637 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2638 | int kmemcg_id; |
d7f25f8a | 2639 | |
f7ce3190 | 2640 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2641 | |
45264778 | 2642 | if (memcg_kmem_bypass()) |
230e9fc2 VD |
2643 | return cachep; |
2644 | ||
d46eb14b | 2645 | memcg = get_mem_cgroup_from_current(); |
4db0c3c2 | 2646 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2647 | if (kmemcg_id < 0) |
ca0dde97 | 2648 | goto out; |
d7f25f8a | 2649 | |
2a4db7eb | 2650 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2651 | if (likely(memcg_cachep)) |
2652 | return memcg_cachep; | |
ca0dde97 LZ |
2653 | |
2654 | /* | |
2655 | * If we are in a safe context (can wait, and not in interrupt | |
2656 | * context), we could be be predictable and return right away. | |
2657 | * This would guarantee that the allocation being performed | |
2658 | * already belongs in the new cache. | |
2659 | * | |
2660 | * However, there are some clashes that can arrive from locking. | |
2661 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2662 | * memcg_create_kmem_cache, this means no further allocation |
2663 | * could happen with the slab_mutex held. So it's better to | |
2664 | * defer everything. | |
ca0dde97 | 2665 | */ |
d5b3cf71 | 2666 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2667 | out: |
8135be5a | 2668 | css_put(&memcg->css); |
ca0dde97 | 2669 | return cachep; |
d7f25f8a | 2670 | } |
d7f25f8a | 2671 | |
45264778 VD |
2672 | /** |
2673 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache | |
2674 | * @cachep: the cache returned by memcg_kmem_get_cache | |
2675 | */ | |
2676 | void memcg_kmem_put_cache(struct kmem_cache *cachep) | |
8135be5a VD |
2677 | { |
2678 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2679 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2680 | } |
2681 | ||
45264778 | 2682 | /** |
60cd4bcd | 2683 | * __memcg_kmem_charge_memcg: charge a kmem page |
45264778 VD |
2684 | * @page: page to charge |
2685 | * @gfp: reclaim mode | |
2686 | * @order: allocation order | |
2687 | * @memcg: memory cgroup to charge | |
2688 | * | |
2689 | * Returns 0 on success, an error code on failure. | |
2690 | */ | |
60cd4bcd | 2691 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
45264778 | 2692 | struct mem_cgroup *memcg) |
7ae1e1d0 | 2693 | { |
f3ccb2c4 VD |
2694 | unsigned int nr_pages = 1 << order; |
2695 | struct page_counter *counter; | |
7ae1e1d0 GC |
2696 | int ret; |
2697 | ||
f3ccb2c4 | 2698 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2699 | if (ret) |
f3ccb2c4 | 2700 | return ret; |
52c29b04 JW |
2701 | |
2702 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2703 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2704 | cancel_charge(memcg, nr_pages); | |
2705 | return -ENOMEM; | |
7ae1e1d0 GC |
2706 | } |
2707 | ||
f3ccb2c4 | 2708 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2709 | |
f3ccb2c4 | 2710 | return 0; |
7ae1e1d0 GC |
2711 | } |
2712 | ||
45264778 | 2713 | /** |
60cd4bcd | 2714 | * __memcg_kmem_charge: charge a kmem page to the current memory cgroup |
45264778 VD |
2715 | * @page: page to charge |
2716 | * @gfp: reclaim mode | |
2717 | * @order: allocation order | |
2718 | * | |
2719 | * Returns 0 on success, an error code on failure. | |
2720 | */ | |
60cd4bcd | 2721 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2722 | { |
f3ccb2c4 | 2723 | struct mem_cgroup *memcg; |
fcff7d7e | 2724 | int ret = 0; |
7ae1e1d0 | 2725 | |
60cd4bcd | 2726 | if (memcg_kmem_bypass()) |
45264778 VD |
2727 | return 0; |
2728 | ||
d46eb14b | 2729 | memcg = get_mem_cgroup_from_current(); |
c4159a75 | 2730 | if (!mem_cgroup_is_root(memcg)) { |
60cd4bcd | 2731 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); |
c4159a75 VD |
2732 | if (!ret) |
2733 | __SetPageKmemcg(page); | |
2734 | } | |
7ae1e1d0 | 2735 | css_put(&memcg->css); |
d05e83a6 | 2736 | return ret; |
7ae1e1d0 | 2737 | } |
45264778 | 2738 | /** |
60cd4bcd | 2739 | * __memcg_kmem_uncharge: uncharge a kmem page |
45264778 VD |
2740 | * @page: page to uncharge |
2741 | * @order: allocation order | |
2742 | */ | |
60cd4bcd | 2743 | void __memcg_kmem_uncharge(struct page *page, int order) |
7ae1e1d0 | 2744 | { |
1306a85a | 2745 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2746 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2747 | |
7ae1e1d0 GC |
2748 | if (!memcg) |
2749 | return; | |
2750 | ||
309381fe | 2751 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2752 | |
52c29b04 JW |
2753 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2754 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2755 | ||
f3ccb2c4 | 2756 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2757 | if (do_memsw_account()) |
f3ccb2c4 | 2758 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2759 | |
1306a85a | 2760 | page->mem_cgroup = NULL; |
c4159a75 VD |
2761 | |
2762 | /* slab pages do not have PageKmemcg flag set */ | |
2763 | if (PageKmemcg(page)) | |
2764 | __ClearPageKmemcg(page); | |
2765 | ||
f3ccb2c4 | 2766 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2767 | } |
84c07d11 | 2768 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 2769 | |
ca3e0214 KH |
2770 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2771 | ||
ca3e0214 KH |
2772 | /* |
2773 | * Because tail pages are not marked as "used", set it. We're under | |
f4b7e272 | 2774 | * pgdat->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2775 | */ |
e94c8a9c | 2776 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2777 | { |
e94c8a9c | 2778 | int i; |
ca3e0214 | 2779 | |
3d37c4a9 KH |
2780 | if (mem_cgroup_disabled()) |
2781 | return; | |
b070e65c | 2782 | |
29833315 | 2783 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2784 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2785 | |
c9019e9b | 2786 | __mod_memcg_state(head->mem_cgroup, MEMCG_RSS_HUGE, -HPAGE_PMD_NR); |
ca3e0214 | 2787 | } |
12d27107 | 2788 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2789 | |
c255a458 | 2790 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
2791 | /** |
2792 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2793 | * @entry: swap entry to be moved | |
2794 | * @from: mem_cgroup which the entry is moved from | |
2795 | * @to: mem_cgroup which the entry is moved to | |
2796 | * | |
2797 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2798 | * as the mem_cgroup's id of @from. | |
2799 | * | |
2800 | * Returns 0 on success, -EINVAL on failure. | |
2801 | * | |
3e32cb2e | 2802 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2803 | * both res and memsw, and called css_get(). |
2804 | */ | |
2805 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2806 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2807 | { |
2808 | unsigned short old_id, new_id; | |
2809 | ||
34c00c31 LZ |
2810 | old_id = mem_cgroup_id(from); |
2811 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2812 | |
2813 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
2814 | mod_memcg_state(from, MEMCG_SWAP, -1); |
2815 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
2816 | return 0; |
2817 | } | |
2818 | return -EINVAL; | |
2819 | } | |
2820 | #else | |
2821 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2822 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2823 | { |
2824 | return -EINVAL; | |
2825 | } | |
8c7c6e34 | 2826 | #endif |
d13d1443 | 2827 | |
bbec2e15 | 2828 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 2829 | |
bbec2e15 RG |
2830 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
2831 | unsigned long max, bool memsw) | |
628f4235 | 2832 | { |
3e32cb2e | 2833 | bool enlarge = false; |
bb4a7ea2 | 2834 | bool drained = false; |
3e32cb2e | 2835 | int ret; |
c054a78c YZ |
2836 | bool limits_invariant; |
2837 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 2838 | |
3e32cb2e | 2839 | do { |
628f4235 KH |
2840 | if (signal_pending(current)) { |
2841 | ret = -EINTR; | |
2842 | break; | |
2843 | } | |
3e32cb2e | 2844 | |
bbec2e15 | 2845 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
2846 | /* |
2847 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 2848 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 2849 | */ |
bbec2e15 RG |
2850 | limits_invariant = memsw ? max >= memcg->memory.max : |
2851 | max <= memcg->memsw.max; | |
c054a78c | 2852 | if (!limits_invariant) { |
bbec2e15 | 2853 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 2854 | ret = -EINVAL; |
8c7c6e34 KH |
2855 | break; |
2856 | } | |
bbec2e15 | 2857 | if (max > counter->max) |
3e32cb2e | 2858 | enlarge = true; |
bbec2e15 RG |
2859 | ret = page_counter_set_max(counter, max); |
2860 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
2861 | |
2862 | if (!ret) | |
2863 | break; | |
2864 | ||
bb4a7ea2 SB |
2865 | if (!drained) { |
2866 | drain_all_stock(memcg); | |
2867 | drained = true; | |
2868 | continue; | |
2869 | } | |
2870 | ||
1ab5c056 AR |
2871 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
2872 | GFP_KERNEL, !memsw)) { | |
2873 | ret = -EBUSY; | |
2874 | break; | |
2875 | } | |
2876 | } while (true); | |
3e32cb2e | 2877 | |
3c11ecf4 KH |
2878 | if (!ret && enlarge) |
2879 | memcg_oom_recover(memcg); | |
3e32cb2e | 2880 | |
628f4235 KH |
2881 | return ret; |
2882 | } | |
2883 | ||
ef8f2327 | 2884 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
2885 | gfp_t gfp_mask, |
2886 | unsigned long *total_scanned) | |
2887 | { | |
2888 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 2889 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
2890 | unsigned long reclaimed; |
2891 | int loop = 0; | |
ef8f2327 | 2892 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 2893 | unsigned long excess; |
0608f43d AM |
2894 | unsigned long nr_scanned; |
2895 | ||
2896 | if (order > 0) | |
2897 | return 0; | |
2898 | ||
ef8f2327 | 2899 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
2900 | |
2901 | /* | |
2902 | * Do not even bother to check the largest node if the root | |
2903 | * is empty. Do it lockless to prevent lock bouncing. Races | |
2904 | * are acceptable as soft limit is best effort anyway. | |
2905 | */ | |
bfc7228b | 2906 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
2907 | return 0; |
2908 | ||
0608f43d AM |
2909 | /* |
2910 | * This loop can run a while, specially if mem_cgroup's continuously | |
2911 | * keep exceeding their soft limit and putting the system under | |
2912 | * pressure | |
2913 | */ | |
2914 | do { | |
2915 | if (next_mz) | |
2916 | mz = next_mz; | |
2917 | else | |
2918 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2919 | if (!mz) | |
2920 | break; | |
2921 | ||
2922 | nr_scanned = 0; | |
ef8f2327 | 2923 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
2924 | gfp_mask, &nr_scanned); |
2925 | nr_reclaimed += reclaimed; | |
2926 | *total_scanned += nr_scanned; | |
0a31bc97 | 2927 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2928 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2929 | |
2930 | /* | |
2931 | * If we failed to reclaim anything from this memory cgroup | |
2932 | * it is time to move on to the next cgroup | |
2933 | */ | |
2934 | next_mz = NULL; | |
bc2f2e7f VD |
2935 | if (!reclaimed) |
2936 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2937 | ||
3e32cb2e | 2938 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2939 | /* |
2940 | * One school of thought says that we should not add | |
2941 | * back the node to the tree if reclaim returns 0. | |
2942 | * But our reclaim could return 0, simply because due | |
2943 | * to priority we are exposing a smaller subset of | |
2944 | * memory to reclaim from. Consider this as a longer | |
2945 | * term TODO. | |
2946 | */ | |
2947 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2948 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2949 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2950 | css_put(&mz->memcg->css); |
2951 | loop++; | |
2952 | /* | |
2953 | * Could not reclaim anything and there are no more | |
2954 | * mem cgroups to try or we seem to be looping without | |
2955 | * reclaiming anything. | |
2956 | */ | |
2957 | if (!nr_reclaimed && | |
2958 | (next_mz == NULL || | |
2959 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2960 | break; | |
2961 | } while (!nr_reclaimed); | |
2962 | if (next_mz) | |
2963 | css_put(&next_mz->memcg->css); | |
2964 | return nr_reclaimed; | |
2965 | } | |
2966 | ||
ea280e7b TH |
2967 | /* |
2968 | * Test whether @memcg has children, dead or alive. Note that this | |
2969 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2970 | * returns %true if there are child csses according to the cgroup | |
2971 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2972 | */ | |
b5f99b53 GC |
2973 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2974 | { | |
ea280e7b TH |
2975 | bool ret; |
2976 | ||
ea280e7b TH |
2977 | rcu_read_lock(); |
2978 | ret = css_next_child(NULL, &memcg->css); | |
2979 | rcu_read_unlock(); | |
2980 | return ret; | |
b5f99b53 GC |
2981 | } |
2982 | ||
c26251f9 | 2983 | /* |
51038171 | 2984 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
2985 | * |
2986 | * Caller is responsible for holding css reference for memcg. | |
2987 | */ | |
2988 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2989 | { | |
2990 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2991 | |
c1e862c1 KH |
2992 | /* we call try-to-free pages for make this cgroup empty */ |
2993 | lru_add_drain_all(); | |
d12c60f6 JS |
2994 | |
2995 | drain_all_stock(memcg); | |
2996 | ||
f817ed48 | 2997 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2998 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2999 | int progress; |
c1e862c1 | 3000 | |
c26251f9 MH |
3001 | if (signal_pending(current)) |
3002 | return -EINTR; | |
3003 | ||
b70a2a21 JW |
3004 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3005 | GFP_KERNEL, true); | |
c1e862c1 | 3006 | if (!progress) { |
f817ed48 | 3007 | nr_retries--; |
c1e862c1 | 3008 | /* maybe some writeback is necessary */ |
8aa7e847 | 3009 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3010 | } |
f817ed48 KH |
3011 | |
3012 | } | |
ab5196c2 MH |
3013 | |
3014 | return 0; | |
cc847582 KH |
3015 | } |
3016 | ||
6770c64e TH |
3017 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3018 | char *buf, size_t nbytes, | |
3019 | loff_t off) | |
c1e862c1 | 3020 | { |
6770c64e | 3021 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3022 | |
d8423011 MH |
3023 | if (mem_cgroup_is_root(memcg)) |
3024 | return -EINVAL; | |
6770c64e | 3025 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3026 | } |
3027 | ||
182446d0 TH |
3028 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3029 | struct cftype *cft) | |
18f59ea7 | 3030 | { |
182446d0 | 3031 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3032 | } |
3033 | ||
182446d0 TH |
3034 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3035 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3036 | { |
3037 | int retval = 0; | |
182446d0 | 3038 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3039 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3040 | |
567fb435 | 3041 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 3042 | return 0; |
567fb435 | 3043 | |
18f59ea7 | 3044 | /* |
af901ca1 | 3045 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3046 | * in the child subtrees. If it is unset, then the change can |
3047 | * occur, provided the current cgroup has no children. | |
3048 | * | |
3049 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3050 | * set if there are no children. | |
3051 | */ | |
c0ff4b85 | 3052 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3053 | (val == 1 || val == 0)) { |
ea280e7b | 3054 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3055 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3056 | else |
3057 | retval = -EBUSY; | |
3058 | } else | |
3059 | retval = -EINVAL; | |
567fb435 | 3060 | |
18f59ea7 BS |
3061 | return retval; |
3062 | } | |
3063 | ||
6f646156 | 3064 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3065 | { |
42a30035 | 3066 | unsigned long val; |
ce00a967 | 3067 | |
3e32cb2e | 3068 | if (mem_cgroup_is_root(memcg)) { |
42a30035 JW |
3069 | val = memcg_page_state(memcg, MEMCG_CACHE) + |
3070 | memcg_page_state(memcg, MEMCG_RSS); | |
3071 | if (swap) | |
3072 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3073 | } else { |
ce00a967 | 3074 | if (!swap) |
3e32cb2e | 3075 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3076 | else |
3e32cb2e | 3077 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3078 | } |
c12176d3 | 3079 | return val; |
ce00a967 JW |
3080 | } |
3081 | ||
3e32cb2e JW |
3082 | enum { |
3083 | RES_USAGE, | |
3084 | RES_LIMIT, | |
3085 | RES_MAX_USAGE, | |
3086 | RES_FAILCNT, | |
3087 | RES_SOFT_LIMIT, | |
3088 | }; | |
ce00a967 | 3089 | |
791badbd | 3090 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3091 | struct cftype *cft) |
8cdea7c0 | 3092 | { |
182446d0 | 3093 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3094 | struct page_counter *counter; |
af36f906 | 3095 | |
3e32cb2e | 3096 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3097 | case _MEM: |
3e32cb2e JW |
3098 | counter = &memcg->memory; |
3099 | break; | |
8c7c6e34 | 3100 | case _MEMSWAP: |
3e32cb2e JW |
3101 | counter = &memcg->memsw; |
3102 | break; | |
510fc4e1 | 3103 | case _KMEM: |
3e32cb2e | 3104 | counter = &memcg->kmem; |
510fc4e1 | 3105 | break; |
d55f90bf | 3106 | case _TCP: |
0db15298 | 3107 | counter = &memcg->tcpmem; |
d55f90bf | 3108 | break; |
8c7c6e34 KH |
3109 | default: |
3110 | BUG(); | |
8c7c6e34 | 3111 | } |
3e32cb2e JW |
3112 | |
3113 | switch (MEMFILE_ATTR(cft->private)) { | |
3114 | case RES_USAGE: | |
3115 | if (counter == &memcg->memory) | |
c12176d3 | 3116 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3117 | if (counter == &memcg->memsw) |
c12176d3 | 3118 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3119 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3120 | case RES_LIMIT: | |
bbec2e15 | 3121 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3122 | case RES_MAX_USAGE: |
3123 | return (u64)counter->watermark * PAGE_SIZE; | |
3124 | case RES_FAILCNT: | |
3125 | return counter->failcnt; | |
3126 | case RES_SOFT_LIMIT: | |
3127 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3128 | default: | |
3129 | BUG(); | |
3130 | } | |
8cdea7c0 | 3131 | } |
510fc4e1 | 3132 | |
84c07d11 | 3133 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3134 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3135 | { |
d6441637 VD |
3136 | int memcg_id; |
3137 | ||
b313aeee VD |
3138 | if (cgroup_memory_nokmem) |
3139 | return 0; | |
3140 | ||
2a4db7eb | 3141 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3142 | BUG_ON(memcg->kmem_state); |
d6441637 | 3143 | |
f3bb3043 | 3144 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3145 | if (memcg_id < 0) |
3146 | return memcg_id; | |
d6441637 | 3147 | |
ef12947c | 3148 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 3149 | /* |
567e9ab2 | 3150 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 3151 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
3152 | * guarantee no one starts accounting before all call sites are |
3153 | * patched. | |
3154 | */ | |
900a38f0 | 3155 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3156 | memcg->kmem_state = KMEM_ONLINE; |
bc2791f8 | 3157 | INIT_LIST_HEAD(&memcg->kmem_caches); |
0b8f73e1 JW |
3158 | |
3159 | return 0; | |
d6441637 VD |
3160 | } |
3161 | ||
8e0a8912 JW |
3162 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3163 | { | |
3164 | struct cgroup_subsys_state *css; | |
3165 | struct mem_cgroup *parent, *child; | |
