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> |
d52aa412 | 42 | #include <linux/smp.h> |
8a9f3ccd | 43 | #include <linux/page-flags.h> |
66e1707b | 44 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
45 | #include <linux/bit_spinlock.h> |
46 | #include <linux/rcupdate.h> | |
e222432b | 47 | #include <linux/limits.h> |
b9e15baf | 48 | #include <linux/export.h> |
8c7c6e34 | 49 | #include <linux/mutex.h> |
bb4cc1a8 | 50 | #include <linux/rbtree.h> |
b6ac57d5 | 51 | #include <linux/slab.h> |
66e1707b | 52 | #include <linux/swap.h> |
02491447 | 53 | #include <linux/swapops.h> |
66e1707b | 54 | #include <linux/spinlock.h> |
2e72b634 | 55 | #include <linux/eventfd.h> |
79bd9814 | 56 | #include <linux/poll.h> |
2e72b634 | 57 | #include <linux/sort.h> |
66e1707b | 58 | #include <linux/fs.h> |
d2ceb9b7 | 59 | #include <linux/seq_file.h> |
70ddf637 | 60 | #include <linux/vmpressure.h> |
b69408e8 | 61 | #include <linux/mm_inline.h> |
5d1ea48b | 62 | #include <linux/swap_cgroup.h> |
cdec2e42 | 63 | #include <linux/cpu.h> |
158e0a2d | 64 | #include <linux/oom.h> |
0056f4e6 | 65 | #include <linux/lockdep.h> |
79bd9814 | 66 | #include <linux/file.h> |
b23afb93 | 67 | #include <linux/tracehook.h> |
08e552c6 | 68 | #include "internal.h" |
d1a4c0b3 | 69 | #include <net/sock.h> |
4bd2c1ee | 70 | #include <net/ip.h> |
f35c3a8e | 71 | #include "slab.h" |
8cdea7c0 | 72 | |
7c0f6ba6 | 73 | #include <linux/uaccess.h> |
8697d331 | 74 | |
cc8e970c KM |
75 | #include <trace/events/vmscan.h> |
76 | ||
073219e9 TH |
77 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
78 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 79 | |
7d828602 JW |
80 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
81 | ||
a181b0e8 | 82 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 83 | |
f7e1cb6e JW |
84 | /* Socket memory accounting disabled? */ |
85 | static bool cgroup_memory_nosocket; | |
86 | ||
04823c83 VD |
87 | /* Kernel memory accounting disabled? */ |
88 | static bool cgroup_memory_nokmem; | |
89 | ||
21afa38e | 90 | /* Whether the swap controller is active */ |
c255a458 | 91 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 92 | int do_swap_account __read_mostly; |
c077719b | 93 | #else |
a0db00fc | 94 | #define do_swap_account 0 |
c077719b KH |
95 | #endif |
96 | ||
7941d214 JW |
97 | /* Whether legacy memory+swap accounting is active */ |
98 | static bool do_memsw_account(void) | |
99 | { | |
100 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; | |
101 | } | |
102 | ||
71cd3113 | 103 | static const char *const mem_cgroup_lru_names[] = { |
58cf188e SZ |
104 | "inactive_anon", |
105 | "active_anon", | |
106 | "inactive_file", | |
107 | "active_file", | |
108 | "unevictable", | |
109 | }; | |
110 | ||
a0db00fc KS |
111 | #define THRESHOLDS_EVENTS_TARGET 128 |
112 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
113 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 114 | |
bb4cc1a8 AM |
115 | /* |
116 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
117 | * their hierarchy representation | |
118 | */ | |
119 | ||
ef8f2327 | 120 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 121 | struct rb_root rb_root; |
fa90b2fd | 122 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
123 | spinlock_t lock; |
124 | }; | |
125 | ||
bb4cc1a8 AM |
126 | struct mem_cgroup_tree { |
127 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
128 | }; | |
129 | ||
130 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
131 | ||
9490ff27 KH |
132 | /* for OOM */ |
133 | struct mem_cgroup_eventfd_list { | |
134 | struct list_head list; | |
135 | struct eventfd_ctx *eventfd; | |
136 | }; | |
2e72b634 | 137 | |
79bd9814 TH |
138 | /* |
139 | * cgroup_event represents events which userspace want to receive. | |
140 | */ | |
3bc942f3 | 141 | struct mem_cgroup_event { |
79bd9814 | 142 | /* |
59b6f873 | 143 | * memcg which the event belongs to. |
79bd9814 | 144 | */ |
59b6f873 | 145 | struct mem_cgroup *memcg; |
79bd9814 TH |
146 | /* |
147 | * eventfd to signal userspace about the event. | |
148 | */ | |
149 | struct eventfd_ctx *eventfd; | |
150 | /* | |
151 | * Each of these stored in a list by the cgroup. | |
152 | */ | |
153 | struct list_head list; | |
fba94807 TH |
154 | /* |
155 | * register_event() callback will be used to add new userspace | |
156 | * waiter for changes related to this event. Use eventfd_signal() | |
157 | * on eventfd to send notification to userspace. | |
158 | */ | |
59b6f873 | 159 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 160 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
161 | /* |
162 | * unregister_event() callback will be called when userspace closes | |
163 | * the eventfd or on cgroup removing. This callback must be set, | |
164 | * if you want provide notification functionality. | |
165 | */ | |
59b6f873 | 166 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 167 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
168 | /* |
169 | * All fields below needed to unregister event when | |
170 | * userspace closes eventfd. | |
171 | */ | |
172 | poll_table pt; | |
173 | wait_queue_head_t *wqh; | |
ac6424b9 | 174 | wait_queue_entry_t wait; |
79bd9814 TH |
175 | struct work_struct remove; |
176 | }; | |
177 | ||
c0ff4b85 R |
178 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
179 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 180 | |
7dc74be0 DN |
181 | /* Stuffs for move charges at task migration. */ |
182 | /* | |
1dfab5ab | 183 | * Types of charges to be moved. |
7dc74be0 | 184 | */ |
1dfab5ab JW |
185 | #define MOVE_ANON 0x1U |
186 | #define MOVE_FILE 0x2U | |
187 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 188 | |
4ffef5fe DN |
189 | /* "mc" and its members are protected by cgroup_mutex */ |
190 | static struct move_charge_struct { | |
b1dd693e | 191 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 192 | struct mm_struct *mm; |
4ffef5fe DN |
193 | struct mem_cgroup *from; |
194 | struct mem_cgroup *to; | |
1dfab5ab | 195 | unsigned long flags; |
4ffef5fe | 196 | unsigned long precharge; |
854ffa8d | 197 | unsigned long moved_charge; |
483c30b5 | 198 | unsigned long moved_swap; |
8033b97c DN |
199 | struct task_struct *moving_task; /* a task moving charges */ |
200 | wait_queue_head_t waitq; /* a waitq for other context */ | |
201 | } mc = { | |
2bd9bb20 | 202 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
203 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
204 | }; | |
4ffef5fe | 205 | |
4e416953 BS |
206 | /* |
207 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
208 | * limit reclaim to prevent infinite loops, if they ever occur. | |
209 | */ | |
a0db00fc | 210 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 211 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 212 | |
217bc319 KH |
213 | enum charge_type { |
214 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 215 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 216 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 217 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
218 | NR_CHARGE_TYPE, |
219 | }; | |
220 | ||
8c7c6e34 | 221 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
222 | enum res_type { |
223 | _MEM, | |
224 | _MEMSWAP, | |
225 | _OOM_TYPE, | |
510fc4e1 | 226 | _KMEM, |
d55f90bf | 227 | _TCP, |
86ae53e1 GC |
228 | }; |
229 | ||
a0db00fc KS |
230 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
231 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 232 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
233 | /* Used for OOM nofiier */ |
234 | #define OOM_CONTROL (0) | |
8c7c6e34 | 235 | |
b05706f1 KT |
236 | /* |
237 | * Iteration constructs for visiting all cgroups (under a tree). If | |
238 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
239 | * be used for reference counting. | |
240 | */ | |
241 | #define for_each_mem_cgroup_tree(iter, root) \ | |
242 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
243 | iter != NULL; \ | |
244 | iter = mem_cgroup_iter(root, iter, NULL)) | |
245 | ||
246 | #define for_each_mem_cgroup(iter) \ | |
247 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
248 | iter != NULL; \ | |
249 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
250 | ||
70ddf637 AV |
251 | /* Some nice accessors for the vmpressure. */ |
252 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
253 | { | |
254 | if (!memcg) | |
255 | memcg = root_mem_cgroup; | |
256 | return &memcg->vmpressure; | |
257 | } | |
258 | ||
259 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
260 | { | |
261 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
262 | } | |
263 | ||
84c07d11 | 264 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 265 | /* |
f7ce3190 | 266 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
267 | * The main reason for not using cgroup id for this: |
268 | * this works better in sparse environments, where we have a lot of memcgs, | |
269 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
270 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
271 | * 200 entry array for that. | |
55007d84 | 272 | * |
dbcf73e2 VD |
273 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
274 | * will double each time we have to increase it. | |
55007d84 | 275 | */ |
dbcf73e2 VD |
276 | static DEFINE_IDA(memcg_cache_ida); |
277 | int memcg_nr_cache_ids; | |
749c5415 | 278 | |
05257a1a VD |
279 | /* Protects memcg_nr_cache_ids */ |
280 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
281 | ||
282 | void memcg_get_cache_ids(void) | |
283 | { | |
284 | down_read(&memcg_cache_ids_sem); | |
285 | } | |
286 | ||
287 | void memcg_put_cache_ids(void) | |
288 | { | |
289 | up_read(&memcg_cache_ids_sem); | |
290 | } | |
291 | ||
55007d84 GC |
292 | /* |
293 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
294 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
295 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
296 | * tunable, but that is strictly not necessary. | |
297 | * | |
b8627835 | 298 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
299 | * this constant directly from cgroup, but it is understandable that this is |
300 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 301 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
302 | * increase ours as well if it increases. |
303 | */ | |
304 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 305 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 306 | |
d7f25f8a GC |
307 | /* |
308 | * A lot of the calls to the cache allocation functions are expected to be | |
309 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
310 | * conditional to this static branch, we'll have to allow modules that does | |
311 | * kmem_cache_alloc and the such to see this symbol as well | |
312 | */ | |
ef12947c | 313 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 314 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 315 | |
17cc4dfe TH |
316 | struct workqueue_struct *memcg_kmem_cache_wq; |
317 | ||
0a4465d3 KT |
318 | static int memcg_shrinker_map_size; |
319 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); | |
320 | ||
321 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) | |
322 | { | |
323 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); | |
324 | } | |
325 | ||
326 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, | |
327 | int size, int old_size) | |
328 | { | |
329 | struct memcg_shrinker_map *new, *old; | |
330 | int nid; | |
331 | ||
332 | lockdep_assert_held(&memcg_shrinker_map_mutex); | |
333 | ||
334 | for_each_node(nid) { | |
335 | old = rcu_dereference_protected( | |
336 | mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); | |
337 | /* Not yet online memcg */ | |
338 | if (!old) | |
339 | return 0; | |
340 | ||
341 | new = kvmalloc(sizeof(*new) + size, GFP_KERNEL); | |
342 | if (!new) | |
343 | return -ENOMEM; | |
344 | ||
345 | /* Set all old bits, clear all new bits */ | |
346 | memset(new->map, (int)0xff, old_size); | |
347 | memset((void *)new->map + old_size, 0, size - old_size); | |
348 | ||
349 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); | |
350 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); | |
351 | } | |
352 | ||
353 | return 0; | |
354 | } | |
355 | ||
356 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) | |
357 | { | |
358 | struct mem_cgroup_per_node *pn; | |
359 | struct memcg_shrinker_map *map; | |
360 | int nid; | |
361 | ||
362 | if (mem_cgroup_is_root(memcg)) | |
363 | return; | |
364 | ||
365 | for_each_node(nid) { | |
366 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
367 | map = rcu_dereference_protected(pn->shrinker_map, true); | |
368 | if (map) | |
369 | kvfree(map); | |
370 | rcu_assign_pointer(pn->shrinker_map, NULL); | |
371 | } | |
372 | } | |
373 | ||
374 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
375 | { | |
376 | struct memcg_shrinker_map *map; | |
377 | int nid, size, ret = 0; | |
378 | ||
379 | if (mem_cgroup_is_root(memcg)) | |
380 | return 0; | |
381 | ||
382 | mutex_lock(&memcg_shrinker_map_mutex); | |
383 | size = memcg_shrinker_map_size; | |
384 | for_each_node(nid) { | |
385 | map = kvzalloc(sizeof(*map) + size, GFP_KERNEL); | |
386 | if (!map) { | |
387 | memcg_free_shrinker_maps(memcg); | |
388 | ret = -ENOMEM; | |
389 | break; | |
390 | } | |
391 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); | |
392 | } | |
393 | mutex_unlock(&memcg_shrinker_map_mutex); | |
394 | ||
395 | return ret; | |
396 | } | |
397 | ||
398 | int memcg_expand_shrinker_maps(int new_id) | |
399 | { | |
400 | int size, old_size, ret = 0; | |
401 | struct mem_cgroup *memcg; | |
402 | ||
403 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); | |
404 | old_size = memcg_shrinker_map_size; | |
405 | if (size <= old_size) | |
406 | return 0; | |
407 | ||
408 | mutex_lock(&memcg_shrinker_map_mutex); | |
409 | if (!root_mem_cgroup) | |
410 | goto unlock; | |
411 | ||
412 | for_each_mem_cgroup(memcg) { | |
413 | if (mem_cgroup_is_root(memcg)) | |
414 | continue; | |
415 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); | |
416 | if (ret) | |
417 | goto unlock; | |
418 | } | |
419 | unlock: | |
420 | if (!ret) | |
421 | memcg_shrinker_map_size = size; | |
422 | mutex_unlock(&memcg_shrinker_map_mutex); | |
423 | return ret; | |
424 | } | |
fae91d6d KT |
425 | |
426 | void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id) | |
427 | { | |
428 | if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) { | |
429 | struct memcg_shrinker_map *map; | |
430 | ||
431 | rcu_read_lock(); | |
432 | map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map); | |
f90280d6 KT |
433 | /* Pairs with smp mb in shrink_slab() */ |
434 | smp_mb__before_atomic(); | |
fae91d6d KT |
435 | set_bit(shrinker_id, map->map); |
436 | rcu_read_unlock(); | |
437 | } | |
438 | } | |
439 | ||
0a4465d3 KT |
440 | #else /* CONFIG_MEMCG_KMEM */ |
441 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
442 | { | |
443 | return 0; | |
444 | } | |
445 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) { } | |
84c07d11 | 446 | #endif /* CONFIG_MEMCG_KMEM */ |
a8964b9b | 447 | |
ad7fa852 TH |
448 | /** |
449 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
450 | * @page: page of interest | |
451 | * | |
452 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
453 | * with @page is returned. The returned css remains associated with @page | |
454 | * until it is released. | |
455 | * | |
456 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
457 | * is returned. | |
ad7fa852 TH |
458 | */ |
459 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
460 | { | |
461 | struct mem_cgroup *memcg; | |
462 | ||
ad7fa852 TH |
463 | memcg = page->mem_cgroup; |
464 | ||
9e10a130 | 465 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
466 | memcg = root_mem_cgroup; |
467 | ||
ad7fa852 TH |
468 | return &memcg->css; |
469 | } | |
470 | ||
2fc04524 VD |
471 | /** |
472 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
473 | * @page: the page | |
474 | * | |
475 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
476 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
477 | * is safe to call this function without holding a reference to @page. | |
478 | * | |
479 | * Note, this function is inherently racy, because there is nothing to prevent | |
480 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
481 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
482 | * do not care (such as procfs interfaces). | |
483 | */ | |
484 | ino_t page_cgroup_ino(struct page *page) | |
485 | { | |
486 | struct mem_cgroup *memcg; | |
487 | unsigned long ino = 0; | |
488 | ||
489 | rcu_read_lock(); | |
490 | memcg = READ_ONCE(page->mem_cgroup); | |
491 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
492 | memcg = parent_mem_cgroup(memcg); | |
493 | if (memcg) | |
494 | ino = cgroup_ino(memcg->css.cgroup); | |
495 | rcu_read_unlock(); | |
496 | return ino; | |
497 | } | |
498 | ||
ef8f2327 MG |
499 | static struct mem_cgroup_per_node * |
500 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 501 | { |
97a6c37b | 502 | int nid = page_to_nid(page); |
f64c3f54 | 503 | |
ef8f2327 | 504 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
505 | } |
506 | ||
ef8f2327 MG |
507 | static struct mem_cgroup_tree_per_node * |
508 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 509 | { |
ef8f2327 | 510 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
511 | } |
512 | ||
ef8f2327 | 513 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
514 | soft_limit_tree_from_page(struct page *page) |
515 | { | |
516 | int nid = page_to_nid(page); | |
bb4cc1a8 | 517 | |
ef8f2327 | 518 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
519 | } |
520 | ||
ef8f2327 MG |
521 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
522 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 523 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
524 | { |
525 | struct rb_node **p = &mctz->rb_root.rb_node; | |
526 | struct rb_node *parent = NULL; | |
ef8f2327 | 527 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 528 | bool rightmost = true; |
bb4cc1a8 AM |
529 | |
530 | if (mz->on_tree) | |
531 | return; | |
532 | ||
533 | mz->usage_in_excess = new_usage_in_excess; | |
534 | if (!mz->usage_in_excess) | |
535 | return; | |
536 | while (*p) { | |
537 | parent = *p; | |
ef8f2327 | 538 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 539 | tree_node); |
fa90b2fd | 540 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 541 | p = &(*p)->rb_left; |
fa90b2fd DB |
542 | rightmost = false; |
543 | } | |
544 | ||
bb4cc1a8 AM |
545 | /* |
546 | * We can't avoid mem cgroups that are over their soft | |
547 | * limit by the same amount | |
548 | */ | |
549 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
550 | p = &(*p)->rb_right; | |
551 | } | |
fa90b2fd DB |
552 | |
553 | if (rightmost) | |
554 | mctz->rb_rightmost = &mz->tree_node; | |
555 | ||
bb4cc1a8 AM |
556 | rb_link_node(&mz->tree_node, parent, p); |
557 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
558 | mz->on_tree = true; | |
559 | } | |
560 | ||
ef8f2327 MG |
561 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
562 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
563 | { |
564 | if (!mz->on_tree) | |
565 | return; | |
fa90b2fd DB |
566 | |
567 | if (&mz->tree_node == mctz->rb_rightmost) | |
568 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
569 | ||
bb4cc1a8 AM |
570 | rb_erase(&mz->tree_node, &mctz->rb_root); |
571 | mz->on_tree = false; | |
572 | } | |
573 | ||
ef8f2327 MG |
574 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
575 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 576 | { |
0a31bc97 JW |
577 | unsigned long flags; |
578 | ||
579 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 580 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 581 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
582 | } |
583 | ||
3e32cb2e JW |
584 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
585 | { | |
586 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 587 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
588 | unsigned long excess = 0; |
589 | ||
590 | if (nr_pages > soft_limit) | |
591 | excess = nr_pages - soft_limit; | |
592 | ||
593 | return excess; | |
594 | } | |
bb4cc1a8 AM |
595 | |
596 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
597 | { | |
3e32cb2e | 598 | unsigned long excess; |
ef8f2327 MG |
599 | struct mem_cgroup_per_node *mz; |
600 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 601 | |
e231875b | 602 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
603 | if (!mctz) |
604 | return; | |
bb4cc1a8 AM |
605 | /* |
606 | * Necessary to update all ancestors when hierarchy is used. | |
607 | * because their event counter is not touched. | |
608 | */ | |
609 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 610 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 611 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
612 | /* |
613 | * We have to update the tree if mz is on RB-tree or | |
614 | * mem is over its softlimit. | |
615 | */ | |
616 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
617 | unsigned long flags; |
618 | ||
619 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
620 | /* if on-tree, remove it */ |
621 | if (mz->on_tree) | |
cf2c8127 | 622 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
623 | /* |
624 | * Insert again. mz->usage_in_excess will be updated. | |
625 | * If excess is 0, no tree ops. | |
626 | */ | |
cf2c8127 | 627 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 628 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
629 | } |
630 | } | |
631 | } | |
632 | ||
633 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
634 | { | |
ef8f2327 MG |
635 | struct mem_cgroup_tree_per_node *mctz; |
636 | struct mem_cgroup_per_node *mz; | |
637 | int nid; | |
bb4cc1a8 | 638 | |
e231875b | 639 | for_each_node(nid) { |
ef8f2327 MG |
640 | mz = mem_cgroup_nodeinfo(memcg, nid); |
641 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
642 | if (mctz) |
643 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
644 | } |
645 | } | |
646 | ||
ef8f2327 MG |
647 | static struct mem_cgroup_per_node * |
648 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 649 | { |
ef8f2327 | 650 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
651 | |
652 | retry: | |
653 | mz = NULL; | |
fa90b2fd | 654 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
655 | goto done; /* Nothing to reclaim from */ |
656 | ||
fa90b2fd DB |
657 | mz = rb_entry(mctz->rb_rightmost, |
658 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
659 | /* |
660 | * Remove the node now but someone else can add it back, | |
661 | * we will to add it back at the end of reclaim to its correct | |
662 | * position in the tree. | |
663 | */ | |
cf2c8127 | 664 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 665 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 666 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
667 | goto retry; |
668 | done: | |
669 | return mz; | |
670 | } | |
671 | ||
ef8f2327 MG |
672 | static struct mem_cgroup_per_node * |
673 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 674 | { |
ef8f2327 | 675 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 676 | |
0a31bc97 | 677 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 678 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 679 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
680 | return mz; |
681 | } | |
682 | ||
ccda7f43 | 683 | static unsigned long memcg_sum_events(struct mem_cgroup *memcg, |
04fecbf5 | 684 | int event) |
e9f8974f | 685 | { |
a983b5eb | 686 | return atomic_long_read(&memcg->events[event]); |
e9f8974f JW |
687 | } |
688 | ||
c0ff4b85 | 689 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 690 | struct page *page, |
f627c2f5 | 691 | bool compound, int nr_pages) |
d52aa412 | 692 | { |
b2402857 KH |
693 | /* |
694 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
695 | * counted as CACHE even if it's on ANON LRU. | |
696 | */ | |
0a31bc97 | 697 | if (PageAnon(page)) |
c9019e9b | 698 | __mod_memcg_state(memcg, MEMCG_RSS, nr_pages); |
9a4caf1e | 699 | else { |
c9019e9b | 700 | __mod_memcg_state(memcg, MEMCG_CACHE, nr_pages); |
9a4caf1e | 701 | if (PageSwapBacked(page)) |
c9019e9b | 702 | __mod_memcg_state(memcg, NR_SHMEM, nr_pages); |
9a4caf1e | 703 | } |
55e462b0 | 704 | |
f627c2f5 KS |
705 | if (compound) { |
706 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
c9019e9b | 707 | __mod_memcg_state(memcg, MEMCG_RSS_HUGE, nr_pages); |
f627c2f5 | 708 | } |
b070e65c | 709 | |
e401f176 KH |
710 | /* pagein of a big page is an event. So, ignore page size */ |
711 | if (nr_pages > 0) | |
c9019e9b | 712 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 713 | else { |
c9019e9b | 714 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
715 | nr_pages = -nr_pages; /* for event */ |
716 | } | |
e401f176 | 717 | |
a983b5eb | 718 | __this_cpu_add(memcg->stat_cpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
719 | } |
720 | ||
0a6b76dd VD |
721 | unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
722 | int nid, unsigned int lru_mask) | |
bb2a0de9 | 723 | { |
b4536f0c | 724 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); |
e231875b | 725 | unsigned long nr = 0; |
ef8f2327 | 726 | enum lru_list lru; |
889976db | 727 | |
e231875b | 728 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 729 | |
ef8f2327 MG |
730 | for_each_lru(lru) { |
731 | if (!(BIT(lru) & lru_mask)) | |
732 | continue; | |
b4536f0c | 733 | nr += mem_cgroup_get_lru_size(lruvec, lru); |
e231875b JZ |
734 | } |
735 | return nr; | |
889976db | 736 | } |
bb2a0de9 | 737 | |
c0ff4b85 | 738 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 739 | unsigned int lru_mask) |
6d12e2d8 | 740 | { |
e231875b | 741 | unsigned long nr = 0; |
889976db | 742 | int nid; |
6d12e2d8 | 743 | |
31aaea4a | 744 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
745 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
746 | return nr; | |
d52aa412 KH |
747 | } |
748 | ||
f53d7ce3 JW |
749 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
750 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
751 | { |
752 | unsigned long val, next; | |
753 | ||
a983b5eb JW |
754 | val = __this_cpu_read(memcg->stat_cpu->nr_page_events); |
755 | next = __this_cpu_read(memcg->stat_cpu->targets[target]); | |
7a159cc9 | 756 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 757 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
758 | switch (target) { |
759 | case MEM_CGROUP_TARGET_THRESH: | |
760 | next = val + THRESHOLDS_EVENTS_TARGET; | |
761 | break; | |
bb4cc1a8 AM |
762 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
763 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
764 | break; | |
f53d7ce3 JW |
765 | case MEM_CGROUP_TARGET_NUMAINFO: |
766 | next = val + NUMAINFO_EVENTS_TARGET; | |
767 | break; | |
768 | default: | |
769 | break; | |
770 | } | |
a983b5eb | 771 | __this_cpu_write(memcg->stat_cpu->targets[target], next); |
f53d7ce3 | 772 | return true; |
7a159cc9 | 773 | } |
f53d7ce3 | 774 | return false; |
d2265e6f KH |
775 | } |
776 | ||
777 | /* | |
778 | * Check events in order. | |
779 | * | |
780 | */ | |
c0ff4b85 | 781 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
782 | { |
783 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
784 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
785 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 786 | bool do_softlimit; |
82b3f2a7 | 787 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 788 | |
bb4cc1a8 AM |
789 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
790 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
791 | #if MAX_NUMNODES > 1 |
792 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
793 | MEM_CGROUP_TARGET_NUMAINFO); | |
794 | #endif | |
c0ff4b85 | 795 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
796 | if (unlikely(do_softlimit)) |
797 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 798 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 799 | if (unlikely(do_numainfo)) |
c0ff4b85 | 800 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 801 | #endif |
0a31bc97 | 802 | } |
d2265e6f KH |
803 | } |
804 | ||
cf475ad2 | 805 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 806 | { |
31a78f23 BS |
807 | /* |
808 | * mm_update_next_owner() may clear mm->owner to NULL | |
809 | * if it races with swapoff, page migration, etc. | |
810 | * So this can be called with p == NULL. | |
811 | */ | |
812 | if (unlikely(!p)) | |
813 | return NULL; | |
814 | ||
073219e9 | 815 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 816 | } |
33398cf2 | 817 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 818 | |
d46eb14b SB |
819 | /** |
820 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
821 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
822 | * | |
823 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
824 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
825 | * returned. | |
826 | */ | |
827 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 828 | { |
d46eb14b SB |
829 | struct mem_cgroup *memcg; |
830 | ||
831 | if (mem_cgroup_disabled()) | |
832 | return NULL; | |
0b7f569e | 833 | |
54595fe2 KH |
834 | rcu_read_lock(); |
835 | do { | |
6f6acb00 MH |
836 | /* |
837 | * Page cache insertions can happen withou an | |
838 | * actual mm context, e.g. during disk probing | |
839 | * on boot, loopback IO, acct() writes etc. | |
840 | */ | |
841 | if (unlikely(!mm)) | |
df381975 | 842 | memcg = root_mem_cgroup; |
6f6acb00 MH |
843 | else { |
844 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
845 | if (unlikely(!memcg)) | |
846 | memcg = root_mem_cgroup; | |
847 | } | |
ec903c0c | 848 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 849 | rcu_read_unlock(); |
c0ff4b85 | 850 | return memcg; |
54595fe2 | 851 | } |
d46eb14b SB |
852 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
853 | ||
f745c6f5 SB |
854 | /** |
855 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
856 | * @page: page from which memcg should be extracted. | |
857 | * | |
858 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
859 | * root_mem_cgroup is returned. | |
860 | */ | |
861 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
862 | { | |
863 | struct mem_cgroup *memcg = page->mem_cgroup; | |
864 | ||
865 | if (mem_cgroup_disabled()) | |
866 | return NULL; | |
867 | ||
868 | rcu_read_lock(); | |
869 | if (!memcg || !css_tryget_online(&memcg->css)) | |
870 | memcg = root_mem_cgroup; | |
871 | rcu_read_unlock(); | |
872 | return memcg; | |
873 | } | |
874 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
875 | ||
d46eb14b SB |
876 | /** |
877 | * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg. | |
878 | */ | |
879 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
880 | { | |
881 | if (unlikely(current->active_memcg)) { | |
882 | struct mem_cgroup *memcg = root_mem_cgroup; | |
883 | ||
884 | rcu_read_lock(); | |
885 | if (css_tryget_online(¤t->active_memcg->css)) | |
886 | memcg = current->active_memcg; | |
887 | rcu_read_unlock(); | |
888 | return memcg; | |
889 | } | |
890 | return get_mem_cgroup_from_mm(current->mm); | |
891 | } | |
54595fe2 | 892 | |
5660048c JW |
893 | /** |
894 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
895 | * @root: hierarchy root | |
896 | * @prev: previously returned memcg, NULL on first invocation | |
897 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
898 | * | |
899 | * Returns references to children of the hierarchy below @root, or | |
900 | * @root itself, or %NULL after a full round-trip. | |
901 | * | |
902 | * Caller must pass the return value in @prev on subsequent | |
903 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
904 | * to cancel a hierarchy walk before the round-trip is complete. | |
905 | * | |
b213b54f | 906 | * Reclaimers can specify a node and a priority level in @reclaim to |
5660048c | 907 | * divide up the memcgs in the hierarchy among all concurrent |
b213b54f | 908 | * reclaimers operating on the same node and priority. |
5660048c | 909 | */ |
694fbc0f | 910 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 911 | struct mem_cgroup *prev, |
694fbc0f | 912 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 913 | { |
33398cf2 | 914 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 915 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 916 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 917 | struct mem_cgroup *pos = NULL; |
711d3d2c | 918 | |
694fbc0f AM |
919 | if (mem_cgroup_disabled()) |
920 | return NULL; | |
5660048c | 921 | |
9f3a0d09 JW |
922 | if (!root) |
923 | root = root_mem_cgroup; | |
7d74b06f | 924 | |
9f3a0d09 | 925 | if (prev && !reclaim) |
5ac8fb31 | 926 | pos = prev; |
14067bb3 | 927 | |
9f3a0d09 JW |
928 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
929 | if (prev) | |
5ac8fb31 | 930 | goto out; |
694fbc0f | 931 | return root; |
9f3a0d09 | 932 | } |
14067bb3 | 933 | |
542f85f9 | 934 | rcu_read_lock(); |
5f578161 | 935 | |
5ac8fb31 | 936 | if (reclaim) { |
ef8f2327 | 937 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 938 | |
ef8f2327 | 939 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
5ac8fb31 JW |
940 | iter = &mz->iter[reclaim->priority]; |
941 | ||
942 | if (prev && reclaim->generation != iter->generation) | |
943 | goto out_unlock; | |
944 | ||
6df38689 | 945 | while (1) { |
4db0c3c2 | 946 | pos = READ_ONCE(iter->position); |
6df38689 VD |
947 | if (!pos || css_tryget(&pos->css)) |
948 | break; | |
5ac8fb31 | 949 | /* |
6df38689 VD |
950 | * css reference reached zero, so iter->position will |
951 | * be cleared by ->css_released. However, we should not | |
952 | * rely on this happening soon, because ->css_released | |
953 | * is called from a work queue, and by busy-waiting we | |
954 | * might block it. So we clear iter->position right | |
955 | * away. | |
5ac8fb31 | 956 | */ |
6df38689 VD |
957 | (void)cmpxchg(&iter->position, pos, NULL); |
958 | } | |
5ac8fb31 JW |
959 | } |
960 | ||
961 | if (pos) | |
962 | css = &pos->css; | |
963 | ||
964 | for (;;) { | |
965 | css = css_next_descendant_pre(css, &root->css); | |
966 | if (!css) { | |
967 | /* | |
968 | * Reclaimers share the hierarchy walk, and a | |
969 | * new one might jump in right at the end of | |
970 | * the hierarchy - make sure they see at least | |
971 | * one group and restart from the beginning. | |
972 | */ | |
973 | if (!prev) | |
974 | continue; | |
975 | break; | |
527a5ec9 | 976 | } |
7d74b06f | 977 | |
5ac8fb31 JW |
978 | /* |
979 | * Verify the css and acquire a reference. The root | |
980 | * is provided by the caller, so we know it's alive | |
981 | * and kicking, and don't take an extra reference. | |
982 | */ | |
983 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 984 | |
5ac8fb31 JW |
985 | if (css == &root->css) |
986 | break; | |
14067bb3 | 987 | |
0b8f73e1 JW |
988 | if (css_tryget(css)) |
989 | break; | |
9f3a0d09 | 990 | |
5ac8fb31 | 991 | memcg = NULL; |
9f3a0d09 | 992 | } |
5ac8fb31 JW |
993 | |
994 | if (reclaim) { | |
5ac8fb31 | 995 | /* |
6df38689 VD |
996 | * The position could have already been updated by a competing |
997 | * thread, so check that the value hasn't changed since we read | |
998 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 999 | */ |
6df38689 VD |
1000 | (void)cmpxchg(&iter->position, pos, memcg); |
1001 | ||
5ac8fb31 JW |
1002 | if (pos) |
1003 | css_put(&pos->css); | |
1004 | ||
1005 | if (!memcg) | |
1006 | iter->generation++; | |
1007 | else if (!prev) | |
1008 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1009 | } |
5ac8fb31 | 1010 | |
542f85f9 MH |
1011 | out_unlock: |
1012 | rcu_read_unlock(); | |
5ac8fb31 | 1013 | out: |
c40046f3 MH |
1014 | if (prev && prev != root) |
1015 | css_put(&prev->css); | |
1016 | ||
9f3a0d09 | 1017 | return memcg; |
14067bb3 | 1018 | } |
7d74b06f | 1019 | |
5660048c JW |
1020 | /** |
1021 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1022 | * @root: hierarchy root | |
1023 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1024 | */ | |
1025 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1026 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1027 | { |
1028 | if (!root) | |
1029 | root = root_mem_cgroup; | |
1030 | if (prev && prev != root) | |
1031 | css_put(&prev->css); | |
1032 | } | |
7d74b06f | 1033 | |
6df38689 VD |
1034 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1035 | { | |
1036 | struct mem_cgroup *memcg = dead_memcg; | |
1037 | struct mem_cgroup_reclaim_iter *iter; | |
ef8f2327 MG |
1038 | struct mem_cgroup_per_node *mz; |
1039 | int nid; | |
6df38689 VD |
1040 | int i; |
1041 | ||
9f15bde6 | 1042 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
6df38689 | 1043 | for_each_node(nid) { |
ef8f2327 MG |
1044 | mz = mem_cgroup_nodeinfo(memcg, nid); |
1045 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
1046 | iter = &mz->iter[i]; | |
1047 | cmpxchg(&iter->position, | |
1048 | dead_memcg, NULL); | |
6df38689 VD |
1049 | } |
1050 | } | |
1051 | } | |
1052 | } | |
1053 | ||
7c5f64f8 VD |
1054 | /** |
1055 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1056 | * @memcg: hierarchy root | |
1057 | * @fn: function to call for each task | |
1058 | * @arg: argument passed to @fn | |
1059 | * | |
1060 | * This function iterates over tasks attached to @memcg or to any of its | |
1061 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1062 | * value, the function breaks the iteration loop and returns the value. | |
1063 | * Otherwise, it will iterate over all tasks and return 0. | |
1064 | * | |
1065 | * This function must not be called for the root memory cgroup. | |
1066 | */ | |
1067 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1068 | int (*fn)(struct task_struct *, void *), void *arg) | |
1069 | { | |
1070 | struct mem_cgroup *iter; | |
1071 | int ret = 0; | |
1072 | ||
1073 | BUG_ON(memcg == root_mem_cgroup); | |
1074 | ||
1075 | for_each_mem_cgroup_tree(iter, memcg) { | |
1076 | struct css_task_iter it; | |
1077 | struct task_struct *task; | |
1078 | ||
bc2fb7ed | 1079 | css_task_iter_start(&iter->css, 0, &it); |
7c5f64f8 VD |
1080 | while (!ret && (task = css_task_iter_next(&it))) |
1081 | ret = fn(task, arg); | |
1082 | css_task_iter_end(&it); | |
1083 | if (ret) { | |
1084 | mem_cgroup_iter_break(memcg, iter); | |
1085 | break; | |
1086 | } | |
1087 | } | |
1088 | return ret; | |
1089 | } | |
1090 | ||
925b7673 | 1091 | /** |
dfe0e773 | 1092 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1093 | * @page: the page |
f144c390 | 1094 | * @pgdat: pgdat of the page |
dfe0e773 JW |
1095 | * |
1096 | * This function is only safe when following the LRU page isolation | |
1097 | * and putback protocol: the LRU lock must be held, and the page must | |
1098 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1099 | */ |
599d0c95 | 1100 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1101 | { |
ef8f2327 | 1102 | struct mem_cgroup_per_node *mz; |
925b7673 | 1103 | struct mem_cgroup *memcg; |
bea8c150 | 1104 | struct lruvec *lruvec; |
6d12e2d8 | 1105 | |
bea8c150 | 1106 | if (mem_cgroup_disabled()) { |
599d0c95 | 1107 | lruvec = &pgdat->lruvec; |
bea8c150 HD |
1108 | goto out; |
1109 | } | |
925b7673 | 1110 | |
1306a85a | 1111 | memcg = page->mem_cgroup; |
7512102c | 1112 | /* |
dfe0e773 | 1113 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1114 | * possibly migrated - before they are charged. |
7512102c | 1115 | */ |
29833315 JW |
1116 | if (!memcg) |
1117 | memcg = root_mem_cgroup; | |
7512102c | 1118 | |
ef8f2327 | 1119 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1120 | lruvec = &mz->lruvec; |
1121 | out: | |
1122 | /* | |
1123 | * Since a node can be onlined after the mem_cgroup was created, | |
1124 | * we have to be prepared to initialize lruvec->zone here; | |
1125 | * and if offlined then reonlined, we need to reinitialize it. | |
1126 | */ | |
599d0c95 MG |
1127 | if (unlikely(lruvec->pgdat != pgdat)) |
1128 | lruvec->pgdat = pgdat; | |
bea8c150 | 1129 | return lruvec; |
08e552c6 | 1130 | } |
b69408e8 | 1131 | |
925b7673 | 1132 | /** |
fa9add64 HD |
1133 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1134 | * @lruvec: mem_cgroup per zone lru vector | |
1135 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1136 | * @zid: zone id of the accounted pages |
fa9add64 | 1137 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1138 | * |
ca707239 HD |
1139 | * This function must be called under lru_lock, just before a page is added |
1140 | * to or just after a page is removed from an lru list (that ordering being | |
1141 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1142 | */ |
fa9add64 | 1143 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1144 | int zid, int nr_pages) |
3f58a829 | 1145 | { |
ef8f2327 | 1146 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1147 | unsigned long *lru_size; |
ca707239 | 1148 | long size; |
3f58a829 MK |
1149 | |
1150 | if (mem_cgroup_disabled()) | |
1151 | return; | |
1152 | ||
ef8f2327 | 1153 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1154 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1155 | |
1156 | if (nr_pages < 0) | |
1157 | *lru_size += nr_pages; | |
1158 | ||
1159 | size = *lru_size; | |
b4536f0c MH |
1160 | if (WARN_ONCE(size < 0, |
1161 | "%s(%p, %d, %d): lru_size %ld\n", | |
1162 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1163 | VM_BUG_ON(1); |
1164 | *lru_size = 0; | |
1165 | } | |
1166 | ||
1167 | if (nr_pages > 0) | |
1168 | *lru_size += nr_pages; | |
08e552c6 | 1169 | } |
544122e5 | 1170 | |
2314b42d | 1171 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1172 | { |
2314b42d | 1173 | struct mem_cgroup *task_memcg; |
158e0a2d | 1174 | struct task_struct *p; |
ffbdccf5 | 1175 | bool ret; |
4c4a2214 | 1176 | |
158e0a2d | 1177 | p = find_lock_task_mm(task); |
de077d22 | 1178 | if (p) { |
2314b42d | 1179 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1180 | task_unlock(p); |
1181 | } else { | |
1182 | /* | |
1183 | * All threads may have already detached their mm's, but the oom | |
1184 | * killer still needs to detect if they have already been oom | |
1185 | * killed to prevent needlessly killing additional tasks. | |
1186 | */ | |
ffbdccf5 | 1187 | rcu_read_lock(); |
2314b42d JW |
1188 | task_memcg = mem_cgroup_from_task(task); |
1189 | css_get(&task_memcg->css); | |
ffbdccf5 | 1190 | rcu_read_unlock(); |
de077d22 | 1191 | } |
2314b42d JW |
1192 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1193 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1194 | return ret; |
1195 | } | |
1196 | ||
19942822 | 1197 | /** |
9d11ea9f | 1198 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1199 | * @memcg: the memory cgroup |
19942822 | 1200 | * |
9d11ea9f | 1201 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1202 | * pages. |
19942822 | 1203 | */ |
c0ff4b85 | 1204 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1205 | { |
3e32cb2e JW |
1206 | unsigned long margin = 0; |
1207 | unsigned long count; | |
1208 | unsigned long limit; | |
9d11ea9f | 1209 | |
3e32cb2e | 1210 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1211 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1212 | if (count < limit) |
1213 | margin = limit - count; | |
1214 | ||
7941d214 | 1215 | if (do_memsw_account()) { |
3e32cb2e | 1216 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1217 | limit = READ_ONCE(memcg->memsw.max); |
3e32cb2e JW |
1218 | if (count <= limit) |
1219 | margin = min(margin, limit - count); | |
cbedbac3 LR |
1220 | else |
1221 | margin = 0; | |
3e32cb2e JW |
1222 | } |
1223 | ||
1224 | return margin; | |
19942822 JW |
1225 | } |
1226 | ||
32047e2a | 1227 | /* |
bdcbb659 | 1228 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1229 | * |
bdcbb659 QH |
1230 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1231 | * moving cgroups. This is for waiting at high-memory pressure | |
1232 | * caused by "move". | |
32047e2a | 1233 | */ |
c0ff4b85 | 1234 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1235 | { |
2bd9bb20 KH |
1236 | struct mem_cgroup *from; |
1237 | struct mem_cgroup *to; | |
4b534334 | 1238 | bool ret = false; |
2bd9bb20 KH |
1239 | /* |
1240 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1241 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1242 | */ | |
1243 | spin_lock(&mc.lock); | |
1244 | from = mc.from; | |
1245 | to = mc.to; | |
1246 | if (!from) | |
1247 | goto unlock; | |
3e92041d | 1248 | |
2314b42d JW |
1249 | ret = mem_cgroup_is_descendant(from, memcg) || |
1250 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1251 | unlock: |
1252 | spin_unlock(&mc.lock); | |
4b534334 KH |
1253 | return ret; |
1254 | } | |
1255 | ||
c0ff4b85 | 1256 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1257 | { |
1258 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1259 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1260 | DEFINE_WAIT(wait); |
1261 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1262 | /* moving charge context might have finished. */ | |
1263 | if (mc.moving_task) | |
1264 | schedule(); | |
1265 | finish_wait(&mc.waitq, &wait); | |
1266 | return true; | |
1267 | } | |
1268 | } | |
1269 | return false; | |
1270 | } | |
1271 | ||
8ad6e404 | 1272 | static const unsigned int memcg1_stats[] = { |
71cd3113 JW |
1273 | MEMCG_CACHE, |
1274 | MEMCG_RSS, | |
1275 | MEMCG_RSS_HUGE, | |
1276 | NR_SHMEM, | |
1277 | NR_FILE_MAPPED, | |
1278 | NR_FILE_DIRTY, | |
1279 | NR_WRITEBACK, | |
1280 | MEMCG_SWAP, | |
1281 | }; | |
1282 | ||
1283 | static const char *const memcg1_stat_names[] = { | |
1284 | "cache", | |
1285 | "rss", | |
1286 | "rss_huge", | |
1287 | "shmem", | |
1288 | "mapped_file", | |
1289 | "dirty", | |
1290 | "writeback", | |
1291 | "swap", | |
1292 | }; | |
1293 | ||
58cf188e | 1294 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1295 | /** |
58cf188e | 1296 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1297 | * @memcg: The memory cgroup that went over limit |
1298 | * @p: Task that is going to be killed | |
1299 | * | |
1300 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1301 | * enabled | |
1302 | */ | |
1303 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1304 | { | |
58cf188e SZ |
1305 | struct mem_cgroup *iter; |
1306 | unsigned int i; | |
e222432b | 1307 | |
e222432b BS |
1308 | rcu_read_lock(); |
1309 | ||
2415b9f5 BV |
1310 | if (p) { |
1311 | pr_info("Task in "); | |
1312 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1313 | pr_cont(" killed as a result of limit of "); | |
1314 | } else { | |
1315 | pr_info("Memory limit reached of cgroup "); | |
1316 | } | |
1317 | ||
e61734c5 | 1318 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1319 | pr_cont("\n"); |
e222432b | 1320 | |
e222432b BS |
1321 | rcu_read_unlock(); |
1322 | ||
3e32cb2e JW |
1323 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1324 | K((u64)page_counter_read(&memcg->memory)), | |
bbec2e15 | 1325 | K((u64)memcg->memory.max), memcg->memory.failcnt); |
3e32cb2e JW |
1326 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", |
1327 | K((u64)page_counter_read(&memcg->memsw)), | |
bbec2e15 | 1328 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); |
3e32cb2e JW |
1329 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", |
1330 | K((u64)page_counter_read(&memcg->kmem)), | |
bbec2e15 | 1331 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); |
58cf188e SZ |
1332 | |
1333 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1334 | pr_info("Memory cgroup stats for "); |
1335 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1336 | pr_cont(":"); |
1337 | ||
71cd3113 JW |
1338 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
1339 | if (memcg1_stats[i] == MEMCG_SWAP && !do_swap_account) | |
58cf188e | 1340 | continue; |
71cd3113 | 1341 | pr_cont(" %s:%luKB", memcg1_stat_names[i], |
ccda7f43 | 1342 | K(memcg_page_state(iter, memcg1_stats[i]))); |
58cf188e SZ |
1343 | } |
1344 | ||
1345 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1346 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1347 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1348 | ||
1349 | pr_cont("\n"); | |
1350 | } | |
e222432b BS |
1351 | } |
1352 | ||
a63d83f4 DR |
1353 | /* |
1354 | * Return the memory (and swap, if configured) limit for a memcg. | |
1355 | */ | |
bbec2e15 | 1356 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1357 | { |
bbec2e15 | 1358 | unsigned long max; |
f3e8eb70 | 1359 | |
bbec2e15 | 1360 | max = memcg->memory.max; |
9a5a8f19 | 1361 | if (mem_cgroup_swappiness(memcg)) { |
bbec2e15 RG |
1362 | unsigned long memsw_max; |
1363 | unsigned long swap_max; | |
9a5a8f19 | 1364 | |
bbec2e15 RG |
1365 | memsw_max = memcg->memsw.max; |
1366 | swap_max = memcg->swap.max; | |
1367 | swap_max = min(swap_max, (unsigned long)total_swap_pages); | |
1368 | max = min(max + swap_max, memsw_max); | |
9a5a8f19 | 1369 | } |
bbec2e15 | 1370 | return max; |
a63d83f4 DR |
1371 | } |
1372 | ||
b6e6edcf | 1373 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1374 | int order) |
9cbb78bb | 1375 | { |
6e0fc46d DR |
1376 | struct oom_control oc = { |
1377 | .zonelist = NULL, | |
1378 | .nodemask = NULL, | |
2a966b77 | 1379 | .memcg = memcg, |
6e0fc46d DR |
1380 | .gfp_mask = gfp_mask, |
1381 | .order = order, | |
6e0fc46d | 1382 | }; |
7c5f64f8 | 1383 | bool ret; |
9cbb78bb | 1384 | |
dc56401f | 1385 | mutex_lock(&oom_lock); |
7c5f64f8 | 1386 | ret = out_of_memory(&oc); |
dc56401f | 1387 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1388 | return ret; |
9cbb78bb DR |
1389 | } |
1390 | ||
ae6e71d3 MC |
1391 | #if MAX_NUMNODES > 1 |
1392 | ||
4d0c066d KH |
1393 | /** |
1394 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1395 | * @memcg: the target memcg |
4d0c066d KH |
1396 | * @nid: the node ID to be checked. |
1397 | * @noswap : specify true here if the user wants flle only information. | |
1398 | * | |
1399 | * This function returns whether the specified memcg contains any | |
1400 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1401 | * pages in the node. | |
1402 | */ | |
c0ff4b85 | 1403 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1404 | int nid, bool noswap) |
1405 | { | |
c0ff4b85 | 1406 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1407 | return true; |
1408 | if (noswap || !total_swap_pages) | |
1409 | return false; | |
c0ff4b85 | 1410 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1411 | return true; |
1412 | return false; | |
1413 | ||
1414 | } | |
889976db YH |
1415 | |
1416 | /* | |
1417 | * Always updating the nodemask is not very good - even if we have an empty | |
1418 | * list or the wrong list here, we can start from some node and traverse all | |
1419 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1420 | * | |
1421 | */ | |
c0ff4b85 | 1422 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1423 | { |
1424 | int nid; | |
453a9bf3 KH |
1425 | /* |
1426 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1427 | * pagein/pageout changes since the last update. | |
1428 | */ | |
c0ff4b85 | 1429 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1430 | return; |
c0ff4b85 | 1431 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1432 | return; |
1433 | ||
889976db | 1434 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1435 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1436 | |
31aaea4a | 1437 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1438 | |
c0ff4b85 R |
1439 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1440 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1441 | } |
453a9bf3 | 1442 | |
c0ff4b85 R |
1443 | atomic_set(&memcg->numainfo_events, 0); |
1444 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1445 | } |
1446 | ||
1447 | /* | |
1448 | * Selecting a node where we start reclaim from. Because what we need is just | |
1449 | * reducing usage counter, start from anywhere is O,K. Considering | |
1450 | * memory reclaim from current node, there are pros. and cons. | |
1451 | * | |
1452 | * Freeing memory from current node means freeing memory from a node which | |
1453 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1454 | * hit limits, it will see a contention on a node. But freeing from remote | |
1455 | * node means more costs for memory reclaim because of memory latency. | |
1456 | * | |
1457 | * Now, we use round-robin. Better algorithm is welcomed. | |
1458 | */ | |
c0ff4b85 | 1459 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1460 | { |
1461 | int node; | |
1462 | ||
c0ff4b85 R |
1463 | mem_cgroup_may_update_nodemask(memcg); |
1464 | node = memcg->last_scanned_node; | |
889976db | 1465 | |
0edaf86c | 1466 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1467 | /* |
fda3d69b MH |
1468 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1469 | * last time it really checked all the LRUs due to rate limiting. | |
1470 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1471 | */ |
1472 | if (unlikely(node == MAX_NUMNODES)) | |
1473 | node = numa_node_id(); | |
1474 | ||
c0ff4b85 | 1475 | memcg->last_scanned_node = node; |
889976db YH |
1476 | return node; |
1477 | } | |
889976db | 1478 | #else |
c0ff4b85 | 1479 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1480 | { |
1481 | return 0; | |
1482 | } | |
1483 | #endif | |
1484 | ||
0608f43d | 1485 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1486 | pg_data_t *pgdat, |
0608f43d AM |
1487 | gfp_t gfp_mask, |
1488 | unsigned long *total_scanned) | |
1489 | { | |
1490 | struct mem_cgroup *victim = NULL; | |
1491 | int total = 0; | |
1492 | int loop = 0; | |
1493 | unsigned long excess; | |
1494 | unsigned long nr_scanned; | |
1495 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1496 | .pgdat = pgdat, |
0608f43d AM |
1497 | .priority = 0, |
1498 | }; | |
1499 | ||
3e32cb2e | 1500 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1501 | |
1502 | while (1) { | |
1503 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1504 | if (!victim) { | |
1505 | loop++; | |
1506 | if (loop >= 2) { | |
1507 | /* | |
1508 | * If we have not been able to reclaim | |
1509 | * anything, it might because there are | |
1510 | * no reclaimable pages under this hierarchy | |
1511 | */ | |
1512 | if (!total) | |
1513 | break; | |
1514 | /* | |
1515 | * We want to do more targeted reclaim. | |
1516 | * excess >> 2 is not to excessive so as to | |
1517 | * reclaim too much, nor too less that we keep | |
1518 | * coming back to reclaim from this cgroup | |
1519 | */ | |
1520 | if (total >= (excess >> 2) || | |
1521 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1522 | break; | |
1523 | } | |
1524 | continue; | |
1525 | } | |
a9dd0a83 | 1526 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1527 | pgdat, &nr_scanned); |
0608f43d | 1528 | *total_scanned += nr_scanned; |
3e32cb2e | 1529 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1530 | break; |
6d61ef40 | 1531 | } |
0608f43d AM |
1532 | mem_cgroup_iter_break(root_memcg, victim); |
1533 | return total; | |
6d61ef40 BS |
1534 | } |
1535 | ||
0056f4e6 JW |
1536 | #ifdef CONFIG_LOCKDEP |
1537 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1538 | .name = "memcg_oom_lock", | |
1539 | }; | |
1540 | #endif | |
1541 | ||
fb2a6fc5 JW |
1542 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1543 | ||
867578cb KH |
1544 | /* |
1545 | * Check OOM-Killer is already running under our hierarchy. | |
1546 | * If someone is running, return false. | |
1547 | */ | |
fb2a6fc5 | 1548 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1549 | { |
79dfdacc | 1550 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1551 | |
fb2a6fc5 JW |
1552 | spin_lock(&memcg_oom_lock); |
1553 | ||
9f3a0d09 | 1554 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1555 | if (iter->oom_lock) { |
79dfdacc MH |
1556 | /* |
1557 | * this subtree of our hierarchy is already locked | |
1558 | * so we cannot give a lock. | |
1559 | */ | |
79dfdacc | 1560 | failed = iter; |
9f3a0d09 JW |
1561 | mem_cgroup_iter_break(memcg, iter); |
1562 | break; | |
23751be0 JW |
1563 | } else |
1564 | iter->oom_lock = true; | |
7d74b06f | 1565 | } |
867578cb | 1566 | |
fb2a6fc5 JW |
1567 | if (failed) { |
1568 | /* | |
1569 | * OK, we failed to lock the whole subtree so we have | |
1570 | * to clean up what we set up to the failing subtree | |
1571 | */ | |
1572 | for_each_mem_cgroup_tree(iter, memcg) { | |
1573 | if (iter == failed) { | |
1574 | mem_cgroup_iter_break(memcg, iter); | |
1575 | break; | |
1576 | } | |
1577 | iter->oom_lock = false; | |
79dfdacc | 1578 | } |
0056f4e6 JW |
1579 | } else |
1580 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1581 | |
1582 | spin_unlock(&memcg_oom_lock); | |
1583 | ||
1584 | return !failed; | |
a636b327 | 1585 | } |
0b7f569e | 1586 | |
fb2a6fc5 | 1587 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1588 | { |
7d74b06f KH |
1589 | struct mem_cgroup *iter; |
1590 | ||
fb2a6fc5 | 1591 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1592 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1593 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1594 | iter->oom_lock = false; |
fb2a6fc5 | 1595 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1596 | } |
1597 | ||
c0ff4b85 | 1598 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1599 | { |
1600 | struct mem_cgroup *iter; | |
1601 | ||
c2b42d3c | 1602 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1603 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1604 | iter->under_oom++; |
1605 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1606 | } |
1607 | ||
c0ff4b85 | 1608 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1609 | { |
1610 | struct mem_cgroup *iter; | |
1611 | ||
867578cb KH |
1612 | /* |
1613 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1614 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1615 | */ |
c2b42d3c | 1616 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1617 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1618 | if (iter->under_oom > 0) |
1619 | iter->under_oom--; | |
1620 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1621 | } |
1622 | ||
867578cb KH |
1623 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1624 | ||
dc98df5a | 1625 | struct oom_wait_info { |
d79154bb | 1626 | struct mem_cgroup *memcg; |
ac6424b9 | 1627 | wait_queue_entry_t wait; |
dc98df5a KH |
1628 | }; |
1629 | ||
ac6424b9 | 1630 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1631 | unsigned mode, int sync, void *arg) |
1632 | { | |
d79154bb HD |
1633 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1634 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1635 | struct oom_wait_info *oom_wait_info; |
1636 | ||
1637 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1638 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1639 | |
2314b42d JW |
1640 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1641 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1642 | return 0; |
dc98df5a KH |
1643 | return autoremove_wake_function(wait, mode, sync, arg); |
1644 | } | |
1645 | ||
c0ff4b85 | 1646 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1647 | { |
c2b42d3c TH |
1648 | /* |
1649 | * For the following lockless ->under_oom test, the only required | |
1650 | * guarantee is that it must see the state asserted by an OOM when | |
1651 | * this function is called as a result of userland actions | |
1652 | * triggered by the notification of the OOM. This is trivially | |