3166 | int kmemcg_id; | |
3167 | ||
3168 | if (memcg->kmem_state != KMEM_ONLINE) | |
3169 | return; | |
3170 | /* | |
3171 | * Clear the online state before clearing memcg_caches array | |
3172 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
3173 | * guarantees that no cache will be created for this cgroup | |
3174 | * after we are done (see memcg_create_kmem_cache()). | |
3175 | */ | |
3176 | memcg->kmem_state = KMEM_ALLOCATED; | |
3177 | ||
3178 | memcg_deactivate_kmem_caches(memcg); | |
3179 | ||
3180 | kmemcg_id = memcg->kmemcg_id; | |
3181 | BUG_ON(kmemcg_id < 0); | |
3182 | ||
3183 | parent = parent_mem_cgroup(memcg); | |
3184 | if (!parent) | |
3185 | parent = root_mem_cgroup; | |
3186 | ||
3187 | /* | |
3188 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3189 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3190 | * to ones of the parent. After we have finished, all list_lrus | |
3191 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3192 | * ordering is imposed by list_lru_node->lock taken by | |
3193 | * memcg_drain_all_list_lrus(). | |
3194 | */ | |
3a06bb78 | 3195 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3196 | css_for_each_descendant_pre(css, &memcg->css) { |
3197 | child = mem_cgroup_from_css(css); | |
3198 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3199 | child->kmemcg_id = parent->kmemcg_id; | |
3200 | if (!memcg->use_hierarchy) | |
3201 | break; | |
3202 | } | |
3a06bb78 TH |
3203 | rcu_read_unlock(); |
3204 | ||
9bec5c35 | 3205 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3206 | |
3207 | memcg_free_cache_id(kmemcg_id); | |
3208 | } | |
3209 | ||
3210 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3211 | { | |
0b8f73e1 JW |
3212 | /* css_alloc() failed, offlining didn't happen */ |
3213 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3214 | memcg_offline_kmem(memcg); | |
3215 | ||
8e0a8912 JW |
3216 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
3217 | memcg_destroy_kmem_caches(memcg); | |
3218 | static_branch_dec(&memcg_kmem_enabled_key); | |
3219 | WARN_ON(page_counter_read(&memcg->kmem)); | |
3220 | } | |
8e0a8912 | 3221 | } |
d6441637 | 3222 | #else |
0b8f73e1 | 3223 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3224 | { |
3225 | return 0; | |
3226 | } | |
3227 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3228 | { | |
3229 | } | |
3230 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3231 | { | |
3232 | } | |
84c07d11 | 3233 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3234 | |
bbec2e15 RG |
3235 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3236 | unsigned long max) | |
d6441637 | 3237 | { |
b313aeee | 3238 | int ret; |
127424c8 | 3239 | |
bbec2e15 RG |
3240 | mutex_lock(&memcg_max_mutex); |
3241 | ret = page_counter_set_max(&memcg->kmem, max); | |
3242 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3243 | return ret; |
d6441637 | 3244 | } |
510fc4e1 | 3245 | |
bbec2e15 | 3246 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3247 | { |
3248 | int ret; | |
3249 | ||
bbec2e15 | 3250 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3251 | |
bbec2e15 | 3252 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3253 | if (ret) |
3254 | goto out; | |
3255 | ||
0db15298 | 3256 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3257 | /* |
3258 | * The active flag needs to be written after the static_key | |
3259 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3260 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3261 | * for details, and note that we don't mark any socket as | |
3262 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3263 | * |
3264 | * We need to do this, because static_keys will span multiple | |
3265 | * sites, but we can't control their order. If we mark a socket | |
3266 | * as accounted, but the accounting functions are not patched in | |
3267 | * yet, we'll lose accounting. | |
3268 | * | |
2d758073 | 3269 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3270 | * because when this value change, the code to process it is not |
3271 | * patched in yet. | |
3272 | */ | |
3273 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3274 | memcg->tcpmem_active = true; |
d55f90bf VD |
3275 | } |
3276 | out: | |
bbec2e15 | 3277 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3278 | return ret; |
3279 | } | |
d55f90bf | 3280 | |
628f4235 KH |
3281 | /* |
3282 | * The user of this function is... | |
3283 | * RES_LIMIT. | |
3284 | */ | |
451af504 TH |
3285 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3286 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3287 | { |
451af504 | 3288 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3289 | unsigned long nr_pages; |
628f4235 KH |
3290 | int ret; |
3291 | ||
451af504 | 3292 | buf = strstrip(buf); |
650c5e56 | 3293 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3294 | if (ret) |
3295 | return ret; | |
af36f906 | 3296 | |
3e32cb2e | 3297 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3298 | case RES_LIMIT: |
4b3bde4c BS |
3299 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3300 | ret = -EINVAL; | |
3301 | break; | |
3302 | } | |
3e32cb2e JW |
3303 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3304 | case _MEM: | |
bbec2e15 | 3305 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3306 | break; |
3e32cb2e | 3307 | case _MEMSWAP: |
bbec2e15 | 3308 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3309 | break; |
3e32cb2e | 3310 | case _KMEM: |
bbec2e15 | 3311 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3312 | break; |
d55f90bf | 3313 | case _TCP: |
bbec2e15 | 3314 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3315 | break; |
3e32cb2e | 3316 | } |
296c81d8 | 3317 | break; |
3e32cb2e JW |
3318 | case RES_SOFT_LIMIT: |
3319 | memcg->soft_limit = nr_pages; | |
3320 | ret = 0; | |
628f4235 KH |
3321 | break; |
3322 | } | |
451af504 | 3323 | return ret ?: nbytes; |
8cdea7c0 BS |
3324 | } |
3325 | ||
6770c64e TH |
3326 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3327 | size_t nbytes, loff_t off) | |
c84872e1 | 3328 | { |
6770c64e | 3329 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3330 | struct page_counter *counter; |
c84872e1 | 3331 | |
3e32cb2e JW |
3332 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3333 | case _MEM: | |
3334 | counter = &memcg->memory; | |
3335 | break; | |
3336 | case _MEMSWAP: | |
3337 | counter = &memcg->memsw; | |
3338 | break; | |
3339 | case _KMEM: | |
3340 | counter = &memcg->kmem; | |
3341 | break; | |
d55f90bf | 3342 | case _TCP: |
0db15298 | 3343 | counter = &memcg->tcpmem; |
d55f90bf | 3344 | break; |
3e32cb2e JW |
3345 | default: |
3346 | BUG(); | |
3347 | } | |
af36f906 | 3348 | |
3e32cb2e | 3349 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3350 | case RES_MAX_USAGE: |
3e32cb2e | 3351 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3352 | break; |
3353 | case RES_FAILCNT: | |
3e32cb2e | 3354 | counter->failcnt = 0; |
29f2a4da | 3355 | break; |
3e32cb2e JW |
3356 | default: |
3357 | BUG(); | |
29f2a4da | 3358 | } |
f64c3f54 | 3359 | |
6770c64e | 3360 | return nbytes; |
c84872e1 PE |
3361 | } |
3362 | ||
182446d0 | 3363 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3364 | struct cftype *cft) |
3365 | { | |
182446d0 | 3366 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3367 | } |
3368 | ||
02491447 | 3369 | #ifdef CONFIG_MMU |
182446d0 | 3370 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3371 | struct cftype *cft, u64 val) |
3372 | { | |
182446d0 | 3373 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3374 | |
1dfab5ab | 3375 | if (val & ~MOVE_MASK) |
7dc74be0 | 3376 | return -EINVAL; |
ee5e8472 | 3377 | |
7dc74be0 | 3378 | /* |
ee5e8472 GC |
3379 | * No kind of locking is needed in here, because ->can_attach() will |
3380 | * check this value once in the beginning of the process, and then carry | |
3381 | * on with stale data. This means that changes to this value will only | |
3382 | * affect task migrations starting after the change. | |
7dc74be0 | 3383 | */ |
c0ff4b85 | 3384 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3385 | return 0; |
3386 | } | |
02491447 | 3387 | #else |
182446d0 | 3388 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3389 | struct cftype *cft, u64 val) |
3390 | { | |
3391 | return -ENOSYS; | |
3392 | } | |
3393 | #endif | |
7dc74be0 | 3394 | |
406eb0c9 | 3395 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3396 | |
3397 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3398 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3399 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3400 | ||
3401 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
3402 | int nid, unsigned int lru_mask) | |
3403 | { | |
3404 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); | |
3405 | unsigned long nr = 0; | |
3406 | enum lru_list lru; | |
3407 | ||
3408 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3409 | ||
3410 | for_each_lru(lru) { | |
3411 | if (!(BIT(lru) & lru_mask)) | |
3412 | continue; | |
205b20cc | 3413 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); |
113b7dfd JW |
3414 | } |
3415 | return nr; | |
3416 | } | |
3417 | ||
3418 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
3419 | unsigned int lru_mask) | |
3420 | { | |
3421 | unsigned long nr = 0; | |
3422 | enum lru_list lru; | |
3423 | ||
3424 | for_each_lru(lru) { | |
3425 | if (!(BIT(lru) & lru_mask)) | |
3426 | continue; | |
205b20cc | 3427 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); |
113b7dfd JW |
3428 | } |
3429 | return nr; | |
3430 | } | |
3431 | ||
2da8ca82 | 3432 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3433 | { |
25485de6 GT |
3434 | struct numa_stat { |
3435 | const char *name; | |
3436 | unsigned int lru_mask; | |
3437 | }; | |
3438 | ||
3439 | static const struct numa_stat stats[] = { | |
3440 | { "total", LRU_ALL }, | |
3441 | { "file", LRU_ALL_FILE }, | |
3442 | { "anon", LRU_ALL_ANON }, | |
3443 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3444 | }; | |
3445 | const struct numa_stat *stat; | |
406eb0c9 | 3446 | int nid; |
25485de6 | 3447 | unsigned long nr; |
aa9694bb | 3448 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3449 | |
25485de6 GT |
3450 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3451 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3452 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3453 | for_each_node_state(nid, N_MEMORY) { | |
3454 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3455 | stat->lru_mask); | |
3456 | seq_printf(m, " N%d=%lu", nid, nr); | |
3457 | } | |
3458 | seq_putc(m, '\n'); | |
406eb0c9 | 3459 | } |
406eb0c9 | 3460 | |
071aee13 YH |
3461 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3462 | struct mem_cgroup *iter; | |
3463 | ||
3464 | nr = 0; | |
3465 | for_each_mem_cgroup_tree(iter, memcg) | |
3466 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3467 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3468 | for_each_node_state(nid, N_MEMORY) { | |
3469 | nr = 0; | |
3470 | for_each_mem_cgroup_tree(iter, memcg) | |
3471 | nr += mem_cgroup_node_nr_lru_pages( | |
3472 | iter, nid, stat->lru_mask); | |
3473 | seq_printf(m, " N%d=%lu", nid, nr); | |
3474 | } | |
3475 | seq_putc(m, '\n'); | |
406eb0c9 | 3476 | } |
406eb0c9 | 3477 | |
406eb0c9 YH |
3478 | return 0; |
3479 | } | |
3480 | #endif /* CONFIG_NUMA */ | |
3481 | ||
df0e53d0 | 3482 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3483 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
3484 | PGPGIN, |
3485 | PGPGOUT, | |
3486 | PGFAULT, | |
3487 | PGMAJFAULT, | |
3488 | }; | |
3489 | ||
3490 | static const char *const memcg1_event_names[] = { | |
3491 | "pgpgin", | |
3492 | "pgpgout", | |
3493 | "pgfault", | |
3494 | "pgmajfault", | |
3495 | }; | |
3496 | ||
2da8ca82 | 3497 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3498 | { |
aa9694bb | 3499 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 3500 | unsigned long memory, memsw; |
af7c4b0e JW |
3501 | struct mem_cgroup *mi; |
3502 | unsigned int i; | |
406eb0c9 | 3503 | |
71cd3113 | 3504 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c RS |
3505 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3506 | ||
71cd3113 JW |
3507 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
3508 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) | |
1dd3a273 | 3509 | continue; |
71cd3113 | 3510 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], |
205b20cc | 3511 | memcg_page_state_local(memcg, memcg1_stats[i]) * |
71cd3113 | 3512 | PAGE_SIZE); |
1dd3a273 | 3513 | } |
7b854121 | 3514 | |
df0e53d0 JW |
3515 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3516 | seq_printf(m, "%s %lu\n", memcg1_event_names[i], | |
205b20cc | 3517 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
3518 | |
3519 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3520 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
205b20cc | 3521 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 3522 | PAGE_SIZE); |
af7c4b0e | 3523 | |
14067bb3 | 3524 | /* Hierarchical information */ |
3e32cb2e JW |
3525 | memory = memsw = PAGE_COUNTER_MAX; |
3526 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
bbec2e15 RG |
3527 | memory = min(memory, mi->memory.max); |
3528 | memsw = min(memsw, mi->memsw.max); | |
fee7b548 | 3529 | } |
3e32cb2e JW |
3530 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3531 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3532 | if (do_memsw_account()) |
3e32cb2e JW |
3533 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3534 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3535 | |
8de7ecc6 | 3536 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
71cd3113 | 3537 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3538 | continue; |
8de7ecc6 | 3539 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
42a30035 | 3540 | (u64)memcg_page_state(memcg, i) * PAGE_SIZE); |
af7c4b0e JW |
3541 | } |
3542 | ||
8de7ecc6 SB |
3543 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3544 | seq_printf(m, "total_%s %llu\n", memcg1_event_names[i], | |
42a30035 | 3545 | (u64)memcg_events(memcg, i)); |
af7c4b0e | 3546 | |
8de7ecc6 SB |
3547 | for (i = 0; i < NR_LRU_LISTS; i++) |
3548 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], | |
42a30035 JW |
3549 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
3550 | PAGE_SIZE); | |
14067bb3 | 3551 | |
7f016ee8 | 3552 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 3553 | { |
ef8f2327 MG |
3554 | pg_data_t *pgdat; |
3555 | struct mem_cgroup_per_node *mz; | |
89abfab1 | 3556 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3557 | unsigned long recent_rotated[2] = {0, 0}; |
3558 | unsigned long recent_scanned[2] = {0, 0}; | |
3559 | ||
ef8f2327 MG |
3560 | for_each_online_pgdat(pgdat) { |
3561 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
3562 | rstat = &mz->lruvec.reclaim_stat; | |
7f016ee8 | 3563 | |
ef8f2327 MG |
3564 | recent_rotated[0] += rstat->recent_rotated[0]; |
3565 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3566 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3567 | recent_scanned[1] += rstat->recent_scanned[1]; | |
3568 | } | |
78ccf5b5 JW |
3569 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3570 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3571 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3572 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3573 | } |
3574 | #endif | |
3575 | ||
d2ceb9b7 KH |
3576 | return 0; |
3577 | } | |
3578 | ||
182446d0 TH |
3579 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3580 | struct cftype *cft) | |
a7885eb8 | 3581 | { |
182446d0 | 3582 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3583 | |
1f4c025b | 3584 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3585 | } |
3586 | ||
182446d0 TH |
3587 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3588 | struct cftype *cft, u64 val) | |
a7885eb8 | 3589 | { |
182446d0 | 3590 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3591 | |
3dae7fec | 3592 | if (val > 100) |
a7885eb8 KM |
3593 | return -EINVAL; |
3594 | ||
14208b0e | 3595 | if (css->parent) |
3dae7fec JW |
3596 | memcg->swappiness = val; |
3597 | else | |
3598 | vm_swappiness = val; | |
068b38c1 | 3599 | |
a7885eb8 KM |
3600 | return 0; |
3601 | } | |
3602 | ||
2e72b634 KS |
3603 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3604 | { | |
3605 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3606 | unsigned long usage; |
2e72b634 KS |
3607 | int i; |
3608 | ||
3609 | rcu_read_lock(); | |
3610 | if (!swap) | |
2c488db2 | 3611 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3612 | else |
2c488db2 | 3613 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3614 | |
3615 | if (!t) | |
3616 | goto unlock; | |
3617 | ||
ce00a967 | 3618 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3619 | |
3620 | /* | |
748dad36 | 3621 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3622 | * If it's not true, a threshold was crossed after last |
3623 | * call of __mem_cgroup_threshold(). | |
3624 | */ | |
5407a562 | 3625 | i = t->current_threshold; |
2e72b634 KS |
3626 | |
3627 | /* | |
3628 | * Iterate backward over array of thresholds starting from | |
3629 | * current_threshold and check if a threshold is crossed. | |
3630 | * If none of thresholds below usage is crossed, we read | |
3631 | * only one element of the array here. | |
3632 | */ | |
3633 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3634 | eventfd_signal(t->entries[i].eventfd, 1); | |
3635 | ||
3636 | /* i = current_threshold + 1 */ | |
3637 | i++; | |
3638 | ||
3639 | /* | |
3640 | * Iterate forward over array of thresholds starting from | |
3641 | * current_threshold+1 and check if a threshold is crossed. | |
3642 | * If none of thresholds above usage is crossed, we read | |
3643 | * only one element of the array here. | |
3644 | */ | |
3645 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3646 | eventfd_signal(t->entries[i].eventfd, 1); | |
3647 | ||
3648 | /* Update current_threshold */ | |
5407a562 | 3649 | t->current_threshold = i - 1; |
2e72b634 KS |
3650 | unlock: |
3651 | rcu_read_unlock(); | |
3652 | } | |
3653 | ||
3654 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3655 | { | |
ad4ca5f4 KS |
3656 | while (memcg) { |
3657 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3658 | if (do_memsw_account()) |
ad4ca5f4 KS |
3659 | __mem_cgroup_threshold(memcg, true); |
3660 | ||
3661 | memcg = parent_mem_cgroup(memcg); | |
3662 | } | |
2e72b634 KS |
3663 | } |
3664 | ||
3665 | static int compare_thresholds(const void *a, const void *b) | |
3666 | { | |
3667 | const struct mem_cgroup_threshold *_a = a; | |
3668 | const struct mem_cgroup_threshold *_b = b; | |
3669 | ||
2bff24a3 GT |
3670 | if (_a->threshold > _b->threshold) |
3671 | return 1; | |
3672 | ||
3673 | if (_a->threshold < _b->threshold) | |
3674 | return -1; | |
3675 | ||
3676 | return 0; | |
2e72b634 KS |
3677 | } |
3678 | ||
c0ff4b85 | 3679 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3680 | { |
3681 | struct mem_cgroup_eventfd_list *ev; | |
3682 | ||
2bcf2e92 MH |
3683 | spin_lock(&memcg_oom_lock); |
3684 | ||
c0ff4b85 | 3685 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3686 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3687 | |
3688 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3689 | return 0; |
3690 | } | |
3691 | ||
c0ff4b85 | 3692 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3693 | { |
7d74b06f KH |
3694 | struct mem_cgroup *iter; |
3695 | ||
c0ff4b85 | 3696 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3697 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3698 | } |