1653 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1654 | * triggering notification. | |
1655 | */ | |
1656 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1657 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1658 | } |
1659 | ||
29ef680a MH |
1660 | enum oom_status { |
1661 | OOM_SUCCESS, | |
1662 | OOM_FAILED, | |
1663 | OOM_ASYNC, | |
1664 | OOM_SKIPPED | |
1665 | }; | |
1666 | ||
1667 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1668 | { |
29ef680a MH |
1669 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1670 | return OOM_SKIPPED; | |
1671 | ||
7a1adfdd RG |
1672 | memcg_memory_event(memcg, MEMCG_OOM); |
1673 | ||
867578cb | 1674 | /* |
49426420 JW |
1675 | * We are in the middle of the charge context here, so we |
1676 | * don't want to block when potentially sitting on a callstack | |
1677 | * that holds all kinds of filesystem and mm locks. | |
1678 | * | |
29ef680a MH |
1679 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1680 | * handling until the charge can succeed; remember the context and put | |
1681 | * the task to sleep at the end of the page fault when all locks are | |
1682 | * released. | |
49426420 | 1683 | * |
29ef680a MH |
1684 | * On the other hand, in-kernel OOM killer allows for an async victim |
1685 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1686 | * relying on the oom victim to make a forward progress and we can | |
1687 | * invoke the oom killer here. | |
1688 | * | |
1689 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1690 | * victim and then we have to bail out from the charge path. | |
867578cb | 1691 | */ |
29ef680a MH |
1692 | if (memcg->oom_kill_disable) { |
1693 | if (!current->in_user_fault) | |
1694 | return OOM_SKIPPED; | |
1695 | css_get(&memcg->css); | |
1696 | current->memcg_in_oom = memcg; | |
1697 | current->memcg_oom_gfp_mask = mask; | |
1698 | current->memcg_oom_order = order; | |
1699 | ||
1700 | return OOM_ASYNC; | |
1701 | } | |
1702 | ||
1703 | if (mem_cgroup_out_of_memory(memcg, mask, order)) | |
1704 | return OOM_SUCCESS; | |
1705 | ||
29ef680a | 1706 | return OOM_FAILED; |
3812c8c8 JW |
1707 | } |
1708 | ||
1709 | /** | |
1710 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1711 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1712 | * |
49426420 JW |
1713 | * This has to be called at the end of a page fault if the memcg OOM |
1714 | * handler was enabled. | |
3812c8c8 | 1715 | * |
49426420 | 1716 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1717 | * sleep on a waitqueue until the userspace task resolves the |
1718 | * situation. Sleeping directly in the charge context with all kinds | |
1719 | * of locks held is not a good idea, instead we remember an OOM state | |
1720 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1721 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1722 | * |
1723 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1724 | * completed, %false otherwise. |
3812c8c8 | 1725 | */ |
49426420 | 1726 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1727 | { |
626ebc41 | 1728 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1729 | struct oom_wait_info owait; |
49426420 | 1730 | bool locked; |
3812c8c8 JW |
1731 | |
1732 | /* OOM is global, do not handle */ | |
3812c8c8 | 1733 | if (!memcg) |
49426420 | 1734 | return false; |
3812c8c8 | 1735 | |
7c5f64f8 | 1736 | if (!handle) |
49426420 | 1737 | goto cleanup; |
3812c8c8 JW |
1738 | |
1739 | owait.memcg = memcg; | |
1740 | owait.wait.flags = 0; | |
1741 | owait.wait.func = memcg_oom_wake_function; | |
1742 | owait.wait.private = current; | |
2055da97 | 1743 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1744 | |
3812c8c8 | 1745 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1746 | mem_cgroup_mark_under_oom(memcg); |
1747 | ||
1748 | locked = mem_cgroup_oom_trylock(memcg); | |
1749 | ||
1750 | if (locked) | |
1751 | mem_cgroup_oom_notify(memcg); | |
1752 | ||
1753 | if (locked && !memcg->oom_kill_disable) { | |
1754 | mem_cgroup_unmark_under_oom(memcg); | |
1755 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1756 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1757 | current->memcg_oom_order); | |
49426420 | 1758 | } else { |
3812c8c8 | 1759 | schedule(); |
49426420 JW |
1760 | mem_cgroup_unmark_under_oom(memcg); |
1761 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1762 | } | |
1763 | ||
1764 | if (locked) { | |
fb2a6fc5 JW |
1765 | mem_cgroup_oom_unlock(memcg); |
1766 | /* | |
1767 | * There is no guarantee that an OOM-lock contender | |
1768 | * sees the wakeups triggered by the OOM kill | |
1769 | * uncharges. Wake any sleepers explicitely. | |
1770 | */ | |
1771 | memcg_oom_recover(memcg); | |
1772 | } | |
49426420 | 1773 | cleanup: |
626ebc41 | 1774 | current->memcg_in_oom = NULL; |
3812c8c8 | 1775 | css_put(&memcg->css); |
867578cb | 1776 | return true; |
0b7f569e KH |
1777 | } |
1778 | ||
3d8b38eb RG |
1779 | /** |
1780 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
1781 | * @victim: task to be killed by the OOM killer | |
1782 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
1783 | * | |
1784 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
1785 | * by killing all belonging OOM-killable tasks. | |
1786 | * | |
1787 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
1788 | */ | |
1789 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
1790 | struct mem_cgroup *oom_domain) | |
1791 | { | |
1792 | struct mem_cgroup *oom_group = NULL; | |
1793 | struct mem_cgroup *memcg; | |
1794 | ||
1795 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
1796 | return NULL; | |
1797 | ||
1798 | if (!oom_domain) | |
1799 | oom_domain = root_mem_cgroup; | |
1800 | ||
1801 | rcu_read_lock(); | |
1802 | ||
1803 | memcg = mem_cgroup_from_task(victim); | |
1804 | if (memcg == root_mem_cgroup) | |
1805 | goto out; | |
1806 | ||
1807 | /* | |
1808 | * Traverse the memory cgroup hierarchy from the victim task's | |
1809 | * cgroup up to the OOMing cgroup (or root) to find the | |
1810 | * highest-level memory cgroup with oom.group set. | |
1811 | */ | |
1812 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
1813 | if (memcg->oom_group) | |
1814 | oom_group = memcg; | |
1815 | ||
1816 | if (memcg == oom_domain) | |
1817 | break; | |
1818 | } | |
1819 | ||
1820 | if (oom_group) | |
1821 | css_get(&oom_group->css); | |
1822 | out: | |
1823 | rcu_read_unlock(); | |
1824 | ||
1825 | return oom_group; | |
1826 | } | |
1827 | ||
1828 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
1829 | { | |
1830 | pr_info("Tasks in "); | |
1831 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1832 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
1833 | } | |
1834 | ||
d7365e78 | 1835 | /** |
81f8c3a4 JW |
1836 | * lock_page_memcg - lock a page->mem_cgroup binding |
1837 | * @page: the page | |
32047e2a | 1838 | * |
81f8c3a4 | 1839 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
1840 | * another cgroup. |
1841 | * | |
1842 | * It ensures lifetime of the returned memcg. Caller is responsible | |
1843 | * for the lifetime of the page; __unlock_page_memcg() is available | |
1844 | * when @page might get freed inside the locked section. | |
d69b042f | 1845 | */ |
739f79fc | 1846 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 KH |
1847 | { |
1848 | struct mem_cgroup *memcg; | |
6de22619 | 1849 | unsigned long flags; |
89c06bd5 | 1850 | |
6de22619 JW |
1851 | /* |
1852 | * The RCU lock is held throughout the transaction. The fast | |
1853 | * path can get away without acquiring the memcg->move_lock | |
1854 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
1855 | * |
1856 | * The RCU lock also protects the memcg from being freed when | |
1857 | * the page state that is going to change is the only thing | |
1858 | * preventing the page itself from being freed. E.g. writeback | |
1859 | * doesn't hold a page reference and relies on PG_writeback to | |
1860 | * keep off truncation, migration and so forth. | |
1861 | */ | |
d7365e78 JW |
1862 | rcu_read_lock(); |
1863 | ||
1864 | if (mem_cgroup_disabled()) | |
739f79fc | 1865 | return NULL; |
89c06bd5 | 1866 | again: |
1306a85a | 1867 | memcg = page->mem_cgroup; |
29833315 | 1868 | if (unlikely(!memcg)) |
739f79fc | 1869 | return NULL; |
d7365e78 | 1870 | |
bdcbb659 | 1871 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 1872 | return memcg; |
89c06bd5 | 1873 | |
6de22619 | 1874 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1875 | if (memcg != page->mem_cgroup) { |
6de22619 | 1876 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1877 | goto again; |
1878 | } | |
6de22619 JW |
1879 | |
1880 | /* | |
1881 | * When charge migration first begins, we can have locked and | |
1882 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 1883 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
1884 | */ |
1885 | memcg->move_lock_task = current; | |
1886 | memcg->move_lock_flags = flags; | |
d7365e78 | 1887 | |
739f79fc | 1888 | return memcg; |
89c06bd5 | 1889 | } |
81f8c3a4 | 1890 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 1891 | |
d7365e78 | 1892 | /** |
739f79fc JW |
1893 | * __unlock_page_memcg - unlock and unpin a memcg |
1894 | * @memcg: the memcg | |
1895 | * | |
1896 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 1897 | */ |
739f79fc | 1898 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 1899 | { |
6de22619 JW |
1900 | if (memcg && memcg->move_lock_task == current) { |
1901 | unsigned long flags = memcg->move_lock_flags; | |
1902 | ||
1903 | memcg->move_lock_task = NULL; | |
1904 | memcg->move_lock_flags = 0; | |
1905 | ||
1906 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1907 | } | |
89c06bd5 | 1908 | |
d7365e78 | 1909 | rcu_read_unlock(); |
89c06bd5 | 1910 | } |
739f79fc JW |
1911 | |
1912 | /** | |
1913 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
1914 | * @page: the page | |
1915 | */ | |
1916 | void unlock_page_memcg(struct page *page) | |
1917 | { | |
1918 | __unlock_page_memcg(page->mem_cgroup); | |
1919 | } | |
81f8c3a4 | 1920 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 1921 | |
cdec2e42 KH |
1922 | struct memcg_stock_pcp { |
1923 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1924 | unsigned int nr_pages; |
cdec2e42 | 1925 | struct work_struct work; |
26fe6168 | 1926 | unsigned long flags; |
a0db00fc | 1927 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1928 | }; |
1929 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1930 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1931 | |
a0956d54 SS |
1932 | /** |
1933 | * consume_stock: Try to consume stocked charge on this cpu. | |
1934 | * @memcg: memcg to consume from. | |
1935 | * @nr_pages: how many pages to charge. | |
1936 | * | |
1937 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1938 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1939 | * service an allocation will refill the stock. | |
1940 | * | |
1941 | * returns true if successful, false otherwise. | |
cdec2e42 | 1942 | */ |
a0956d54 | 1943 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1944 | { |
1945 | struct memcg_stock_pcp *stock; | |
db2ba40c | 1946 | unsigned long flags; |
3e32cb2e | 1947 | bool ret = false; |
cdec2e42 | 1948 | |
a983b5eb | 1949 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 1950 | return ret; |
a0956d54 | 1951 | |
db2ba40c JW |
1952 | local_irq_save(flags); |
1953 | ||
1954 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 1955 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1956 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1957 | ret = true; |
1958 | } | |
db2ba40c JW |
1959 | |
1960 | local_irq_restore(flags); | |
1961 | ||
cdec2e42 KH |
1962 | return ret; |
1963 | } | |
1964 | ||
1965 | /* | |
3e32cb2e | 1966 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1967 | */ |
1968 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1969 | { | |
1970 | struct mem_cgroup *old = stock->cached; | |
1971 | ||
11c9ea4e | 1972 | if (stock->nr_pages) { |
3e32cb2e | 1973 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 1974 | if (do_memsw_account()) |
3e32cb2e | 1975 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1976 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1977 | stock->nr_pages = 0; |
cdec2e42 KH |
1978 | } |
1979 | stock->cached = NULL; | |
cdec2e42 KH |
1980 | } |
1981 | ||
cdec2e42 KH |
1982 | static void drain_local_stock(struct work_struct *dummy) |
1983 | { | |
db2ba40c JW |
1984 | struct memcg_stock_pcp *stock; |
1985 | unsigned long flags; | |
1986 | ||
72f0184c MH |
1987 | /* |
1988 | * The only protection from memory hotplug vs. drain_stock races is | |
1989 | * that we always operate on local CPU stock here with IRQ disabled | |
1990 | */ | |
db2ba40c JW |
1991 | local_irq_save(flags); |
1992 | ||
1993 | stock = this_cpu_ptr(&memcg_stock); | |
cdec2e42 | 1994 | drain_stock(stock); |
26fe6168 | 1995 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
1996 | |
1997 | local_irq_restore(flags); | |
cdec2e42 KH |
1998 | } |
1999 | ||
2000 | /* | |
3e32cb2e | 2001 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2002 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2003 | */ |
c0ff4b85 | 2004 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2005 | { |
db2ba40c JW |
2006 | struct memcg_stock_pcp *stock; |
2007 | unsigned long flags; | |
2008 | ||
2009 | local_irq_save(flags); | |
cdec2e42 | 2010 | |
db2ba40c | 2011 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2012 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2013 | drain_stock(stock); |
c0ff4b85 | 2014 | stock->cached = memcg; |
cdec2e42 | 2015 | } |
11c9ea4e | 2016 | stock->nr_pages += nr_pages; |
db2ba40c | 2017 | |
a983b5eb | 2018 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2019 | drain_stock(stock); |
2020 | ||
db2ba40c | 2021 | local_irq_restore(flags); |
cdec2e42 KH |
2022 | } |
2023 | ||
2024 | /* | |
c0ff4b85 | 2025 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2026 | * of the hierarchy under it. |
cdec2e42 | 2027 | */ |
6d3d6aa2 | 2028 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2029 | { |
26fe6168 | 2030 | int cpu, curcpu; |
d38144b7 | 2031 | |
6d3d6aa2 JW |
2032 | /* If someone's already draining, avoid adding running more workers. */ |
2033 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2034 | return; | |
72f0184c MH |
2035 | /* |
2036 | * Notify other cpus that system-wide "drain" is running | |
2037 | * We do not care about races with the cpu hotplug because cpu down | |
2038 | * as well as workers from this path always operate on the local | |
2039 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2040 | */ | |
5af12d0e | 2041 | curcpu = get_cpu(); |
cdec2e42 KH |
2042 | for_each_online_cpu(cpu) { |
2043 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2044 | struct mem_cgroup *memcg; |
26fe6168 | 2045 | |
c0ff4b85 | 2046 | memcg = stock->cached; |
72f0184c | 2047 | if (!memcg || !stock->nr_pages || !css_tryget(&memcg->css)) |
26fe6168 | 2048 | continue; |
72f0184c MH |
2049 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) { |
2050 | css_put(&memcg->css); | |
3e92041d | 2051 | continue; |
72f0184c | 2052 | } |
d1a05b69 MH |
2053 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2054 | if (cpu == curcpu) | |
2055 | drain_local_stock(&stock->work); | |
2056 | else | |
2057 | schedule_work_on(cpu, &stock->work); | |
2058 | } | |
72f0184c | 2059 | css_put(&memcg->css); |
cdec2e42 | 2060 | } |
5af12d0e | 2061 | put_cpu(); |
9f50fad6 | 2062 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2063 | } |
2064 | ||
308167fc | 2065 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2066 | { |
cdec2e42 | 2067 | struct memcg_stock_pcp *stock; |
a983b5eb | 2068 | struct mem_cgroup *memcg; |
cdec2e42 | 2069 | |
cdec2e42 KH |
2070 | stock = &per_cpu(memcg_stock, cpu); |
2071 | drain_stock(stock); | |
a983b5eb JW |
2072 | |
2073 | for_each_mem_cgroup(memcg) { | |
2074 | int i; | |
2075 | ||
2076 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2077 | int nid; | |
2078 | long x; | |
2079 | ||
2080 | x = this_cpu_xchg(memcg->stat_cpu->count[i], 0); | |
2081 | if (x) | |
2082 | atomic_long_add(x, &memcg->stat[i]); | |
2083 | ||
2084 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2085 | continue; | |
2086 | ||
2087 | for_each_node(nid) { | |
2088 | struct mem_cgroup_per_node *pn; | |
2089 | ||
2090 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2091 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
2092 | if (x) | |
2093 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2094 | } | |
2095 | } | |
2096 | ||
e27be240 | 2097 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2098 | long x; |
2099 | ||
2100 | x = this_cpu_xchg(memcg->stat_cpu->events[i], 0); | |
2101 | if (x) | |
2102 | atomic_long_add(x, &memcg->events[i]); | |
2103 | } | |
2104 | } | |
2105 | ||
308167fc | 2106 | return 0; |
cdec2e42 KH |
2107 | } |
2108 | ||
f7e1cb6e JW |
2109 | static void reclaim_high(struct mem_cgroup *memcg, |
2110 | unsigned int nr_pages, | |
2111 | gfp_t gfp_mask) | |
2112 | { | |
2113 | do { | |
2114 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
2115 | continue; | |
e27be240 | 2116 | memcg_memory_event(memcg, MEMCG_HIGH); |
f7e1cb6e JW |
2117 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
2118 | } while ((memcg = parent_mem_cgroup(memcg))); | |
2119 | } | |
2120 | ||
2121 | static void high_work_func(struct work_struct *work) | |
2122 | { | |
2123 | struct mem_cgroup *memcg; | |
2124 | ||
2125 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2126 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2127 | } |
2128 | ||
b23afb93 TH |
2129 | /* |
2130 | * Scheduled by try_charge() to be executed from the userland return path | |
2131 | * and reclaims memory over the high limit. | |
2132 | */ | |
2133 | void mem_cgroup_handle_over_high(void) | |
2134 | { | |
2135 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 2136 | struct mem_cgroup *memcg; |
b23afb93 TH |
2137 | |
2138 | if (likely(!nr_pages)) | |
2139 | return; | |
2140 | ||
f7e1cb6e JW |
2141 | memcg = get_mem_cgroup_from_mm(current->mm); |
2142 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
2143 | css_put(&memcg->css); |
2144 | current->memcg_nr_pages_over_high = 0; | |
2145 | } | |
2146 | ||
00501b53 JW |
2147 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2148 | unsigned int nr_pages) | |
8a9f3ccd | 2149 | { |
a983b5eb | 2150 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
9b130619 | 2151 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2152 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2153 | struct page_counter *counter; |
6539cc05 | 2154 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2155 | bool may_swap = true; |
2156 | bool drained = false; | |
29ef680a MH |
2157 | bool oomed = false; |
2158 | enum oom_status oom_status; | |
a636b327 | 2159 | |
ce00a967 | 2160 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2161 | return 0; |
6539cc05 | 2162 | retry: |
b6b6cc72 | 2163 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2164 | return 0; |
8a9f3ccd | 2165 | |
7941d214 | 2166 | if (!do_memsw_account() || |
6071ca52 JW |
2167 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2168 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2169 | goto done_restock; |
7941d214 | 2170 | if (do_memsw_account()) |
3e32cb2e JW |
2171 | page_counter_uncharge(&memcg->memsw, batch); |
2172 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2173 | } else { |
3e32cb2e | 2174 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2175 | may_swap = false; |
3fbe7244 | 2176 | } |
7a81b88c | 2177 | |
6539cc05 JW |
2178 | if (batch > nr_pages) { |
2179 | batch = nr_pages; | |
2180 | goto retry; | |
2181 | } | |
6d61ef40 | 2182 | |
06b078fc JW |
2183 | /* |
2184 | * Unlike in global OOM situations, memcg is not in a physical | |
2185 | * memory shortage. Allow dying and OOM-killed tasks to | |
2186 | * bypass the last charges so that they can exit quickly and | |
2187 | * free their memory. | |
2188 | */ | |
da99ecf1 | 2189 | if (unlikely(tsk_is_oom_victim(current) || |
06b078fc JW |
2190 | fatal_signal_pending(current) || |
2191 | current->flags & PF_EXITING)) | |
10d53c74 | 2192 | goto force; |
06b078fc | 2193 | |
89a28483 JW |
2194 | /* |
2195 | * Prevent unbounded recursion when reclaim operations need to | |
2196 | * allocate memory. This might exceed the limits temporarily, | |
2197 | * but we prefer facilitating memory reclaim and getting back | |
2198 | * under the limit over triggering OOM kills in these cases. | |
2199 | */ | |
2200 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2201 | goto force; | |
2202 | ||
06b078fc JW |
2203 | if (unlikely(task_in_memcg_oom(current))) |
2204 | goto nomem; | |
2205 | ||
d0164adc | 2206 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2207 | goto nomem; |
4b534334 | 2208 | |
e27be240 | 2209 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2210 | |
b70a2a21 JW |
2211 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2212 | gfp_mask, may_swap); | |
6539cc05 | 2213 | |
61e02c74 | 2214 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2215 | goto retry; |
28c34c29 | 2216 | |
b70a2a21 | 2217 | if (!drained) { |
6d3d6aa2 | 2218 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2219 | drained = true; |
2220 | goto retry; | |
2221 | } | |
2222 | ||
28c34c29 JW |
2223 | if (gfp_mask & __GFP_NORETRY) |
2224 | goto nomem; | |
6539cc05 JW |
2225 | /* |
2226 | * Even though the limit is exceeded at this point, reclaim | |
2227 | * may have been able to free some pages. Retry the charge | |
2228 | * before killing the task. | |
2229 | * | |
2230 | * Only for regular pages, though: huge pages are rather | |
2231 | * unlikely to succeed so close to the limit, and we fall back | |
2232 | * to regular pages anyway in case of failure. | |
2233 | */ | |
61e02c74 | 2234 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2235 | goto retry; |
2236 | /* | |
2237 | * At task move, charge accounts can be doubly counted. So, it's | |
2238 | * better to wait until the end of task_move if something is going on. | |
2239 | */ | |
2240 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2241 | goto retry; | |
2242 | ||
9b130619 JW |
2243 | if (nr_retries--) |
2244 | goto retry; | |
2245 | ||
29ef680a MH |
2246 | if (gfp_mask & __GFP_RETRY_MAYFAIL && oomed) |
2247 | goto nomem; | |
2248 | ||
06b078fc | 2249 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2250 | goto force; |
06b078fc | 2251 | |
6539cc05 | 2252 | if (fatal_signal_pending(current)) |
10d53c74 | 2253 | goto force; |
6539cc05 | 2254 | |
29ef680a MH |
2255 | /* |
2256 | * keep retrying as long as the memcg oom killer is able to make | |
2257 | * a forward progress or bypass the charge if the oom killer | |
2258 | * couldn't make any progress. | |
2259 | */ | |
2260 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2261 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2262 | switch (oom_status) { |
2263 | case OOM_SUCCESS: | |
2264 | nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
2265 | oomed = true; | |
2266 | goto retry; | |
2267 | case OOM_FAILED: | |
2268 | goto force; | |
2269 | default: | |
2270 | goto nomem; | |
2271 | } | |
7a81b88c | 2272 | nomem: |
6d1fdc48 | 2273 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2274 | return -ENOMEM; |
10d53c74 TH |
2275 | force: |
2276 | /* | |
2277 | * The allocation either can't fail or will lead to more memory | |
2278 | * being freed very soon. Allow memory usage go over the limit | |
2279 | * temporarily by force charging it. | |
2280 | */ | |
2281 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2282 | if (do_memsw_account()) |
10d53c74 TH |
2283 | page_counter_charge(&memcg->memsw, nr_pages); |
2284 | css_get_many(&memcg->css, nr_pages); | |
2285 | ||
2286 | return 0; | |
6539cc05 JW |
2287 | |
2288 | done_restock: | |
e8ea14cc | 2289 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2290 | if (batch > nr_pages) |
2291 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2292 | |
241994ed | 2293 | /* |
b23afb93 TH |
2294 | * If the hierarchy is above the normal consumption range, schedule |
2295 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2296 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2297 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2298 | * not recorded as it most likely matches current's and won't | |
2299 | * change in the meantime. As high limit is checked again before | |
2300 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2301 | */ |
2302 | do { | |
b23afb93 | 2303 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2304 | /* Don't bother a random interrupted task */ |
2305 | if (in_interrupt()) { | |
2306 | schedule_work(&memcg->high_work); | |
2307 | break; | |
2308 | } | |
9516a18a | 2309 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2310 | set_notify_resume(current); |
2311 | break; | |
2312 | } | |
241994ed | 2313 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2314 | |
2315 | return 0; | |
7a81b88c | 2316 | } |
8a9f3ccd | 2317 | |
00501b53 | 2318 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2319 | { |
ce00a967 JW |
2320 | if (mem_cgroup_is_root(memcg)) |
2321 | return; | |
2322 | ||
3e32cb2e | 2323 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2324 | if (do_memsw_account()) |
3e32cb2e | 2325 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2326 | |
e8ea14cc | 2327 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2328 | } |
2329 | ||
0a31bc97 JW |
2330 | static void lock_page_lru(struct page *page, int *isolated) |
2331 | { | |
2332 | struct zone *zone = page_zone(page); | |
2333 | ||
a52633d8 | 2334 | spin_lock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2335 | if (PageLRU(page)) { |
2336 | struct lruvec *lruvec; | |
2337 | ||
599d0c95 | 2338 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2339 | ClearPageLRU(page); |
2340 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2341 | *isolated = 1; | |
2342 | } else | |
2343 | *isolated = 0; | |
2344 | } | |
2345 | ||
2346 | static void unlock_page_lru(struct page *page, int isolated) | |
2347 | { | |
2348 | struct zone *zone = page_zone(page); | |
2349 | ||
2350 | if (isolated) { | |
2351 | struct lruvec *lruvec; | |
2352 | ||
599d0c95 | 2353 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2354 | VM_BUG_ON_PAGE(PageLRU(page), page); |
2355 | SetPageLRU(page); | |
2356 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2357 | } | |
a52633d8 | 2358 | spin_unlock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2359 | } |
2360 | ||
00501b53 | 2361 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2362 | bool lrucare) |
7a81b88c | 2363 | { |
0a31bc97 | 2364 | int isolated; |
9ce70c02 | 2365 | |
1306a85a | 2366 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2367 | |
2368 | /* | |
2369 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2370 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2371 | */ | |
0a31bc97 JW |
2372 | if (lrucare) |
2373 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2374 | |
0a31bc97 JW |
2375 | /* |
2376 | * Nobody should be changing or seriously looking at | |
1306a85a | 2377 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2378 | * |
2379 | * - the page is uncharged | |
2380 | * | |
2381 | * - the page is off-LRU | |
2382 | * | |
2383 | * - an anonymous fault has exclusive page access, except for | |
2384 | * a locked page table | |
2385 | * | |
2386 | * - a page cache insertion, a swapin fault, or a migration | |
2387 | * have the page locked | |
2388 | */ | |
1306a85a | 2389 | page->mem_cgroup = memcg; |
9ce70c02 | 2390 | |
0a31bc97 JW |
2391 | if (lrucare) |
2392 | unlock_page_lru(page, isolated); | |
7a81b88c | 2393 | } |
66e1707b | 2394 | |
84c07d11 | 2395 | #ifdef CONFIG_MEMCG_KMEM |
f3bb3043 | 2396 | static int memcg_alloc_cache_id(void) |
55007d84 | 2397 | { |
f3bb3043 VD |
2398 | int id, size; |
2399 | int err; | |
2400 | ||
dbcf73e2 | 2401 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2402 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2403 | if (id < 0) | |
2404 | return id; | |
55007d84 | 2405 | |
dbcf73e2 | 2406 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2407 | return id; |
2408 | ||
2409 | /* | |
2410 | * There's no space for the new id in memcg_caches arrays, | |
2411 | * so we have to grow them. | |
2412 | */ | |
05257a1a | 2413 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2414 | |
2415 | size = 2 * (id + 1); | |
55007d84 GC |
2416 | if (size < MEMCG_CACHES_MIN_SIZE) |
2417 | size = MEMCG_CACHES_MIN_SIZE; | |
2418 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2419 | size = MEMCG_CACHES_MAX_SIZE; | |
2420 | ||
f3bb3043 | 2421 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2422 | if (!err) |
2423 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2424 | if (!err) |
2425 | memcg_nr_cache_ids = size; | |
2426 | ||
2427 | up_write(&memcg_cache_ids_sem); | |
2428 | ||
f3bb3043 | 2429 | if (err) { |
dbcf73e2 | 2430 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2431 | return err; |
2432 | } | |
2433 | return id; | |
2434 | } | |
2435 | ||
2436 | static void memcg_free_cache_id(int id) | |
2437 | { | |
dbcf73e2 | 2438 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2439 | } |
2440 | ||
d5b3cf71 | 2441 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2442 | struct mem_cgroup *memcg; |
2443 | struct kmem_cache *cachep; | |
2444 | struct work_struct work; | |
2445 | }; | |
2446 | ||
d5b3cf71 | 2447 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2448 | { |
d5b3cf71 VD |
2449 | struct memcg_kmem_cache_create_work *cw = |
2450 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2451 | struct mem_cgroup *memcg = cw->memcg; |
2452 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2453 | |
d5b3cf71 | 2454 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2455 | |
5722d094 | 2456 | css_put(&memcg->css); |
d7f25f8a GC |
2457 | kfree(cw); |
2458 | } | |
2459 | ||
2460 | /* | |
2461 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2462 | */ |
85cfb245 | 2463 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
d5b3cf71 | 2464 | struct kmem_cache *cachep) |
d7f25f8a | 2465 | { |
d5b3cf71 | 2466 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2467 | |
c892fd82 | 2468 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT | __GFP_NOWARN); |
8135be5a | 2469 | if (!cw) |
d7f25f8a | 2470 | return; |
8135be5a VD |
2471 | |
2472 | css_get(&memcg->css); | |
d7f25f8a GC |
2473 | |
2474 | cw->memcg = memcg; | |
2475 | cw->cachep = cachep; | |
d5b3cf71 | 2476 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2477 | |
17cc4dfe | 2478 | queue_work(memcg_kmem_cache_wq, &cw->work); |
d7f25f8a GC |
2479 | } |
2480 | ||
45264778 VD |
2481 | static inline bool memcg_kmem_bypass(void) |
2482 | { | |
2483 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
2484 | return true; | |
2485 | return false; | |
2486 | } | |
2487 | ||
2488 | /** | |
2489 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation | |
2490 | * @cachep: the original global kmem cache | |
2491 | * | |
d7f25f8a GC |
2492 | * Return the kmem_cache we're supposed to use for a slab allocation. |
2493 | * We try to use the current memcg's version of the cache. | |
2494 | * | |
45264778 VD |
2495 | * If the cache does not exist yet, if we are the first user of it, we |
2496 | * create it asynchronously in a workqueue and let the current allocation | |
2497 | * go through with the original cache. | |
d7f25f8a | 2498 | * |
45264778 VD |
2499 | * This function takes a reference to the cache it returns to assure it |
2500 | * won't get destroyed while we are working with it. Once the caller is | |
2501 | * done with it, memcg_kmem_put_cache() must be called to release the | |
2502 | * reference. | |
d7f25f8a | 2503 | */ |
45264778 | 2504 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2505 | { |
2506 | struct mem_cgroup *memcg; | |
959c8963 | 2507 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2508 | int kmemcg_id; |
d7f25f8a | 2509 | |
f7ce3190 | 2510 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2511 | |
45264778 | 2512 | if (memcg_kmem_bypass()) |
230e9fc2 VD |
2513 | return cachep; |
2514 | ||
d46eb14b | 2515 | memcg = get_mem_cgroup_from_current(); |
4db0c3c2 | 2516 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2517 | if (kmemcg_id < 0) |
ca0dde97 | 2518 | goto out; |
d7f25f8a | 2519 | |
2a4db7eb | 2520 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2521 | if (likely(memcg_cachep)) |
2522 | return memcg_cachep; | |
ca0dde97 LZ |
2523 | |
2524 | /* | |
2525 | * If we are in a safe context (can wait, and not in interrupt | |
2526 | * context), we could be be predictable and return right away. | |
2527 | * This would guarantee that the allocation being performed | |
2528 | * already belongs in the new cache. | |
2529 | * | |
2530 | * However, there are some clashes that can arrive from locking. | |
2531 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2532 | * memcg_create_kmem_cache, this means no further allocation |
2533 | * could happen with the slab_mutex held. So it's better to | |
2534 | * defer everything. | |
ca0dde97 | 2535 | */ |
d5b3cf71 | 2536 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2537 | out: |
8135be5a | 2538 | css_put(&memcg->css); |
ca0dde97 | 2539 | return cachep; |
d7f25f8a | 2540 | } |
d7f25f8a | 2541 | |
45264778 VD |
2542 | /** |
2543 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache | |
2544 | * @cachep: the cache returned by memcg_kmem_get_cache | |
2545 | */ | |
2546 | void memcg_kmem_put_cache(struct kmem_cache *cachep) | |
8135be5a VD |
2547 | { |
2548 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2549 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2550 | } |
2551 | ||
45264778 | 2552 | /** |
b213b54f | 2553 | * memcg_kmem_charge_memcg: charge a kmem page |
45264778 VD |
2554 | * @page: page to charge |
2555 | * @gfp: reclaim mode | |
2556 | * @order: allocation order | |
2557 | * @memcg: memory cgroup to charge | |
2558 | * | |
2559 | * Returns 0 on success, an error code on failure. | |
2560 | */ | |
2561 | int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, | |
2562 | struct mem_cgroup *memcg) | |
7ae1e1d0 | 2563 | { |
f3ccb2c4 VD |
2564 | unsigned int nr_pages = 1 << order; |
2565 | struct page_counter *counter; | |
7ae1e1d0 GC |
2566 | int ret; |
2567 | ||
f3ccb2c4 | 2568 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2569 | if (ret) |
f3ccb2c4 | 2570 | return ret; |
52c29b04 JW |
2571 | |
2572 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2573 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2574 | cancel_charge(memcg, nr_pages); | |
2575 | return -ENOMEM; | |
7ae1e1d0 GC |
2576 | } |
2577 | ||
f3ccb2c4 | 2578 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2579 | |
f3ccb2c4 | 2580 | return 0; |
7ae1e1d0 GC |
2581 | } |
2582 | ||
45264778 VD |
2583 | /** |
2584 | * memcg_kmem_charge: charge a kmem page to the current memory cgroup | |
2585 | * @page: page to charge | |
2586 | * @gfp: reclaim mode | |
2587 | * @order: allocation order | |
2588 | * | |
2589 | * Returns 0 on success, an error code on failure. | |
2590 | */ | |
2591 | int memcg_kmem_charge(struct page *page, gfp_t gfp, int order) | |
7ae1e1d0 | 2592 | { |
f3ccb2c4 | 2593 | struct mem_cgroup *memcg; |
fcff7d7e | 2594 | int ret = 0; |
7ae1e1d0 | 2595 | |
e68599a3 | 2596 | if (mem_cgroup_disabled() || memcg_kmem_bypass()) |
45264778 VD |
2597 | return 0; |
2598 | ||
d46eb14b | 2599 | memcg = get_mem_cgroup_from_current(); |
c4159a75 | 2600 | if (!mem_cgroup_is_root(memcg)) { |
45264778 | 2601 | ret = memcg_kmem_charge_memcg(page, gfp, order, memcg); |
c4159a75 VD |
2602 | if (!ret) |
2603 | __SetPageKmemcg(page); | |
2604 | } | |
7ae1e1d0 | 2605 | css_put(&memcg->css); |
d05e83a6 | 2606 | return ret; |
7ae1e1d0 | 2607 | } |
45264778 VD |
2608 | /** |
2609 | * memcg_kmem_uncharge: uncharge a kmem page | |
2610 | * @page: page to uncharge | |
2611 | * @order: allocation order | |
2612 | */ | |
2613 | void memcg_kmem_uncharge(struct page *page, int order) | |
7ae1e1d0 | 2614 | { |
1306a85a | 2615 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2616 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2617 | |
7ae1e1d0 GC |
2618 | if (!memcg) |
2619 | return; | |
2620 | ||
309381fe | 2621 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2622 | |
52c29b04 JW |
2623 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2624 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2625 | ||
f3ccb2c4 | 2626 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2627 | if (do_memsw_account()) |
f3ccb2c4 | 2628 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2629 | |
1306a85a | 2630 | page->mem_cgroup = NULL; |
c4159a75 VD |
2631 | |
2632 | /* slab pages do not have PageKmemcg flag set */ | |
2633 | if (PageKmemcg(page)) | |
2634 | __ClearPageKmemcg(page); | |
2635 | ||
f3ccb2c4 | 2636 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2637 | } |
84c07d11 | 2638 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 2639 | |
ca3e0214 KH |
2640 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2641 | ||
ca3e0214 KH |
2642 | /* |
2643 | * Because tail pages are not marked as "used", set it. We're under | |
a52633d8 | 2644 | * zone_lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2645 | */ |
e94c8a9c | 2646 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2647 | { |
e94c8a9c | 2648 | int i; |
ca3e0214 | 2649 | |
3d37c4a9 KH |
2650 | if (mem_cgroup_disabled()) |
2651 | return; | |
b070e65c | 2652 | |
29833315 | 2653 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2654 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2655 | |
c9019e9b | 2656 | __mod_memcg_state(head->mem_cgroup, MEMCG_RSS_HUGE, -HPAGE_PMD_NR); |
ca3e0214 | 2657 | } |
12d27107 | 2658 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2659 | |
c255a458 | 2660 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
2661 | /** |
2662 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2663 | * @entry: swap entry to be moved | |
2664 | * @from: mem_cgroup which the entry is moved from | |
2665 | * @to: mem_cgroup which the entry is moved to | |
2666 | * | |
2667 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2668 | * as the mem_cgroup's id of @from. | |
2669 | * | |
2670 | * Returns 0 on success, -EINVAL on failure. | |
2671 | * | |
3e32cb2e | 2672 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2673 | * both res and memsw, and called css_get(). |
2674 | */ | |
2675 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2676 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2677 | { |
2678 | unsigned short old_id, new_id; | |
2679 | ||
34c00c31 LZ |
2680 | old_id = mem_cgroup_id(from); |
2681 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2682 | |
2683 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
2684 | mod_memcg_state(from, MEMCG_SWAP, -1); |
2685 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
2686 | return 0; |
2687 | } | |
2688 | return -EINVAL; | |
2689 | } | |
2690 | #else | |
2691 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2692 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2693 | { |
2694 | return -EINVAL; | |
2695 | } | |
8c7c6e34 | 2696 | #endif |
d13d1443 | 2697 | |
bbec2e15 | 2698 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 2699 | |
bbec2e15 RG |
2700 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
2701 | unsigned long max, bool memsw) | |
628f4235 | 2702 | { |
3e32cb2e | 2703 | bool enlarge = false; |
bb4a7ea2 | 2704 | bool drained = false; |
3e32cb2e | 2705 | int ret; |
c054a78c YZ |
2706 | bool limits_invariant; |
2707 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 2708 | |
3e32cb2e | 2709 | do { |
628f4235 KH |
2710 | if (signal_pending(current)) { |
2711 | ret = -EINTR; | |
2712 | break; | |
2713 | } | |
3e32cb2e | 2714 | |
bbec2e15 | 2715 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
2716 | /* |
2717 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 2718 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 2719 | */ |
bbec2e15 RG |
2720 | limits_invariant = memsw ? max >= memcg->memory.max : |
2721 | max <= memcg->memsw.max; | |
c054a78c | 2722 | if (!limits_invariant) { |
bbec2e15 | 2723 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 2724 | ret = -EINVAL; |
8c7c6e34 KH |
2725 | break; |
2726 | } | |
bbec2e15 | 2727 | if (max > counter->max) |
3e32cb2e | 2728 | enlarge = true; |
bbec2e15 RG |
2729 | ret = page_counter_set_max(counter, max); |
2730 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
2731 | |
2732 | if (!ret) | |
2733 | break; | |
2734 | ||
bb4a7ea2 SB |
2735 | if (!drained) { |
2736 | drain_all_stock(memcg); | |
2737 | drained = true; | |
2738 | continue; | |
2739 | } | |
2740 | ||
1ab5c056 AR |
2741 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
2742 | GFP_KERNEL, !memsw)) { | |
2743 | ret = -EBUSY; | |
2744 | break; | |
2745 | } | |
2746 | } while (true); | |
3e32cb2e | 2747 | |
3c11ecf4 KH |
2748 | if (!ret && enlarge) |
2749 | memcg_oom_recover(memcg); | |
3e32cb2e | 2750 | |
628f4235 KH |
2751 | return ret; |
2752 | } | |
2753 | ||
ef8f2327 | 2754 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
2755 | gfp_t gfp_mask, |
2756 | unsigned long *total_scanned) | |
2757 | { | |
2758 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 2759 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
2760 | unsigned long reclaimed; |
2761 | int loop = 0; | |
ef8f2327 | 2762 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 2763 | unsigned long excess; |
0608f43d AM |
2764 | unsigned long nr_scanned; |
2765 | ||
2766 | if (order > 0) | |
2767 | return 0; | |
2768 | ||
ef8f2327 | 2769 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
2770 | |
2771 | /* | |
2772 | * Do not even bother to check the largest node if the root | |
2773 | * is empty. Do it lockless to prevent lock bouncing. Races | |
2774 | * are acceptable as soft limit is best effort anyway. | |
2775 | */ | |
bfc7228b | 2776 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
2777 | return 0; |
2778 | ||
0608f43d AM |
2779 | /* |
2780 | * This loop can run a while, specially if mem_cgroup's continuously | |
2781 | * keep exceeding their soft limit and putting the system under | |
2782 | * pressure | |
2783 | */ | |
2784 | do { | |
2785 | if (next_mz) | |
2786 | mz = next_mz; | |
2787 | else | |
2788 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2789 | if (!mz) | |
2790 | break; | |
2791 | ||
2792 | nr_scanned = 0; | |
ef8f2327 | 2793 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
2794 | gfp_mask, &nr_scanned); |
2795 | nr_reclaimed += reclaimed; | |
2796 | *total_scanned += nr_scanned; | |
0a31bc97 | 2797 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2798 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2799 | |
2800 | /* | |
2801 | * If we failed to reclaim anything from this memory cgroup | |
2802 | * it is time to move on to the next cgroup | |
2803 | */ | |
2804 | next_mz = NULL; | |
bc2f2e7f VD |
2805 | if (!reclaimed) |
2806 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2807 | ||
3e32cb2e | 2808 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2809 | /* |
2810 | * One school of thought says that we should not add | |
2811 | * back the node to the tree if reclaim returns 0. | |
2812 | * But our reclaim could return 0, simply because due | |
2813 | * to priority we are exposing a smaller subset of | |
2814 | * memory to reclaim from. Consider this as a longer | |
2815 | * term TODO. | |
2816 | */ | |
2817 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2818 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2819 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2820 | css_put(&mz->memcg->css); |
2821 | loop++; | |
2822 | /* | |
2823 | * Could not reclaim anything and there are no more | |
2824 | * mem cgroups to try or we seem to be looping without | |
2825 | * reclaiming anything. | |
2826 | */ | |
2827 | if (!nr_reclaimed && | |
2828 | (next_mz == NULL || | |
2829 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2830 | break; | |
2831 | } while (!nr_reclaimed); | |
2832 | if (next_mz) | |
2833 | css_put(&next_mz->memcg->css); | |
2834 | return nr_reclaimed; | |
2835 | } | |
2836 | ||
ea280e7b TH |
2837 | /* |
2838 | * Test whether @memcg has children, dead or alive. Note that this | |
2839 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2840 | * returns %true if there are child csses according to the cgroup | |
2841 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2842 | */ | |
b5f99b53 GC |
2843 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2844 | { | |
ea280e7b TH |
2845 | bool ret; |
2846 | ||
ea280e7b TH |
2847 | rcu_read_lock(); |
2848 | ret = css_next_child(NULL, &memcg->css); | |
2849 | rcu_read_unlock(); | |
2850 | return ret; | |
b5f99b53 GC |
2851 | } |
2852 | ||
c26251f9 | 2853 | /* |
51038171 | 2854 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
2855 | * |
2856 | * Caller is responsible for holding css reference for memcg. | |
2857 | */ | |
2858 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2859 | { | |
2860 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2861 | |
c1e862c1 KH |
2862 | /* we call try-to-free pages for make this cgroup empty */ |
2863 | lru_add_drain_all(); | |
d12c60f6 JS |
2864 | |
2865 | drain_all_stock(memcg); | |
2866 | ||
f817ed48 | 2867 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2868 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2869 | int progress; |
c1e862c1 | 2870 | |
c26251f9 MH |
2871 | if (signal_pending(current)) |
2872 | return -EINTR; | |
2873 | ||
b70a2a21 JW |
2874 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2875 | GFP_KERNEL, true); | |
c1e862c1 | 2876 | if (!progress) { |
f817ed48 | 2877 | nr_retries--; |
c1e862c1 | 2878 | /* maybe some writeback is necessary */ |
8aa7e847 | 2879 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2880 | } |
f817ed48 KH |
2881 | |
2882 | } | |
ab5196c2 MH |
2883 | |
2884 | return 0; | |
cc847582 KH |
2885 | } |
2886 | ||
6770c64e TH |
2887 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2888 | char *buf, size_t nbytes, | |
2889 | loff_t off) | |
c1e862c1 | 2890 | { |
6770c64e | 2891 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2892 | |
d8423011 MH |
2893 | if (mem_cgroup_is_root(memcg)) |
2894 | return -EINVAL; | |
6770c64e | 2895 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2896 | } |
2897 | ||
182446d0 TH |
2898 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2899 | struct cftype *cft) | |
18f59ea7 | 2900 | { |
182446d0 | 2901 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2902 | } |
2903 | ||
182446d0 TH |
2904 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2905 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2906 | { |
2907 | int retval = 0; | |
182446d0 | 2908 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2909 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2910 | |
567fb435 | 2911 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 2912 | return 0; |
567fb435 | 2913 | |
18f59ea7 | 2914 | /* |
af901ca1 | 2915 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2916 | * in the child subtrees. If it is unset, then the change can |
2917 | * occur, provided the current cgroup has no children. | |
2918 | * | |
2919 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2920 | * set if there are no children. | |
2921 | */ | |
c0ff4b85 | 2922 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2923 | (val == 1 || val == 0)) { |
ea280e7b | 2924 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2925 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2926 | else |
2927 | retval = -EBUSY; | |
2928 | } else | |
2929 | retval = -EINVAL; | |
567fb435 | 2930 | |
18f59ea7 BS |
2931 | return retval; |
2932 | } | |
2933 | ||
8de7ecc6 SB |
2934 | struct accumulated_stats { |
2935 | unsigned long stat[MEMCG_NR_STAT]; | |
2936 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
2937 | unsigned long lru_pages[NR_LRU_LISTS]; | |
2938 | const unsigned int *stats_array; | |
2939 | const unsigned int *events_array; | |
2940 | int stats_size; | |
2941 | int events_size; | |
2942 | }; | |
ce00a967 | 2943 | |
8de7ecc6 SB |
2944 | static void accumulate_memcg_tree(struct mem_cgroup *memcg, |
2945 | struct accumulated_stats *acc) | |
587d9f72 | 2946 | { |
8de7ecc6 | 2947 | struct mem_cgroup *mi; |
72b54e73 | 2948 | int i; |
587d9f72 | 2949 | |
8de7ecc6 SB |
2950 | for_each_mem_cgroup_tree(mi, memcg) { |
2951 | for (i = 0; i < acc->stats_size; i++) | |
2952 | acc->stat[i] += memcg_page_state(mi, | |
2953 | acc->stats_array ? acc->stats_array[i] : i); | |
587d9f72 | 2954 | |
8de7ecc6 SB |
2955 | for (i = 0; i < acc->events_size; i++) |
2956 | acc->events[i] += memcg_sum_events(mi, | |
2957 | acc->events_array ? acc->events_array[i] : i); | |
2958 | ||
2959 | for (i = 0; i < NR_LRU_LISTS; i++) | |
2960 | acc->lru_pages[i] += | |
2961 | mem_cgroup_nr_lru_pages(mi, BIT(i)); | |
72b54e73 | 2962 | } |
587d9f72 JW |
2963 | } |
2964 | ||
6f646156 | 2965 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2966 | { |
72b54e73 | 2967 | unsigned long val = 0; |
ce00a967 | 2968 | |
3e32cb2e | 2969 | if (mem_cgroup_is_root(memcg)) { |
72b54e73 VD |
2970 | struct mem_cgroup *iter; |
2971 | ||
2972 | for_each_mem_cgroup_tree(iter, memcg) { | |
ccda7f43 JW |
2973 | val += memcg_page_state(iter, MEMCG_CACHE); |
2974 | val += memcg_page_state(iter, MEMCG_RSS); | |
72b54e73 | 2975 | if (swap) |
ccda7f43 | 2976 | val += memcg_page_state(iter, MEMCG_SWAP); |
72b54e73 | 2977 | } |
3e32cb2e | 2978 | } else { |
ce00a967 | 2979 | if (!swap) |
3e32cb2e | 2980 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2981 | else |
3e32cb2e | 2982 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2983 | } |
c12176d3 | 2984 | return val; |
ce00a967 JW |
2985 | } |
2986 | ||
3e32cb2e JW |
2987 | enum { |
2988 | RES_USAGE, | |
2989 | RES_LIMIT, | |
2990 | RES_MAX_USAGE, | |
2991 | RES_FAILCNT, | |
2992 | RES_SOFT_LIMIT, | |
2993 | }; | |
ce00a967 | 2994 | |
791badbd | 2995 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2996 | struct cftype *cft) |
8cdea7c0 | 2997 | { |
182446d0 | 2998 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2999 | struct page_counter *counter; |
af36f906 | 3000 | |
3e32cb2e | 3001 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3002 | case _MEM: |
3e32cb2e JW |
3003 | counter = &memcg->memory; |
3004 | break; | |
8c7c6e34 | 3005 | case _MEMSWAP: |
3e32cb2e JW |
3006 | counter = &memcg->memsw; |
3007 | break; | |
510fc4e1 | 3008 | case _KMEM: |
3e32cb2e | 3009 | counter = &memcg->kmem; |
510fc4e1 | 3010 | break; |
d55f90bf | 3011 | case _TCP: |
0db15298 | 3012 | counter = &memcg->tcpmem; |
d55f90bf | 3013 | break; |
8c7c6e34 KH |
3014 | default: |
3015 | BUG(); | |
8c7c6e34 | 3016 | } |
3e32cb2e JW |
3017 | |
3018 | switch (MEMFILE_ATTR(cft->private)) { | |
3019 | case RES_USAGE: | |
3020 | if (counter == &memcg->memory) | |
c12176d3 | 3021 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3022 | if (counter == &memcg->memsw) |
c12176d3 | 3023 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3024 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3025 | case RES_LIMIT: | |
bbec2e15 | 3026 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3027 | case RES_MAX_USAGE: |
3028 | return (u64)counter->watermark * PAGE_SIZE; | |
3029 | case RES_FAILCNT: | |
3030 | return counter->failcnt; | |
3031 | case RES_SOFT_LIMIT: | |
3032 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3033 | default: | |
3034 | BUG(); | |
3035 | } | |
8cdea7c0 | 3036 | } |
510fc4e1 | 3037 | |
84c07d11 | 3038 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3039 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3040 | { |
d6441637 VD |
3041 | int memcg_id; |
3042 | ||
b313aeee VD |
3043 | if (cgroup_memory_nokmem) |
3044 | return 0; | |
3045 | ||
2a4db7eb | 3046 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3047 | BUG_ON(memcg->kmem_state); |
d6441637 | 3048 | |
f3bb3043 | 3049 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3050 | if (memcg_id < 0) |
3051 | return memcg_id; | |
d6441637 | 3052 | |
ef12947c | 3053 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 3054 | /* |
567e9ab2 | 3055 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 3056 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
3057 | * guarantee no one starts accounting before all call sites are |
3058 | * patched. | |
3059 | */ | |
900a38f0 | 3060 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3061 | memcg->kmem_state = KMEM_ONLINE; |
bc2791f8 | 3062 | INIT_LIST_HEAD(&memcg->kmem_caches); |
0b8f73e1 JW |
3063 | |
3064 | return 0; | |
d6441637 VD |
3065 | } |
3066 | ||
8e0a8912 JW |
3067 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3068 | { | |
3069 | struct cgroup_subsys_state *css; | |
3070 | struct mem_cgroup *parent, *child; | |
3071 | int kmemcg_id; | |
3072 | ||
3073 | if (memcg->kmem_state != KMEM_ONLINE) | |
3074 | return; | |
3075 | /* | |
3076 | * Clear the online state before clearing memcg_caches array | |
3077 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
3078 | * guarantees that no cache will be created for this cgroup | |
3079 | * after we are done (see memcg_create_kmem_cache()). | |
3080 | */ | |
3081 | memcg->kmem_state = KMEM_ALLOCATED; | |
3082 | ||
3083 | memcg_deactivate_kmem_caches(memcg); | |
3084 | ||
3085 | kmemcg_id = memcg->kmemcg_id; | |
3086 | BUG_ON(kmemcg_id < 0); | |
3087 | ||
3088 | parent = parent_mem_cgroup(memcg); | |
3089 | if (!parent) | |
3090 | parent = root_mem_cgroup; | |
3091 | ||
3092 | /* | |
3093 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3094 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3095 | * to ones of the parent. After we have finished, all list_lrus | |
3096 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3097 | * ordering is imposed by list_lru_node->lock taken by | |
3098 | * memcg_drain_all_list_lrus(). | |
3099 | */ | |
3a06bb78 | 3100 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3101 | css_for_each_descendant_pre(css, &memcg->css) { |
3102 | child = mem_cgroup_from_css(css); | |
3103 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3104 | child->kmemcg_id = parent->kmemcg_id; | |
3105 | if (!memcg->use_hierarchy) | |
3106 | break; | |
3107 | } | |
3a06bb78 TH |
3108 | rcu_read_unlock(); |
3109 | ||
9bec5c35 | 3110 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3111 | |
3112 | memcg_free_cache_id(kmemcg_id); | |
3113 | } | |
3114 | ||
3115 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3116 | { | |
0b8f73e1 JW |
3117 | /* css_alloc() failed, offlining didn't happen */ |
3118 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3119 | memcg_offline_kmem(memcg); | |
3120 | ||
8e0a8912 JW |
3121 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
3122 | memcg_destroy_kmem_caches(memcg); | |
3123 | static_branch_dec(&memcg_kmem_enabled_key); | |
3124 | WARN_ON(page_counter_read(&memcg->kmem)); | |
3125 | } | |
8e0a8912 | 3126 | } |
d6441637 | 3127 | #else |
0b8f73e1 | 3128 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3129 | { |
3130 | return 0; | |
3131 | } | |
3132 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3133 | { | |
3134 | } | |
3135 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3136 | { | |
3137 | } | |
84c07d11 | 3138 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3139 | |
bbec2e15 RG |
3140 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3141 | unsigned long max) | |
d6441637 | 3142 | { |
b313aeee | 3143 | int ret; |
127424c8 | 3144 | |
bbec2e15 RG |
3145 | mutex_lock(&memcg_max_mutex); |
3146 | ret = page_counter_set_max(&memcg->kmem, max); | |
3147 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3148 | return ret; |
d6441637 | 3149 | } |
510fc4e1 | 3150 | |
bbec2e15 | 3151 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3152 | { |
3153 | int ret; | |
3154 | ||
bbec2e15 | 3155 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3156 | |
bbec2e15 | 3157 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3158 | if (ret) |
3159 | goto out; | |
3160 | ||
0db15298 | 3161 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3162 | /* |
3163 | * The active flag needs to be written after the static_key | |
3164 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3165 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3166 | * for details, and note that we don't mark any socket as | |
3167 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3168 | * |
3169 | * We need to do this, because static_keys will span multiple | |
3170 | * sites, but we can't control their order. If we mark a socket | |
3171 | * as accounted, but the accounting functions are not patched in | |
3172 | * yet, we'll lose accounting. | |
3173 | * | |
2d758073 | 3174 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3175 | * because when this value change, the code to process it is not |
3176 | * patched in yet. | |
3177 | */ | |
3178 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3179 | memcg->tcpmem_active = true; |
d55f90bf VD |
3180 | } |
3181 | out: | |
bbec2e15 | 3182 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3183 | return ret; |
3184 | } | |
d55f90bf | 3185 | |
628f4235 KH |
3186 | /* |
3187 | * The user of this function is... | |
3188 | * RES_LIMIT. | |
3189 | */ | |
451af504 TH |
3190 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3191 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3192 | { |
451af504 | 3193 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3194 | unsigned long nr_pages; |
628f4235 KH |
3195 | int ret; |
3196 | ||
451af504 | 3197 | buf = strstrip(buf); |
650c5e56 | 3198 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3199 | if (ret) |
3200 | return ret; | |
af36f906 | 3201 | |
3e32cb2e | 3202 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3203 | case RES_LIMIT: |
4b3bde4c BS |
3204 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3205 | ret = -EINVAL; | |
3206 | break; | |
3207 | } | |
3e32cb2e JW |
3208 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3209 | case _MEM: | |
bbec2e15 | 3210 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3211 | break; |
3e32cb2e | 3212 | case _MEMSWAP: |
bbec2e15 | 3213 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3214 | break; |
3e32cb2e | 3215 | case _KMEM: |
bbec2e15 | 3216 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3217 | break; |
d55f90bf | 3218 | case _TCP: |
bbec2e15 | 3219 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3220 | break; |
3e32cb2e | 3221 | } |
296c81d8 | 3222 | break; |
3e32cb2e JW |
3223 | case RES_SOFT_LIMIT: |
3224 | memcg->soft_limit = nr_pages; | |
3225 | ret = 0; | |
628f4235 KH |
3226 | break; |
3227 | } | |
451af504 | 3228 | return ret ?: nbytes; |
8cdea7c0 BS |
3229 | } |
3230 | ||
6770c64e TH |
3231 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3232 | size_t nbytes, loff_t off) | |
c84872e1 | 3233 | { |
6770c64e | 3234 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3235 | struct page_counter *counter; |
c84872e1 | 3236 | |
3e32cb2e JW |
3237 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3238 | case _MEM: | |
3239 | counter = &memcg->memory; | |
3240 | break; | |
3241 | case _MEMSWAP: | |
3242 | counter = &memcg->memsw; | |
3243 | break; | |
3244 | case _KMEM: | |
3245 | counter = &memcg->kmem; | |