3699 | ||
59b6f873 | 3700 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3701 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3702 | { |
2c488db2 KS |
3703 | struct mem_cgroup_thresholds *thresholds; |
3704 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3705 | unsigned long threshold; |
3706 | unsigned long usage; | |
2c488db2 | 3707 | int i, size, ret; |
2e72b634 | 3708 | |
650c5e56 | 3709 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3710 | if (ret) |
3711 | return ret; | |
3712 | ||
3713 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3714 | |
05b84301 | 3715 | if (type == _MEM) { |
2c488db2 | 3716 | thresholds = &memcg->thresholds; |
ce00a967 | 3717 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3718 | } else if (type == _MEMSWAP) { |
2c488db2 | 3719 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3720 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3721 | } else |
2e72b634 KS |
3722 | BUG(); |
3723 | ||
2e72b634 | 3724 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3725 | if (thresholds->primary) |
2e72b634 KS |
3726 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3727 | ||
2c488db2 | 3728 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3729 | |
3730 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 3731 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 3732 | if (!new) { |
2e72b634 KS |
3733 | ret = -ENOMEM; |
3734 | goto unlock; | |
3735 | } | |
2c488db2 | 3736 | new->size = size; |
2e72b634 KS |
3737 | |
3738 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3739 | if (thresholds->primary) { |
3740 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3741 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3742 | } |
3743 | ||
2e72b634 | 3744 | /* Add new threshold */ |
2c488db2 KS |
3745 | new->entries[size - 1].eventfd = eventfd; |
3746 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3747 | |
3748 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3749 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3750 | compare_thresholds, NULL); |
3751 | ||
3752 | /* Find current threshold */ | |
2c488db2 | 3753 | new->current_threshold = -1; |
2e72b634 | 3754 | for (i = 0; i < size; i++) { |
748dad36 | 3755 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3756 | /* |
2c488db2 KS |
3757 | * new->current_threshold will not be used until |
3758 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3759 | * it here. |
3760 | */ | |
2c488db2 | 3761 | ++new->current_threshold; |
748dad36 SZ |
3762 | } else |
3763 | break; | |
2e72b634 KS |
3764 | } |
3765 | ||
2c488db2 KS |
3766 | /* Free old spare buffer and save old primary buffer as spare */ |
3767 | kfree(thresholds->spare); | |
3768 | thresholds->spare = thresholds->primary; | |
3769 | ||
3770 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3771 | |
907860ed | 3772 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3773 | synchronize_rcu(); |
3774 | ||
2e72b634 KS |
3775 | unlock: |
3776 | mutex_unlock(&memcg->thresholds_lock); | |
3777 | ||
3778 | return ret; | |
3779 | } | |
3780 | ||
59b6f873 | 3781 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3782 | struct eventfd_ctx *eventfd, const char *args) |
3783 | { | |
59b6f873 | 3784 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3785 | } |
3786 | ||
59b6f873 | 3787 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3788 | struct eventfd_ctx *eventfd, const char *args) |
3789 | { | |
59b6f873 | 3790 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3791 | } |
3792 | ||
59b6f873 | 3793 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3794 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3795 | { |
2c488db2 KS |
3796 | struct mem_cgroup_thresholds *thresholds; |
3797 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3798 | unsigned long usage; |
2c488db2 | 3799 | int i, j, size; |
2e72b634 KS |
3800 | |
3801 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3802 | |
3803 | if (type == _MEM) { | |
2c488db2 | 3804 | thresholds = &memcg->thresholds; |
ce00a967 | 3805 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3806 | } else if (type == _MEMSWAP) { |
2c488db2 | 3807 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3808 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3809 | } else |
2e72b634 KS |
3810 | BUG(); |
3811 | ||
371528ca AV |
3812 | if (!thresholds->primary) |
3813 | goto unlock; | |
3814 | ||
2e72b634 KS |
3815 | /* Check if a threshold crossed before removing */ |
3816 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3817 | ||
3818 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3819 | size = 0; |
3820 | for (i = 0; i < thresholds->primary->size; i++) { | |
3821 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3822 | size++; |
3823 | } | |
3824 | ||
2c488db2 | 3825 | new = thresholds->spare; |
907860ed | 3826 | |
2e72b634 KS |
3827 | /* Set thresholds array to NULL if we don't have thresholds */ |
3828 | if (!size) { | |
2c488db2 KS |
3829 | kfree(new); |
3830 | new = NULL; | |
907860ed | 3831 | goto swap_buffers; |
2e72b634 KS |
3832 | } |
3833 | ||
2c488db2 | 3834 | new->size = size; |
2e72b634 KS |
3835 | |
3836 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3837 | new->current_threshold = -1; |
3838 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3839 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3840 | continue; |
3841 | ||
2c488db2 | 3842 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3843 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3844 | /* |
2c488db2 | 3845 | * new->current_threshold will not be used |
2e72b634 KS |
3846 | * until rcu_assign_pointer(), so it's safe to increment |
3847 | * it here. | |
3848 | */ | |
2c488db2 | 3849 | ++new->current_threshold; |
2e72b634 KS |
3850 | } |
3851 | j++; | |
3852 | } | |
3853 | ||
907860ed | 3854 | swap_buffers: |
2c488db2 KS |
3855 | /* Swap primary and spare array */ |
3856 | thresholds->spare = thresholds->primary; | |
8c757763 | 3857 | |
2c488db2 | 3858 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3859 | |
907860ed | 3860 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3861 | synchronize_rcu(); |
6611d8d7 MC |
3862 | |
3863 | /* If all events are unregistered, free the spare array */ | |
3864 | if (!new) { | |
3865 | kfree(thresholds->spare); | |
3866 | thresholds->spare = NULL; | |
3867 | } | |
371528ca | 3868 | unlock: |
2e72b634 | 3869 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3870 | } |
c1e862c1 | 3871 | |
59b6f873 | 3872 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3873 | struct eventfd_ctx *eventfd) |
3874 | { | |
59b6f873 | 3875 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3876 | } |
3877 | ||
59b6f873 | 3878 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3879 | struct eventfd_ctx *eventfd) |
3880 | { | |
59b6f873 | 3881 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3882 | } |
3883 | ||
59b6f873 | 3884 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3885 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3886 | { |
9490ff27 | 3887 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3888 | |
9490ff27 KH |
3889 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3890 | if (!event) | |
3891 | return -ENOMEM; | |
3892 | ||
1af8efe9 | 3893 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3894 | |
3895 | event->eventfd = eventfd; | |
3896 | list_add(&event->list, &memcg->oom_notify); | |
3897 | ||
3898 | /* already in OOM ? */ | |
c2b42d3c | 3899 | if (memcg->under_oom) |
9490ff27 | 3900 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3901 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3902 | |
3903 | return 0; | |
3904 | } | |
3905 | ||
59b6f873 | 3906 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3907 | struct eventfd_ctx *eventfd) |
9490ff27 | 3908 | { |
9490ff27 | 3909 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3910 | |
1af8efe9 | 3911 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3912 | |
c0ff4b85 | 3913 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3914 | if (ev->eventfd == eventfd) { |
3915 | list_del(&ev->list); | |
3916 | kfree(ev); | |
3917 | } | |
3918 | } | |
3919 | ||
1af8efe9 | 3920 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3921 | } |
3922 | ||
2da8ca82 | 3923 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3924 | { |
aa9694bb | 3925 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 3926 | |
791badbd | 3927 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3928 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
3929 | seq_printf(sf, "oom_kill %lu\n", |
3930 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
3931 | return 0; |
3932 | } | |
3933 | ||
182446d0 | 3934 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3935 | struct cftype *cft, u64 val) |
3936 | { | |
182446d0 | 3937 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3938 | |
3939 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3940 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3941 | return -EINVAL; |
3942 | ||
c0ff4b85 | 3943 | memcg->oom_kill_disable = val; |
4d845ebf | 3944 | if (!val) |
c0ff4b85 | 3945 | memcg_oom_recover(memcg); |
3dae7fec | 3946 | |
3c11ecf4 KH |
3947 | return 0; |
3948 | } | |
3949 | ||
52ebea74 TH |
3950 | #ifdef CONFIG_CGROUP_WRITEBACK |
3951 | ||
841710aa TH |
3952 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3953 | { | |
3954 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3955 | } | |
3956 | ||
3957 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3958 | { | |
3959 | wb_domain_exit(&memcg->cgwb_domain); | |
3960 | } | |
3961 | ||
2529bb3a TH |
3962 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3963 | { | |
3964 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3965 | } | |
3966 | ||
841710aa TH |
3967 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3968 | { | |
3969 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3970 | ||
3971 | if (!memcg->css.parent) | |
3972 | return NULL; | |
3973 | ||
3974 | return &memcg->cgwb_domain; | |
3975 | } | |
3976 | ||
0b3d6e6f GT |
3977 | /* |
3978 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
3979 | * Keep in sync with memcg_exact_page(). | |
3980 | */ | |
3981 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
3982 | { | |
871789d4 | 3983 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
3984 | int cpu; |
3985 | ||
3986 | for_each_online_cpu(cpu) | |
871789d4 | 3987 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
3988 | if (x < 0) |
3989 | x = 0; | |
3990 | return x; | |
3991 | } | |
3992 | ||
c2aa723a TH |
3993 | /** |
3994 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3995 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3996 | * @pfilepages: out parameter for number of file pages |
3997 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3998 | * @pdirty: out parameter for number of dirty pages |
3999 | * @pwriteback: out parameter for number of pages under writeback | |
4000 | * | |
c5edf9cd TH |
4001 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4002 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4003 | * is a bit more involved. | |
c2aa723a | 4004 | * |
c5edf9cd TH |
4005 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4006 | * headroom is calculated as the lowest headroom of itself and the | |
4007 | * ancestors. Note that this doesn't consider the actual amount of | |
4008 | * available memory in the system. The caller should further cap | |
4009 | * *@pheadroom accordingly. | |
c2aa723a | 4010 | */ |
c5edf9cd TH |
4011 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4012 | unsigned long *pheadroom, unsigned long *pdirty, | |
4013 | unsigned long *pwriteback) | |
c2aa723a TH |
4014 | { |
4015 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4016 | struct mem_cgroup *parent; | |
c2aa723a | 4017 | |
0b3d6e6f | 4018 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a TH |
4019 | |
4020 | /* this should eventually include NR_UNSTABLE_NFS */ | |
0b3d6e6f | 4021 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4022 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4023 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4024 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4025 | |
c2aa723a | 4026 | while ((parent = parent_mem_cgroup(memcg))) { |
bbec2e15 | 4027 | unsigned long ceiling = min(memcg->memory.max, memcg->high); |
c2aa723a TH |
4028 | unsigned long used = page_counter_read(&memcg->memory); |
4029 | ||
c5edf9cd | 4030 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4031 | memcg = parent; |
4032 | } | |
c2aa723a TH |
4033 | } |
4034 | ||
841710aa TH |
4035 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4036 | ||
4037 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4038 | { | |
4039 | return 0; | |
4040 | } | |
4041 | ||
4042 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4043 | { | |
4044 | } | |
4045 | ||
2529bb3a TH |
4046 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4047 | { | |
4048 | } | |
4049 | ||
52ebea74 TH |
4050 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4051 | ||
3bc942f3 TH |
4052 | /* |
4053 | * DO NOT USE IN NEW FILES. | |
4054 | * | |
4055 | * "cgroup.event_control" implementation. | |
4056 | * | |
4057 | * This is way over-engineered. It tries to support fully configurable | |
4058 | * events for each user. Such level of flexibility is completely | |
4059 | * unnecessary especially in the light of the planned unified hierarchy. | |
4060 | * | |
4061 | * Please deprecate this and replace with something simpler if at all | |
4062 | * possible. | |
4063 | */ | |
4064 | ||
79bd9814 TH |
4065 | /* |
4066 | * Unregister event and free resources. | |
4067 | * | |
4068 | * Gets called from workqueue. | |
4069 | */ | |
3bc942f3 | 4070 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4071 | { |
3bc942f3 TH |
4072 | struct mem_cgroup_event *event = |
4073 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4074 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4075 | |
4076 | remove_wait_queue(event->wqh, &event->wait); | |
4077 | ||
59b6f873 | 4078 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4079 | |
4080 | /* Notify userspace the event is going away. */ | |
4081 | eventfd_signal(event->eventfd, 1); | |
4082 | ||
4083 | eventfd_ctx_put(event->eventfd); | |
4084 | kfree(event); | |
59b6f873 | 4085 | css_put(&memcg->css); |
79bd9814 TH |
4086 | } |
4087 | ||
4088 | /* | |
a9a08845 | 4089 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4090 | * |
4091 | * Called with wqh->lock held and interrupts disabled. | |
4092 | */ | |
ac6424b9 | 4093 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4094 | int sync, void *key) |
79bd9814 | 4095 | { |
3bc942f3 TH |
4096 | struct mem_cgroup_event *event = |
4097 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4098 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4099 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4100 | |
a9a08845 | 4101 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4102 | /* |
4103 | * If the event has been detached at cgroup removal, we | |
4104 | * can simply return knowing the other side will cleanup | |
4105 | * for us. | |
4106 | * | |
4107 | * We can't race against event freeing since the other | |
4108 | * side will require wqh->lock via remove_wait_queue(), | |
4109 | * which we hold. | |
4110 | */ | |
fba94807 | 4111 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4112 | if (!list_empty(&event->list)) { |
4113 | list_del_init(&event->list); | |
4114 | /* | |
4115 | * We are in atomic context, but cgroup_event_remove() | |
4116 | * may sleep, so we have to call it in workqueue. | |
4117 | */ | |
4118 | schedule_work(&event->remove); | |
4119 | } | |
fba94807 | 4120 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4121 | } |
4122 | ||
4123 | return 0; | |
4124 | } | |
4125 | ||
3bc942f3 | 4126 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4127 | wait_queue_head_t *wqh, poll_table *pt) |
4128 | { | |
3bc942f3 TH |
4129 | struct mem_cgroup_event *event = |
4130 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4131 | |
4132 | event->wqh = wqh; | |
4133 | add_wait_queue(wqh, &event->wait); | |
4134 | } | |
4135 | ||
4136 | /* | |
3bc942f3 TH |
4137 | * DO NOT USE IN NEW FILES. |
4138 | * | |
79bd9814 TH |
4139 | * Parse input and register new cgroup event handler. |
4140 | * | |
4141 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4142 | * Interpretation of args is defined by control file implementation. | |
4143 | */ | |
451af504 TH |
4144 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4145 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4146 | { |
451af504 | 4147 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4148 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4149 | struct mem_cgroup_event *event; |
79bd9814 TH |
4150 | struct cgroup_subsys_state *cfile_css; |
4151 | unsigned int efd, cfd; | |
4152 | struct fd efile; | |
4153 | struct fd cfile; | |
fba94807 | 4154 | const char *name; |
79bd9814 TH |
4155 | char *endp; |
4156 | int ret; | |
4157 | ||
451af504 TH |
4158 | buf = strstrip(buf); |
4159 | ||
4160 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4161 | if (*endp != ' ') |
4162 | return -EINVAL; | |
451af504 | 4163 | buf = endp + 1; |
79bd9814 | 4164 | |
451af504 | 4165 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4166 | if ((*endp != ' ') && (*endp != '\0')) |
4167 | return -EINVAL; | |
451af504 | 4168 | buf = endp + 1; |
79bd9814 TH |
4169 | |
4170 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4171 | if (!event) | |
4172 | return -ENOMEM; | |
4173 | ||
59b6f873 | 4174 | event->memcg = memcg; |
79bd9814 | 4175 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4176 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4177 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4178 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4179 | |
4180 | efile = fdget(efd); | |
4181 | if (!efile.file) { | |
4182 | ret = -EBADF; | |
4183 | goto out_kfree; | |
4184 | } | |
4185 | ||
4186 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4187 | if (IS_ERR(event->eventfd)) { | |
4188 | ret = PTR_ERR(event->eventfd); | |
4189 | goto out_put_efile; | |
4190 | } | |
4191 | ||
4192 | cfile = fdget(cfd); | |
4193 | if (!cfile.file) { | |
4194 | ret = -EBADF; | |
4195 | goto out_put_eventfd; | |
4196 | } | |
4197 | ||
4198 | /* the process need read permission on control file */ | |
4199 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4200 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4201 | if (ret < 0) | |
4202 | goto out_put_cfile; | |
4203 | ||
fba94807 TH |
4204 | /* |
4205 | * Determine the event callbacks and set them in @event. This used | |
4206 | * to be done via struct cftype but cgroup core no longer knows | |
4207 | * about these events. The following is crude but the whole thing | |
4208 | * is for compatibility anyway. | |
3bc942f3 TH |
4209 | * |
4210 | * DO NOT ADD NEW FILES. | |
fba94807 | 4211 | */ |
b583043e | 4212 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4213 | |
4214 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4215 | event->register_event = mem_cgroup_usage_register_event; | |
4216 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4217 | } else if (!strcmp(name, "memory.oom_control")) { | |
4218 | event->register_event = mem_cgroup_oom_register_event; | |
4219 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4220 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4221 | event->register_event = vmpressure_register_event; | |
4222 | event->unregister_event = vmpressure_unregister_event; | |
4223 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4224 | event->register_event = memsw_cgroup_usage_register_event; |
4225 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4226 | } else { |
4227 | ret = -EINVAL; | |
4228 | goto out_put_cfile; | |
4229 | } | |
4230 | ||
79bd9814 | 4231 | /* |
b5557c4c TH |
4232 | * Verify @cfile should belong to @css. Also, remaining events are |
4233 | * automatically removed on cgroup destruction but the removal is | |
4234 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4235 | */ |
b583043e | 4236 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4237 | &memory_cgrp_subsys); |
79bd9814 | 4238 | ret = -EINVAL; |
5a17f543 | 4239 | if (IS_ERR(cfile_css)) |
79bd9814 | 4240 | goto out_put_cfile; |
5a17f543 TH |
4241 | if (cfile_css != css) { |
4242 | css_put(cfile_css); | |
79bd9814 | 4243 | goto out_put_cfile; |
5a17f543 | 4244 | } |
79bd9814 | 4245 | |
451af504 | 4246 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4247 | if (ret) |
4248 | goto out_put_css; | |
4249 | ||
9965ed17 | 4250 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4251 | |
fba94807 TH |
4252 | spin_lock(&memcg->event_list_lock); |
4253 | list_add(&event->list, &memcg->event_list); | |
4254 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4255 | |
4256 | fdput(cfile); | |
4257 | fdput(efile); | |
4258 | ||
451af504 | 4259 | return nbytes; |
79bd9814 TH |
4260 | |
4261 | out_put_css: | |
b5557c4c | 4262 | css_put(css); |
79bd9814 TH |
4263 | out_put_cfile: |
4264 | fdput(cfile); | |
4265 | out_put_eventfd: | |
4266 | eventfd_ctx_put(event->eventfd); | |
4267 | out_put_efile: | |
4268 | fdput(efile); | |
4269 | out_kfree: | |
4270 | kfree(event); | |
4271 | ||
4272 | return ret; | |
4273 | } | |
4274 | ||
241994ed | 4275 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4276 | { |
0eea1030 | 4277 | .name = "usage_in_bytes", |
8c7c6e34 | 4278 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4279 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4280 | }, |
c84872e1 PE |
4281 | { |
4282 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4283 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4284 | .write = mem_cgroup_reset, |
791badbd | 4285 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4286 | }, |
8cdea7c0 | 4287 | { |
0eea1030 | 4288 | .name = "limit_in_bytes", |
8c7c6e34 | 4289 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4290 | .write = mem_cgroup_write, |
791badbd | 4291 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4292 | }, |
296c81d8 BS |
4293 | { |
4294 | .name = "soft_limit_in_bytes", | |
4295 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4296 | .write = mem_cgroup_write, |
791badbd | 4297 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4298 | }, |
8cdea7c0 BS |
4299 | { |
4300 | .name = "failcnt", | |
8c7c6e34 | 4301 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4302 | .write = mem_cgroup_reset, |
791badbd | 4303 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4304 | }, |
d2ceb9b7 KH |
4305 | { |
4306 | .name = "stat", | |
2da8ca82 | 4307 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4308 | }, |
c1e862c1 KH |
4309 | { |
4310 | .name = "force_empty", | |
6770c64e | 4311 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4312 | }, |
18f59ea7 BS |
4313 | { |
4314 | .name = "use_hierarchy", | |
4315 | .write_u64 = mem_cgroup_hierarchy_write, | |
4316 | .read_u64 = mem_cgroup_hierarchy_read, | |
4317 | }, | |
79bd9814 | 4318 | { |
3bc942f3 | 4319 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4320 | .write = memcg_write_event_control, |
7dbdb199 | 4321 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4322 | }, |
a7885eb8 KM |
4323 | { |
4324 | .name = "swappiness", | |
4325 | .read_u64 = mem_cgroup_swappiness_read, | |
4326 | .write_u64 = mem_cgroup_swappiness_write, | |
4327 | }, | |
7dc74be0 DN |
4328 | { |
4329 | .name = "move_charge_at_immigrate", | |
4330 | .read_u64 = mem_cgroup_move_charge_read, | |
4331 | .write_u64 = mem_cgroup_move_charge_write, | |
4332 | }, | |
9490ff27 KH |
4333 | { |
4334 | .name = "oom_control", | |
2da8ca82 | 4335 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4336 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4337 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4338 | }, | |
70ddf637 AV |
4339 | { |
4340 | .name = "pressure_level", | |
70ddf637 | 4341 | }, |
406eb0c9 YH |
4342 | #ifdef CONFIG_NUMA |
4343 | { | |
4344 | .name = "numa_stat", | |
2da8ca82 | 4345 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4346 | }, |
4347 | #endif | |
510fc4e1 GC |
4348 | { |
4349 | .name = "kmem.limit_in_bytes", | |
4350 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4351 | .write = mem_cgroup_write, |
791badbd | 4352 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4353 | }, |
4354 | { | |
4355 | .name = "kmem.usage_in_bytes", | |
4356 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4357 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4358 | }, |
4359 | { | |
4360 | .name = "kmem.failcnt", | |
4361 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4362 | .write = mem_cgroup_reset, |
791badbd | 4363 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4364 | }, |
4365 | { | |
4366 | .name = "kmem.max_usage_in_bytes", | |
4367 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4368 | .write = mem_cgroup_reset, |
791badbd | 4369 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4370 | }, |
5b365771 | 4371 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
749c5415 GC |
4372 | { |
4373 | .name = "kmem.slabinfo", | |
bc2791f8 TH |
4374 | .seq_start = memcg_slab_start, |
4375 | .seq_next = memcg_slab_next, | |
4376 | .seq_stop = memcg_slab_stop, | |
b047501c | 4377 | .seq_show = memcg_slab_show, |
749c5415 GC |
4378 | }, |
4379 | #endif | |
d55f90bf VD |
4380 | { |
4381 | .name = "kmem.tcp.limit_in_bytes", | |
4382 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4383 | .write = mem_cgroup_write, | |
4384 | .read_u64 = mem_cgroup_read_u64, | |
4385 | }, | |
4386 | { | |
4387 | .name = "kmem.tcp.usage_in_bytes", | |
4388 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4389 | .read_u64 = mem_cgroup_read_u64, | |
4390 | }, | |
4391 | { | |
4392 | .name = "kmem.tcp.failcnt", | |
4393 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4394 | .write = mem_cgroup_reset, | |
4395 | .read_u64 = mem_cgroup_read_u64, | |
4396 | }, | |
4397 | { | |
4398 | .name = "kmem.tcp.max_usage_in_bytes", | |
4399 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4400 | .write = mem_cgroup_reset, | |
4401 | .read_u64 = mem_cgroup_read_u64, | |
4402 | }, | |
6bc10349 | 4403 | { }, /* terminate */ |
af36f906 | 4404 | }; |
8c7c6e34 | 4405 | |
73f576c0 JW |
4406 | /* |
4407 | * Private memory cgroup IDR | |
4408 | * | |
4409 | * Swap-out records and page cache shadow entries need to store memcg | |
4410 | * references in constrained space, so we maintain an ID space that is | |
4411 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4412 | * memory-controlled cgroups to 64k. | |
4413 | * | |
4414 | * However, there usually are many references to the oflline CSS after | |
4415 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4416 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4417 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4418 | * relatively small ID space and prevent the creation of new cgroups | |
4419 | * even when there are much fewer than 64k cgroups - possibly none. | |
4420 | * | |
4421 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4422 | * be freed and recycled when it's no longer needed, which is usually | |
4423 | * when the CSS is offlined. | |
4424 | * | |
4425 | * The only exception to that are records of swapped out tmpfs/shmem | |
4426 | * pages that need to be attributed to live ancestors on swapin. But | |
4427 | * those references are manageable from userspace. | |
4428 | */ | |
4429 | ||
4430 | static DEFINE_IDR(mem_cgroup_idr); | |
4431 | ||
7e97de0b KT |
4432 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
4433 | { | |
4434 | if (memcg->id.id > 0) { | |
4435 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
4436 | memcg->id.id = 0; | |
4437 | } | |
4438 | } | |
4439 | ||
615d66c3 | 4440 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4441 | { |
1c2d479a | 4442 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
4443 | } |
4444 | ||
615d66c3 | 4445 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4446 | { |
1c2d479a | 4447 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 4448 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
4449 | |
4450 | /* Memcg ID pins CSS */ | |
4451 | css_put(&memcg->css); | |
4452 | } | |
4453 | } | |
4454 | ||
615d66c3 VD |
4455 | static inline void mem_cgroup_id_get(struct mem_cgroup *memcg) |
4456 | { | |
4457 | mem_cgroup_id_get_many(memcg, 1); | |
4458 | } | |
4459 | ||
4460 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) | |
4461 | { | |
4462 | mem_cgroup_id_put_many(memcg, 1); | |
4463 | } | |
4464 | ||
73f576c0 JW |
4465 | /** |
4466 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4467 | * @id: the memcg id to look up | |
4468 | * | |
4469 | * Caller must hold rcu_read_lock(). | |
4470 | */ | |
4471 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4472 | { | |
4473 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4474 | return idr_find(&mem_cgroup_idr, id); | |
4475 | } | |
4476 | ||
ef8f2327 | 4477 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4478 | { |
4479 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 4480 | int tmp = node; |
1ecaab2b KH |
4481 | /* |
4482 | * This routine is called against possible nodes. | |
4483 | * But it's BUG to call kmalloc() against offline node. | |
4484 | * | |
4485 | * TODO: this routine can waste much memory for nodes which will | |
4486 | * never be onlined. It's better to use memory hotplug callback | |
4487 | * function. | |
4488 | */ | |
41e3355d KH |
4489 | if (!node_state(node, N_NORMAL_MEMORY)) |
4490 | tmp = -1; | |
17295c88 | 4491 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4492 | if (!pn) |
4493 | return 1; | |
1ecaab2b | 4494 | |
a983b5eb JW |
4495 | pn->lruvec_stat_cpu = alloc_percpu(struct lruvec_stat); |
4496 | if (!pn->lruvec_stat_cpu) { | |
00f3ca2c JW |
4497 | kfree(pn); |
4498 | return 1; | |
4499 | } | |
4500 | ||
ef8f2327 MG |
4501 | lruvec_init(&pn->lruvec); |
4502 | pn->usage_in_excess = 0; | |
4503 | pn->on_tree = false; | |
4504 | pn->memcg = memcg; | |
4505 | ||
54f72fe0 | 4506 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4507 | return 0; |
4508 | } | |
4509 | ||
ef8f2327 | 4510 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4511 | { |
00f3ca2c JW |
4512 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
4513 | ||
4eaf431f MH |
4514 | if (!pn) |
4515 | return; | |
4516 | ||
a983b5eb | 4517 | free_percpu(pn->lruvec_stat_cpu); |
00f3ca2c | 4518 | kfree(pn); |
1ecaab2b KH |
4519 | } |
4520 | ||
40e952f9 | 4521 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4522 | { |
c8b2a36f | 4523 | int node; |
59927fb9 | 4524 | |
c8b2a36f | 4525 | for_each_node(node) |
ef8f2327 | 4526 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 4527 | free_percpu(memcg->vmstats_percpu); |
8ff69e2c | 4528 | kfree(memcg); |
59927fb9 | 4529 | } |
3afe36b1 | 4530 | |
40e952f9 TE |
4531 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
4532 | { | |
4533 | memcg_wb_domain_exit(memcg); | |
4534 | __mem_cgroup_free(memcg); | |
4535 | } | |
4536 | ||
0b8f73e1 | 4537 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4538 | { |
d142e3e6 | 4539 | struct mem_cgroup *memcg; |
b9726c26 | 4540 | unsigned int size; |
6d12e2d8 | 4541 | int node; |
8cdea7c0 | 4542 | |
0b8f73e1 JW |
4543 | size = sizeof(struct mem_cgroup); |
4544 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4545 | ||
4546 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4547 | if (!memcg) |
0b8f73e1 JW |
4548 | return NULL; |
4549 | ||
73f576c0 JW |
4550 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
4551 | 1, MEM_CGROUP_ID_MAX, | |
4552 | GFP_KERNEL); | |
4553 | if (memcg->id.id < 0) | |
4554 | goto fail; | |
4555 | ||
871789d4 CD |
4556 | memcg->vmstats_percpu = alloc_percpu(struct memcg_vmstats_percpu); |
4557 | if (!memcg->vmstats_percpu) | |
0b8f73e1 | 4558 | goto fail; |
78fb7466 | 4559 | |
3ed28fa1 | 4560 | for_each_node(node) |
ef8f2327 | 4561 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 4562 | goto fail; |
f64c3f54 | 4563 | |
0b8f73e1 JW |
4564 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4565 | goto fail; | |
28dbc4b6 | 4566 | |
f7e1cb6e | 4567 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4568 | memcg->last_scanned_node = MAX_NUMNODES; |
4569 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4570 | mutex_init(&memcg->thresholds_lock); |
4571 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4572 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4573 | INIT_LIST_HEAD(&memcg->event_list); |
4574 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4575 | memcg->socket_pressure = jiffies; |
84c07d11 | 4576 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 4577 | memcg->kmemcg_id = -1; |
900a38f0 | 4578 | #endif |
52ebea74 TH |
4579 | #ifdef CONFIG_CGROUP_WRITEBACK |
4580 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4581 | #endif | |
73f576c0 | 4582 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
4583 | return memcg; |
4584 | fail: | |
7e97de0b | 4585 | mem_cgroup_id_remove(memcg); |
40e952f9 | 4586 | __mem_cgroup_free(memcg); |
0b8f73e1 | 4587 | return NULL; |
d142e3e6 GC |
4588 | } |
4589 | ||
0b8f73e1 JW |
4590 | static struct cgroup_subsys_state * __ref |
4591 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4592 | { |
0b8f73e1 JW |
4593 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4594 | struct mem_cgroup *memcg; | |
4595 | long error = -ENOMEM; | |
d142e3e6 | 4596 | |
0b8f73e1 JW |
4597 | memcg = mem_cgroup_alloc(); |
4598 | if (!memcg) | |
4599 | return ERR_PTR(error); | |
d142e3e6 | 4600 | |
0b8f73e1 JW |
4601 | memcg->high = PAGE_COUNTER_MAX; |
4602 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4603 | if (parent) { | |
4604 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4605 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4606 | } | |
4607 | if (parent && parent->use_hierarchy) { | |
4608 | memcg->use_hierarchy = true; | |
3e32cb2e | 4609 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4610 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4611 | page_counter_init(&memcg->memsw, &parent->memsw); |
4612 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4613 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4614 | } else { |
3e32cb2e | 4615 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4616 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4617 | page_counter_init(&memcg->memsw, NULL); |
4618 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4619 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4620 | /* |
4621 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4622 | * much sense so let cgroup subsystem know about this | |
4623 | * unfortunate state in our controller. | |
4624 | */ | |
d142e3e6 | 4625 | if (parent != root_mem_cgroup) |
073219e9 | 4626 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4627 | } |
d6441637 | 4628 | |
0b8f73e1 JW |
4629 | /* The following stuff does not apply to the root */ |
4630 | if (!parent) { | |
4631 | root_mem_cgroup = memcg; | |
4632 | return &memcg->css; | |
4633 | } | |
4634 | ||
b313aeee | 4635 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
4636 | if (error) |
4637 | goto fail; | |
127424c8 | 4638 | |
f7e1cb6e | 4639 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4640 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4641 | |
0b8f73e1 JW |
4642 | return &memcg->css; |
4643 | fail: | |
7e97de0b | 4644 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 4645 | mem_cgroup_free(memcg); |
ea3a9645 | 4646 | return ERR_PTR(-ENOMEM); |
0b8f73e1 JW |
4647 | } |
4648 | ||
73f576c0 | 4649 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 4650 | { |
58fa2a55 VD |
4651 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
4652 | ||
0a4465d3 KT |
4653 | /* |
4654 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
4655 | * by the time the maps are allocated. So, we allocate maps | |
4656 | * here, when for_each_mem_cgroup() can't skip it. | |
4657 | */ | |
4658 | if (memcg_alloc_shrinker_maps(memcg)) { | |
4659 | mem_cgroup_id_remove(memcg); | |
4660 | return -ENOMEM; | |
4661 | } | |
4662 | ||
73f576c0 | 4663 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 4664 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 4665 | css_get(css); |
2f7dd7a4 | 4666 | return 0; |
8cdea7c0 BS |
4667 | } |
4668 | ||
eb95419b | 4669 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4670 | { |
eb95419b | 4671 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4672 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4673 | |
4674 | /* | |
4675 | * Unregister events and notify userspace. | |
4676 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4677 | * directory to avoid race between userspace and kernelspace. | |
4678 | */ | |
fba94807 TH |
4679 | spin_lock(&memcg->event_list_lock); |
4680 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4681 | list_del_init(&event->list); |
4682 | schedule_work(&event->remove); | |
4683 | } | |
fba94807 | 4684 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4685 | |
bf8d5d52 | 4686 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4687 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 4688 | |
567e9ab2 | 4689 | memcg_offline_kmem(memcg); |
52ebea74 | 4690 | wb_memcg_offline(memcg); |
73f576c0 | 4691 | |
591edfb1 RG |
4692 | drain_all_stock(memcg); |
4693 | ||
73f576c0 | 4694 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
4695 | } |
4696 | ||
6df38689 VD |
4697 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4698 | { | |
4699 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4700 | ||
4701 | invalidate_reclaim_iterators(memcg); | |
4702 | } | |
4703 | ||
eb95419b | 4704 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4705 | { |
eb95419b | 4706 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4707 | |
f7e1cb6e | 4708 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4709 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4710 | |
0db15298 | 4711 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4712 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4713 | |
0b8f73e1 JW |
4714 | vmpressure_cleanup(&memcg->vmpressure); |
4715 | cancel_work_sync(&memcg->high_work); | |
4716 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 4717 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 4718 | memcg_free_kmem(memcg); |
0b8f73e1 | 4719 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4720 | } |
4721 | ||
1ced953b TH |
4722 | /** |
4723 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4724 | * @css: the target css | |
4725 | * | |
4726 | * Reset the states of the mem_cgroup associated with @css. This is | |
4727 | * invoked when the userland requests disabling on the default hierarchy | |
4728 | * but the memcg is pinned through dependency. The memcg should stop | |
4729 | * applying policies and should revert to the vanilla state as it may be | |
4730 | * made visible again. | |
4731 | * | |
4732 | * The current implementation only resets the essential configurations. | |
4733 | * This needs to be expanded to cover all the visible parts. | |
4734 | */ | |
4735 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4736 | { | |