3246 | break; | |
d55f90bf | 3247 | case _TCP: |
0db15298 | 3248 | counter = &memcg->tcpmem; |
d55f90bf | 3249 | break; |
3e32cb2e JW |
3250 | default: |
3251 | BUG(); | |
3252 | } | |
af36f906 | 3253 | |
3e32cb2e | 3254 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3255 | case RES_MAX_USAGE: |
3e32cb2e | 3256 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3257 | break; |
3258 | case RES_FAILCNT: | |
3e32cb2e | 3259 | counter->failcnt = 0; |
29f2a4da | 3260 | break; |
3e32cb2e JW |
3261 | default: |
3262 | BUG(); | |
29f2a4da | 3263 | } |
f64c3f54 | 3264 | |
6770c64e | 3265 | return nbytes; |
c84872e1 PE |
3266 | } |
3267 | ||
182446d0 | 3268 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3269 | struct cftype *cft) |
3270 | { | |
182446d0 | 3271 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3272 | } |
3273 | ||
02491447 | 3274 | #ifdef CONFIG_MMU |
182446d0 | 3275 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3276 | struct cftype *cft, u64 val) |
3277 | { | |
182446d0 | 3278 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3279 | |
1dfab5ab | 3280 | if (val & ~MOVE_MASK) |
7dc74be0 | 3281 | return -EINVAL; |
ee5e8472 | 3282 | |
7dc74be0 | 3283 | /* |
ee5e8472 GC |
3284 | * No kind of locking is needed in here, because ->can_attach() will |
3285 | * check this value once in the beginning of the process, and then carry | |
3286 | * on with stale data. This means that changes to this value will only | |
3287 | * affect task migrations starting after the change. | |
7dc74be0 | 3288 | */ |
c0ff4b85 | 3289 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3290 | return 0; |
3291 | } | |
02491447 | 3292 | #else |
182446d0 | 3293 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3294 | struct cftype *cft, u64 val) |
3295 | { | |
3296 | return -ENOSYS; | |
3297 | } | |
3298 | #endif | |
7dc74be0 | 3299 | |
406eb0c9 | 3300 | #ifdef CONFIG_NUMA |
2da8ca82 | 3301 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3302 | { |
25485de6 GT |
3303 | struct numa_stat { |
3304 | const char *name; | |
3305 | unsigned int lru_mask; | |
3306 | }; | |
3307 | ||
3308 | static const struct numa_stat stats[] = { | |
3309 | { "total", LRU_ALL }, | |
3310 | { "file", LRU_ALL_FILE }, | |
3311 | { "anon", LRU_ALL_ANON }, | |
3312 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3313 | }; | |
3314 | const struct numa_stat *stat; | |
406eb0c9 | 3315 | int nid; |
25485de6 | 3316 | unsigned long nr; |
2da8ca82 | 3317 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3318 | |
25485de6 GT |
3319 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3320 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3321 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3322 | for_each_node_state(nid, N_MEMORY) { | |
3323 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3324 | stat->lru_mask); | |
3325 | seq_printf(m, " N%d=%lu", nid, nr); | |
3326 | } | |
3327 | seq_putc(m, '\n'); | |
406eb0c9 | 3328 | } |
406eb0c9 | 3329 | |
071aee13 YH |
3330 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3331 | struct mem_cgroup *iter; | |
3332 | ||
3333 | nr = 0; | |
3334 | for_each_mem_cgroup_tree(iter, memcg) | |
3335 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3336 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3337 | for_each_node_state(nid, N_MEMORY) { | |
3338 | nr = 0; | |
3339 | for_each_mem_cgroup_tree(iter, memcg) | |
3340 | nr += mem_cgroup_node_nr_lru_pages( | |
3341 | iter, nid, stat->lru_mask); | |
3342 | seq_printf(m, " N%d=%lu", nid, nr); | |
3343 | } | |
3344 | seq_putc(m, '\n'); | |
406eb0c9 | 3345 | } |
406eb0c9 | 3346 | |
406eb0c9 YH |
3347 | return 0; |
3348 | } | |
3349 | #endif /* CONFIG_NUMA */ | |
3350 | ||
df0e53d0 | 3351 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3352 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
3353 | PGPGIN, |
3354 | PGPGOUT, | |
3355 | PGFAULT, | |
3356 | PGMAJFAULT, | |
3357 | }; | |
3358 | ||
3359 | static const char *const memcg1_event_names[] = { | |
3360 | "pgpgin", | |
3361 | "pgpgout", | |
3362 | "pgfault", | |
3363 | "pgmajfault", | |
3364 | }; | |
3365 | ||
2da8ca82 | 3366 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3367 | { |
2da8ca82 | 3368 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3369 | unsigned long memory, memsw; |
af7c4b0e JW |
3370 | struct mem_cgroup *mi; |
3371 | unsigned int i; | |
8de7ecc6 | 3372 | struct accumulated_stats acc; |
406eb0c9 | 3373 | |
71cd3113 | 3374 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c RS |
3375 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3376 | ||
71cd3113 JW |
3377 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
3378 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) | |
1dd3a273 | 3379 | continue; |
71cd3113 | 3380 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], |
ccda7f43 | 3381 | memcg_page_state(memcg, memcg1_stats[i]) * |
71cd3113 | 3382 | PAGE_SIZE); |
1dd3a273 | 3383 | } |
7b854121 | 3384 | |
df0e53d0 JW |
3385 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3386 | seq_printf(m, "%s %lu\n", memcg1_event_names[i], | |
ccda7f43 | 3387 | memcg_sum_events(memcg, memcg1_events[i])); |
af7c4b0e JW |
3388 | |
3389 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3390 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3391 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3392 | ||
14067bb3 | 3393 | /* Hierarchical information */ |
3e32cb2e JW |
3394 | memory = memsw = PAGE_COUNTER_MAX; |
3395 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
bbec2e15 RG |
3396 | memory = min(memory, mi->memory.max); |
3397 | memsw = min(memsw, mi->memsw.max); | |
fee7b548 | 3398 | } |
3e32cb2e JW |
3399 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3400 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3401 | if (do_memsw_account()) |
3e32cb2e JW |
3402 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3403 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3404 | |
8de7ecc6 SB |
3405 | memset(&acc, 0, sizeof(acc)); |
3406 | acc.stats_size = ARRAY_SIZE(memcg1_stats); | |
3407 | acc.stats_array = memcg1_stats; | |
3408 | acc.events_size = ARRAY_SIZE(memcg1_events); | |
3409 | acc.events_array = memcg1_events; | |
3410 | accumulate_memcg_tree(memcg, &acc); | |
af7c4b0e | 3411 | |
8de7ecc6 | 3412 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
71cd3113 | 3413 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3414 | continue; |
8de7ecc6 SB |
3415 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
3416 | (u64)acc.stat[i] * PAGE_SIZE); | |
af7c4b0e JW |
3417 | } |
3418 | ||
8de7ecc6 SB |
3419 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3420 | seq_printf(m, "total_%s %llu\n", memcg1_event_names[i], | |
3421 | (u64)acc.events[i]); | |
af7c4b0e | 3422 | |
8de7ecc6 SB |
3423 | for (i = 0; i < NR_LRU_LISTS; i++) |
3424 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], | |
3425 | (u64)acc.lru_pages[i] * PAGE_SIZE); | |
14067bb3 | 3426 | |
7f016ee8 | 3427 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 3428 | { |
ef8f2327 MG |
3429 | pg_data_t *pgdat; |
3430 | struct mem_cgroup_per_node *mz; | |
89abfab1 | 3431 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3432 | unsigned long recent_rotated[2] = {0, 0}; |
3433 | unsigned long recent_scanned[2] = {0, 0}; | |
3434 | ||
ef8f2327 MG |
3435 | for_each_online_pgdat(pgdat) { |
3436 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
3437 | rstat = &mz->lruvec.reclaim_stat; | |
7f016ee8 | 3438 | |
ef8f2327 MG |
3439 | recent_rotated[0] += rstat->recent_rotated[0]; |
3440 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3441 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3442 | recent_scanned[1] += rstat->recent_scanned[1]; | |
3443 | } | |
78ccf5b5 JW |
3444 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3445 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3446 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3447 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3448 | } |
3449 | #endif | |
3450 | ||
d2ceb9b7 KH |
3451 | return 0; |
3452 | } | |
3453 | ||
182446d0 TH |
3454 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3455 | struct cftype *cft) | |
a7885eb8 | 3456 | { |
182446d0 | 3457 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3458 | |
1f4c025b | 3459 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3460 | } |
3461 | ||
182446d0 TH |
3462 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3463 | struct cftype *cft, u64 val) | |
a7885eb8 | 3464 | { |
182446d0 | 3465 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3466 | |
3dae7fec | 3467 | if (val > 100) |
a7885eb8 KM |
3468 | return -EINVAL; |
3469 | ||
14208b0e | 3470 | if (css->parent) |
3dae7fec JW |
3471 | memcg->swappiness = val; |
3472 | else | |
3473 | vm_swappiness = val; | |
068b38c1 | 3474 | |
a7885eb8 KM |
3475 | return 0; |
3476 | } | |
3477 | ||
2e72b634 KS |
3478 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3479 | { | |
3480 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3481 | unsigned long usage; |
2e72b634 KS |
3482 | int i; |
3483 | ||
3484 | rcu_read_lock(); | |
3485 | if (!swap) | |
2c488db2 | 3486 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3487 | else |
2c488db2 | 3488 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3489 | |
3490 | if (!t) | |
3491 | goto unlock; | |
3492 | ||
ce00a967 | 3493 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3494 | |
3495 | /* | |
748dad36 | 3496 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3497 | * If it's not true, a threshold was crossed after last |
3498 | * call of __mem_cgroup_threshold(). | |
3499 | */ | |
5407a562 | 3500 | i = t->current_threshold; |
2e72b634 KS |
3501 | |
3502 | /* | |
3503 | * Iterate backward over array of thresholds starting from | |
3504 | * current_threshold and check if a threshold is crossed. | |
3505 | * If none of thresholds below usage is crossed, we read | |
3506 | * only one element of the array here. | |
3507 | */ | |
3508 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3509 | eventfd_signal(t->entries[i].eventfd, 1); | |
3510 | ||
3511 | /* i = current_threshold + 1 */ | |
3512 | i++; | |
3513 | ||
3514 | /* | |
3515 | * Iterate forward over array of thresholds starting from | |
3516 | * current_threshold+1 and check if a threshold is crossed. | |
3517 | * If none of thresholds above usage is crossed, we read | |
3518 | * only one element of the array here. | |
3519 | */ | |
3520 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3521 | eventfd_signal(t->entries[i].eventfd, 1); | |
3522 | ||
3523 | /* Update current_threshold */ | |
5407a562 | 3524 | t->current_threshold = i - 1; |
2e72b634 KS |
3525 | unlock: |
3526 | rcu_read_unlock(); | |
3527 | } | |
3528 | ||
3529 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3530 | { | |
ad4ca5f4 KS |
3531 | while (memcg) { |
3532 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3533 | if (do_memsw_account()) |
ad4ca5f4 KS |
3534 | __mem_cgroup_threshold(memcg, true); |
3535 | ||
3536 | memcg = parent_mem_cgroup(memcg); | |
3537 | } | |
2e72b634 KS |
3538 | } |
3539 | ||
3540 | static int compare_thresholds(const void *a, const void *b) | |
3541 | { | |
3542 | const struct mem_cgroup_threshold *_a = a; | |
3543 | const struct mem_cgroup_threshold *_b = b; | |
3544 | ||
2bff24a3 GT |
3545 | if (_a->threshold > _b->threshold) |
3546 | return 1; | |
3547 | ||
3548 | if (_a->threshold < _b->threshold) | |
3549 | return -1; | |
3550 | ||
3551 | return 0; | |
2e72b634 KS |
3552 | } |
3553 | ||
c0ff4b85 | 3554 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3555 | { |
3556 | struct mem_cgroup_eventfd_list *ev; | |
3557 | ||
2bcf2e92 MH |
3558 | spin_lock(&memcg_oom_lock); |
3559 | ||
c0ff4b85 | 3560 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3561 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3562 | |
3563 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3564 | return 0; |
3565 | } | |
3566 | ||
c0ff4b85 | 3567 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3568 | { |
7d74b06f KH |
3569 | struct mem_cgroup *iter; |
3570 | ||
c0ff4b85 | 3571 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3572 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3573 | } |
3574 | ||
59b6f873 | 3575 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3576 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3577 | { |
2c488db2 KS |
3578 | struct mem_cgroup_thresholds *thresholds; |
3579 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3580 | unsigned long threshold; |
3581 | unsigned long usage; | |
2c488db2 | 3582 | int i, size, ret; |
2e72b634 | 3583 | |
650c5e56 | 3584 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3585 | if (ret) |
3586 | return ret; | |
3587 | ||
3588 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3589 | |
05b84301 | 3590 | if (type == _MEM) { |
2c488db2 | 3591 | thresholds = &memcg->thresholds; |
ce00a967 | 3592 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3593 | } else if (type == _MEMSWAP) { |
2c488db2 | 3594 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3595 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3596 | } else |
2e72b634 KS |
3597 | BUG(); |
3598 | ||
2e72b634 | 3599 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3600 | if (thresholds->primary) |
2e72b634 KS |
3601 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3602 | ||
2c488db2 | 3603 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3604 | |
3605 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3606 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3607 | GFP_KERNEL); |
2c488db2 | 3608 | if (!new) { |
2e72b634 KS |
3609 | ret = -ENOMEM; |
3610 | goto unlock; | |
3611 | } | |
2c488db2 | 3612 | new->size = size; |
2e72b634 KS |
3613 | |
3614 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3615 | if (thresholds->primary) { |
3616 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3617 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3618 | } |
3619 | ||
2e72b634 | 3620 | /* Add new threshold */ |
2c488db2 KS |
3621 | new->entries[size - 1].eventfd = eventfd; |
3622 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3623 | |
3624 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3625 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3626 | compare_thresholds, NULL); |
3627 | ||
3628 | /* Find current threshold */ | |
2c488db2 | 3629 | new->current_threshold = -1; |
2e72b634 | 3630 | for (i = 0; i < size; i++) { |
748dad36 | 3631 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3632 | /* |
2c488db2 KS |
3633 | * new->current_threshold will not be used until |
3634 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3635 | * it here. |
3636 | */ | |
2c488db2 | 3637 | ++new->current_threshold; |
748dad36 SZ |
3638 | } else |
3639 | break; | |
2e72b634 KS |
3640 | } |
3641 | ||
2c488db2 KS |
3642 | /* Free old spare buffer and save old primary buffer as spare */ |
3643 | kfree(thresholds->spare); | |
3644 | thresholds->spare = thresholds->primary; | |
3645 | ||
3646 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3647 | |
907860ed | 3648 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3649 | synchronize_rcu(); |
3650 | ||
2e72b634 KS |
3651 | unlock: |
3652 | mutex_unlock(&memcg->thresholds_lock); | |
3653 | ||
3654 | return ret; | |
3655 | } | |
3656 | ||
59b6f873 | 3657 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3658 | struct eventfd_ctx *eventfd, const char *args) |
3659 | { | |
59b6f873 | 3660 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3661 | } |
3662 | ||
59b6f873 | 3663 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3664 | struct eventfd_ctx *eventfd, const char *args) |
3665 | { | |
59b6f873 | 3666 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3667 | } |
3668 | ||
59b6f873 | 3669 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3670 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3671 | { |
2c488db2 KS |
3672 | struct mem_cgroup_thresholds *thresholds; |
3673 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3674 | unsigned long usage; |
2c488db2 | 3675 | int i, j, size; |
2e72b634 KS |
3676 | |
3677 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3678 | |
3679 | if (type == _MEM) { | |
2c488db2 | 3680 | thresholds = &memcg->thresholds; |
ce00a967 | 3681 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3682 | } else if (type == _MEMSWAP) { |
2c488db2 | 3683 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3684 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3685 | } else |
2e72b634 KS |
3686 | BUG(); |
3687 | ||
371528ca AV |
3688 | if (!thresholds->primary) |
3689 | goto unlock; | |
3690 | ||
2e72b634 KS |
3691 | /* Check if a threshold crossed before removing */ |
3692 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3693 | ||
3694 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3695 | size = 0; |
3696 | for (i = 0; i < thresholds->primary->size; i++) { | |
3697 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3698 | size++; |
3699 | } | |
3700 | ||
2c488db2 | 3701 | new = thresholds->spare; |
907860ed | 3702 | |
2e72b634 KS |
3703 | /* Set thresholds array to NULL if we don't have thresholds */ |
3704 | if (!size) { | |
2c488db2 KS |
3705 | kfree(new); |
3706 | new = NULL; | |
907860ed | 3707 | goto swap_buffers; |
2e72b634 KS |
3708 | } |
3709 | ||
2c488db2 | 3710 | new->size = size; |
2e72b634 KS |
3711 | |
3712 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3713 | new->current_threshold = -1; |
3714 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3715 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3716 | continue; |
3717 | ||
2c488db2 | 3718 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3719 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3720 | /* |
2c488db2 | 3721 | * new->current_threshold will not be used |
2e72b634 KS |
3722 | * until rcu_assign_pointer(), so it's safe to increment |
3723 | * it here. | |
3724 | */ | |
2c488db2 | 3725 | ++new->current_threshold; |
2e72b634 KS |
3726 | } |
3727 | j++; | |
3728 | } | |
3729 | ||
907860ed | 3730 | swap_buffers: |
2c488db2 KS |
3731 | /* Swap primary and spare array */ |
3732 | thresholds->spare = thresholds->primary; | |
8c757763 | 3733 | |
2c488db2 | 3734 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3735 | |
907860ed | 3736 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3737 | synchronize_rcu(); |
6611d8d7 MC |
3738 | |
3739 | /* If all events are unregistered, free the spare array */ | |
3740 | if (!new) { | |
3741 | kfree(thresholds->spare); | |
3742 | thresholds->spare = NULL; | |
3743 | } | |
371528ca | 3744 | unlock: |
2e72b634 | 3745 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3746 | } |
c1e862c1 | 3747 | |
59b6f873 | 3748 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3749 | struct eventfd_ctx *eventfd) |
3750 | { | |
59b6f873 | 3751 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3752 | } |
3753 | ||
59b6f873 | 3754 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3755 | struct eventfd_ctx *eventfd) |
3756 | { | |
59b6f873 | 3757 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3758 | } |
3759 | ||
59b6f873 | 3760 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3761 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3762 | { |
9490ff27 | 3763 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3764 | |
9490ff27 KH |
3765 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3766 | if (!event) | |
3767 | return -ENOMEM; | |
3768 | ||
1af8efe9 | 3769 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3770 | |
3771 | event->eventfd = eventfd; | |
3772 | list_add(&event->list, &memcg->oom_notify); | |
3773 | ||
3774 | /* already in OOM ? */ | |
c2b42d3c | 3775 | if (memcg->under_oom) |
9490ff27 | 3776 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3777 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3778 | |
3779 | return 0; | |
3780 | } | |
3781 | ||
59b6f873 | 3782 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3783 | struct eventfd_ctx *eventfd) |
9490ff27 | 3784 | { |
9490ff27 | 3785 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3786 | |
1af8efe9 | 3787 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3788 | |
c0ff4b85 | 3789 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3790 | if (ev->eventfd == eventfd) { |
3791 | list_del(&ev->list); | |
3792 | kfree(ev); | |
3793 | } | |
3794 | } | |
3795 | ||
1af8efe9 | 3796 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3797 | } |
3798 | ||
2da8ca82 | 3799 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3800 | { |
2da8ca82 | 3801 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3802 | |
791badbd | 3803 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3804 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
3805 | seq_printf(sf, "oom_kill %lu\n", |
3806 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
3807 | return 0; |
3808 | } | |
3809 | ||
182446d0 | 3810 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3811 | struct cftype *cft, u64 val) |
3812 | { | |
182446d0 | 3813 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3814 | |
3815 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3816 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3817 | return -EINVAL; |
3818 | ||
c0ff4b85 | 3819 | memcg->oom_kill_disable = val; |
4d845ebf | 3820 | if (!val) |
c0ff4b85 | 3821 | memcg_oom_recover(memcg); |
3dae7fec | 3822 | |
3c11ecf4 KH |
3823 | return 0; |
3824 | } | |
3825 | ||
52ebea74 TH |
3826 | #ifdef CONFIG_CGROUP_WRITEBACK |
3827 | ||
841710aa TH |
3828 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3829 | { | |
3830 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3831 | } | |
3832 | ||
3833 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3834 | { | |
3835 | wb_domain_exit(&memcg->cgwb_domain); | |
3836 | } | |
3837 | ||
2529bb3a TH |
3838 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3839 | { | |
3840 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3841 | } | |
3842 | ||
841710aa TH |
3843 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3844 | { | |
3845 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3846 | ||
3847 | if (!memcg->css.parent) | |
3848 | return NULL; | |
3849 | ||
3850 | return &memcg->cgwb_domain; | |
3851 | } | |
3852 | ||
c2aa723a TH |
3853 | /** |
3854 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3855 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3856 | * @pfilepages: out parameter for number of file pages |
3857 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3858 | * @pdirty: out parameter for number of dirty pages |
3859 | * @pwriteback: out parameter for number of pages under writeback | |
3860 | * | |
c5edf9cd TH |
3861 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3862 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3863 | * is a bit more involved. | |
c2aa723a | 3864 | * |
c5edf9cd TH |
3865 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3866 | * headroom is calculated as the lowest headroom of itself and the | |
3867 | * ancestors. Note that this doesn't consider the actual amount of | |
3868 | * available memory in the system. The caller should further cap | |
3869 | * *@pheadroom accordingly. | |
c2aa723a | 3870 | */ |
c5edf9cd TH |
3871 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3872 | unsigned long *pheadroom, unsigned long *pdirty, | |
3873 | unsigned long *pwriteback) | |
c2aa723a TH |
3874 | { |
3875 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3876 | struct mem_cgroup *parent; | |
c2aa723a | 3877 | |
ccda7f43 | 3878 | *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a TH |
3879 | |
3880 | /* this should eventually include NR_UNSTABLE_NFS */ | |
ccda7f43 | 3881 | *pwriteback = memcg_page_state(memcg, NR_WRITEBACK); |
c5edf9cd TH |
3882 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3883 | (1 << LRU_ACTIVE_FILE)); | |
3884 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3885 | |
c2aa723a | 3886 | while ((parent = parent_mem_cgroup(memcg))) { |
bbec2e15 | 3887 | unsigned long ceiling = min(memcg->memory.max, memcg->high); |
c2aa723a TH |
3888 | unsigned long used = page_counter_read(&memcg->memory); |
3889 | ||
c5edf9cd | 3890 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3891 | memcg = parent; |
3892 | } | |
c2aa723a TH |
3893 | } |
3894 | ||
841710aa TH |
3895 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3896 | ||
3897 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3898 | { | |
3899 | return 0; | |
3900 | } | |
3901 | ||
3902 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3903 | { | |
3904 | } | |
3905 | ||
2529bb3a TH |
3906 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3907 | { | |
3908 | } | |
3909 | ||
52ebea74 TH |
3910 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3911 | ||
3bc942f3 TH |
3912 | /* |
3913 | * DO NOT USE IN NEW FILES. | |
3914 | * | |
3915 | * "cgroup.event_control" implementation. | |
3916 | * | |
3917 | * This is way over-engineered. It tries to support fully configurable | |
3918 | * events for each user. Such level of flexibility is completely | |
3919 | * unnecessary especially in the light of the planned unified hierarchy. | |
3920 | * | |
3921 | * Please deprecate this and replace with something simpler if at all | |
3922 | * possible. | |
3923 | */ | |
3924 | ||
79bd9814 TH |
3925 | /* |
3926 | * Unregister event and free resources. | |
3927 | * | |
3928 | * Gets called from workqueue. | |
3929 | */ | |
3bc942f3 | 3930 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3931 | { |
3bc942f3 TH |
3932 | struct mem_cgroup_event *event = |
3933 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3934 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3935 | |
3936 | remove_wait_queue(event->wqh, &event->wait); | |
3937 | ||
59b6f873 | 3938 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3939 | |
3940 | /* Notify userspace the event is going away. */ | |
3941 | eventfd_signal(event->eventfd, 1); | |
3942 | ||
3943 | eventfd_ctx_put(event->eventfd); | |
3944 | kfree(event); | |
59b6f873 | 3945 | css_put(&memcg->css); |
79bd9814 TH |
3946 | } |
3947 | ||
3948 | /* | |
a9a08845 | 3949 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
3950 | * |
3951 | * Called with wqh->lock held and interrupts disabled. | |
3952 | */ | |
ac6424b9 | 3953 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 3954 | int sync, void *key) |
79bd9814 | 3955 | { |
3bc942f3 TH |
3956 | struct mem_cgroup_event *event = |
3957 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3958 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 3959 | __poll_t flags = key_to_poll(key); |
79bd9814 | 3960 | |
a9a08845 | 3961 | if (flags & EPOLLHUP) { |
79bd9814 TH |
3962 | /* |
3963 | * If the event has been detached at cgroup removal, we | |
3964 | * can simply return knowing the other side will cleanup | |
3965 | * for us. | |
3966 | * | |
3967 | * We can't race against event freeing since the other | |
3968 | * side will require wqh->lock via remove_wait_queue(), | |
3969 | * which we hold. | |
3970 | */ | |
fba94807 | 3971 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3972 | if (!list_empty(&event->list)) { |
3973 | list_del_init(&event->list); | |
3974 | /* | |
3975 | * We are in atomic context, but cgroup_event_remove() | |
3976 | * may sleep, so we have to call it in workqueue. | |
3977 | */ | |
3978 | schedule_work(&event->remove); | |
3979 | } | |
fba94807 | 3980 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3981 | } |
3982 | ||
3983 | return 0; | |
3984 | } | |
3985 | ||
3bc942f3 | 3986 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3987 | wait_queue_head_t *wqh, poll_table *pt) |
3988 | { | |
3bc942f3 TH |
3989 | struct mem_cgroup_event *event = |
3990 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3991 | |
3992 | event->wqh = wqh; | |
3993 | add_wait_queue(wqh, &event->wait); | |
3994 | } | |
3995 | ||
3996 | /* | |
3bc942f3 TH |
3997 | * DO NOT USE IN NEW FILES. |
3998 | * | |
79bd9814 TH |
3999 | * Parse input and register new cgroup event handler. |
4000 | * | |
4001 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4002 | * Interpretation of args is defined by control file implementation. | |
4003 | */ | |
451af504 TH |
4004 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4005 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4006 | { |
451af504 | 4007 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4008 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4009 | struct mem_cgroup_event *event; |
79bd9814 TH |
4010 | struct cgroup_subsys_state *cfile_css; |
4011 | unsigned int efd, cfd; | |
4012 | struct fd efile; | |
4013 | struct fd cfile; | |
fba94807 | 4014 | const char *name; |
79bd9814 TH |
4015 | char *endp; |
4016 | int ret; | |
4017 | ||
451af504 TH |
4018 | buf = strstrip(buf); |
4019 | ||
4020 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4021 | if (*endp != ' ') |
4022 | return -EINVAL; | |
451af504 | 4023 | buf = endp + 1; |
79bd9814 | 4024 | |
451af504 | 4025 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4026 | if ((*endp != ' ') && (*endp != '\0')) |
4027 | return -EINVAL; | |
451af504 | 4028 | buf = endp + 1; |
79bd9814 TH |
4029 | |
4030 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4031 | if (!event) | |
4032 | return -ENOMEM; | |
4033 | ||
59b6f873 | 4034 | event->memcg = memcg; |
79bd9814 | 4035 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4036 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4037 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4038 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4039 | |
4040 | efile = fdget(efd); | |
4041 | if (!efile.file) { | |
4042 | ret = -EBADF; | |
4043 | goto out_kfree; | |