4737 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4738 | ||
bbec2e15 RG |
4739 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
4740 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
4741 | page_counter_set_max(&memcg->memsw, PAGE_COUNTER_MAX); | |
4742 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); | |
4743 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 4744 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4745 | page_counter_set_low(&memcg->memory, 0); |
241994ed | 4746 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4747 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4748 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4749 | } |
4750 | ||
02491447 | 4751 | #ifdef CONFIG_MMU |
7dc74be0 | 4752 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4753 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4754 | { |
05b84301 | 4755 | int ret; |
9476db97 | 4756 | |
d0164adc MG |
4757 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4758 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4759 | if (!ret) { |
854ffa8d | 4760 | mc.precharge += count; |
854ffa8d DN |
4761 | return ret; |
4762 | } | |
9476db97 | 4763 | |
3674534b | 4764 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 4765 | while (count--) { |
3674534b | 4766 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 4767 | if (ret) |
38c5d72f | 4768 | return ret; |
854ffa8d | 4769 | mc.precharge++; |
9476db97 | 4770 | cond_resched(); |
854ffa8d | 4771 | } |
9476db97 | 4772 | return 0; |
4ffef5fe DN |
4773 | } |
4774 | ||
4ffef5fe DN |
4775 | union mc_target { |
4776 | struct page *page; | |
02491447 | 4777 | swp_entry_t ent; |
4ffef5fe DN |
4778 | }; |
4779 | ||
4ffef5fe | 4780 | enum mc_target_type { |
8d32ff84 | 4781 | MC_TARGET_NONE = 0, |
4ffef5fe | 4782 | MC_TARGET_PAGE, |
02491447 | 4783 | MC_TARGET_SWAP, |
c733a828 | 4784 | MC_TARGET_DEVICE, |
4ffef5fe DN |
4785 | }; |
4786 | ||
90254a65 DN |
4787 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4788 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4789 | { |
c733a828 | 4790 | struct page *page = _vm_normal_page(vma, addr, ptent, true); |
4ffef5fe | 4791 | |
90254a65 DN |
4792 | if (!page || !page_mapped(page)) |
4793 | return NULL; | |
4794 | if (PageAnon(page)) { | |
1dfab5ab | 4795 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4796 | return NULL; |
1dfab5ab JW |
4797 | } else { |
4798 | if (!(mc.flags & MOVE_FILE)) | |
4799 | return NULL; | |
4800 | } | |
90254a65 DN |
4801 | if (!get_page_unless_zero(page)) |
4802 | return NULL; | |
4803 | ||
4804 | return page; | |
4805 | } | |
4806 | ||
c733a828 | 4807 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 4808 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 4809 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 4810 | { |
90254a65 DN |
4811 | struct page *page = NULL; |
4812 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4813 | ||
1dfab5ab | 4814 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4815 | return NULL; |
c733a828 JG |
4816 | |
4817 | /* | |
4818 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
4819 | * a device and because they are not accessible by CPU they are store | |
4820 | * as special swap entry in the CPU page table. | |
4821 | */ | |
4822 | if (is_device_private_entry(ent)) { | |
4823 | page = device_private_entry_to_page(ent); | |
4824 | /* | |
4825 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
4826 | * a refcount of 1 when free (unlike normal page) | |
4827 | */ | |
4828 | if (!page_ref_add_unless(page, 1, 1)) | |
4829 | return NULL; | |
4830 | return page; | |
4831 | } | |
4832 | ||
4b91355e KH |
4833 | /* |
4834 | * Because lookup_swap_cache() updates some statistics counter, | |
4835 | * we call find_get_page() with swapper_space directly. | |
4836 | */ | |
f6ab1f7f | 4837 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
7941d214 | 4838 | if (do_memsw_account()) |
90254a65 DN |
4839 | entry->val = ent.val; |
4840 | ||
4841 | return page; | |
4842 | } | |
4b91355e KH |
4843 | #else |
4844 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 4845 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
4846 | { |
4847 | return NULL; | |
4848 | } | |
4849 | #endif | |
90254a65 | 4850 | |
87946a72 DN |
4851 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4852 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4853 | { | |
4854 | struct page *page = NULL; | |
87946a72 DN |
4855 | struct address_space *mapping; |
4856 | pgoff_t pgoff; | |
4857 | ||
4858 | if (!vma->vm_file) /* anonymous vma */ | |
4859 | return NULL; | |
1dfab5ab | 4860 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4861 | return NULL; |
4862 | ||
87946a72 | 4863 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4864 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4865 | |
4866 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4867 | #ifdef CONFIG_SWAP |
4868 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4869 | if (shmem_mapping(mapping)) { |
4870 | page = find_get_entry(mapping, pgoff); | |
3159f943 | 4871 | if (xa_is_value(page)) { |
139b6a6f | 4872 | swp_entry_t swp = radix_to_swp_entry(page); |
7941d214 | 4873 | if (do_memsw_account()) |
139b6a6f | 4874 | *entry = swp; |
f6ab1f7f HY |
4875 | page = find_get_page(swap_address_space(swp), |
4876 | swp_offset(swp)); | |
139b6a6f JW |
4877 | } |
4878 | } else | |
4879 | page = find_get_page(mapping, pgoff); | |
4880 | #else | |
4881 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4882 | #endif |
87946a72 DN |
4883 | return page; |
4884 | } | |
4885 | ||
b1b0deab CG |
4886 | /** |
4887 | * mem_cgroup_move_account - move account of the page | |
4888 | * @page: the page | |
25843c2b | 4889 | * @compound: charge the page as compound or small page |
b1b0deab CG |
4890 | * @from: mem_cgroup which the page is moved from. |
4891 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4892 | * | |
3ac808fd | 4893 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
4894 | * |
4895 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4896 | * from old cgroup. | |
4897 | */ | |
4898 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 4899 | bool compound, |
b1b0deab CG |
4900 | struct mem_cgroup *from, |
4901 | struct mem_cgroup *to) | |
4902 | { | |
4903 | unsigned long flags; | |
f627c2f5 | 4904 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 4905 | int ret; |
c4843a75 | 4906 | bool anon; |
b1b0deab CG |
4907 | |
4908 | VM_BUG_ON(from == to); | |
4909 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 4910 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
4911 | |
4912 | /* | |
6a93ca8f | 4913 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 4914 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 4915 | */ |
f627c2f5 | 4916 | ret = -EBUSY; |
b1b0deab CG |
4917 | if (!trylock_page(page)) |
4918 | goto out; | |
4919 | ||
4920 | ret = -EINVAL; | |
4921 | if (page->mem_cgroup != from) | |
4922 | goto out_unlock; | |
4923 | ||
c4843a75 GT |
4924 | anon = PageAnon(page); |
4925 | ||
b1b0deab CG |
4926 | spin_lock_irqsave(&from->move_lock, flags); |
4927 | ||
c4843a75 | 4928 | if (!anon && page_mapped(page)) { |
c9019e9b JW |
4929 | __mod_memcg_state(from, NR_FILE_MAPPED, -nr_pages); |
4930 | __mod_memcg_state(to, NR_FILE_MAPPED, nr_pages); | |
b1b0deab CG |
4931 | } |
4932 | ||
c4843a75 GT |
4933 | /* |
4934 | * move_lock grabbed above and caller set from->moving_account, so | |
ccda7f43 | 4935 | * mod_memcg_page_state will serialize updates to PageDirty. |
c4843a75 GT |
4936 | * So mapping should be stable for dirty pages. |
4937 | */ | |
4938 | if (!anon && PageDirty(page)) { | |
4939 | struct address_space *mapping = page_mapping(page); | |
4940 | ||
4941 | if (mapping_cap_account_dirty(mapping)) { | |
c9019e9b JW |
4942 | __mod_memcg_state(from, NR_FILE_DIRTY, -nr_pages); |
4943 | __mod_memcg_state(to, NR_FILE_DIRTY, nr_pages); | |
c4843a75 GT |
4944 | } |
4945 | } | |
4946 | ||
b1b0deab | 4947 | if (PageWriteback(page)) { |
c9019e9b JW |
4948 | __mod_memcg_state(from, NR_WRITEBACK, -nr_pages); |
4949 | __mod_memcg_state(to, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
4950 | } |
4951 | ||
4952 | /* | |
4953 | * It is safe to change page->mem_cgroup here because the page | |
4954 | * is referenced, charged, and isolated - we can't race with | |
4955 | * uncharging, charging, migration, or LRU putback. | |
4956 | */ | |
4957 | ||
4958 | /* caller should have done css_get */ | |
4959 | page->mem_cgroup = to; | |
4960 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4961 | ||
4962 | ret = 0; | |
4963 | ||
4964 | local_irq_disable(); | |
f627c2f5 | 4965 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 4966 | memcg_check_events(to, page); |
f627c2f5 | 4967 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
4968 | memcg_check_events(from, page); |
4969 | local_irq_enable(); | |
4970 | out_unlock: | |
4971 | unlock_page(page); | |
4972 | out: | |
4973 | return ret; | |
4974 | } | |
4975 | ||
7cf7806c LR |
4976 | /** |
4977 | * get_mctgt_type - get target type of moving charge | |
4978 | * @vma: the vma the pte to be checked belongs | |
4979 | * @addr: the address corresponding to the pte to be checked | |
4980 | * @ptent: the pte to be checked | |
4981 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
4982 | * | |
4983 | * Returns | |
4984 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4985 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4986 | * move charge. if @target is not NULL, the page is stored in target->page | |
4987 | * with extra refcnt got(Callers should handle it). | |
4988 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
4989 | * target for charge migration. if @target is not NULL, the entry is stored | |
4990 | * in target->ent. | |
df6ad698 JG |
4991 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PUBLIC |
4992 | * or MEMORY_DEVICE_PRIVATE (so ZONE_DEVICE page and thus not on the lru). | |
4993 | * For now we such page is charge like a regular page would be as for all | |
4994 | * intent and purposes it is just special memory taking the place of a | |
4995 | * regular page. | |
c733a828 JG |
4996 | * |
4997 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
4998 | * |
4999 | * Called with pte lock held. | |
5000 | */ | |
5001 | ||
8d32ff84 | 5002 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5003 | unsigned long addr, pte_t ptent, union mc_target *target) |
5004 | { | |
5005 | struct page *page = NULL; | |
8d32ff84 | 5006 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5007 | swp_entry_t ent = { .val = 0 }; |
5008 | ||
5009 | if (pte_present(ptent)) | |
5010 | page = mc_handle_present_pte(vma, addr, ptent); | |
5011 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5012 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5013 | else if (pte_none(ptent)) |
87946a72 | 5014 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5015 | |
5016 | if (!page && !ent.val) | |
8d32ff84 | 5017 | return ret; |
02491447 | 5018 | if (page) { |
02491447 | 5019 | /* |
0a31bc97 | 5020 | * Do only loose check w/o serialization. |
1306a85a | 5021 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5022 | * not under LRU exclusion. |
02491447 | 5023 | */ |
1306a85a | 5024 | if (page->mem_cgroup == mc.from) { |
02491447 | 5025 | ret = MC_TARGET_PAGE; |
df6ad698 JG |
5026 | if (is_device_private_page(page) || |
5027 | is_device_public_page(page)) | |
c733a828 | 5028 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5029 | if (target) |
5030 | target->page = page; | |
5031 | } | |
5032 | if (!ret || !target) | |
5033 | put_page(page); | |
5034 | } | |
3e14a57b HY |
5035 | /* |
5036 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5037 | * But we cannot move a tail-page in a THP. | |
5038 | */ | |
5039 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5040 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5041 | ret = MC_TARGET_SWAP; |
5042 | if (target) | |
5043 | target->ent = ent; | |
4ffef5fe | 5044 | } |
4ffef5fe DN |
5045 | return ret; |
5046 | } | |
5047 | ||
12724850 NH |
5048 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5049 | /* | |
d6810d73 HY |
5050 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5051 | * not support them for now. | |
12724850 NH |
5052 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5053 | */ | |
5054 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5055 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5056 | { | |
5057 | struct page *page = NULL; | |
12724850 NH |
5058 | enum mc_target_type ret = MC_TARGET_NONE; |
5059 | ||
84c3fc4e ZY |
5060 | if (unlikely(is_swap_pmd(pmd))) { |
5061 | VM_BUG_ON(thp_migration_supported() && | |
5062 | !is_pmd_migration_entry(pmd)); | |
5063 | return ret; | |
5064 | } | |
12724850 | 5065 | page = pmd_page(pmd); |
309381fe | 5066 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5067 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5068 | return ret; |
1306a85a | 5069 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
5070 | ret = MC_TARGET_PAGE; |
5071 | if (target) { | |
5072 | get_page(page); | |
5073 | target->page = page; | |
5074 | } | |
5075 | } | |
5076 | return ret; | |
5077 | } | |
5078 | #else | |
5079 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5080 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5081 | { | |
5082 | return MC_TARGET_NONE; | |
5083 | } | |
5084 | #endif | |
5085 | ||
4ffef5fe DN |
5086 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5087 | unsigned long addr, unsigned long end, | |
5088 | struct mm_walk *walk) | |
5089 | { | |
26bcd64a | 5090 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5091 | pte_t *pte; |
5092 | spinlock_t *ptl; | |
5093 | ||
b6ec57f4 KS |
5094 | ptl = pmd_trans_huge_lock(pmd, vma); |
5095 | if (ptl) { | |
c733a828 JG |
5096 | /* |
5097 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
5098 | * support transparent huge page with MEMORY_DEVICE_PUBLIC or | |
5099 | * MEMORY_DEVICE_PRIVATE but this might change. | |
5100 | */ | |
12724850 NH |
5101 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5102 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5103 | spin_unlock(ptl); |
1a5a9906 | 5104 | return 0; |
12724850 | 5105 | } |
03319327 | 5106 | |
45f83cef AA |
5107 | if (pmd_trans_unstable(pmd)) |
5108 | return 0; | |
4ffef5fe DN |
5109 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5110 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5111 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5112 | mc.precharge++; /* increment precharge temporarily */ |
5113 | pte_unmap_unlock(pte - 1, ptl); | |
5114 | cond_resched(); | |
5115 | ||
7dc74be0 DN |
5116 | return 0; |
5117 | } | |
5118 | ||
4ffef5fe DN |
5119 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5120 | { | |
5121 | unsigned long precharge; | |
4ffef5fe | 5122 | |
26bcd64a NH |
5123 | struct mm_walk mem_cgroup_count_precharge_walk = { |
5124 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5125 | .mm = mm, | |
5126 | }; | |
dfe076b0 | 5127 | down_read(&mm->mmap_sem); |
0247f3f4 JM |
5128 | walk_page_range(0, mm->highest_vm_end, |
5129 | &mem_cgroup_count_precharge_walk); | |
dfe076b0 | 5130 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5131 | |
5132 | precharge = mc.precharge; | |
5133 | mc.precharge = 0; | |
5134 | ||
5135 | return precharge; | |
5136 | } | |
5137 | ||
4ffef5fe DN |
5138 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5139 | { | |
dfe076b0 DN |
5140 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5141 | ||
5142 | VM_BUG_ON(mc.moving_task); | |
5143 | mc.moving_task = current; | |
5144 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5145 | } |
5146 | ||
dfe076b0 DN |
5147 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5148 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5149 | { |
2bd9bb20 KH |
5150 | struct mem_cgroup *from = mc.from; |
5151 | struct mem_cgroup *to = mc.to; | |
5152 | ||
4ffef5fe | 5153 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5154 | if (mc.precharge) { |
00501b53 | 5155 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5156 | mc.precharge = 0; |
5157 | } | |
5158 | /* | |
5159 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5160 | * we must uncharge here. | |
5161 | */ | |
5162 | if (mc.moved_charge) { | |
00501b53 | 5163 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5164 | mc.moved_charge = 0; |
4ffef5fe | 5165 | } |
483c30b5 DN |
5166 | /* we must fixup refcnts and charges */ |
5167 | if (mc.moved_swap) { | |
483c30b5 | 5168 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5169 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5170 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5171 | |
615d66c3 VD |
5172 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5173 | ||
05b84301 | 5174 | /* |
3e32cb2e JW |
5175 | * we charged both to->memory and to->memsw, so we |
5176 | * should uncharge to->memory. | |
05b84301 | 5177 | */ |
ce00a967 | 5178 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5179 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5180 | ||
615d66c3 VD |
5181 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
5182 | css_put_many(&mc.to->css, mc.moved_swap); | |
3e32cb2e | 5183 | |
483c30b5 DN |
5184 | mc.moved_swap = 0; |
5185 | } | |
dfe076b0 DN |
5186 | memcg_oom_recover(from); |
5187 | memcg_oom_recover(to); | |
5188 | wake_up_all(&mc.waitq); | |
5189 | } | |
5190 | ||
5191 | static void mem_cgroup_clear_mc(void) | |
5192 | { | |
264a0ae1 TH |
5193 | struct mm_struct *mm = mc.mm; |
5194 | ||
dfe076b0 DN |
5195 | /* |
5196 | * we must clear moving_task before waking up waiters at the end of | |
5197 | * task migration. | |
5198 | */ | |
5199 | mc.moving_task = NULL; | |
5200 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5201 | spin_lock(&mc.lock); |
4ffef5fe DN |
5202 | mc.from = NULL; |
5203 | mc.to = NULL; | |
264a0ae1 | 5204 | mc.mm = NULL; |
2bd9bb20 | 5205 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5206 | |
5207 | mmput(mm); | |
4ffef5fe DN |
5208 | } |
5209 | ||
1f7dd3e5 | 5210 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5211 | { |
1f7dd3e5 | 5212 | struct cgroup_subsys_state *css; |
eed67d75 | 5213 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5214 | struct mem_cgroup *from; |
4530eddb | 5215 | struct task_struct *leader, *p; |
9f2115f9 | 5216 | struct mm_struct *mm; |
1dfab5ab | 5217 | unsigned long move_flags; |
9f2115f9 | 5218 | int ret = 0; |
7dc74be0 | 5219 | |
1f7dd3e5 TH |
5220 | /* charge immigration isn't supported on the default hierarchy */ |
5221 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5222 | return 0; |
5223 | ||
4530eddb TH |
5224 | /* |
5225 | * Multi-process migrations only happen on the default hierarchy | |
5226 | * where charge immigration is not used. Perform charge | |
5227 | * immigration if @tset contains a leader and whine if there are | |
5228 | * multiple. | |
5229 | */ | |
5230 | p = NULL; | |
1f7dd3e5 | 5231 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5232 | WARN_ON_ONCE(p); |
5233 | p = leader; | |
1f7dd3e5 | 5234 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5235 | } |
5236 | if (!p) | |
5237 | return 0; | |
5238 | ||
1f7dd3e5 TH |
5239 | /* |
5240 | * We are now commited to this value whatever it is. Changes in this | |
5241 | * tunable will only affect upcoming migrations, not the current one. | |
5242 | * So we need to save it, and keep it going. | |
5243 | */ | |
5244 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5245 | if (!move_flags) | |
5246 | return 0; | |
5247 | ||
9f2115f9 TH |
5248 | from = mem_cgroup_from_task(p); |
5249 | ||
5250 | VM_BUG_ON(from == memcg); | |
5251 | ||
5252 | mm = get_task_mm(p); | |
5253 | if (!mm) | |
5254 | return 0; | |
5255 | /* We move charges only when we move a owner of the mm */ | |
5256 | if (mm->owner == p) { | |
5257 | VM_BUG_ON(mc.from); | |
5258 | VM_BUG_ON(mc.to); | |
5259 | VM_BUG_ON(mc.precharge); | |
5260 | VM_BUG_ON(mc.moved_charge); | |
5261 | VM_BUG_ON(mc.moved_swap); | |
5262 | ||
5263 | spin_lock(&mc.lock); | |
264a0ae1 | 5264 | mc.mm = mm; |
9f2115f9 TH |
5265 | mc.from = from; |
5266 | mc.to = memcg; | |
5267 | mc.flags = move_flags; | |
5268 | spin_unlock(&mc.lock); | |
5269 | /* We set mc.moving_task later */ | |
5270 | ||
5271 | ret = mem_cgroup_precharge_mc(mm); | |
5272 | if (ret) | |
5273 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5274 | } else { |
5275 | mmput(mm); | |
7dc74be0 DN |
5276 | } |
5277 | return ret; | |
5278 | } | |
5279 | ||
1f7dd3e5 | 5280 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5281 | { |
4e2f245d JW |
5282 | if (mc.to) |
5283 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5284 | } |
5285 | ||
4ffef5fe DN |
5286 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5287 | unsigned long addr, unsigned long end, | |
5288 | struct mm_walk *walk) | |
7dc74be0 | 5289 | { |
4ffef5fe | 5290 | int ret = 0; |
26bcd64a | 5291 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5292 | pte_t *pte; |
5293 | spinlock_t *ptl; | |
12724850 NH |
5294 | enum mc_target_type target_type; |
5295 | union mc_target target; | |
5296 | struct page *page; | |
4ffef5fe | 5297 | |
b6ec57f4 KS |
5298 | ptl = pmd_trans_huge_lock(pmd, vma); |
5299 | if (ptl) { | |
62ade86a | 5300 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5301 | spin_unlock(ptl); |
12724850 NH |
5302 | return 0; |
5303 | } | |
5304 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5305 | if (target_type == MC_TARGET_PAGE) { | |
5306 | page = target.page; | |
5307 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5308 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5309 | mc.from, mc.to)) { |
12724850 NH |
5310 | mc.precharge -= HPAGE_PMD_NR; |
5311 | mc.moved_charge += HPAGE_PMD_NR; | |
5312 | } | |
5313 | putback_lru_page(page); | |
5314 | } | |
5315 | put_page(page); | |
c733a828 JG |
5316 | } else if (target_type == MC_TARGET_DEVICE) { |
5317 | page = target.page; | |
5318 | if (!mem_cgroup_move_account(page, true, | |
5319 | mc.from, mc.to)) { | |
5320 | mc.precharge -= HPAGE_PMD_NR; | |
5321 | mc.moved_charge += HPAGE_PMD_NR; | |
5322 | } | |
5323 | put_page(page); | |
12724850 | 5324 | } |
bf929152 | 5325 | spin_unlock(ptl); |
1a5a9906 | 5326 | return 0; |
12724850 NH |
5327 | } |
5328 | ||
45f83cef AA |
5329 | if (pmd_trans_unstable(pmd)) |
5330 | return 0; | |
4ffef5fe DN |
5331 | retry: |
5332 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5333 | for (; addr != end; addr += PAGE_SIZE) { | |
5334 | pte_t ptent = *(pte++); | |
c733a828 | 5335 | bool device = false; |
02491447 | 5336 | swp_entry_t ent; |
4ffef5fe DN |
5337 | |
5338 | if (!mc.precharge) | |
5339 | break; | |
5340 | ||
8d32ff84 | 5341 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
5342 | case MC_TARGET_DEVICE: |
5343 | device = true; | |
5344 | /* fall through */ | |
4ffef5fe DN |
5345 | case MC_TARGET_PAGE: |
5346 | page = target.page; | |
53f9263b KS |
5347 | /* |
5348 | * We can have a part of the split pmd here. Moving it | |
5349 | * can be done but it would be too convoluted so simply | |
5350 | * ignore such a partial THP and keep it in original | |
5351 | * memcg. There should be somebody mapping the head. | |
5352 | */ | |
5353 | if (PageTransCompound(page)) | |
5354 | goto put; | |
c733a828 | 5355 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 5356 | goto put; |
f627c2f5 KS |
5357 | if (!mem_cgroup_move_account(page, false, |
5358 | mc.from, mc.to)) { | |
4ffef5fe | 5359 | mc.precharge--; |
854ffa8d DN |
5360 | /* we uncharge from mc.from later. */ |
5361 | mc.moved_charge++; | |
4ffef5fe | 5362 | } |
c733a828 JG |
5363 | if (!device) |
5364 | putback_lru_page(page); | |
8d32ff84 | 5365 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5366 | put_page(page); |
5367 | break; | |
02491447 DN |
5368 | case MC_TARGET_SWAP: |
5369 | ent = target.ent; | |
e91cbb42 | 5370 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5371 | mc.precharge--; |
483c30b5 DN |
5372 | /* we fixup refcnts and charges later. */ |
5373 | mc.moved_swap++; | |
5374 | } | |
02491447 | 5375 | break; |
4ffef5fe DN |
5376 | default: |
5377 | break; | |
5378 | } | |
5379 | } | |
5380 | pte_unmap_unlock(pte - 1, ptl); | |
5381 | cond_resched(); | |
5382 | ||
5383 | if (addr != end) { | |
5384 | /* | |
5385 | * We have consumed all precharges we got in can_attach(). | |
5386 | * We try charge one by one, but don't do any additional | |
5387 | * charges to mc.to if we have failed in charge once in attach() | |
5388 | * phase. | |
5389 | */ | |
854ffa8d | 5390 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5391 | if (!ret) |
5392 | goto retry; | |
5393 | } | |
5394 | ||
5395 | return ret; | |
5396 | } | |
5397 | ||
264a0ae1 | 5398 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 5399 | { |
26bcd64a NH |
5400 | struct mm_walk mem_cgroup_move_charge_walk = { |
5401 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
264a0ae1 | 5402 | .mm = mc.mm, |
26bcd64a | 5403 | }; |
4ffef5fe DN |
5404 | |
5405 | lru_add_drain_all(); | |
312722cb | 5406 | /* |
81f8c3a4 JW |
5407 | * Signal lock_page_memcg() to take the memcg's move_lock |
5408 | * while we're moving its pages to another memcg. Then wait | |
5409 | * for already started RCU-only updates to finish. | |
312722cb JW |
5410 | */ |
5411 | atomic_inc(&mc.from->moving_account); | |
5412 | synchronize_rcu(); | |
dfe076b0 | 5413 | retry: |
264a0ae1 | 5414 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
5415 | /* |
5416 | * Someone who are holding the mmap_sem might be waiting in | |
5417 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5418 | * and retry. Because we cancel precharges, we might not be able | |
5419 | * to move enough charges, but moving charge is a best-effort | |
5420 | * feature anyway, so it wouldn't be a big problem. | |
5421 | */ | |
5422 | __mem_cgroup_clear_mc(); | |
5423 | cond_resched(); | |
5424 | goto retry; | |
5425 | } | |
26bcd64a NH |
5426 | /* |
5427 | * When we have consumed all precharges and failed in doing | |
5428 | * additional charge, the page walk just aborts. | |
5429 | */ | |
0247f3f4 JM |
5430 | walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk); |
5431 | ||
264a0ae1 | 5432 | up_read(&mc.mm->mmap_sem); |
312722cb | 5433 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5434 | } |
5435 | ||
264a0ae1 | 5436 | static void mem_cgroup_move_task(void) |
67e465a7 | 5437 | { |
264a0ae1 TH |
5438 | if (mc.to) { |
5439 | mem_cgroup_move_charge(); | |
a433658c | 5440 | mem_cgroup_clear_mc(); |
264a0ae1 | 5441 | } |
67e465a7 | 5442 | } |
5cfb80a7 | 5443 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5444 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5445 | { |
5446 | return 0; | |
5447 | } | |
1f7dd3e5 | 5448 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5449 | { |
5450 | } | |
264a0ae1 | 5451 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
5452 | { |
5453 | } | |
5454 | #endif | |
67e465a7 | 5455 | |
f00baae7 TH |
5456 | /* |
5457 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5458 | * to verify whether we're attached to the default hierarchy on each mount |
5459 | * attempt. | |
f00baae7 | 5460 | */ |
eb95419b | 5461 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5462 | { |
5463 | /* | |
aa6ec29b | 5464 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5465 | * guarantees that @root doesn't have any children, so turning it |
5466 | * on for the root memcg is enough. | |
5467 | */ | |
9e10a130 | 5468 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5469 | root_mem_cgroup->use_hierarchy = true; |
5470 | else | |
5471 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5472 | } |
5473 | ||
677dc973 CD |
5474 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
5475 | { | |
5476 | if (value == PAGE_COUNTER_MAX) | |
5477 | seq_puts(m, "max\n"); | |
5478 | else | |
5479 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
5480 | ||
5481 | return 0; | |
5482 | } | |
5483 | ||
241994ed JW |
5484 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5485 | struct cftype *cft) | |
5486 | { | |
f5fc3c5d JW |
5487 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5488 | ||
5489 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5490 | } |
5491 | ||
bf8d5d52 RG |
5492 | static int memory_min_show(struct seq_file *m, void *v) |
5493 | { | |
677dc973 CD |
5494 | return seq_puts_memcg_tunable(m, |
5495 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
5496 | } |
5497 | ||
5498 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
5499 | char *buf, size_t nbytes, loff_t off) | |
5500 | { | |
5501 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5502 | unsigned long min; | |
5503 | int err; | |
5504 | ||
5505 | buf = strstrip(buf); | |
5506 | err = page_counter_memparse(buf, "max", &min); | |
5507 | if (err) | |
5508 | return err; | |
5509 | ||
5510 | page_counter_set_min(&memcg->memory, min); | |
5511 | ||
5512 | return nbytes; | |
5513 | } | |
5514 | ||
241994ed JW |
5515 | static int memory_low_show(struct seq_file *m, void *v) |
5516 | { | |
677dc973 CD |
5517 | return seq_puts_memcg_tunable(m, |
5518 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
5519 | } |
5520 | ||
5521 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5522 | char *buf, size_t nbytes, loff_t off) | |
5523 | { | |
5524 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5525 | unsigned long low; | |
5526 | int err; | |
5527 | ||
5528 | buf = strstrip(buf); | |
d2973697 | 5529 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5530 | if (err) |
5531 | return err; | |
5532 | ||
23067153 | 5533 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
5534 | |
5535 | return nbytes; | |
5536 | } | |
5537 | ||
5538 | static int memory_high_show(struct seq_file *m, void *v) | |
5539 | { | |
677dc973 | 5540 | return seq_puts_memcg_tunable(m, READ_ONCE(mem_cgroup_from_seq(m)->high)); |
241994ed JW |
5541 | } |
5542 | ||
5543 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5544 | char *buf, size_t nbytes, loff_t off) | |
5545 | { | |
5546 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 5547 | unsigned long nr_pages; |
241994ed JW |
5548 | unsigned long high; |
5549 | int err; | |
5550 | ||
5551 | buf = strstrip(buf); | |
d2973697 | 5552 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5553 | if (err) |
5554 | return err; | |
5555 | ||
5556 | memcg->high = high; | |
5557 | ||
588083bb JW |
5558 | nr_pages = page_counter_read(&memcg->memory); |
5559 | if (nr_pages > high) | |
5560 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5561 | GFP_KERNEL, true); | |
5562 | ||
2529bb3a | 5563 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5564 | return nbytes; |
5565 | } | |
5566 | ||
5567 | static int memory_max_show(struct seq_file *m, void *v) | |
5568 | { | |
677dc973 CD |
5569 | return seq_puts_memcg_tunable(m, |
5570 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
5571 | } |
5572 | ||
5573 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5574 | char *buf, size_t nbytes, loff_t off) | |
5575 | { | |
5576 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
5577 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5578 | bool drained = false; | |
241994ed JW |
5579 | unsigned long max; |
5580 | int err; | |
5581 | ||
5582 | buf = strstrip(buf); | |
d2973697 | 5583 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5584 | if (err) |
5585 | return err; | |
5586 | ||
bbec2e15 | 5587 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
5588 | |
5589 | for (;;) { | |
5590 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5591 | ||
5592 | if (nr_pages <= max) | |
5593 | break; | |
5594 | ||
5595 | if (signal_pending(current)) { | |
5596 | err = -EINTR; | |
5597 | break; | |
5598 | } | |
5599 | ||
5600 | if (!drained) { | |
5601 | drain_all_stock(memcg); | |
5602 | drained = true; | |
5603 | continue; | |
5604 | } | |
5605 | ||
5606 | if (nr_reclaims) { | |
5607 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5608 | GFP_KERNEL, true)) | |
5609 | nr_reclaims--; | |
5610 | continue; | |
5611 | } | |
5612 | ||
e27be240 | 5613 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
5614 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
5615 | break; | |
5616 | } | |
241994ed | 5617 | |
2529bb3a | 5618 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5619 | return nbytes; |
5620 | } | |
5621 | ||
5622 | static int memory_events_show(struct seq_file *m, void *v) | |
5623 | { | |
aa9694bb | 5624 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 5625 | |
e27be240 JW |
5626 | seq_printf(m, "low %lu\n", |
5627 | atomic_long_read(&memcg->memory_events[MEMCG_LOW])); | |
5628 | seq_printf(m, "high %lu\n", | |
5629 | atomic_long_read(&memcg->memory_events[MEMCG_HIGH])); | |
5630 | seq_printf(m, "max %lu\n", | |
5631 | atomic_long_read(&memcg->memory_events[MEMCG_MAX])); | |
5632 | seq_printf(m, "oom %lu\n", | |
5633 | atomic_long_read(&memcg->memory_events[MEMCG_OOM])); | |
fe6bdfc8 RG |
5634 | seq_printf(m, "oom_kill %lu\n", |
5635 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
241994ed JW |
5636 | |
5637 | return 0; | |
5638 | } | |
5639 | ||
587d9f72 JW |
5640 | static int memory_stat_show(struct seq_file *m, void *v) |
5641 | { | |
aa9694bb | 5642 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
587d9f72 JW |
5643 | int i; |
5644 | ||
5645 | /* | |
5646 | * Provide statistics on the state of the memory subsystem as | |
5647 | * well as cumulative event counters that show past behavior. | |
5648 | * | |
5649 | * This list is ordered following a combination of these gradients: | |
5650 | * 1) generic big picture -> specifics and details | |
5651 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
5652 | * | |
5653 | * Current memory state: | |
5654 | */ | |
5655 | ||
5656 | seq_printf(m, "anon %llu\n", | |
42a30035 | 5657 | (u64)memcg_page_state(memcg, MEMCG_RSS) * PAGE_SIZE); |
587d9f72 | 5658 | seq_printf(m, "file %llu\n", |
42a30035 | 5659 | (u64)memcg_page_state(memcg, MEMCG_CACHE) * PAGE_SIZE); |
12580e4b | 5660 | seq_printf(m, "kernel_stack %llu\n", |
42a30035 | 5661 | (u64)memcg_page_state(memcg, MEMCG_KERNEL_STACK_KB) * 1024); |
27ee57c9 | 5662 | seq_printf(m, "slab %llu\n", |
42a30035 JW |
5663 | (u64)(memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) + |
5664 | memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE)) * | |
5665 | PAGE_SIZE); | |
b2807f07 | 5666 | seq_printf(m, "sock %llu\n", |
42a30035 | 5667 | (u64)memcg_page_state(memcg, MEMCG_SOCK) * PAGE_SIZE); |
587d9f72 | 5668 | |
9a4caf1e | 5669 | seq_printf(m, "shmem %llu\n", |
42a30035 | 5670 | (u64)memcg_page_state(memcg, NR_SHMEM) * PAGE_SIZE); |
587d9f72 | 5671 | seq_printf(m, "file_mapped %llu\n", |
42a30035 | 5672 | (u64)memcg_page_state(memcg, NR_FILE_MAPPED) * PAGE_SIZE); |
587d9f72 | 5673 | seq_printf(m, "file_dirty %llu\n", |
42a30035 | 5674 | (u64)memcg_page_state(memcg, NR_FILE_DIRTY) * PAGE_SIZE); |
587d9f72 | 5675 | seq_printf(m, "file_writeback %llu\n", |
42a30035 | 5676 | (u64)memcg_page_state(memcg, NR_WRITEBACK) * PAGE_SIZE); |
587d9f72 | 5677 | |
1ff9e6e1 CD |
5678 | /* |
5679 | * TODO: We should eventually replace our own MEMCG_RSS_HUGE counter | |
5680 | * with the NR_ANON_THP vm counter, but right now it's a pain in the | |
5681 | * arse because it requires migrating the work out of rmap to a place | |
5682 | * where the page->mem_cgroup is set up and stable. | |
5683 | */ | |
5684 | seq_printf(m, "anon_thp %llu\n", | |
42a30035 | 5685 | (u64)memcg_page_state(memcg, MEMCG_RSS_HUGE) * PAGE_SIZE); |
1ff9e6e1 | 5686 | |
8de7ecc6 SB |
5687 | for (i = 0; i < NR_LRU_LISTS; i++) |
5688 | seq_printf(m, "%s %llu\n", mem_cgroup_lru_names[i], | |
42a30035 JW |
5689 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
5690 | PAGE_SIZE); | |
587d9f72 | 5691 | |
27ee57c9 | 5692 | seq_printf(m, "slab_reclaimable %llu\n", |
42a30035 JW |
5693 | (u64)memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) * |
5694 | PAGE_SIZE); | |
27ee57c9 | 5695 | seq_printf(m, "slab_unreclaimable %llu\n", |
42a30035 JW |
5696 | (u64)memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE) * |
5697 | PAGE_SIZE); | |
27ee57c9 | 5698 | |
587d9f72 JW |
5699 | /* Accumulated memory events */ |
5700 | ||
42a30035 JW |
5701 | seq_printf(m, "pgfault %lu\n", memcg_events(memcg, PGFAULT)); |
5702 | seq_printf(m, "pgmajfault %lu\n", memcg_events(memcg, PGMAJFAULT)); | |
587d9f72 | 5703 | |
e9b257ed | 5704 | seq_printf(m, "workingset_refault %lu\n", |
42a30035 | 5705 | memcg_page_state(memcg, WORKINGSET_REFAULT)); |
e9b257ed | 5706 | seq_printf(m, "workingset_activate %lu\n", |
42a30035 | 5707 | memcg_page_state(memcg, WORKINGSET_ACTIVATE)); |
e9b257ed | 5708 | seq_printf(m, "workingset_nodereclaim %lu\n", |
42a30035 JW |
5709 | memcg_page_state(memcg, WORKINGSET_NODERECLAIM)); |
5710 | ||
5711 | seq_printf(m, "pgrefill %lu\n", memcg_events(memcg, PGREFILL)); | |
5712 | seq_printf(m, "pgscan %lu\n", memcg_events(memcg, PGSCAN_KSWAPD) + | |
5713 | memcg_events(memcg, PGSCAN_DIRECT)); | |
5714 | seq_printf(m, "pgsteal %lu\n", memcg_events(memcg, PGSTEAL_KSWAPD) + | |
5715 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
5716 | seq_printf(m, "pgactivate %lu\n", memcg_events(memcg, PGACTIVATE)); | |
5717 | seq_printf(m, "pgdeactivate %lu\n", memcg_events(memcg, PGDEACTIVATE)); | |
5718 | seq_printf(m, "pglazyfree %lu\n", memcg_events(memcg, PGLAZYFREE)); | |
5719 | seq_printf(m, "pglazyfreed %lu\n", memcg_events(memcg, PGLAZYFREED)); | |
2262185c | 5720 | |
1ff9e6e1 | 5721 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
42a30035 JW |
5722 | seq_printf(m, "thp_fault_alloc %lu\n", |
5723 | memcg_events(memcg, THP_FAULT_ALLOC)); | |
1ff9e6e1 | 5724 | seq_printf(m, "thp_collapse_alloc %lu\n", |
42a30035 | 5725 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1ff9e6e1 CD |
5726 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
5727 | ||
587d9f72 JW |
5728 | return 0; |
5729 | } | |
5730 | ||
3d8b38eb RG |
5731 | static int memory_oom_group_show(struct seq_file *m, void *v) |
5732 | { | |
aa9694bb | 5733 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
5734 | |
5735 | seq_printf(m, "%d\n", memcg->oom_group); | |
5736 | ||
5737 | return 0; | |
5738 | } | |
5739 | ||
5740 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
5741 | char *buf, size_t nbytes, loff_t off) | |
5742 | { | |
5743 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5744 | int ret, oom_group; | |
5745 | ||
5746 | buf = strstrip(buf); | |
5747 | if (!buf) | |
5748 | return -EINVAL; | |
5749 | ||
5750 | ret = kstrtoint(buf, 0, &oom_group); | |
5751 | if (ret) | |
5752 | return ret; | |
5753 | ||
5754 | if (oom_group != 0 && oom_group != 1) | |
5755 | return -EINVAL; | |
5756 | ||
5757 | memcg->oom_group = oom_group; | |
5758 | ||
5759 | return nbytes; | |
5760 | } | |
5761 | ||
241994ed JW |
5762 | static struct cftype memory_files[] = { |
5763 | { | |
5764 | .name = "current", | |
f5fc3c5d | 5765 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5766 | .read_u64 = memory_current_read, |
5767 | }, | |
bf8d5d52 RG |
5768 | { |
5769 | .name = "min", | |
5770 | .flags = CFTYPE_NOT_ON_ROOT, | |
5771 | .seq_show = memory_min_show, | |
5772 | .write = memory_min_write, | |
5773 | }, | |
241994ed JW |
5774 | { |
5775 | .name = "low", | |
5776 | .flags = CFTYPE_NOT_ON_ROOT, | |