4044 | } | |
4045 | ||
4046 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4047 | if (IS_ERR(event->eventfd)) { | |
4048 | ret = PTR_ERR(event->eventfd); | |
4049 | goto out_put_efile; | |
4050 | } | |
4051 | ||
4052 | cfile = fdget(cfd); | |
4053 | if (!cfile.file) { | |
4054 | ret = -EBADF; | |
4055 | goto out_put_eventfd; | |
4056 | } | |
4057 | ||
4058 | /* the process need read permission on control file */ | |
4059 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4060 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4061 | if (ret < 0) | |
4062 | goto out_put_cfile; | |
4063 | ||
fba94807 TH |
4064 | /* |
4065 | * Determine the event callbacks and set them in @event. This used | |
4066 | * to be done via struct cftype but cgroup core no longer knows | |
4067 | * about these events. The following is crude but the whole thing | |
4068 | * is for compatibility anyway. | |
3bc942f3 TH |
4069 | * |
4070 | * DO NOT ADD NEW FILES. | |
fba94807 | 4071 | */ |
b583043e | 4072 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4073 | |
4074 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4075 | event->register_event = mem_cgroup_usage_register_event; | |
4076 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4077 | } else if (!strcmp(name, "memory.oom_control")) { | |
4078 | event->register_event = mem_cgroup_oom_register_event; | |
4079 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4080 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4081 | event->register_event = vmpressure_register_event; | |
4082 | event->unregister_event = vmpressure_unregister_event; | |
4083 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4084 | event->register_event = memsw_cgroup_usage_register_event; |
4085 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4086 | } else { |
4087 | ret = -EINVAL; | |
4088 | goto out_put_cfile; | |
4089 | } | |
4090 | ||
79bd9814 | 4091 | /* |
b5557c4c TH |
4092 | * Verify @cfile should belong to @css. Also, remaining events are |
4093 | * automatically removed on cgroup destruction but the removal is | |
4094 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4095 | */ |
b583043e | 4096 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4097 | &memory_cgrp_subsys); |
79bd9814 | 4098 | ret = -EINVAL; |
5a17f543 | 4099 | if (IS_ERR(cfile_css)) |
79bd9814 | 4100 | goto out_put_cfile; |
5a17f543 TH |
4101 | if (cfile_css != css) { |
4102 | css_put(cfile_css); | |
79bd9814 | 4103 | goto out_put_cfile; |
5a17f543 | 4104 | } |
79bd9814 | 4105 | |
451af504 | 4106 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4107 | if (ret) |
4108 | goto out_put_css; | |
4109 | ||
9965ed17 | 4110 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4111 | |
fba94807 TH |
4112 | spin_lock(&memcg->event_list_lock); |
4113 | list_add(&event->list, &memcg->event_list); | |
4114 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4115 | |
4116 | fdput(cfile); | |
4117 | fdput(efile); | |
4118 | ||
451af504 | 4119 | return nbytes; |
79bd9814 TH |
4120 | |
4121 | out_put_css: | |
b5557c4c | 4122 | css_put(css); |
79bd9814 TH |
4123 | out_put_cfile: |
4124 | fdput(cfile); | |
4125 | out_put_eventfd: | |
4126 | eventfd_ctx_put(event->eventfd); | |
4127 | out_put_efile: | |
4128 | fdput(efile); | |
4129 | out_kfree: | |
4130 | kfree(event); | |
4131 | ||
4132 | return ret; | |
4133 | } | |
4134 | ||
241994ed | 4135 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4136 | { |
0eea1030 | 4137 | .name = "usage_in_bytes", |
8c7c6e34 | 4138 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4139 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4140 | }, |
c84872e1 PE |
4141 | { |
4142 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4143 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4144 | .write = mem_cgroup_reset, |
791badbd | 4145 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4146 | }, |
8cdea7c0 | 4147 | { |
0eea1030 | 4148 | .name = "limit_in_bytes", |
8c7c6e34 | 4149 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4150 | .write = mem_cgroup_write, |
791badbd | 4151 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4152 | }, |
296c81d8 BS |
4153 | { |
4154 | .name = "soft_limit_in_bytes", | |
4155 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4156 | .write = mem_cgroup_write, |
791badbd | 4157 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4158 | }, |
8cdea7c0 BS |
4159 | { |
4160 | .name = "failcnt", | |
8c7c6e34 | 4161 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4162 | .write = mem_cgroup_reset, |
791badbd | 4163 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4164 | }, |
d2ceb9b7 KH |
4165 | { |
4166 | .name = "stat", | |
2da8ca82 | 4167 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4168 | }, |
c1e862c1 KH |
4169 | { |
4170 | .name = "force_empty", | |
6770c64e | 4171 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4172 | }, |
18f59ea7 BS |
4173 | { |
4174 | .name = "use_hierarchy", | |
4175 | .write_u64 = mem_cgroup_hierarchy_write, | |
4176 | .read_u64 = mem_cgroup_hierarchy_read, | |
4177 | }, | |
79bd9814 | 4178 | { |
3bc942f3 | 4179 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4180 | .write = memcg_write_event_control, |
7dbdb199 | 4181 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4182 | }, |
a7885eb8 KM |
4183 | { |
4184 | .name = "swappiness", | |
4185 | .read_u64 = mem_cgroup_swappiness_read, | |
4186 | .write_u64 = mem_cgroup_swappiness_write, | |
4187 | }, | |
7dc74be0 DN |
4188 | { |
4189 | .name = "move_charge_at_immigrate", | |
4190 | .read_u64 = mem_cgroup_move_charge_read, | |
4191 | .write_u64 = mem_cgroup_move_charge_write, | |
4192 | }, | |
9490ff27 KH |
4193 | { |
4194 | .name = "oom_control", | |
2da8ca82 | 4195 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4196 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4197 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4198 | }, | |
70ddf637 AV |
4199 | { |
4200 | .name = "pressure_level", | |
70ddf637 | 4201 | }, |
406eb0c9 YH |
4202 | #ifdef CONFIG_NUMA |
4203 | { | |
4204 | .name = "numa_stat", | |
2da8ca82 | 4205 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4206 | }, |
4207 | #endif | |
510fc4e1 GC |
4208 | { |
4209 | .name = "kmem.limit_in_bytes", | |
4210 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4211 | .write = mem_cgroup_write, |
791badbd | 4212 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4213 | }, |
4214 | { | |
4215 | .name = "kmem.usage_in_bytes", | |
4216 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4217 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4218 | }, |
4219 | { | |
4220 | .name = "kmem.failcnt", | |
4221 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4222 | .write = mem_cgroup_reset, |
791badbd | 4223 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4224 | }, |
4225 | { | |
4226 | .name = "kmem.max_usage_in_bytes", | |
4227 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4228 | .write = mem_cgroup_reset, |
791badbd | 4229 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4230 | }, |
5b365771 | 4231 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
749c5415 GC |
4232 | { |
4233 | .name = "kmem.slabinfo", | |
bc2791f8 TH |
4234 | .seq_start = memcg_slab_start, |
4235 | .seq_next = memcg_slab_next, | |
4236 | .seq_stop = memcg_slab_stop, | |
b047501c | 4237 | .seq_show = memcg_slab_show, |
749c5415 GC |
4238 | }, |
4239 | #endif | |
d55f90bf VD |
4240 | { |
4241 | .name = "kmem.tcp.limit_in_bytes", | |
4242 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4243 | .write = mem_cgroup_write, | |
4244 | .read_u64 = mem_cgroup_read_u64, | |
4245 | }, | |
4246 | { | |
4247 | .name = "kmem.tcp.usage_in_bytes", | |
4248 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4249 | .read_u64 = mem_cgroup_read_u64, | |
4250 | }, | |
4251 | { | |
4252 | .name = "kmem.tcp.failcnt", | |
4253 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4254 | .write = mem_cgroup_reset, | |
4255 | .read_u64 = mem_cgroup_read_u64, | |
4256 | }, | |
4257 | { | |
4258 | .name = "kmem.tcp.max_usage_in_bytes", | |
4259 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4260 | .write = mem_cgroup_reset, | |
4261 | .read_u64 = mem_cgroup_read_u64, | |
4262 | }, | |
6bc10349 | 4263 | { }, /* terminate */ |
af36f906 | 4264 | }; |
8c7c6e34 | 4265 | |
73f576c0 JW |
4266 | /* |
4267 | * Private memory cgroup IDR | |
4268 | * | |
4269 | * Swap-out records and page cache shadow entries need to store memcg | |
4270 | * references in constrained space, so we maintain an ID space that is | |
4271 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4272 | * memory-controlled cgroups to 64k. | |
4273 | * | |
4274 | * However, there usually are many references to the oflline CSS after | |
4275 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4276 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4277 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4278 | * relatively small ID space and prevent the creation of new cgroups | |
4279 | * even when there are much fewer than 64k cgroups - possibly none. | |
4280 | * | |
4281 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4282 | * be freed and recycled when it's no longer needed, which is usually | |
4283 | * when the CSS is offlined. | |
4284 | * | |
4285 | * The only exception to that are records of swapped out tmpfs/shmem | |
4286 | * pages that need to be attributed to live ancestors on swapin. But | |
4287 | * those references are manageable from userspace. | |
4288 | */ | |
4289 | ||
4290 | static DEFINE_IDR(mem_cgroup_idr); | |
4291 | ||
7e97de0b KT |
4292 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
4293 | { | |
4294 | if (memcg->id.id > 0) { | |
4295 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
4296 | memcg->id.id = 0; | |
4297 | } | |
4298 | } | |
4299 | ||
615d66c3 | 4300 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4301 | { |
1c2d479a | 4302 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
4303 | } |
4304 | ||
615d66c3 | 4305 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4306 | { |
1c2d479a | 4307 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 4308 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
4309 | |
4310 | /* Memcg ID pins CSS */ | |
4311 | css_put(&memcg->css); | |
4312 | } | |
4313 | } | |
4314 | ||
615d66c3 VD |
4315 | static inline void mem_cgroup_id_get(struct mem_cgroup *memcg) |
4316 | { | |
4317 | mem_cgroup_id_get_many(memcg, 1); | |
4318 | } | |
4319 | ||
4320 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) | |
4321 | { | |
4322 | mem_cgroup_id_put_many(memcg, 1); | |
4323 | } | |
4324 | ||
73f576c0 JW |
4325 | /** |
4326 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4327 | * @id: the memcg id to look up | |
4328 | * | |
4329 | * Caller must hold rcu_read_lock(). | |
4330 | */ | |
4331 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4332 | { | |
4333 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4334 | return idr_find(&mem_cgroup_idr, id); | |
4335 | } | |
4336 | ||
ef8f2327 | 4337 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4338 | { |
4339 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 4340 | int tmp = node; |
1ecaab2b KH |
4341 | /* |
4342 | * This routine is called against possible nodes. | |
4343 | * But it's BUG to call kmalloc() against offline node. | |
4344 | * | |
4345 | * TODO: this routine can waste much memory for nodes which will | |
4346 | * never be onlined. It's better to use memory hotplug callback | |
4347 | * function. | |
4348 | */ | |
41e3355d KH |
4349 | if (!node_state(node, N_NORMAL_MEMORY)) |
4350 | tmp = -1; | |
17295c88 | 4351 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4352 | if (!pn) |
4353 | return 1; | |
1ecaab2b | 4354 | |
a983b5eb JW |
4355 | pn->lruvec_stat_cpu = alloc_percpu(struct lruvec_stat); |
4356 | if (!pn->lruvec_stat_cpu) { | |
00f3ca2c JW |
4357 | kfree(pn); |
4358 | return 1; | |
4359 | } | |
4360 | ||
ef8f2327 MG |
4361 | lruvec_init(&pn->lruvec); |
4362 | pn->usage_in_excess = 0; | |
4363 | pn->on_tree = false; | |
4364 | pn->memcg = memcg; | |
4365 | ||
54f72fe0 | 4366 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4367 | return 0; |
4368 | } | |
4369 | ||
ef8f2327 | 4370 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4371 | { |
00f3ca2c JW |
4372 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
4373 | ||
4eaf431f MH |
4374 | if (!pn) |
4375 | return; | |
4376 | ||
a983b5eb | 4377 | free_percpu(pn->lruvec_stat_cpu); |
00f3ca2c | 4378 | kfree(pn); |
1ecaab2b KH |
4379 | } |
4380 | ||
40e952f9 | 4381 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4382 | { |
c8b2a36f | 4383 | int node; |
59927fb9 | 4384 | |
c8b2a36f | 4385 | for_each_node(node) |
ef8f2327 | 4386 | free_mem_cgroup_per_node_info(memcg, node); |
a983b5eb | 4387 | free_percpu(memcg->stat_cpu); |
8ff69e2c | 4388 | kfree(memcg); |
59927fb9 | 4389 | } |
3afe36b1 | 4390 | |
40e952f9 TE |
4391 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
4392 | { | |
4393 | memcg_wb_domain_exit(memcg); | |
4394 | __mem_cgroup_free(memcg); | |
4395 | } | |
4396 | ||
0b8f73e1 | 4397 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4398 | { |
d142e3e6 | 4399 | struct mem_cgroup *memcg; |
0b8f73e1 | 4400 | size_t size; |
6d12e2d8 | 4401 | int node; |
8cdea7c0 | 4402 | |
0b8f73e1 JW |
4403 | size = sizeof(struct mem_cgroup); |
4404 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4405 | ||
4406 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4407 | if (!memcg) |
0b8f73e1 JW |
4408 | return NULL; |
4409 | ||
73f576c0 JW |
4410 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
4411 | 1, MEM_CGROUP_ID_MAX, | |
4412 | GFP_KERNEL); | |
4413 | if (memcg->id.id < 0) | |
4414 | goto fail; | |
4415 | ||
a983b5eb JW |
4416 | memcg->stat_cpu = alloc_percpu(struct mem_cgroup_stat_cpu); |
4417 | if (!memcg->stat_cpu) | |
0b8f73e1 | 4418 | goto fail; |
78fb7466 | 4419 | |
3ed28fa1 | 4420 | for_each_node(node) |
ef8f2327 | 4421 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 4422 | goto fail; |
f64c3f54 | 4423 | |
0b8f73e1 JW |
4424 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4425 | goto fail; | |
28dbc4b6 | 4426 | |
f7e1cb6e | 4427 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4428 | memcg->last_scanned_node = MAX_NUMNODES; |
4429 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4430 | mutex_init(&memcg->thresholds_lock); |
4431 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4432 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4433 | INIT_LIST_HEAD(&memcg->event_list); |
4434 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4435 | memcg->socket_pressure = jiffies; |
84c07d11 | 4436 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 4437 | memcg->kmemcg_id = -1; |
900a38f0 | 4438 | #endif |
52ebea74 TH |
4439 | #ifdef CONFIG_CGROUP_WRITEBACK |
4440 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4441 | #endif | |
73f576c0 | 4442 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
4443 | return memcg; |
4444 | fail: | |
7e97de0b | 4445 | mem_cgroup_id_remove(memcg); |
40e952f9 | 4446 | __mem_cgroup_free(memcg); |
0b8f73e1 | 4447 | return NULL; |
d142e3e6 GC |
4448 | } |
4449 | ||
0b8f73e1 JW |
4450 | static struct cgroup_subsys_state * __ref |
4451 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4452 | { |
0b8f73e1 JW |
4453 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4454 | struct mem_cgroup *memcg; | |
4455 | long error = -ENOMEM; | |
d142e3e6 | 4456 | |
0b8f73e1 JW |
4457 | memcg = mem_cgroup_alloc(); |
4458 | if (!memcg) | |
4459 | return ERR_PTR(error); | |
d142e3e6 | 4460 | |
0b8f73e1 JW |
4461 | memcg->high = PAGE_COUNTER_MAX; |
4462 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4463 | if (parent) { | |
4464 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4465 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4466 | } | |
4467 | if (parent && parent->use_hierarchy) { | |
4468 | memcg->use_hierarchy = true; | |
3e32cb2e | 4469 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4470 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4471 | page_counter_init(&memcg->memsw, &parent->memsw); |
4472 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4473 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4474 | } else { |
3e32cb2e | 4475 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4476 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4477 | page_counter_init(&memcg->memsw, NULL); |
4478 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4479 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4480 | /* |
4481 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4482 | * much sense so let cgroup subsystem know about this | |
4483 | * unfortunate state in our controller. | |
4484 | */ | |
d142e3e6 | 4485 | if (parent != root_mem_cgroup) |
073219e9 | 4486 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4487 | } |
d6441637 | 4488 | |
0b8f73e1 JW |
4489 | /* The following stuff does not apply to the root */ |
4490 | if (!parent) { | |
4491 | root_mem_cgroup = memcg; | |
4492 | return &memcg->css; | |
4493 | } | |
4494 | ||
b313aeee | 4495 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
4496 | if (error) |
4497 | goto fail; | |
127424c8 | 4498 | |
f7e1cb6e | 4499 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4500 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4501 | |
0b8f73e1 JW |
4502 | return &memcg->css; |
4503 | fail: | |
7e97de0b | 4504 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 4505 | mem_cgroup_free(memcg); |
ea3a9645 | 4506 | return ERR_PTR(-ENOMEM); |
0b8f73e1 JW |
4507 | } |
4508 | ||
73f576c0 | 4509 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 4510 | { |
58fa2a55 VD |
4511 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
4512 | ||
0a4465d3 KT |
4513 | /* |
4514 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
4515 | * by the time the maps are allocated. So, we allocate maps | |
4516 | * here, when for_each_mem_cgroup() can't skip it. | |
4517 | */ | |
4518 | if (memcg_alloc_shrinker_maps(memcg)) { | |
4519 | mem_cgroup_id_remove(memcg); | |
4520 | return -ENOMEM; | |
4521 | } | |
4522 | ||
73f576c0 | 4523 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 4524 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 4525 | css_get(css); |
2f7dd7a4 | 4526 | return 0; |
8cdea7c0 BS |
4527 | } |
4528 | ||
eb95419b | 4529 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4530 | { |
eb95419b | 4531 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4532 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4533 | |
4534 | /* | |
4535 | * Unregister events and notify userspace. | |
4536 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4537 | * directory to avoid race between userspace and kernelspace. | |
4538 | */ | |
fba94807 TH |
4539 | spin_lock(&memcg->event_list_lock); |
4540 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4541 | list_del_init(&event->list); |
4542 | schedule_work(&event->remove); | |
4543 | } | |
fba94807 | 4544 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4545 | |
bf8d5d52 | 4546 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4547 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 4548 | |
567e9ab2 | 4549 | memcg_offline_kmem(memcg); |
52ebea74 | 4550 | wb_memcg_offline(memcg); |
73f576c0 | 4551 | |
591edfb1 RG |
4552 | drain_all_stock(memcg); |
4553 | ||
73f576c0 | 4554 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
4555 | } |
4556 | ||
6df38689 VD |
4557 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4558 | { | |
4559 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4560 | ||
4561 | invalidate_reclaim_iterators(memcg); | |
4562 | } | |
4563 | ||
eb95419b | 4564 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4565 | { |
eb95419b | 4566 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4567 | |
f7e1cb6e | 4568 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4569 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4570 | |
0db15298 | 4571 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4572 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4573 | |
0b8f73e1 JW |
4574 | vmpressure_cleanup(&memcg->vmpressure); |
4575 | cancel_work_sync(&memcg->high_work); | |
4576 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 4577 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 4578 | memcg_free_kmem(memcg); |
0b8f73e1 | 4579 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4580 | } |
4581 | ||
1ced953b TH |
4582 | /** |
4583 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4584 | * @css: the target css | |
4585 | * | |
4586 | * Reset the states of the mem_cgroup associated with @css. This is | |
4587 | * invoked when the userland requests disabling on the default hierarchy | |
4588 | * but the memcg is pinned through dependency. The memcg should stop | |
4589 | * applying policies and should revert to the vanilla state as it may be | |
4590 | * made visible again. | |
4591 | * | |
4592 | * The current implementation only resets the essential configurations. | |
4593 | * This needs to be expanded to cover all the visible parts. | |
4594 | */ | |
4595 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4596 | { | |
4597 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4598 | ||
bbec2e15 RG |
4599 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
4600 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
4601 | page_counter_set_max(&memcg->memsw, PAGE_COUNTER_MAX); | |
4602 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); | |
4603 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 4604 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4605 | page_counter_set_low(&memcg->memory, 0); |
241994ed | 4606 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4607 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4608 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4609 | } |
4610 | ||
02491447 | 4611 | #ifdef CONFIG_MMU |
7dc74be0 | 4612 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4613 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4614 | { |
05b84301 | 4615 | int ret; |
9476db97 | 4616 | |
d0164adc MG |
4617 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4618 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4619 | if (!ret) { |
854ffa8d | 4620 | mc.precharge += count; |
854ffa8d DN |
4621 | return ret; |
4622 | } | |
9476db97 | 4623 | |
3674534b | 4624 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 4625 | while (count--) { |
3674534b | 4626 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 4627 | if (ret) |
38c5d72f | 4628 | return ret; |
854ffa8d | 4629 | mc.precharge++; |
9476db97 | 4630 | cond_resched(); |
854ffa8d | 4631 | } |
9476db97 | 4632 | return 0; |
4ffef5fe DN |
4633 | } |
4634 | ||
4ffef5fe DN |
4635 | union mc_target { |
4636 | struct page *page; | |
02491447 | 4637 | swp_entry_t ent; |
4ffef5fe DN |
4638 | }; |
4639 | ||
4ffef5fe | 4640 | enum mc_target_type { |
8d32ff84 | 4641 | MC_TARGET_NONE = 0, |
4ffef5fe | 4642 | MC_TARGET_PAGE, |
02491447 | 4643 | MC_TARGET_SWAP, |
c733a828 | 4644 | MC_TARGET_DEVICE, |
4ffef5fe DN |
4645 | }; |
4646 | ||
90254a65 DN |
4647 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4648 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4649 | { |
c733a828 | 4650 | struct page *page = _vm_normal_page(vma, addr, ptent, true); |
4ffef5fe | 4651 | |
90254a65 DN |
4652 | if (!page || !page_mapped(page)) |
4653 | return NULL; | |
4654 | if (PageAnon(page)) { | |
1dfab5ab | 4655 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4656 | return NULL; |
1dfab5ab JW |
4657 | } else { |
4658 | if (!(mc.flags & MOVE_FILE)) | |
4659 | return NULL; | |
4660 | } | |
90254a65 DN |
4661 | if (!get_page_unless_zero(page)) |
4662 | return NULL; | |
4663 | ||
4664 | return page; | |
4665 | } | |
4666 | ||
c733a828 | 4667 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 4668 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 4669 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 4670 | { |
90254a65 DN |
4671 | struct page *page = NULL; |
4672 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4673 | ||
1dfab5ab | 4674 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4675 | return NULL; |
c733a828 JG |
4676 | |
4677 | /* | |
4678 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
4679 | * a device and because they are not accessible by CPU they are store | |
4680 | * as special swap entry in the CPU page table. | |
4681 | */ | |
4682 | if (is_device_private_entry(ent)) { | |
4683 | page = device_private_entry_to_page(ent); | |
4684 | /* | |
4685 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
4686 | * a refcount of 1 when free (unlike normal page) | |
4687 | */ | |
4688 | if (!page_ref_add_unless(page, 1, 1)) | |
4689 | return NULL; | |
4690 | return page; | |
4691 | } | |
4692 | ||
4b91355e KH |
4693 | /* |
4694 | * Because lookup_swap_cache() updates some statistics counter, | |
4695 | * we call find_get_page() with swapper_space directly. | |
4696 | */ | |
f6ab1f7f | 4697 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
7941d214 | 4698 | if (do_memsw_account()) |
90254a65 DN |
4699 | entry->val = ent.val; |
4700 | ||
4701 | return page; | |
4702 | } | |
4b91355e KH |
4703 | #else |
4704 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 4705 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
4706 | { |
4707 | return NULL; | |
4708 | } | |
4709 | #endif | |
90254a65 | 4710 | |
87946a72 DN |
4711 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4712 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4713 | { | |
4714 | struct page *page = NULL; | |
87946a72 DN |
4715 | struct address_space *mapping; |
4716 | pgoff_t pgoff; | |
4717 | ||
4718 | if (!vma->vm_file) /* anonymous vma */ | |
4719 | return NULL; | |
1dfab5ab | 4720 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4721 | return NULL; |
4722 | ||
87946a72 | 4723 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4724 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4725 | |
4726 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4727 | #ifdef CONFIG_SWAP |
4728 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4729 | if (shmem_mapping(mapping)) { |
4730 | page = find_get_entry(mapping, pgoff); | |
3159f943 | 4731 | if (xa_is_value(page)) { |
139b6a6f | 4732 | swp_entry_t swp = radix_to_swp_entry(page); |
7941d214 | 4733 | if (do_memsw_account()) |
139b6a6f | 4734 | *entry = swp; |
f6ab1f7f HY |
4735 | page = find_get_page(swap_address_space(swp), |
4736 | swp_offset(swp)); | |
139b6a6f JW |
4737 | } |
4738 | } else | |
4739 | page = find_get_page(mapping, pgoff); | |
4740 | #else | |
4741 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4742 | #endif |
87946a72 DN |
4743 | return page; |
4744 | } | |
4745 | ||
b1b0deab CG |
4746 | /** |
4747 | * mem_cgroup_move_account - move account of the page | |
4748 | * @page: the page | |
25843c2b | 4749 | * @compound: charge the page as compound or small page |
b1b0deab CG |
4750 | * @from: mem_cgroup which the page is moved from. |
4751 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4752 | * | |
3ac808fd | 4753 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
4754 | * |
4755 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4756 | * from old cgroup. | |
4757 | */ | |
4758 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 4759 | bool compound, |
b1b0deab CG |
4760 | struct mem_cgroup *from, |
4761 | struct mem_cgroup *to) | |
4762 | { | |
4763 | unsigned long flags; | |
f627c2f5 | 4764 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 4765 | int ret; |
c4843a75 | 4766 | bool anon; |
b1b0deab CG |
4767 | |
4768 | VM_BUG_ON(from == to); | |
4769 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 4770 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
4771 | |
4772 | /* | |
6a93ca8f | 4773 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 4774 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 4775 | */ |
f627c2f5 | 4776 | ret = -EBUSY; |
b1b0deab CG |
4777 | if (!trylock_page(page)) |
4778 | goto out; | |
4779 | ||
4780 | ret = -EINVAL; | |
4781 | if (page->mem_cgroup != from) | |
4782 | goto out_unlock; | |
4783 | ||
c4843a75 GT |
4784 | anon = PageAnon(page); |
4785 | ||
b1b0deab CG |
4786 | spin_lock_irqsave(&from->move_lock, flags); |
4787 | ||
c4843a75 | 4788 | if (!anon && page_mapped(page)) { |
c9019e9b JW |
4789 | __mod_memcg_state(from, NR_FILE_MAPPED, -nr_pages); |
4790 | __mod_memcg_state(to, NR_FILE_MAPPED, nr_pages); | |
b1b0deab CG |
4791 | } |
4792 | ||
c4843a75 GT |
4793 | /* |
4794 | * move_lock grabbed above and caller set from->moving_account, so | |
ccda7f43 | 4795 | * mod_memcg_page_state will serialize updates to PageDirty. |
c4843a75 GT |
4796 | * So mapping should be stable for dirty pages. |
4797 | */ | |
4798 | if (!anon && PageDirty(page)) { | |
4799 | struct address_space *mapping = page_mapping(page); | |
4800 | ||
4801 | if (mapping_cap_account_dirty(mapping)) { | |
c9019e9b JW |
4802 | __mod_memcg_state(from, NR_FILE_DIRTY, -nr_pages); |
4803 | __mod_memcg_state(to, NR_FILE_DIRTY, nr_pages); | |
c4843a75 GT |
4804 | } |
4805 | } | |
4806 | ||
b1b0deab | 4807 | if (PageWriteback(page)) { |
c9019e9b JW |
4808 | __mod_memcg_state(from, NR_WRITEBACK, -nr_pages); |
4809 | __mod_memcg_state(to, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
4810 | } |
4811 | ||
4812 | /* | |
4813 | * It is safe to change page->mem_cgroup here because the page | |
4814 | * is referenced, charged, and isolated - we can't race with | |
4815 | * uncharging, charging, migration, or LRU putback. | |
4816 | */ | |
4817 | ||
4818 | /* caller should have done css_get */ | |
4819 | page->mem_cgroup = to; | |