5777 | .seq_show = memory_low_show, | |
5778 | .write = memory_low_write, | |
5779 | }, | |
5780 | { | |
5781 | .name = "high", | |
5782 | .flags = CFTYPE_NOT_ON_ROOT, | |
5783 | .seq_show = memory_high_show, | |
5784 | .write = memory_high_write, | |
5785 | }, | |
5786 | { | |
5787 | .name = "max", | |
5788 | .flags = CFTYPE_NOT_ON_ROOT, | |
5789 | .seq_show = memory_max_show, | |
5790 | .write = memory_max_write, | |
5791 | }, | |
5792 | { | |
5793 | .name = "events", | |
5794 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5795 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5796 | .seq_show = memory_events_show, |
5797 | }, | |
587d9f72 JW |
5798 | { |
5799 | .name = "stat", | |
5800 | .flags = CFTYPE_NOT_ON_ROOT, | |
5801 | .seq_show = memory_stat_show, | |
5802 | }, | |
3d8b38eb RG |
5803 | { |
5804 | .name = "oom.group", | |
5805 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
5806 | .seq_show = memory_oom_group_show, | |
5807 | .write = memory_oom_group_write, | |
5808 | }, | |
241994ed JW |
5809 | { } /* terminate */ |
5810 | }; | |
5811 | ||
073219e9 | 5812 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5813 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5814 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5815 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5816 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5817 | .css_free = mem_cgroup_css_free, |
1ced953b | 5818 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5819 | .can_attach = mem_cgroup_can_attach, |
5820 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 5821 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5822 | .bind = mem_cgroup_bind, |
241994ed JW |
5823 | .dfl_cftypes = memory_files, |
5824 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5825 | .early_init = 0, |
8cdea7c0 | 5826 | }; |
c077719b | 5827 | |
241994ed | 5828 | /** |
bf8d5d52 | 5829 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 5830 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
5831 | * @memcg: the memory cgroup to check |
5832 | * | |
23067153 RG |
5833 | * WARNING: This function is not stateless! It can only be used as part |
5834 | * of a top-down tree iteration, not for isolated queries. | |
34c81057 | 5835 | * |
bf8d5d52 RG |
5836 | * Returns one of the following: |
5837 | * MEMCG_PROT_NONE: cgroup memory is not protected | |
5838 | * MEMCG_PROT_LOW: cgroup memory is protected as long there is | |
5839 | * an unprotected supply of reclaimable memory from other cgroups. | |
5840 | * MEMCG_PROT_MIN: cgroup memory is protected | |
34c81057 | 5841 | * |
bf8d5d52 | 5842 | * @root is exclusive; it is never protected when looked at directly |
34c81057 | 5843 | * |
bf8d5d52 RG |
5844 | * To provide a proper hierarchical behavior, effective memory.min/low values |
5845 | * are used. Below is the description of how effective memory.low is calculated. | |
5846 | * Effective memory.min values is calculated in the same way. | |
34c81057 | 5847 | * |
23067153 RG |
5848 | * Effective memory.low is always equal or less than the original memory.low. |
5849 | * If there is no memory.low overcommittment (which is always true for | |
5850 | * top-level memory cgroups), these two values are equal. | |
5851 | * Otherwise, it's a part of parent's effective memory.low, | |
5852 | * calculated as a cgroup's memory.low usage divided by sum of sibling's | |
5853 | * memory.low usages, where memory.low usage is the size of actually | |
5854 | * protected memory. | |
34c81057 | 5855 | * |
23067153 RG |
5856 | * low_usage |
5857 | * elow = min( memory.low, parent->elow * ------------------ ), | |
5858 | * siblings_low_usage | |
34c81057 | 5859 | * |
23067153 RG |
5860 | * | memory.current, if memory.current < memory.low |
5861 | * low_usage = | | |
82ede7ee | 5862 | * | 0, otherwise. |
34c81057 | 5863 | * |
23067153 RG |
5864 | * |
5865 | * Such definition of the effective memory.low provides the expected | |
5866 | * hierarchical behavior: parent's memory.low value is limiting | |
5867 | * children, unprotected memory is reclaimed first and cgroups, | |
5868 | * which are not using their guarantee do not affect actual memory | |
5869 | * distribution. | |
5870 | * | |
5871 | * For example, if there are memcgs A, A/B, A/C, A/D and A/E: | |
5872 | * | |
5873 | * A A/memory.low = 2G, A/memory.current = 6G | |
5874 | * //\\ | |
5875 | * BC DE B/memory.low = 3G B/memory.current = 2G | |
5876 | * C/memory.low = 1G C/memory.current = 2G | |
5877 | * D/memory.low = 0 D/memory.current = 2G | |
5878 | * E/memory.low = 10G E/memory.current = 0 | |
5879 | * | |
5880 | * and the memory pressure is applied, the following memory distribution | |
5881 | * is expected (approximately): | |
5882 | * | |
5883 | * A/memory.current = 2G | |
5884 | * | |
5885 | * B/memory.current = 1.3G | |
5886 | * C/memory.current = 0.6G | |
5887 | * D/memory.current = 0 | |
5888 | * E/memory.current = 0 | |
5889 | * | |
5890 | * These calculations require constant tracking of the actual low usages | |
bf8d5d52 RG |
5891 | * (see propagate_protected_usage()), as well as recursive calculation of |
5892 | * effective memory.low values. But as we do call mem_cgroup_protected() | |
23067153 RG |
5893 | * path for each memory cgroup top-down from the reclaim, |
5894 | * it's possible to optimize this part, and save calculated elow | |
5895 | * for next usage. This part is intentionally racy, but it's ok, | |
5896 | * as memory.low is a best-effort mechanism. | |
241994ed | 5897 | */ |
bf8d5d52 RG |
5898 | enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root, |
5899 | struct mem_cgroup *memcg) | |
241994ed | 5900 | { |
23067153 | 5901 | struct mem_cgroup *parent; |
bf8d5d52 RG |
5902 | unsigned long emin, parent_emin; |
5903 | unsigned long elow, parent_elow; | |
5904 | unsigned long usage; | |
23067153 | 5905 | |
241994ed | 5906 | if (mem_cgroup_disabled()) |
bf8d5d52 | 5907 | return MEMCG_PROT_NONE; |
241994ed | 5908 | |
34c81057 SC |
5909 | if (!root) |
5910 | root = root_mem_cgroup; | |
5911 | if (memcg == root) | |
bf8d5d52 | 5912 | return MEMCG_PROT_NONE; |
241994ed | 5913 | |
23067153 | 5914 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 RG |
5915 | if (!usage) |
5916 | return MEMCG_PROT_NONE; | |
5917 | ||
5918 | emin = memcg->memory.min; | |
5919 | elow = memcg->memory.low; | |
34c81057 | 5920 | |
bf8d5d52 | 5921 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
5922 | /* No parent means a non-hierarchical mode on v1 memcg */ |
5923 | if (!parent) | |
5924 | return MEMCG_PROT_NONE; | |
5925 | ||
23067153 RG |
5926 | if (parent == root) |
5927 | goto exit; | |
5928 | ||
bf8d5d52 RG |
5929 | parent_emin = READ_ONCE(parent->memory.emin); |
5930 | emin = min(emin, parent_emin); | |
5931 | if (emin && parent_emin) { | |
5932 | unsigned long min_usage, siblings_min_usage; | |
5933 | ||
5934 | min_usage = min(usage, memcg->memory.min); | |
5935 | siblings_min_usage = atomic_long_read( | |
5936 | &parent->memory.children_min_usage); | |
5937 | ||
5938 | if (min_usage && siblings_min_usage) | |
5939 | emin = min(emin, parent_emin * min_usage / | |
5940 | siblings_min_usage); | |
5941 | } | |
5942 | ||
23067153 RG |
5943 | parent_elow = READ_ONCE(parent->memory.elow); |
5944 | elow = min(elow, parent_elow); | |
bf8d5d52 RG |
5945 | if (elow && parent_elow) { |
5946 | unsigned long low_usage, siblings_low_usage; | |
23067153 | 5947 | |
bf8d5d52 RG |
5948 | low_usage = min(usage, memcg->memory.low); |
5949 | siblings_low_usage = atomic_long_read( | |
5950 | &parent->memory.children_low_usage); | |
23067153 | 5951 | |
bf8d5d52 RG |
5952 | if (low_usage && siblings_low_usage) |
5953 | elow = min(elow, parent_elow * low_usage / | |
5954 | siblings_low_usage); | |
5955 | } | |
23067153 | 5956 | |
23067153 | 5957 | exit: |
bf8d5d52 | 5958 | memcg->memory.emin = emin; |
23067153 | 5959 | memcg->memory.elow = elow; |
bf8d5d52 RG |
5960 | |
5961 | if (usage <= emin) | |
5962 | return MEMCG_PROT_MIN; | |
5963 | else if (usage <= elow) | |
5964 | return MEMCG_PROT_LOW; | |
5965 | else | |
5966 | return MEMCG_PROT_NONE; | |
241994ed JW |
5967 | } |
5968 | ||
00501b53 JW |
5969 | /** |
5970 | * mem_cgroup_try_charge - try charging a page | |
5971 | * @page: page to charge | |
5972 | * @mm: mm context of the victim | |
5973 | * @gfp_mask: reclaim mode | |
5974 | * @memcgp: charged memcg return | |
25843c2b | 5975 | * @compound: charge the page as compound or small page |
00501b53 JW |
5976 | * |
5977 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5978 | * pages according to @gfp_mask if necessary. | |
5979 | * | |
5980 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5981 | * Otherwise, an error code is returned. | |
5982 | * | |
5983 | * After page->mapping has been set up, the caller must finalize the | |
5984 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5985 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5986 | */ | |
5987 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
5988 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
5989 | bool compound) | |
00501b53 JW |
5990 | { |
5991 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 5992 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5993 | int ret = 0; |
5994 | ||
5995 | if (mem_cgroup_disabled()) | |
5996 | goto out; | |
5997 | ||
5998 | if (PageSwapCache(page)) { | |
00501b53 JW |
5999 | /* |
6000 | * Every swap fault against a single page tries to charge the | |
6001 | * page, bail as early as possible. shmem_unuse() encounters | |
6002 | * already charged pages, too. The USED bit is protected by | |
6003 | * the page lock, which serializes swap cache removal, which | |
6004 | * in turn serializes uncharging. | |
6005 | */ | |
e993d905 | 6006 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 6007 | if (compound_head(page)->mem_cgroup) |
00501b53 | 6008 | goto out; |
e993d905 | 6009 | |
37e84351 | 6010 | if (do_swap_account) { |
e993d905 VD |
6011 | swp_entry_t ent = { .val = page_private(page), }; |
6012 | unsigned short id = lookup_swap_cgroup_id(ent); | |
6013 | ||
6014 | rcu_read_lock(); | |
6015 | memcg = mem_cgroup_from_id(id); | |
6016 | if (memcg && !css_tryget_online(&memcg->css)) | |
6017 | memcg = NULL; | |
6018 | rcu_read_unlock(); | |
6019 | } | |
00501b53 JW |
6020 | } |
6021 | ||
00501b53 JW |
6022 | if (!memcg) |
6023 | memcg = get_mem_cgroup_from_mm(mm); | |
6024 | ||
6025 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
6026 | ||
6027 | css_put(&memcg->css); | |
00501b53 JW |
6028 | out: |
6029 | *memcgp = memcg; | |
6030 | return ret; | |
6031 | } | |
6032 | ||
2cf85583 TH |
6033 | int mem_cgroup_try_charge_delay(struct page *page, struct mm_struct *mm, |
6034 | gfp_t gfp_mask, struct mem_cgroup **memcgp, | |
6035 | bool compound) | |
6036 | { | |
6037 | struct mem_cgroup *memcg; | |
6038 | int ret; | |
6039 | ||
6040 | ret = mem_cgroup_try_charge(page, mm, gfp_mask, memcgp, compound); | |
6041 | memcg = *memcgp; | |
6042 | mem_cgroup_throttle_swaprate(memcg, page_to_nid(page), gfp_mask); | |
6043 | return ret; | |
6044 | } | |
6045 | ||
00501b53 JW |
6046 | /** |
6047 | * mem_cgroup_commit_charge - commit a page charge | |
6048 | * @page: page to charge | |
6049 | * @memcg: memcg to charge the page to | |
6050 | * @lrucare: page might be on LRU already | |
25843c2b | 6051 | * @compound: charge the page as compound or small page |
00501b53 JW |
6052 | * |
6053 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
6054 | * after page->mapping has been set up. This must happen atomically | |
6055 | * as part of the page instantiation, i.e. under the page table lock | |
6056 | * for anonymous pages, under the page lock for page and swap cache. | |
6057 | * | |
6058 | * In addition, the page must not be on the LRU during the commit, to | |
6059 | * prevent racing with task migration. If it might be, use @lrucare. | |
6060 | * | |
6061 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
6062 | */ | |
6063 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 6064 | bool lrucare, bool compound) |
00501b53 | 6065 | { |
f627c2f5 | 6066 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6067 | |
6068 | VM_BUG_ON_PAGE(!page->mapping, page); | |
6069 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
6070 | ||
6071 | if (mem_cgroup_disabled()) | |
6072 | return; | |
6073 | /* | |
6074 | * Swap faults will attempt to charge the same page multiple | |
6075 | * times. But reuse_swap_page() might have removed the page | |
6076 | * from swapcache already, so we can't check PageSwapCache(). | |
6077 | */ | |
6078 | if (!memcg) | |
6079 | return; | |
6080 | ||
6abb5a86 JW |
6081 | commit_charge(page, memcg, lrucare); |
6082 | ||
6abb5a86 | 6083 | local_irq_disable(); |
f627c2f5 | 6084 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
6085 | memcg_check_events(memcg, page); |
6086 | local_irq_enable(); | |
00501b53 | 6087 | |
7941d214 | 6088 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
6089 | swp_entry_t entry = { .val = page_private(page) }; |
6090 | /* | |
6091 | * The swap entry might not get freed for a long time, | |
6092 | * let's not wait for it. The page already received a | |
6093 | * memory+swap charge, drop the swap entry duplicate. | |
6094 | */ | |
38d8b4e6 | 6095 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 JW |
6096 | } |
6097 | } | |
6098 | ||
6099 | /** | |
6100 | * mem_cgroup_cancel_charge - cancel a page charge | |
6101 | * @page: page to charge | |
6102 | * @memcg: memcg to charge the page to | |
25843c2b | 6103 | * @compound: charge the page as compound or small page |
00501b53 JW |
6104 | * |
6105 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
6106 | */ | |
f627c2f5 KS |
6107 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
6108 | bool compound) | |
00501b53 | 6109 | { |
f627c2f5 | 6110 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6111 | |
6112 | if (mem_cgroup_disabled()) | |
6113 | return; | |
6114 | /* | |
6115 | * Swap faults will attempt to charge the same page multiple | |
6116 | * times. But reuse_swap_page() might have removed the page | |
6117 | * from swapcache already, so we can't check PageSwapCache(). | |
6118 | */ | |
6119 | if (!memcg) | |
6120 | return; | |
6121 | ||
00501b53 JW |
6122 | cancel_charge(memcg, nr_pages); |
6123 | } | |
6124 | ||
a9d5adee JG |
6125 | struct uncharge_gather { |
6126 | struct mem_cgroup *memcg; | |
6127 | unsigned long pgpgout; | |
6128 | unsigned long nr_anon; | |
6129 | unsigned long nr_file; | |
6130 | unsigned long nr_kmem; | |
6131 | unsigned long nr_huge; | |
6132 | unsigned long nr_shmem; | |
6133 | struct page *dummy_page; | |
6134 | }; | |
6135 | ||
6136 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6137 | { |
a9d5adee JG |
6138 | memset(ug, 0, sizeof(*ug)); |
6139 | } | |
6140 | ||
6141 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6142 | { | |
6143 | unsigned long nr_pages = ug->nr_anon + ug->nr_file + ug->nr_kmem; | |
747db954 JW |
6144 | unsigned long flags; |
6145 | ||
a9d5adee JG |
6146 | if (!mem_cgroup_is_root(ug->memcg)) { |
6147 | page_counter_uncharge(&ug->memcg->memory, nr_pages); | |
7941d214 | 6148 | if (do_memsw_account()) |
a9d5adee JG |
6149 | page_counter_uncharge(&ug->memcg->memsw, nr_pages); |
6150 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) | |
6151 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6152 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6153 | } |
747db954 JW |
6154 | |
6155 | local_irq_save(flags); | |
c9019e9b JW |
6156 | __mod_memcg_state(ug->memcg, MEMCG_RSS, -ug->nr_anon); |
6157 | __mod_memcg_state(ug->memcg, MEMCG_CACHE, -ug->nr_file); | |
6158 | __mod_memcg_state(ug->memcg, MEMCG_RSS_HUGE, -ug->nr_huge); | |
6159 | __mod_memcg_state(ug->memcg, NR_SHMEM, -ug->nr_shmem); | |
6160 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); | |
871789d4 | 6161 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, nr_pages); |
a9d5adee | 6162 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6163 | local_irq_restore(flags); |
e8ea14cc | 6164 | |
a9d5adee JG |
6165 | if (!mem_cgroup_is_root(ug->memcg)) |
6166 | css_put_many(&ug->memcg->css, nr_pages); | |
6167 | } | |
6168 | ||
6169 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6170 | { | |
6171 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
3f2eb028 JG |
6172 | VM_BUG_ON_PAGE(page_count(page) && !is_zone_device_page(page) && |
6173 | !PageHWPoison(page) , page); | |
a9d5adee JG |
6174 | |
6175 | if (!page->mem_cgroup) | |
6176 | return; | |
6177 | ||
6178 | /* | |
6179 | * Nobody should be changing or seriously looking at | |
6180 | * page->mem_cgroup at this point, we have fully | |
6181 | * exclusive access to the page. | |
6182 | */ | |
6183 | ||
6184 | if (ug->memcg != page->mem_cgroup) { | |
6185 | if (ug->memcg) { | |
6186 | uncharge_batch(ug); | |
6187 | uncharge_gather_clear(ug); | |
6188 | } | |
6189 | ug->memcg = page->mem_cgroup; | |
6190 | } | |
6191 | ||
6192 | if (!PageKmemcg(page)) { | |
6193 | unsigned int nr_pages = 1; | |
6194 | ||
6195 | if (PageTransHuge(page)) { | |
6196 | nr_pages <<= compound_order(page); | |
6197 | ug->nr_huge += nr_pages; | |
6198 | } | |
6199 | if (PageAnon(page)) | |
6200 | ug->nr_anon += nr_pages; | |
6201 | else { | |
6202 | ug->nr_file += nr_pages; | |
6203 | if (PageSwapBacked(page)) | |
6204 | ug->nr_shmem += nr_pages; | |
6205 | } | |
6206 | ug->pgpgout++; | |
6207 | } else { | |
6208 | ug->nr_kmem += 1 << compound_order(page); | |
6209 | __ClearPageKmemcg(page); | |
6210 | } | |
6211 | ||
6212 | ug->dummy_page = page; | |
6213 | page->mem_cgroup = NULL; | |
747db954 JW |
6214 | } |
6215 | ||
6216 | static void uncharge_list(struct list_head *page_list) | |
6217 | { | |
a9d5adee | 6218 | struct uncharge_gather ug; |
747db954 | 6219 | struct list_head *next; |
a9d5adee JG |
6220 | |
6221 | uncharge_gather_clear(&ug); | |
747db954 | 6222 | |
8b592656 JW |
6223 | /* |
6224 | * Note that the list can be a single page->lru; hence the | |
6225 | * do-while loop instead of a simple list_for_each_entry(). | |
6226 | */ | |
747db954 JW |
6227 | next = page_list->next; |
6228 | do { | |
a9d5adee JG |
6229 | struct page *page; |
6230 | ||
747db954 JW |
6231 | page = list_entry(next, struct page, lru); |
6232 | next = page->lru.next; | |
6233 | ||
a9d5adee | 6234 | uncharge_page(page, &ug); |
747db954 JW |
6235 | } while (next != page_list); |
6236 | ||
a9d5adee JG |
6237 | if (ug.memcg) |
6238 | uncharge_batch(&ug); | |
747db954 JW |
6239 | } |
6240 | ||
0a31bc97 JW |
6241 | /** |
6242 | * mem_cgroup_uncharge - uncharge a page | |
6243 | * @page: page to uncharge | |
6244 | * | |
6245 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
6246 | * mem_cgroup_commit_charge(). | |
6247 | */ | |
6248 | void mem_cgroup_uncharge(struct page *page) | |
6249 | { | |
a9d5adee JG |
6250 | struct uncharge_gather ug; |
6251 | ||
0a31bc97 JW |
6252 | if (mem_cgroup_disabled()) |
6253 | return; | |
6254 | ||
747db954 | 6255 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6256 | if (!page->mem_cgroup) |
0a31bc97 JW |
6257 | return; |
6258 | ||
a9d5adee JG |
6259 | uncharge_gather_clear(&ug); |
6260 | uncharge_page(page, &ug); | |
6261 | uncharge_batch(&ug); | |
747db954 | 6262 | } |
0a31bc97 | 6263 | |
747db954 JW |
6264 | /** |
6265 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6266 | * @page_list: list of pages to uncharge | |
6267 | * | |
6268 | * Uncharge a list of pages previously charged with | |
6269 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
6270 | */ | |
6271 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6272 | { | |
6273 | if (mem_cgroup_disabled()) | |
6274 | return; | |