4820 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4821 | ||
4822 | ret = 0; | |
4823 | ||
4824 | local_irq_disable(); | |
f627c2f5 | 4825 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 4826 | memcg_check_events(to, page); |
f627c2f5 | 4827 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
4828 | memcg_check_events(from, page); |
4829 | local_irq_enable(); | |
4830 | out_unlock: | |
4831 | unlock_page(page); | |
4832 | out: | |
4833 | return ret; | |
4834 | } | |
4835 | ||
7cf7806c LR |
4836 | /** |
4837 | * get_mctgt_type - get target type of moving charge | |
4838 | * @vma: the vma the pte to be checked belongs | |
4839 | * @addr: the address corresponding to the pte to be checked | |
4840 | * @ptent: the pte to be checked | |
4841 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
4842 | * | |
4843 | * Returns | |
4844 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4845 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4846 | * move charge. if @target is not NULL, the page is stored in target->page | |
4847 | * with extra refcnt got(Callers should handle it). | |
4848 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
4849 | * target for charge migration. if @target is not NULL, the entry is stored | |
4850 | * in target->ent. | |
df6ad698 JG |
4851 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PUBLIC |
4852 | * or MEMORY_DEVICE_PRIVATE (so ZONE_DEVICE page and thus not on the lru). | |
4853 | * For now we such page is charge like a regular page would be as for all | |
4854 | * intent and purposes it is just special memory taking the place of a | |
4855 | * regular page. | |
c733a828 JG |
4856 | * |
4857 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
4858 | * |
4859 | * Called with pte lock held. | |
4860 | */ | |
4861 | ||
8d32ff84 | 4862 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4863 | unsigned long addr, pte_t ptent, union mc_target *target) |
4864 | { | |
4865 | struct page *page = NULL; | |
8d32ff84 | 4866 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4867 | swp_entry_t ent = { .val = 0 }; |
4868 | ||
4869 | if (pte_present(ptent)) | |
4870 | page = mc_handle_present_pte(vma, addr, ptent); | |
4871 | else if (is_swap_pte(ptent)) | |
48406ef8 | 4872 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 4873 | else if (pte_none(ptent)) |
87946a72 | 4874 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4875 | |
4876 | if (!page && !ent.val) | |
8d32ff84 | 4877 | return ret; |
02491447 | 4878 | if (page) { |
02491447 | 4879 | /* |
0a31bc97 | 4880 | * Do only loose check w/o serialization. |
1306a85a | 4881 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4882 | * not under LRU exclusion. |
02491447 | 4883 | */ |
1306a85a | 4884 | if (page->mem_cgroup == mc.from) { |
02491447 | 4885 | ret = MC_TARGET_PAGE; |
df6ad698 JG |
4886 | if (is_device_private_page(page) || |
4887 | is_device_public_page(page)) | |
c733a828 | 4888 | ret = MC_TARGET_DEVICE; |
02491447 DN |
4889 | if (target) |
4890 | target->page = page; | |
4891 | } | |
4892 | if (!ret || !target) | |
4893 | put_page(page); | |
4894 | } | |
3e14a57b HY |
4895 | /* |
4896 | * There is a swap entry and a page doesn't exist or isn't charged. | |
4897 | * But we cannot move a tail-page in a THP. | |
4898 | */ | |
4899 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 4900 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4901 | ret = MC_TARGET_SWAP; |
4902 | if (target) | |
4903 | target->ent = ent; | |
4ffef5fe | 4904 | } |
4ffef5fe DN |
4905 | return ret; |
4906 | } | |
4907 | ||
12724850 NH |
4908 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4909 | /* | |
d6810d73 HY |
4910 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
4911 | * not support them for now. | |
12724850 NH |
4912 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
4913 | */ | |
4914 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4915 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4916 | { | |
4917 | struct page *page = NULL; | |
12724850 NH |
4918 | enum mc_target_type ret = MC_TARGET_NONE; |
4919 | ||
84c3fc4e ZY |
4920 | if (unlikely(is_swap_pmd(pmd))) { |
4921 | VM_BUG_ON(thp_migration_supported() && | |
4922 | !is_pmd_migration_entry(pmd)); | |
4923 | return ret; | |
4924 | } | |
12724850 | 4925 | page = pmd_page(pmd); |
309381fe | 4926 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4927 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4928 | return ret; |
1306a85a | 4929 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4930 | ret = MC_TARGET_PAGE; |
4931 | if (target) { | |
4932 | get_page(page); | |
4933 | target->page = page; | |
4934 | } | |
4935 | } | |
4936 | return ret; | |
4937 | } | |
4938 | #else | |
4939 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4940 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4941 | { | |
4942 | return MC_TARGET_NONE; | |
4943 | } | |
4944 | #endif | |
4945 | ||
4ffef5fe DN |
4946 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4947 | unsigned long addr, unsigned long end, | |
4948 | struct mm_walk *walk) | |
4949 | { | |
26bcd64a | 4950 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4951 | pte_t *pte; |
4952 | spinlock_t *ptl; | |
4953 | ||
b6ec57f4 KS |
4954 | ptl = pmd_trans_huge_lock(pmd, vma); |
4955 | if (ptl) { | |
c733a828 JG |
4956 | /* |
4957 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
4958 | * support transparent huge page with MEMORY_DEVICE_PUBLIC or | |
4959 | * MEMORY_DEVICE_PRIVATE but this might change. | |
4960 | */ | |
12724850 NH |
4961 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4962 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4963 | spin_unlock(ptl); |
1a5a9906 | 4964 | return 0; |
12724850 | 4965 | } |
03319327 | 4966 | |
45f83cef AA |
4967 | if (pmd_trans_unstable(pmd)) |
4968 | return 0; | |
4ffef5fe DN |
4969 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4970 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4971 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4972 | mc.precharge++; /* increment precharge temporarily */ |
4973 | pte_unmap_unlock(pte - 1, ptl); | |
4974 | cond_resched(); | |
4975 | ||
7dc74be0 DN |
4976 | return 0; |
4977 | } | |
4978 | ||
4ffef5fe DN |
4979 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4980 | { | |
4981 | unsigned long precharge; | |
4ffef5fe | 4982 | |
26bcd64a NH |
4983 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4984 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4985 | .mm = mm, | |
4986 | }; | |
dfe076b0 | 4987 | down_read(&mm->mmap_sem); |
0247f3f4 JM |
4988 | walk_page_range(0, mm->highest_vm_end, |
4989 | &mem_cgroup_count_precharge_walk); | |
dfe076b0 | 4990 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4991 | |
4992 | precharge = mc.precharge; | |
4993 | mc.precharge = 0; | |
4994 | ||
4995 | return precharge; | |
4996 | } | |
4997 | ||
4ffef5fe DN |
4998 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4999 | { | |
dfe076b0 DN |
5000 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5001 | ||
5002 | VM_BUG_ON(mc.moving_task); | |
5003 | mc.moving_task = current; | |
5004 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5005 | } |
5006 | ||
dfe076b0 DN |
5007 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5008 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5009 | { |
2bd9bb20 KH |
5010 | struct mem_cgroup *from = mc.from; |
5011 | struct mem_cgroup *to = mc.to; | |
5012 | ||
4ffef5fe | 5013 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5014 | if (mc.precharge) { |
00501b53 | 5015 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5016 | mc.precharge = 0; |
5017 | } | |
5018 | /* | |
5019 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5020 | * we must uncharge here. | |
5021 | */ | |
5022 | if (mc.moved_charge) { | |
00501b53 | 5023 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5024 | mc.moved_charge = 0; |
4ffef5fe | 5025 | } |
483c30b5 DN |
5026 | /* we must fixup refcnts and charges */ |
5027 | if (mc.moved_swap) { | |
483c30b5 | 5028 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5029 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5030 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5031 | |
615d66c3 VD |
5032 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5033 | ||
05b84301 | 5034 | /* |
3e32cb2e JW |
5035 | * we charged both to->memory and to->memsw, so we |
5036 | * should uncharge to->memory. | |
05b84301 | 5037 | */ |
ce00a967 | 5038 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5039 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5040 | ||
615d66c3 VD |
5041 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
5042 | css_put_many(&mc.to->css, mc.moved_swap); | |
3e32cb2e | 5043 | |
483c30b5 DN |
5044 | mc.moved_swap = 0; |
5045 | } | |
dfe076b0 DN |
5046 | memcg_oom_recover(from); |
5047 | memcg_oom_recover(to); | |
5048 | wake_up_all(&mc.waitq); | |
5049 | } | |
5050 | ||
5051 | static void mem_cgroup_clear_mc(void) | |
5052 | { | |
264a0ae1 TH |
5053 | struct mm_struct *mm = mc.mm; |
5054 | ||
dfe076b0 DN |
5055 | /* |
5056 | * we must clear moving_task before waking up waiters at the end of | |
5057 | * task migration. | |
5058 | */ | |
5059 | mc.moving_task = NULL; | |
5060 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5061 | spin_lock(&mc.lock); |
4ffef5fe DN |
5062 | mc.from = NULL; |
5063 | mc.to = NULL; | |
264a0ae1 | 5064 | mc.mm = NULL; |
2bd9bb20 | 5065 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5066 | |
5067 | mmput(mm); | |
4ffef5fe DN |
5068 | } |
5069 | ||
1f7dd3e5 | 5070 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5071 | { |
1f7dd3e5 | 5072 | struct cgroup_subsys_state *css; |
eed67d75 | 5073 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5074 | struct mem_cgroup *from; |
4530eddb | 5075 | struct task_struct *leader, *p; |
9f2115f9 | 5076 | struct mm_struct *mm; |
1dfab5ab | 5077 | unsigned long move_flags; |
9f2115f9 | 5078 | int ret = 0; |
7dc74be0 | 5079 | |
1f7dd3e5 TH |
5080 | /* charge immigration isn't supported on the default hierarchy */ |
5081 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5082 | return 0; |
5083 | ||
4530eddb TH |
5084 | /* |
5085 | * Multi-process migrations only happen on the default hierarchy | |
5086 | * where charge immigration is not used. Perform charge | |
5087 | * immigration if @tset contains a leader and whine if there are | |
5088 | * multiple. | |
5089 | */ | |
5090 | p = NULL; | |
1f7dd3e5 | 5091 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5092 | WARN_ON_ONCE(p); |
5093 | p = leader; | |
1f7dd3e5 | 5094 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5095 | } |
5096 | if (!p) | |
5097 | return 0; | |
5098 | ||
1f7dd3e5 TH |
5099 | /* |
5100 | * We are now commited to this value whatever it is. Changes in this | |
5101 | * tunable will only affect upcoming migrations, not the current one. | |
5102 | * So we need to save it, and keep it going. | |
5103 | */ | |
5104 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5105 | if (!move_flags) | |
5106 | return 0; | |
5107 | ||
9f2115f9 TH |
5108 | from = mem_cgroup_from_task(p); |
5109 | ||
5110 | VM_BUG_ON(from == memcg); | |
5111 | ||
5112 | mm = get_task_mm(p); | |
5113 | if (!mm) | |
5114 | return 0; | |
5115 | /* We move charges only when we move a owner of the mm */ | |
5116 | if (mm->owner == p) { | |
5117 | VM_BUG_ON(mc.from); | |
5118 | VM_BUG_ON(mc.to); | |
5119 | VM_BUG_ON(mc.precharge); | |
5120 | VM_BUG_ON(mc.moved_charge); | |
5121 | VM_BUG_ON(mc.moved_swap); | |
5122 | ||
5123 | spin_lock(&mc.lock); | |
264a0ae1 | 5124 | mc.mm = mm; |
9f2115f9 TH |
5125 | mc.from = from; |
5126 | mc.to = memcg; | |
5127 | mc.flags = move_flags; | |
5128 | spin_unlock(&mc.lock); | |
5129 | /* We set mc.moving_task later */ | |
5130 | ||
5131 | ret = mem_cgroup_precharge_mc(mm); | |
5132 | if (ret) | |
5133 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5134 | } else { |
5135 | mmput(mm); | |
7dc74be0 DN |
5136 | } |
5137 | return ret; | |
5138 | } | |
5139 | ||
1f7dd3e5 | 5140 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5141 | { |
4e2f245d JW |
5142 | if (mc.to) |
5143 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5144 | } |
5145 | ||
4ffef5fe DN |
5146 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5147 | unsigned long addr, unsigned long end, | |
5148 | struct mm_walk *walk) | |
7dc74be0 | 5149 | { |
4ffef5fe | 5150 | int ret = 0; |
26bcd64a | 5151 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5152 | pte_t *pte; |
5153 | spinlock_t *ptl; | |
12724850 NH |
5154 | enum mc_target_type target_type; |
5155 | union mc_target target; | |
5156 | struct page *page; | |
4ffef5fe | 5157 | |
b6ec57f4 KS |
5158 | ptl = pmd_trans_huge_lock(pmd, vma); |
5159 | if (ptl) { | |
62ade86a | 5160 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5161 | spin_unlock(ptl); |
12724850 NH |
5162 | return 0; |
5163 | } | |
5164 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5165 | if (target_type == MC_TARGET_PAGE) { | |
5166 | page = target.page; | |
5167 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5168 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5169 | mc.from, mc.to)) { |
12724850 NH |
5170 | mc.precharge -= HPAGE_PMD_NR; |
5171 | mc.moved_charge += HPAGE_PMD_NR; | |
5172 | } | |
5173 | putback_lru_page(page); | |
5174 | } | |
5175 | put_page(page); | |
c733a828 JG |
5176 | } else if (target_type == MC_TARGET_DEVICE) { |
5177 | page = target.page; | |
5178 | if (!mem_cgroup_move_account(page, true, | |
5179 | mc.from, mc.to)) { | |
5180 | mc.precharge -= HPAGE_PMD_NR; | |
5181 | mc.moved_charge += HPAGE_PMD_NR; | |
5182 | } | |
5183 | put_page(page); | |
12724850 | 5184 | } |
bf929152 | 5185 | spin_unlock(ptl); |
1a5a9906 | 5186 | return 0; |
12724850 NH |
5187 | } |
5188 | ||
45f83cef AA |
5189 | if (pmd_trans_unstable(pmd)) |
5190 | return 0; | |
4ffef5fe DN |
5191 | retry: |
5192 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5193 | for (; addr != end; addr += PAGE_SIZE) { | |
5194 | pte_t ptent = *(pte++); | |
c733a828 | 5195 | bool device = false; |
02491447 | 5196 | swp_entry_t ent; |
4ffef5fe DN |
5197 | |
5198 | if (!mc.precharge) | |
5199 | break; | |
5200 | ||
8d32ff84 | 5201 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
5202 | case MC_TARGET_DEVICE: |
5203 | device = true; | |
5204 | /* fall through */ | |
4ffef5fe DN |
5205 | case MC_TARGET_PAGE: |
5206 | page = target.page; | |
53f9263b KS |
5207 | /* |
5208 | * We can have a part of the split pmd here. Moving it | |
5209 | * can be done but it would be too convoluted so simply | |
5210 | * ignore such a partial THP and keep it in original | |
5211 | * memcg. There should be somebody mapping the head. | |
5212 | */ | |
5213 | if (PageTransCompound(page)) | |
5214 | goto put; | |
c733a828 | 5215 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 5216 | goto put; |
f627c2f5 KS |
5217 | if (!mem_cgroup_move_account(page, false, |
5218 | mc.from, mc.to)) { | |
4ffef5fe | 5219 | mc.precharge--; |
854ffa8d DN |
5220 | /* we uncharge from mc.from later. */ |
5221 | mc.moved_charge++; | |
4ffef5fe | 5222 | } |
c733a828 JG |
5223 | if (!device) |
5224 | putback_lru_page(page); | |
8d32ff84 | 5225 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5226 | put_page(page); |
5227 | break; | |
02491447 DN |
5228 | case MC_TARGET_SWAP: |
5229 | ent = target.ent; | |
e91cbb42 | 5230 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5231 | mc.precharge--; |
483c30b5 DN |
5232 | /* we fixup refcnts and charges later. */ |
5233 | mc.moved_swap++; | |
5234 | } | |
02491447 | 5235 | break; |
4ffef5fe DN |
5236 | default: |
5237 | break; | |
5238 | } | |
5239 | } | |
5240 | pte_unmap_unlock(pte - 1, ptl); | |
5241 | cond_resched(); | |
5242 | ||
5243 | if (addr != end) { | |
5244 | /* | |
5245 | * We have consumed all precharges we got in can_attach(). | |
5246 | * We try charge one by one, but don't do any additional | |
5247 | * charges to mc.to if we have failed in charge once in attach() | |
5248 | * phase. | |
5249 | */ | |
854ffa8d | 5250 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5251 | if (!ret) |
5252 | goto retry; | |
5253 | } | |
5254 | ||
5255 | return ret; | |
5256 | } | |
5257 | ||
264a0ae1 | 5258 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 5259 | { |
26bcd64a NH |
5260 | struct mm_walk mem_cgroup_move_charge_walk = { |
5261 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
264a0ae1 | 5262 | .mm = mc.mm, |
26bcd64a | 5263 | }; |
4ffef5fe DN |
5264 | |
5265 | lru_add_drain_all(); | |
312722cb | 5266 | /* |
81f8c3a4 JW |
5267 | * Signal lock_page_memcg() to take the memcg's move_lock |
5268 | * while we're moving its pages to another memcg. Then wait | |
5269 | * for already started RCU-only updates to finish. | |
312722cb JW |
5270 | */ |
5271 | atomic_inc(&mc.from->moving_account); | |
5272 | synchronize_rcu(); | |
dfe076b0 | 5273 | retry: |
264a0ae1 | 5274 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
5275 | /* |
5276 | * Someone who are holding the mmap_sem might be waiting in | |
5277 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5278 | * and retry. Because we cancel precharges, we might not be able | |
5279 | * to move enough charges, but moving charge is a best-effort | |
5280 | * feature anyway, so it wouldn't be a big problem. | |
5281 | */ | |
5282 | __mem_cgroup_clear_mc(); | |
5283 | cond_resched(); | |
5284 | goto retry; | |
5285 | } | |
26bcd64a NH |
5286 | /* |
5287 | * When we have consumed all precharges and failed in doing | |
5288 | * additional charge, the page walk just aborts. | |
5289 | */ | |
0247f3f4 JM |
5290 | walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk); |
5291 | ||
264a0ae1 | 5292 | up_read(&mc.mm->mmap_sem); |
312722cb | 5293 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5294 | } |
5295 | ||
264a0ae1 | 5296 | static void mem_cgroup_move_task(void) |
67e465a7 | 5297 | { |
264a0ae1 TH |
5298 | if (mc.to) { |
5299 | mem_cgroup_move_charge(); | |
a433658c | 5300 | mem_cgroup_clear_mc(); |
264a0ae1 | 5301 | } |
67e465a7 | 5302 | } |
5cfb80a7 | 5303 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5304 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5305 | { |
5306 | return 0; | |
5307 | } | |
1f7dd3e5 | 5308 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5309 | { |
5310 | } | |
264a0ae1 | 5311 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
5312 | { |
5313 | } | |
5314 | #endif | |
67e465a7 | 5315 | |
f00baae7 TH |
5316 | /* |
5317 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5318 | * to verify whether we're attached to the default hierarchy on each mount |
5319 | * attempt. | |
f00baae7 | 5320 | */ |
eb95419b | 5321 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5322 | { |
5323 | /* | |
aa6ec29b | 5324 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5325 | * guarantees that @root doesn't have any children, so turning it |
5326 | * on for the root memcg is enough. | |
5327 | */ | |
9e10a130 | 5328 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5329 | root_mem_cgroup->use_hierarchy = true; |
5330 | else | |
5331 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5332 | } |
5333 | ||
241994ed JW |
5334 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5335 | struct cftype *cft) | |
5336 | { | |
f5fc3c5d JW |
5337 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5338 | ||
5339 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5340 | } |
5341 | ||
bf8d5d52 RG |
5342 | static int memory_min_show(struct seq_file *m, void *v) |
5343 | { | |
5344 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5345 | unsigned long min = READ_ONCE(memcg->memory.min); | |
5346 | ||
5347 | if (min == PAGE_COUNTER_MAX) | |
5348 | seq_puts(m, "max\n"); | |
5349 | else | |
5350 | seq_printf(m, "%llu\n", (u64)min * PAGE_SIZE); | |
5351 | ||
5352 | return 0; | |
5353 | } | |
5354 | ||
5355 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
5356 | char *buf, size_t nbytes, loff_t off) | |
5357 | { | |
5358 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5359 | unsigned long min; | |
5360 | int err; | |
5361 | ||
5362 | buf = strstrip(buf); | |
5363 | err = page_counter_memparse(buf, "max", &min); | |
5364 | if (err) | |
5365 | return err; | |
5366 | ||
5367 | page_counter_set_min(&memcg->memory, min); | |
5368 | ||
5369 | return nbytes; | |
5370 | } | |
5371 | ||
241994ed JW |
5372 | static int memory_low_show(struct seq_file *m, void *v) |
5373 | { | |
5374 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
23067153 | 5375 | unsigned long low = READ_ONCE(memcg->memory.low); |
241994ed JW |
5376 | |
5377 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 5378 | seq_puts(m, "max\n"); |
241994ed JW |
5379 | else |
5380 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5381 | ||
5382 | return 0; | |
5383 | } | |
5384 | ||
5385 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5386 | char *buf, size_t nbytes, loff_t off) | |
5387 | { | |
5388 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5389 | unsigned long low; | |
5390 | int err; | |
5391 | ||
5392 | buf = strstrip(buf); | |
d2973697 | 5393 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5394 | if (err) |
5395 | return err; | |
5396 | ||
23067153 | 5397 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
5398 | |
5399 | return nbytes; | |
5400 | } | |
5401 | ||
5402 | static int memory_high_show(struct seq_file *m, void *v) | |
5403 | { | |
5404 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5405 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
5406 | |
5407 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5408 | seq_puts(m, "max\n"); |
241994ed JW |
5409 | else |
5410 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5411 | ||
5412 | return 0; | |
5413 | } | |
5414 | ||
5415 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5416 | char *buf, size_t nbytes, loff_t off) | |
5417 | { | |
5418 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 5419 | unsigned long nr_pages; |
241994ed JW |
5420 | unsigned long high; |
5421 | int err; | |
5422 | ||
5423 | buf = strstrip(buf); | |
d2973697 | 5424 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5425 | if (err) |
5426 | return err; | |
5427 | ||
5428 | memcg->high = high; | |
5429 | ||
588083bb JW |
5430 | nr_pages = page_counter_read(&memcg->memory); |
5431 | if (nr_pages > high) | |
5432 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5433 | GFP_KERNEL, true); | |
5434 | ||
2529bb3a | 5435 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5436 | return nbytes; |
5437 | } | |
5438 | ||
5439 | static int memory_max_show(struct seq_file *m, void *v) | |
5440 | { | |
5441 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
bbec2e15 | 5442 | unsigned long max = READ_ONCE(memcg->memory.max); |
241994ed JW |
5443 | |
5444 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5445 | seq_puts(m, "max\n"); |
241994ed JW |
5446 | else |
5447 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5448 | ||
5449 | return 0; | |
5450 | } | |
5451 | ||
5452 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5453 | char *buf, size_t nbytes, loff_t off) | |
5454 | { | |
5455 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
5456 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5457 | bool drained = false; | |
241994ed JW |
5458 | unsigned long max; |
5459 | int err; | |
5460 | ||
5461 | buf = strstrip(buf); | |
d2973697 | 5462 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5463 | if (err) |
5464 | return err; | |
5465 | ||
bbec2e15 | 5466 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
5467 | |
5468 | for (;;) { | |
5469 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5470 | ||
5471 | if (nr_pages <= max) | |
5472 | break; | |
5473 | ||
5474 | if (signal_pending(current)) { | |
5475 | err = -EINTR; | |
5476 | break; | |
5477 | } | |
5478 | ||
5479 | if (!drained) { | |
5480 | drain_all_stock(memcg); | |
5481 | drained = true; | |
5482 | continue; | |
5483 | } | |
5484 | ||
5485 | if (nr_reclaims) { | |
5486 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5487 | GFP_KERNEL, true)) | |
5488 | nr_reclaims--; | |
5489 | continue; | |
5490 | } | |
5491 | ||
e27be240 | 5492 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
5493 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
5494 | break; | |
5495 | } | |
241994ed | 5496 | |
2529bb3a | 5497 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5498 | return nbytes; |
5499 | } | |
5500 | ||
5501 | static int memory_events_show(struct seq_file *m, void *v) | |
5502 | { | |
5503 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5504 | ||
e27be240 JW |
5505 | seq_printf(m, "low %lu\n", |
5506 | atomic_long_read(&memcg->memory_events[MEMCG_LOW])); | |
5507 | seq_printf(m, "high %lu\n", | |
5508 | atomic_long_read(&memcg->memory_events[MEMCG_HIGH])); | |
5509 | seq_printf(m, "max %lu\n", | |
5510 | atomic_long_read(&memcg->memory_events[MEMCG_MAX])); | |
5511 | seq_printf(m, "oom %lu\n", | |
5512 | atomic_long_read(&memcg->memory_events[MEMCG_OOM])); | |
fe6bdfc8 RG |
5513 | seq_printf(m, "oom_kill %lu\n", |
5514 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
241994ed JW |
5515 | |
5516 | return 0; | |
5517 | } | |
5518 | ||
587d9f72 JW |
5519 | static int memory_stat_show(struct seq_file *m, void *v) |
5520 | { | |
5521 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
8de7ecc6 | 5522 | struct accumulated_stats acc; |
587d9f72 JW |
5523 | int i; |
5524 | ||
5525 | /* | |
5526 | * Provide statistics on the state of the memory subsystem as | |
5527 | * well as cumulative event counters that show past behavior. | |
5528 | * | |
5529 | * This list is ordered following a combination of these gradients: | |
5530 | * 1) generic big picture -> specifics and details | |
5531 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
5532 | * | |
5533 | * Current memory state: | |
5534 | */ | |
5535 | ||
8de7ecc6 SB |
5536 | memset(&acc, 0, sizeof(acc)); |
5537 | acc.stats_size = MEMCG_NR_STAT; | |
5538 | acc.events_size = NR_VM_EVENT_ITEMS; | |
5539 | accumulate_memcg_tree(memcg, &acc); | |
72b54e73 | 5540 | |
587d9f72 | 5541 | seq_printf(m, "anon %llu\n", |
8de7ecc6 | 5542 | (u64)acc.stat[MEMCG_RSS] * PAGE_SIZE); |
587d9f72 | 5543 | seq_printf(m, "file %llu\n", |
8de7ecc6 | 5544 | (u64)acc.stat[MEMCG_CACHE] * PAGE_SIZE); |
12580e4b | 5545 | seq_printf(m, "kernel_stack %llu\n", |
8de7ecc6 | 5546 | (u64)acc.stat[MEMCG_KERNEL_STACK_KB] * 1024); |
27ee57c9 | 5547 | seq_printf(m, "slab %llu\n", |
8de7ecc6 SB |
5548 | (u64)(acc.stat[NR_SLAB_RECLAIMABLE] + |
5549 | acc.stat[NR_SLAB_UNRECLAIMABLE]) * PAGE_SIZE); | |
b2807f07 | 5550 | seq_printf(m, "sock %llu\n", |
8de7ecc6 | 5551 | (u64)acc.stat[MEMCG_SOCK] * PAGE_SIZE); |
587d9f72 | 5552 | |
9a4caf1e | 5553 | seq_printf(m, "shmem %llu\n", |
8de7ecc6 | 5554 | (u64)acc.stat[NR_SHMEM] * PAGE_SIZE); |
587d9f72 | 5555 | seq_printf(m, "file_mapped %llu\n", |
8de7ecc6 | 5556 | (u64)acc.stat[NR_FILE_MAPPED] * PAGE_SIZE); |
587d9f72 | 5557 | seq_printf(m, "file_dirty %llu\n", |
8de7ecc6 | 5558 | (u64)acc.stat[NR_FILE_DIRTY] * PAGE_SIZE); |
587d9f72 | 5559 | seq_printf(m, "file_writeback %llu\n", |
8de7ecc6 | 5560 | (u64)acc.stat[NR_WRITEBACK] * PAGE_SIZE); |
587d9f72 | 5561 | |
8de7ecc6 SB |
5562 | for (i = 0; i < NR_LRU_LISTS; i++) |
5563 | seq_printf(m, "%s %llu\n", mem_cgroup_lru_names[i], | |
5564 | (u64)acc.lru_pages[i] * PAGE_SIZE); | |
587d9f72 | 5565 | |
27ee57c9 | 5566 | seq_printf(m, "slab_reclaimable %llu\n", |
8de7ecc6 | 5567 | (u64)acc.stat[NR_SLAB_RECLAIMABLE] * PAGE_SIZE); |
27ee57c9 | 5568 | seq_printf(m, "slab_unreclaimable %llu\n", |
8de7ecc6 | 5569 | (u64)acc.stat[NR_SLAB_UNRECLAIMABLE] * PAGE_SIZE); |
27ee57c9 | 5570 | |
587d9f72 JW |
5571 | /* Accumulated memory events */ |
5572 | ||
8de7ecc6 SB |
5573 | seq_printf(m, "pgfault %lu\n", acc.events[PGFAULT]); |
5574 | seq_printf(m, "pgmajfault %lu\n", acc.events[PGMAJFAULT]); | |
587d9f72 | 5575 | |
e9b257ed JW |
5576 | seq_printf(m, "workingset_refault %lu\n", |
5577 | acc.stat[WORKINGSET_REFAULT]); | |
5578 | seq_printf(m, "workingset_activate %lu\n", | |
5579 | acc.stat[WORKINGSET_ACTIVATE]); | |
5580 | seq_printf(m, "workingset_nodereclaim %lu\n", | |
5581 | acc.stat[WORKINGSET_NODERECLAIM]); | |
5582 | ||
8de7ecc6 SB |
5583 | seq_printf(m, "pgrefill %lu\n", acc.events[PGREFILL]); |
5584 | seq_printf(m, "pgscan %lu\n", acc.events[PGSCAN_KSWAPD] + | |
5585 | acc.events[PGSCAN_DIRECT]); | |
5586 | seq_printf(m, "pgsteal %lu\n", acc.events[PGSTEAL_KSWAPD] + | |
5587 | acc.events[PGSTEAL_DIRECT]); | |
5588 | seq_printf(m, "pgactivate %lu\n", acc.events[PGACTIVATE]); | |
5589 | seq_printf(m, "pgdeactivate %lu\n", acc.events[PGDEACTIVATE]); | |
5590 | seq_printf(m, "pglazyfree %lu\n", acc.events[PGLAZYFREE]); | |
5591 | seq_printf(m, "pglazyfreed %lu\n", acc.events[PGLAZYFREED]); | |
2262185c | 5592 | |
587d9f72 JW |
5593 | return 0; |
5594 | } | |
5595 | ||
3d8b38eb RG |
5596 | static int memory_oom_group_show(struct seq_file *m, void *v) |
5597 | { | |
5598 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5599 | ||
5600 | seq_printf(m, "%d\n", memcg->oom_group); | |
5601 | ||
5602 | return 0; | |
5603 | } | |
5604 | ||
5605 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
5606 | char *buf, size_t nbytes, loff_t off) | |
5607 | { | |
5608 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5609 | int ret, oom_group; | |
5610 | ||
5611 | buf = strstrip(buf); | |
5612 | if (!buf) | |
5613 | return -EINVAL; | |
5614 | ||
5615 | ret = kstrtoint(buf, 0, &oom_group); | |
5616 | if (ret) | |
5617 | return ret; | |
5618 | ||
5619 | if (oom_group != 0 && oom_group != 1) | |
5620 | return -EINVAL; | |
5621 | ||
5622 | memcg->oom_group = oom_group; | |