0a31bc97 | 6275 | |
747db954 JW |
6276 | if (!list_empty(page_list)) |
6277 | uncharge_list(page_list); | |
0a31bc97 JW |
6278 | } |
6279 | ||
6280 | /** | |
6a93ca8f JW |
6281 | * mem_cgroup_migrate - charge a page's replacement |
6282 | * @oldpage: currently circulating page | |
6283 | * @newpage: replacement page | |
0a31bc97 | 6284 | * |
6a93ca8f JW |
6285 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6286 | * be uncharged upon free. | |
0a31bc97 JW |
6287 | * |
6288 | * Both pages must be locked, @newpage->mapping must be set up. | |
6289 | */ | |
6a93ca8f | 6290 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6291 | { |
29833315 | 6292 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
6293 | unsigned int nr_pages; |
6294 | bool compound; | |
d93c4130 | 6295 | unsigned long flags; |
0a31bc97 JW |
6296 | |
6297 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6298 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6299 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6300 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6301 | newpage); | |
0a31bc97 JW |
6302 | |
6303 | if (mem_cgroup_disabled()) | |
6304 | return; | |
6305 | ||
6306 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6307 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6308 | return; |
6309 | ||
45637bab | 6310 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6311 | memcg = oldpage->mem_cgroup; |
29833315 | 6312 | if (!memcg) |
0a31bc97 JW |
6313 | return; |
6314 | ||
44b7a8d3 JW |
6315 | /* Force-charge the new page. The old one will be freed soon */ |
6316 | compound = PageTransHuge(newpage); | |
6317 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
6318 | ||
6319 | page_counter_charge(&memcg->memory, nr_pages); | |
6320 | if (do_memsw_account()) | |
6321 | page_counter_charge(&memcg->memsw, nr_pages); | |
6322 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 6323 | |
9cf7666a | 6324 | commit_charge(newpage, memcg, false); |
44b7a8d3 | 6325 | |
d93c4130 | 6326 | local_irq_save(flags); |
44b7a8d3 JW |
6327 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
6328 | memcg_check_events(memcg, newpage); | |
d93c4130 | 6329 | local_irq_restore(flags); |
0a31bc97 JW |
6330 | } |
6331 | ||
ef12947c | 6332 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6333 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6334 | ||
2d758073 | 6335 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6336 | { |
6337 | struct mem_cgroup *memcg; | |
6338 | ||
2d758073 JW |
6339 | if (!mem_cgroup_sockets_enabled) |
6340 | return; | |
6341 | ||
edbe69ef RG |
6342 | /* |
6343 | * Socket cloning can throw us here with sk_memcg already | |
6344 | * filled. It won't however, necessarily happen from | |
6345 | * process context. So the test for root memcg given | |
6346 | * the current task's memcg won't help us in this case. | |
6347 | * | |
6348 | * Respecting the original socket's memcg is a better | |
6349 | * decision in this case. | |
6350 | */ | |
6351 | if (sk->sk_memcg) { | |
6352 | css_get(&sk->sk_memcg->css); | |
6353 | return; | |
6354 | } | |
6355 | ||
11092087 JW |
6356 | rcu_read_lock(); |
6357 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6358 | if (memcg == root_mem_cgroup) |
6359 | goto out; | |
0db15298 | 6360 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6361 | goto out; |
f7e1cb6e | 6362 | if (css_tryget_online(&memcg->css)) |
11092087 | 6363 | sk->sk_memcg = memcg; |
f7e1cb6e | 6364 | out: |
11092087 JW |
6365 | rcu_read_unlock(); |
6366 | } | |
11092087 | 6367 | |
2d758073 | 6368 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6369 | { |
2d758073 JW |
6370 | if (sk->sk_memcg) |
6371 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6372 | } |
6373 | ||
6374 | /** | |
6375 | * mem_cgroup_charge_skmem - charge socket memory | |
6376 | * @memcg: memcg to charge | |
6377 | * @nr_pages: number of pages to charge | |
6378 | * | |
6379 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6380 | * @memcg's configured limit, %false if the charge had to be forced. | |
6381 | */ | |
6382 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6383 | { | |
f7e1cb6e | 6384 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6385 | |
f7e1cb6e | 6386 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6387 | struct page_counter *fail; |
f7e1cb6e | 6388 | |
0db15298 JW |
6389 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6390 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6391 | return true; |
6392 | } | |
0db15298 JW |
6393 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6394 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6395 | return false; |
11092087 | 6396 | } |
d886f4e4 | 6397 | |
f7e1cb6e JW |
6398 | /* Don't block in the packet receive path */ |
6399 | if (in_softirq()) | |
6400 | gfp_mask = GFP_NOWAIT; | |
6401 | ||
c9019e9b | 6402 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6403 | |
f7e1cb6e JW |
6404 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6405 | return true; | |
6406 | ||
6407 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6408 | return false; |
6409 | } | |
6410 | ||
6411 | /** | |
6412 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
6413 | * @memcg: memcg to uncharge |
6414 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
6415 | */ |
6416 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6417 | { | |
f7e1cb6e | 6418 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6419 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
6420 | return; |
6421 | } | |
d886f4e4 | 6422 | |
c9019e9b | 6423 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 6424 | |
475d0487 | 6425 | refill_stock(memcg, nr_pages); |
11092087 JW |
6426 | } |
6427 | ||
f7e1cb6e JW |
6428 | static int __init cgroup_memory(char *s) |
6429 | { | |
6430 | char *token; | |
6431 | ||
6432 | while ((token = strsep(&s, ",")) != NULL) { | |
6433 | if (!*token) | |
6434 | continue; | |
6435 | if (!strcmp(token, "nosocket")) | |
6436 | cgroup_memory_nosocket = true; | |
04823c83 VD |
6437 | if (!strcmp(token, "nokmem")) |
6438 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
6439 | } |
6440 | return 0; | |
6441 | } | |
6442 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 6443 | |
2d11085e | 6444 | /* |
1081312f MH |
6445 | * subsys_initcall() for memory controller. |
6446 | * | |
308167fc SAS |
6447 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
6448 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
6449 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
6450 | * should be initialized from here. | |
2d11085e MH |
6451 | */ |
6452 | static int __init mem_cgroup_init(void) | |
6453 | { | |
95a045f6 JW |
6454 | int cpu, node; |
6455 | ||
84c07d11 | 6456 | #ifdef CONFIG_MEMCG_KMEM |
13583c3d VD |
6457 | /* |
6458 | * Kmem cache creation is mostly done with the slab_mutex held, | |
17cc4dfe TH |
6459 | * so use a workqueue with limited concurrency to avoid stalling |
6460 | * all worker threads in case lots of cgroups are created and | |
6461 | * destroyed simultaneously. | |
13583c3d | 6462 | */ |
17cc4dfe TH |
6463 | memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1); |
6464 | BUG_ON(!memcg_kmem_cache_wq); | |
13583c3d VD |
6465 | #endif |
6466 | ||
308167fc SAS |
6467 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
6468 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
6469 | |
6470 | for_each_possible_cpu(cpu) | |
6471 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
6472 | drain_local_stock); | |
6473 | ||
6474 | for_each_node(node) { | |
6475 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
6476 | |
6477 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
6478 | node_online(node) ? node : NUMA_NO_NODE); | |
6479 | ||
ef8f2327 | 6480 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 6481 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 6482 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
6483 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
6484 | } | |
6485 | ||
2d11085e MH |
6486 | return 0; |
6487 | } | |
6488 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
6489 | |
6490 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
6491 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
6492 | { | |
1c2d479a | 6493 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
6494 | /* |
6495 | * The root cgroup cannot be destroyed, so it's refcount must | |
6496 | * always be >= 1. | |
6497 | */ | |
6498 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
6499 | VM_BUG_ON(1); | |
6500 | break; | |
6501 | } | |
6502 | memcg = parent_mem_cgroup(memcg); | |
6503 | if (!memcg) | |
6504 | memcg = root_mem_cgroup; | |
6505 | } | |
6506 | return memcg; | |
6507 | } | |
6508 | ||
21afa38e JW |
6509 | /** |
6510 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
6511 | * @page: page whose memsw charge to transfer | |
6512 | * @entry: swap entry to move the charge to | |
6513 | * | |
6514 | * Transfer the memsw charge of @page to @entry. | |
6515 | */ | |
6516 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
6517 | { | |
1f47b61f | 6518 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 6519 | unsigned int nr_entries; |
21afa38e JW |
6520 | unsigned short oldid; |
6521 | ||
6522 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6523 | VM_BUG_ON_PAGE(page_count(page), page); | |
6524 | ||
7941d214 | 6525 | if (!do_memsw_account()) |
21afa38e JW |
6526 | return; |
6527 | ||
6528 | memcg = page->mem_cgroup; | |
6529 | ||
6530 | /* Readahead page, never charged */ | |
6531 | if (!memcg) | |
6532 | return; | |
6533 | ||
1f47b61f VD |
6534 | /* |
6535 | * In case the memcg owning these pages has been offlined and doesn't | |
6536 | * have an ID allocated to it anymore, charge the closest online | |
6537 | * ancestor for the swap instead and transfer the memory+swap charge. | |
6538 | */ | |
6539 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
d6810d73 HY |
6540 | nr_entries = hpage_nr_pages(page); |
6541 | /* Get references for the tail pages, too */ | |
6542 | if (nr_entries > 1) | |
6543 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
6544 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
6545 | nr_entries); | |
21afa38e | 6546 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6547 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
6548 | |
6549 | page->mem_cgroup = NULL; | |
6550 | ||
6551 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 6552 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 6553 | |
1f47b61f VD |
6554 | if (memcg != swap_memcg) { |
6555 | if (!mem_cgroup_is_root(swap_memcg)) | |
d6810d73 HY |
6556 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
6557 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
6558 | } |
6559 | ||
ce9ce665 SAS |
6560 | /* |
6561 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 6562 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 6563 | * important here to have the interrupts disabled because it is the |
b93b0163 | 6564 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
6565 | */ |
6566 | VM_BUG_ON(!irqs_disabled()); | |
d6810d73 HY |
6567 | mem_cgroup_charge_statistics(memcg, page, PageTransHuge(page), |
6568 | -nr_entries); | |
21afa38e | 6569 | memcg_check_events(memcg, page); |
73f576c0 JW |
6570 | |
6571 | if (!mem_cgroup_is_root(memcg)) | |
d08afa14 | 6572 | css_put_many(&memcg->css, nr_entries); |
21afa38e JW |
6573 | } |
6574 | ||
38d8b4e6 HY |
6575 | /** |
6576 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
6577 | * @page: page being added to swap |
6578 | * @entry: swap entry to charge | |
6579 | * | |
38d8b4e6 | 6580 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
6581 | * |
6582 | * Returns 0 on success, -ENOMEM on failure. | |
6583 | */ | |
6584 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
6585 | { | |
38d8b4e6 | 6586 | unsigned int nr_pages = hpage_nr_pages(page); |
37e84351 | 6587 | struct page_counter *counter; |
38d8b4e6 | 6588 | struct mem_cgroup *memcg; |
37e84351 VD |
6589 | unsigned short oldid; |
6590 | ||
6591 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
6592 | return 0; | |
6593 | ||
6594 | memcg = page->mem_cgroup; | |
6595 | ||
6596 | /* Readahead page, never charged */ | |
6597 | if (!memcg) | |
6598 | return 0; | |
6599 | ||
f3a53a3a TH |
6600 | if (!entry.val) { |
6601 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 6602 | return 0; |
f3a53a3a | 6603 | } |
bb98f2c5 | 6604 | |
1f47b61f VD |
6605 | memcg = mem_cgroup_id_get_online(memcg); |
6606 | ||
37e84351 | 6607 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 6608 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
6609 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
6610 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 6611 | mem_cgroup_id_put(memcg); |
37e84351 | 6612 | return -ENOMEM; |
1f47b61f | 6613 | } |
37e84351 | 6614 | |
38d8b4e6 HY |
6615 | /* Get references for the tail pages, too */ |
6616 | if (nr_pages > 1) | |
6617 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
6618 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 6619 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6620 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 6621 | |
37e84351 VD |
6622 | return 0; |
6623 | } | |
6624 | ||
21afa38e | 6625 | /** |
38d8b4e6 | 6626 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 6627 | * @entry: swap entry to uncharge |
38d8b4e6 | 6628 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 6629 | */ |
38d8b4e6 | 6630 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
6631 | { |
6632 | struct mem_cgroup *memcg; | |
6633 | unsigned short id; | |
6634 | ||
37e84351 | 6635 | if (!do_swap_account) |
21afa38e JW |
6636 | return; |
6637 | ||
38d8b4e6 | 6638 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 6639 | rcu_read_lock(); |
adbe427b | 6640 | memcg = mem_cgroup_from_id(id); |
21afa38e | 6641 | if (memcg) { |
37e84351 VD |
6642 | if (!mem_cgroup_is_root(memcg)) { |
6643 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
38d8b4e6 | 6644 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 6645 | else |
38d8b4e6 | 6646 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 6647 | } |
c9019e9b | 6648 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 6649 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
6650 | } |
6651 | rcu_read_unlock(); | |
6652 | } | |
6653 | ||
d8b38438 VD |
6654 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
6655 | { | |
6656 | long nr_swap_pages = get_nr_swap_pages(); | |
6657 | ||
6658 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6659 | return nr_swap_pages; | |
6660 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
6661 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 6662 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
6663 | page_counter_read(&memcg->swap)); |
6664 | return nr_swap_pages; | |
6665 | } | |
6666 | ||
5ccc5aba VD |
6667 | bool mem_cgroup_swap_full(struct page *page) |
6668 | { | |
6669 | struct mem_cgroup *memcg; | |
6670 | ||
6671 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
6672 | ||
6673 | if (vm_swap_full()) | |
6674 | return true; | |
6675 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6676 | return false; | |
6677 | ||
6678 | memcg = page->mem_cgroup; | |
6679 | if (!memcg) | |
6680 | return false; | |
6681 | ||
6682 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
bbec2e15 | 6683 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.max) |
5ccc5aba VD |
6684 | return true; |
6685 | ||
6686 | return false; | |
6687 | } | |
6688 | ||
21afa38e JW |
6689 | /* for remember boot option*/ |
6690 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
6691 | static int really_do_swap_account __initdata = 1; | |
6692 | #else | |
6693 | static int really_do_swap_account __initdata; | |
6694 | #endif | |
6695 | ||
6696 | static int __init enable_swap_account(char *s) | |
6697 | { | |
6698 | if (!strcmp(s, "1")) | |
6699 | really_do_swap_account = 1; | |
6700 | else if (!strcmp(s, "0")) | |
6701 | really_do_swap_account = 0; | |
6702 | return 1; | |
6703 | } | |
6704 | __setup("swapaccount=", enable_swap_account); | |
6705 | ||
37e84351 VD |
6706 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
6707 | struct cftype *cft) | |
6708 | { | |
6709 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6710 | ||
6711 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
6712 | } | |
6713 | ||
6714 | static int swap_max_show(struct seq_file *m, void *v) | |
6715 | { | |
677dc973 CD |
6716 | return seq_puts_memcg_tunable(m, |
6717 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
6718 | } |
6719 | ||
6720 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
6721 | char *buf, size_t nbytes, loff_t off) | |
6722 | { | |
6723 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6724 | unsigned long max; | |
6725 | int err; | |
6726 | ||
6727 | buf = strstrip(buf); | |
6728 | err = page_counter_memparse(buf, "max", &max); | |
6729 | if (err) | |
6730 | return err; | |
6731 | ||
be09102b | 6732 | xchg(&memcg->swap.max, max); |
37e84351 VD |
6733 | |
6734 | return nbytes; | |
6735 | } | |
6736 | ||
f3a53a3a TH |
6737 | static int swap_events_show(struct seq_file *m, void *v) |
6738 | { | |
aa9694bb | 6739 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a TH |
6740 | |
6741 | seq_printf(m, "max %lu\n", | |
6742 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
6743 | seq_printf(m, "fail %lu\n", | |
6744 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
6745 | ||
6746 | return 0; | |
6747 | } | |
6748 | ||
37e84351 VD |
6749 | static struct cftype swap_files[] = { |
6750 | { | |
6751 | .name = "swap.current", | |
6752 | .flags = CFTYPE_NOT_ON_ROOT, | |
6753 | .read_u64 = swap_current_read, | |
6754 | }, | |
6755 | { | |
6756 | .name = "swap.max", | |
6757 | .flags = CFTYPE_NOT_ON_ROOT, | |
6758 | .seq_show = swap_max_show, | |
6759 | .write = swap_max_write, | |
6760 | }, | |
f3a53a3a TH |
6761 | { |
6762 | .name = "swap.events", | |
6763 | .flags = CFTYPE_NOT_ON_ROOT, | |
6764 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
6765 | .seq_show = swap_events_show, | |
6766 | }, | |
37e84351 VD |
6767 | { } /* terminate */ |
6768 | }; | |
6769 | ||
21afa38e JW |
6770 | static struct cftype memsw_cgroup_files[] = { |
6771 | { | |
6772 | .name = "memsw.usage_in_bytes", | |
6773 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6774 | .read_u64 = mem_cgroup_read_u64, | |
6775 | }, | |
6776 | { | |
6777 | .name = "memsw.max_usage_in_bytes", | |
6778 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6779 | .write = mem_cgroup_reset, | |
6780 | .read_u64 = mem_cgroup_read_u64, | |
6781 | }, | |
6782 | { | |
6783 | .name = "memsw.limit_in_bytes", | |
6784 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6785 | .write = mem_cgroup_write, | |
6786 | .read_u64 = mem_cgroup_read_u64, | |
6787 | }, | |
6788 | { | |
6789 | .name = "memsw.failcnt", | |
6790 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6791 | .write = mem_cgroup_reset, | |
6792 | .read_u64 = mem_cgroup_read_u64, | |
6793 | }, | |
6794 | { }, /* terminate */ | |
6795 | }; | |
6796 | ||
6797 | static int __init mem_cgroup_swap_init(void) | |
6798 | { | |
6799 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6800 | do_swap_account = 1; | |
37e84351 VD |
6801 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
6802 | swap_files)); | |
21afa38e JW |
6803 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6804 | memsw_cgroup_files)); | |
6805 | } | |
6806 | return 0; | |
6807 | } | |
6808 | subsys_initcall(mem_cgroup_swap_init); | |
6809 | ||
6810 | #endif /* CONFIG_MEMCG_SWAP */ |