5623 | ||
5624 | return nbytes; | |
5625 | } | |
5626 | ||
241994ed JW |
5627 | static struct cftype memory_files[] = { |
5628 | { | |
5629 | .name = "current", | |
f5fc3c5d | 5630 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5631 | .read_u64 = memory_current_read, |
5632 | }, | |
bf8d5d52 RG |
5633 | { |
5634 | .name = "min", | |
5635 | .flags = CFTYPE_NOT_ON_ROOT, | |
5636 | .seq_show = memory_min_show, | |
5637 | .write = memory_min_write, | |
5638 | }, | |
241994ed JW |
5639 | { |
5640 | .name = "low", | |
5641 | .flags = CFTYPE_NOT_ON_ROOT, | |
5642 | .seq_show = memory_low_show, | |
5643 | .write = memory_low_write, | |
5644 | }, | |
5645 | { | |
5646 | .name = "high", | |
5647 | .flags = CFTYPE_NOT_ON_ROOT, | |
5648 | .seq_show = memory_high_show, | |
5649 | .write = memory_high_write, | |
5650 | }, | |
5651 | { | |
5652 | .name = "max", | |
5653 | .flags = CFTYPE_NOT_ON_ROOT, | |
5654 | .seq_show = memory_max_show, | |
5655 | .write = memory_max_write, | |
5656 | }, | |
5657 | { | |
5658 | .name = "events", | |
5659 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5660 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5661 | .seq_show = memory_events_show, |
5662 | }, | |
587d9f72 JW |
5663 | { |
5664 | .name = "stat", | |
5665 | .flags = CFTYPE_NOT_ON_ROOT, | |
5666 | .seq_show = memory_stat_show, | |
5667 | }, | |
3d8b38eb RG |
5668 | { |
5669 | .name = "oom.group", | |
5670 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
5671 | .seq_show = memory_oom_group_show, | |
5672 | .write = memory_oom_group_write, | |
5673 | }, | |
241994ed JW |
5674 | { } /* terminate */ |
5675 | }; | |
5676 | ||
073219e9 | 5677 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5678 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5679 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5680 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5681 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5682 | .css_free = mem_cgroup_css_free, |
1ced953b | 5683 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5684 | .can_attach = mem_cgroup_can_attach, |
5685 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 5686 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5687 | .bind = mem_cgroup_bind, |
241994ed JW |
5688 | .dfl_cftypes = memory_files, |
5689 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5690 | .early_init = 0, |
8cdea7c0 | 5691 | }; |
c077719b | 5692 | |
241994ed | 5693 | /** |
bf8d5d52 | 5694 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 5695 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
5696 | * @memcg: the memory cgroup to check |
5697 | * | |
23067153 RG |
5698 | * WARNING: This function is not stateless! It can only be used as part |
5699 | * of a top-down tree iteration, not for isolated queries. | |
34c81057 | 5700 | * |
bf8d5d52 RG |
5701 | * Returns one of the following: |
5702 | * MEMCG_PROT_NONE: cgroup memory is not protected | |
5703 | * MEMCG_PROT_LOW: cgroup memory is protected as long there is | |
5704 | * an unprotected supply of reclaimable memory from other cgroups. | |
5705 | * MEMCG_PROT_MIN: cgroup memory is protected | |
34c81057 | 5706 | * |
bf8d5d52 | 5707 | * @root is exclusive; it is never protected when looked at directly |
34c81057 | 5708 | * |
bf8d5d52 RG |
5709 | * To provide a proper hierarchical behavior, effective memory.min/low values |
5710 | * are used. Below is the description of how effective memory.low is calculated. | |
5711 | * Effective memory.min values is calculated in the same way. | |
34c81057 | 5712 | * |
23067153 RG |
5713 | * Effective memory.low is always equal or less than the original memory.low. |
5714 | * If there is no memory.low overcommittment (which is always true for | |
5715 | * top-level memory cgroups), these two values are equal. | |
5716 | * Otherwise, it's a part of parent's effective memory.low, | |
5717 | * calculated as a cgroup's memory.low usage divided by sum of sibling's | |
5718 | * memory.low usages, where memory.low usage is the size of actually | |
5719 | * protected memory. | |
34c81057 | 5720 | * |
23067153 RG |
5721 | * low_usage |
5722 | * elow = min( memory.low, parent->elow * ------------------ ), | |
5723 | * siblings_low_usage | |
34c81057 | 5724 | * |
23067153 RG |
5725 | * | memory.current, if memory.current < memory.low |
5726 | * low_usage = | | |
5727 | | 0, otherwise. | |
34c81057 | 5728 | * |
23067153 RG |
5729 | * |
5730 | * Such definition of the effective memory.low provides the expected | |
5731 | * hierarchical behavior: parent's memory.low value is limiting | |
5732 | * children, unprotected memory is reclaimed first and cgroups, | |
5733 | * which are not using their guarantee do not affect actual memory | |
5734 | * distribution. | |
5735 | * | |
5736 | * For example, if there are memcgs A, A/B, A/C, A/D and A/E: | |
5737 | * | |
5738 | * A A/memory.low = 2G, A/memory.current = 6G | |
5739 | * //\\ | |
5740 | * BC DE B/memory.low = 3G B/memory.current = 2G | |
5741 | * C/memory.low = 1G C/memory.current = 2G | |
5742 | * D/memory.low = 0 D/memory.current = 2G | |
5743 | * E/memory.low = 10G E/memory.current = 0 | |
5744 | * | |
5745 | * and the memory pressure is applied, the following memory distribution | |
5746 | * is expected (approximately): | |
5747 | * | |
5748 | * A/memory.current = 2G | |
5749 | * | |
5750 | * B/memory.current = 1.3G | |
5751 | * C/memory.current = 0.6G | |
5752 | * D/memory.current = 0 | |
5753 | * E/memory.current = 0 | |
5754 | * | |
5755 | * These calculations require constant tracking of the actual low usages | |
bf8d5d52 RG |
5756 | * (see propagate_protected_usage()), as well as recursive calculation of |
5757 | * effective memory.low values. But as we do call mem_cgroup_protected() | |
23067153 RG |
5758 | * path for each memory cgroup top-down from the reclaim, |
5759 | * it's possible to optimize this part, and save calculated elow | |
5760 | * for next usage. This part is intentionally racy, but it's ok, | |
5761 | * as memory.low is a best-effort mechanism. | |
241994ed | 5762 | */ |
bf8d5d52 RG |
5763 | enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root, |
5764 | struct mem_cgroup *memcg) | |
241994ed | 5765 | { |
23067153 | 5766 | struct mem_cgroup *parent; |
bf8d5d52 RG |
5767 | unsigned long emin, parent_emin; |
5768 | unsigned long elow, parent_elow; | |
5769 | unsigned long usage; | |
23067153 | 5770 | |
241994ed | 5771 | if (mem_cgroup_disabled()) |
bf8d5d52 | 5772 | return MEMCG_PROT_NONE; |
241994ed | 5773 | |
34c81057 SC |
5774 | if (!root) |
5775 | root = root_mem_cgroup; | |
5776 | if (memcg == root) | |
bf8d5d52 | 5777 | return MEMCG_PROT_NONE; |
241994ed | 5778 | |
23067153 | 5779 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 RG |
5780 | if (!usage) |
5781 | return MEMCG_PROT_NONE; | |
5782 | ||
5783 | emin = memcg->memory.min; | |
5784 | elow = memcg->memory.low; | |
34c81057 | 5785 | |
bf8d5d52 | 5786 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
5787 | /* No parent means a non-hierarchical mode on v1 memcg */ |
5788 | if (!parent) | |
5789 | return MEMCG_PROT_NONE; | |
5790 | ||
23067153 RG |
5791 | if (parent == root) |
5792 | goto exit; | |
5793 | ||
bf8d5d52 RG |
5794 | parent_emin = READ_ONCE(parent->memory.emin); |
5795 | emin = min(emin, parent_emin); | |
5796 | if (emin && parent_emin) { | |
5797 | unsigned long min_usage, siblings_min_usage; | |
5798 | ||
5799 | min_usage = min(usage, memcg->memory.min); | |
5800 | siblings_min_usage = atomic_long_read( | |
5801 | &parent->memory.children_min_usage); | |
5802 | ||
5803 | if (min_usage && siblings_min_usage) | |
5804 | emin = min(emin, parent_emin * min_usage / | |
5805 | siblings_min_usage); | |
5806 | } | |
5807 | ||
23067153 RG |
5808 | parent_elow = READ_ONCE(parent->memory.elow); |
5809 | elow = min(elow, parent_elow); | |
bf8d5d52 RG |
5810 | if (elow && parent_elow) { |
5811 | unsigned long low_usage, siblings_low_usage; | |
23067153 | 5812 | |
bf8d5d52 RG |
5813 | low_usage = min(usage, memcg->memory.low); |
5814 | siblings_low_usage = atomic_long_read( | |
5815 | &parent->memory.children_low_usage); | |
23067153 | 5816 | |
bf8d5d52 RG |
5817 | if (low_usage && siblings_low_usage) |
5818 | elow = min(elow, parent_elow * low_usage / | |
5819 | siblings_low_usage); | |
5820 | } | |
23067153 | 5821 | |
23067153 | 5822 | exit: |
bf8d5d52 | 5823 | memcg->memory.emin = emin; |
23067153 | 5824 | memcg->memory.elow = elow; |
bf8d5d52 RG |
5825 | |
5826 | if (usage <= emin) | |
5827 | return MEMCG_PROT_MIN; | |
5828 | else if (usage <= elow) | |
5829 | return MEMCG_PROT_LOW; | |
5830 | else | |
5831 | return MEMCG_PROT_NONE; | |
241994ed JW |
5832 | } |
5833 | ||
00501b53 JW |
5834 | /** |
5835 | * mem_cgroup_try_charge - try charging a page | |
5836 | * @page: page to charge | |
5837 | * @mm: mm context of the victim | |
5838 | * @gfp_mask: reclaim mode | |
5839 | * @memcgp: charged memcg return | |
25843c2b | 5840 | * @compound: charge the page as compound or small page |
00501b53 JW |
5841 | * |
5842 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5843 | * pages according to @gfp_mask if necessary. | |
5844 | * | |
5845 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5846 | * Otherwise, an error code is returned. | |
5847 | * | |
5848 | * After page->mapping has been set up, the caller must finalize the | |
5849 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5850 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5851 | */ | |
5852 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
5853 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
5854 | bool compound) | |
00501b53 JW |
5855 | { |
5856 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 5857 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5858 | int ret = 0; |
5859 | ||
5860 | if (mem_cgroup_disabled()) | |
5861 | goto out; | |
5862 | ||
5863 | if (PageSwapCache(page)) { | |
00501b53 JW |
5864 | /* |
5865 | * Every swap fault against a single page tries to charge the | |
5866 | * page, bail as early as possible. shmem_unuse() encounters | |
5867 | * already charged pages, too. The USED bit is protected by | |
5868 | * the page lock, which serializes swap cache removal, which | |
5869 | * in turn serializes uncharging. | |
5870 | */ | |
e993d905 | 5871 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 5872 | if (compound_head(page)->mem_cgroup) |
00501b53 | 5873 | goto out; |
e993d905 | 5874 | |
37e84351 | 5875 | if (do_swap_account) { |
e993d905 VD |
5876 | swp_entry_t ent = { .val = page_private(page), }; |
5877 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5878 | ||
5879 | rcu_read_lock(); | |
5880 | memcg = mem_cgroup_from_id(id); | |
5881 | if (memcg && !css_tryget_online(&memcg->css)) | |
5882 | memcg = NULL; | |
5883 | rcu_read_unlock(); | |
5884 | } | |
00501b53 JW |
5885 | } |
5886 | ||
00501b53 JW |
5887 | if (!memcg) |
5888 | memcg = get_mem_cgroup_from_mm(mm); | |
5889 | ||
5890 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5891 | ||
5892 | css_put(&memcg->css); | |
00501b53 JW |
5893 | out: |
5894 | *memcgp = memcg; | |
5895 | return ret; | |
5896 | } | |
5897 | ||
2cf85583 TH |
5898 | int mem_cgroup_try_charge_delay(struct page *page, struct mm_struct *mm, |
5899 | gfp_t gfp_mask, struct mem_cgroup **memcgp, | |
5900 | bool compound) | |
5901 | { | |
5902 | struct mem_cgroup *memcg; | |
5903 | int ret; | |
5904 | ||
5905 | ret = mem_cgroup_try_charge(page, mm, gfp_mask, memcgp, compound); | |
5906 | memcg = *memcgp; | |
5907 | mem_cgroup_throttle_swaprate(memcg, page_to_nid(page), gfp_mask); | |
5908 | return ret; | |
5909 | } | |
5910 | ||
00501b53 JW |
5911 | /** |
5912 | * mem_cgroup_commit_charge - commit a page charge | |
5913 | * @page: page to charge | |
5914 | * @memcg: memcg to charge the page to | |
5915 | * @lrucare: page might be on LRU already | |
25843c2b | 5916 | * @compound: charge the page as compound or small page |
00501b53 JW |
5917 | * |
5918 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5919 | * after page->mapping has been set up. This must happen atomically | |
5920 | * as part of the page instantiation, i.e. under the page table lock | |
5921 | * for anonymous pages, under the page lock for page and swap cache. | |
5922 | * | |
5923 | * In addition, the page must not be on the LRU during the commit, to | |
5924 | * prevent racing with task migration. If it might be, use @lrucare. | |
5925 | * | |
5926 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5927 | */ | |
5928 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 5929 | bool lrucare, bool compound) |
00501b53 | 5930 | { |
f627c2f5 | 5931 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5932 | |
5933 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5934 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5935 | ||
5936 | if (mem_cgroup_disabled()) | |
5937 | return; | |
5938 | /* | |
5939 | * Swap faults will attempt to charge the same page multiple | |
5940 | * times. But reuse_swap_page() might have removed the page | |
5941 | * from swapcache already, so we can't check PageSwapCache(). | |
5942 | */ | |
5943 | if (!memcg) | |
5944 | return; | |
5945 | ||
6abb5a86 JW |
5946 | commit_charge(page, memcg, lrucare); |
5947 | ||
6abb5a86 | 5948 | local_irq_disable(); |
f627c2f5 | 5949 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
5950 | memcg_check_events(memcg, page); |
5951 | local_irq_enable(); | |
00501b53 | 5952 | |
7941d214 | 5953 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
5954 | swp_entry_t entry = { .val = page_private(page) }; |
5955 | /* | |
5956 | * The swap entry might not get freed for a long time, | |
5957 | * let's not wait for it. The page already received a | |
5958 | * memory+swap charge, drop the swap entry duplicate. | |
5959 | */ | |
38d8b4e6 | 5960 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 JW |
5961 | } |
5962 | } | |
5963 | ||
5964 | /** | |
5965 | * mem_cgroup_cancel_charge - cancel a page charge | |
5966 | * @page: page to charge | |
5967 | * @memcg: memcg to charge the page to | |
25843c2b | 5968 | * @compound: charge the page as compound or small page |
00501b53 JW |
5969 | * |
5970 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5971 | */ | |
f627c2f5 KS |
5972 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
5973 | bool compound) | |
00501b53 | 5974 | { |
f627c2f5 | 5975 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5976 | |
5977 | if (mem_cgroup_disabled()) | |
5978 | return; | |
5979 | /* | |
5980 | * Swap faults will attempt to charge the same page multiple | |
5981 | * times. But reuse_swap_page() might have removed the page | |
5982 | * from swapcache already, so we can't check PageSwapCache(). | |
5983 | */ | |
5984 | if (!memcg) | |
5985 | return; | |
5986 | ||
00501b53 JW |
5987 | cancel_charge(memcg, nr_pages); |
5988 | } | |
5989 | ||
a9d5adee JG |
5990 | struct uncharge_gather { |
5991 | struct mem_cgroup *memcg; | |
5992 | unsigned long pgpgout; | |
5993 | unsigned long nr_anon; | |
5994 | unsigned long nr_file; | |
5995 | unsigned long nr_kmem; | |
5996 | unsigned long nr_huge; | |
5997 | unsigned long nr_shmem; | |
5998 | struct page *dummy_page; | |
5999 | }; | |
6000 | ||
6001 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6002 | { |
a9d5adee JG |
6003 | memset(ug, 0, sizeof(*ug)); |
6004 | } | |
6005 | ||
6006 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6007 | { | |
6008 | unsigned long nr_pages = ug->nr_anon + ug->nr_file + ug->nr_kmem; | |
747db954 JW |
6009 | unsigned long flags; |
6010 | ||
a9d5adee JG |
6011 | if (!mem_cgroup_is_root(ug->memcg)) { |
6012 | page_counter_uncharge(&ug->memcg->memory, nr_pages); | |
7941d214 | 6013 | if (do_memsw_account()) |
a9d5adee JG |
6014 | page_counter_uncharge(&ug->memcg->memsw, nr_pages); |
6015 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) | |
6016 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6017 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6018 | } |
747db954 JW |
6019 | |
6020 | local_irq_save(flags); | |
c9019e9b JW |
6021 | __mod_memcg_state(ug->memcg, MEMCG_RSS, -ug->nr_anon); |
6022 | __mod_memcg_state(ug->memcg, MEMCG_CACHE, -ug->nr_file); | |
6023 | __mod_memcg_state(ug->memcg, MEMCG_RSS_HUGE, -ug->nr_huge); | |
6024 | __mod_memcg_state(ug->memcg, NR_SHMEM, -ug->nr_shmem); | |
6025 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); | |
a983b5eb | 6026 | __this_cpu_add(ug->memcg->stat_cpu->nr_page_events, nr_pages); |
a9d5adee | 6027 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6028 | local_irq_restore(flags); |
e8ea14cc | 6029 | |
a9d5adee JG |
6030 | if (!mem_cgroup_is_root(ug->memcg)) |
6031 | css_put_many(&ug->memcg->css, nr_pages); | |
6032 | } | |
6033 | ||
6034 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6035 | { | |
6036 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
3f2eb028 JG |
6037 | VM_BUG_ON_PAGE(page_count(page) && !is_zone_device_page(page) && |
6038 | !PageHWPoison(page) , page); | |
a9d5adee JG |
6039 | |
6040 | if (!page->mem_cgroup) | |
6041 | return; | |
6042 | ||
6043 | /* | |
6044 | * Nobody should be changing or seriously looking at | |
6045 | * page->mem_cgroup at this point, we have fully | |
6046 | * exclusive access to the page. | |
6047 | */ | |
6048 | ||
6049 | if (ug->memcg != page->mem_cgroup) { | |
6050 | if (ug->memcg) { | |
6051 | uncharge_batch(ug); | |
6052 | uncharge_gather_clear(ug); | |
6053 | } | |
6054 | ug->memcg = page->mem_cgroup; | |
6055 | } | |
6056 | ||
6057 | if (!PageKmemcg(page)) { | |
6058 | unsigned int nr_pages = 1; | |
6059 | ||
6060 | if (PageTransHuge(page)) { | |
6061 | nr_pages <<= compound_order(page); | |
6062 | ug->nr_huge += nr_pages; | |
6063 | } | |
6064 | if (PageAnon(page)) | |
6065 | ug->nr_anon += nr_pages; | |
6066 | else { | |
6067 | ug->nr_file += nr_pages; | |
6068 | if (PageSwapBacked(page)) | |
6069 | ug->nr_shmem += nr_pages; | |
6070 | } | |
6071 | ug->pgpgout++; | |
6072 | } else { | |
6073 | ug->nr_kmem += 1 << compound_order(page); | |
6074 | __ClearPageKmemcg(page); | |
6075 | } | |
6076 | ||
6077 | ug->dummy_page = page; | |
6078 | page->mem_cgroup = NULL; | |
747db954 JW |
6079 | } |
6080 | ||
6081 | static void uncharge_list(struct list_head *page_list) | |
6082 | { | |
a9d5adee | 6083 | struct uncharge_gather ug; |
747db954 | 6084 | struct list_head *next; |
a9d5adee JG |
6085 | |
6086 | uncharge_gather_clear(&ug); | |
747db954 | 6087 | |
8b592656 JW |
6088 | /* |
6089 | * Note that the list can be a single page->lru; hence the | |
6090 | * do-while loop instead of a simple list_for_each_entry(). | |
6091 | */ | |
747db954 JW |
6092 | next = page_list->next; |
6093 | do { | |
a9d5adee JG |
6094 | struct page *page; |
6095 | ||
747db954 JW |
6096 | page = list_entry(next, struct page, lru); |
6097 | next = page->lru.next; | |
6098 | ||
a9d5adee | 6099 | uncharge_page(page, &ug); |
747db954 JW |
6100 | } while (next != page_list); |
6101 | ||
a9d5adee JG |
6102 | if (ug.memcg) |
6103 | uncharge_batch(&ug); | |
747db954 JW |
6104 | } |
6105 | ||
0a31bc97 JW |
6106 | /** |
6107 | * mem_cgroup_uncharge - uncharge a page | |
6108 | * @page: page to uncharge | |
6109 | * | |
6110 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
6111 | * mem_cgroup_commit_charge(). | |
6112 | */ | |
6113 | void mem_cgroup_uncharge(struct page *page) | |
6114 | { | |
a9d5adee JG |
6115 | struct uncharge_gather ug; |
6116 | ||
0a31bc97 JW |
6117 | if (mem_cgroup_disabled()) |
6118 | return; | |
6119 | ||
747db954 | 6120 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6121 | if (!page->mem_cgroup) |
0a31bc97 JW |
6122 | return; |
6123 | ||
a9d5adee JG |
6124 | uncharge_gather_clear(&ug); |
6125 | uncharge_page(page, &ug); | |
6126 | uncharge_batch(&ug); | |
747db954 | 6127 | } |
0a31bc97 | 6128 | |
747db954 JW |
6129 | /** |
6130 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6131 | * @page_list: list of pages to uncharge | |
6132 | * | |
6133 | * Uncharge a list of pages previously charged with | |
6134 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
6135 | */ | |
6136 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6137 | { | |
6138 | if (mem_cgroup_disabled()) | |
6139 | return; | |
0a31bc97 | 6140 | |
747db954 JW |
6141 | if (!list_empty(page_list)) |
6142 | uncharge_list(page_list); | |
0a31bc97 JW |
6143 | } |
6144 | ||
6145 | /** | |
6a93ca8f JW |
6146 | * mem_cgroup_migrate - charge a page's replacement |
6147 | * @oldpage: currently circulating page | |
6148 | * @newpage: replacement page | |
0a31bc97 | 6149 | * |
6a93ca8f JW |
6150 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6151 | * be uncharged upon free. | |
0a31bc97 JW |
6152 | * |
6153 | * Both pages must be locked, @newpage->mapping must be set up. | |
6154 | */ | |
6a93ca8f | 6155 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6156 | { |
29833315 | 6157 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
6158 | unsigned int nr_pages; |
6159 | bool compound; | |
d93c4130 | 6160 | unsigned long flags; |
0a31bc97 JW |
6161 | |
6162 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6163 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6164 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6165 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6166 | newpage); | |
0a31bc97 JW |
6167 | |
6168 | if (mem_cgroup_disabled()) | |
6169 | return; | |
6170 | ||
6171 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6172 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6173 | return; |
6174 | ||
45637bab | 6175 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6176 | memcg = oldpage->mem_cgroup; |
29833315 | 6177 | if (!memcg) |
0a31bc97 JW |
6178 | return; |
6179 | ||
44b7a8d3 JW |
6180 | /* Force-charge the new page. The old one will be freed soon */ |
6181 | compound = PageTransHuge(newpage); | |
6182 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
6183 | ||
6184 | page_counter_charge(&memcg->memory, nr_pages); | |
6185 | if (do_memsw_account()) | |
6186 | page_counter_charge(&memcg->memsw, nr_pages); | |
6187 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 6188 | |
9cf7666a | 6189 | commit_charge(newpage, memcg, false); |
44b7a8d3 | 6190 | |
d93c4130 | 6191 | local_irq_save(flags); |
44b7a8d3 JW |
6192 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
6193 | memcg_check_events(memcg, newpage); | |
d93c4130 | 6194 | local_irq_restore(flags); |
0a31bc97 JW |
6195 | } |
6196 | ||
ef12947c | 6197 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6198 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6199 | ||
2d758073 | 6200 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6201 | { |
6202 | struct mem_cgroup *memcg; | |
6203 | ||
2d758073 JW |
6204 | if (!mem_cgroup_sockets_enabled) |
6205 | return; | |
6206 | ||
edbe69ef RG |
6207 | /* |
6208 | * Socket cloning can throw us here with sk_memcg already | |
6209 | * filled. It won't however, necessarily happen from | |
6210 | * process context. So the test for root memcg given | |
6211 | * the current task's memcg won't help us in this case. | |
6212 | * | |
6213 | * Respecting the original socket's memcg is a better | |
6214 | * decision in this case. | |
6215 | */ | |
6216 | if (sk->sk_memcg) { | |
6217 | css_get(&sk->sk_memcg->css); | |
6218 | return; | |
6219 | } | |
6220 | ||
11092087 JW |
6221 | rcu_read_lock(); |
6222 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6223 | if (memcg == root_mem_cgroup) |
6224 | goto out; | |
0db15298 | 6225 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6226 | goto out; |
f7e1cb6e | 6227 | if (css_tryget_online(&memcg->css)) |
11092087 | 6228 | sk->sk_memcg = memcg; |
f7e1cb6e | 6229 | out: |
11092087 JW |
6230 | rcu_read_unlock(); |
6231 | } | |
11092087 | 6232 | |
2d758073 | 6233 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6234 | { |
2d758073 JW |
6235 | if (sk->sk_memcg) |
6236 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6237 | } |
6238 | ||
6239 | /** | |
6240 | * mem_cgroup_charge_skmem - charge socket memory | |
6241 | * @memcg: memcg to charge | |
6242 | * @nr_pages: number of pages to charge | |
6243 | * | |
6244 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6245 | * @memcg's configured limit, %false if the charge had to be forced. | |
6246 | */ | |
6247 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6248 | { | |
f7e1cb6e | 6249 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6250 | |
f7e1cb6e | 6251 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6252 | struct page_counter *fail; |
f7e1cb6e | 6253 | |
0db15298 JW |
6254 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6255 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6256 | return true; |
6257 | } | |
0db15298 JW |
6258 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6259 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6260 | return false; |
11092087 | 6261 | } |
d886f4e4 | 6262 | |
f7e1cb6e JW |
6263 | /* Don't block in the packet receive path */ |
6264 | if (in_softirq()) | |
6265 | gfp_mask = GFP_NOWAIT; | |
6266 | ||
c9019e9b | 6267 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6268 | |
f7e1cb6e JW |
6269 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6270 | return true; | |
6271 | ||
6272 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6273 | return false; |
6274 | } | |
6275 | ||
6276 | /** | |
6277 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
6278 | * @memcg: memcg to uncharge |
6279 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
6280 | */ |
6281 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6282 | { | |
f7e1cb6e | 6283 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6284 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
6285 | return; |
6286 | } | |
d886f4e4 | 6287 | |
c9019e9b | 6288 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 6289 | |
475d0487 | 6290 | refill_stock(memcg, nr_pages); |
11092087 JW |
6291 | } |
6292 | ||
f7e1cb6e JW |
6293 | static int __init cgroup_memory(char *s) |
6294 | { | |
6295 | char *token; | |
6296 | ||
6297 | while ((token = strsep(&s, ",")) != NULL) { | |
6298 | if (!*token) | |
6299 | continue; | |
6300 | if (!strcmp(token, "nosocket")) | |
6301 | cgroup_memory_nosocket = true; | |
04823c83 VD |
6302 | if (!strcmp(token, "nokmem")) |
6303 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
6304 | } |
6305 | return 0; | |
6306 | } | |
6307 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 6308 | |
2d11085e | 6309 | /* |
1081312f MH |
6310 | * subsys_initcall() for memory controller. |
6311 | * | |
308167fc SAS |
6312 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
6313 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
6314 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
6315 | * should be initialized from here. | |
2d11085e MH |
6316 | */ |
6317 | static int __init mem_cgroup_init(void) | |
6318 | { | |
95a045f6 JW |
6319 | int cpu, node; |
6320 | ||
84c07d11 | 6321 | #ifdef CONFIG_MEMCG_KMEM |
13583c3d VD |
6322 | /* |
6323 | * Kmem cache creation is mostly done with the slab_mutex held, | |
17cc4dfe TH |
6324 | * so use a workqueue with limited concurrency to avoid stalling |
6325 | * all worker threads in case lots of cgroups are created and | |
6326 | * destroyed simultaneously. | |
13583c3d | 6327 | */ |
17cc4dfe TH |
6328 | memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1); |
6329 | BUG_ON(!memcg_kmem_cache_wq); | |
13583c3d VD |
6330 | #endif |
6331 | ||
308167fc SAS |
6332 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
6333 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
6334 | |
6335 | for_each_possible_cpu(cpu) | |
6336 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
6337 | drain_local_stock); | |
6338 | ||
6339 | for_each_node(node) { | |
6340 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
6341 | |
6342 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
6343 | node_online(node) ? node : NUMA_NO_NODE); | |
6344 | ||
ef8f2327 | 6345 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 6346 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 6347 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
6348 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
6349 | } | |
6350 | ||
2d11085e MH |
6351 | return 0; |
6352 | } | |
6353 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
6354 | |
6355 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
6356 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
6357 | { | |
1c2d479a | 6358 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
6359 | /* |
6360 | * The root cgroup cannot be destroyed, so it's refcount must | |
6361 | * always be >= 1. | |
6362 | */ | |
6363 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
6364 | VM_BUG_ON(1); | |
6365 | break; | |
6366 | } | |
6367 | memcg = parent_mem_cgroup(memcg); | |
6368 | if (!memcg) | |
6369 | memcg = root_mem_cgroup; | |
6370 | } | |
6371 | return memcg; | |
6372 | } | |
6373 | ||
21afa38e JW |
6374 | /** |
6375 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
6376 | * @page: page whose memsw charge to transfer | |
6377 | * @entry: swap entry to move the charge to | |
6378 | * | |
6379 | * Transfer the memsw charge of @page to @entry. | |
6380 | */ | |
6381 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
6382 | { | |
1f47b61f | 6383 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 6384 | unsigned int nr_entries; |
21afa38e JW |
6385 | unsigned short oldid; |
6386 | ||
6387 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6388 | VM_BUG_ON_PAGE(page_count(page), page); | |
6389 | ||
7941d214 | 6390 | if (!do_memsw_account()) |
21afa38e JW |
6391 | return; |
6392 | ||
6393 | memcg = page->mem_cgroup; | |
6394 | ||
6395 | /* Readahead page, never charged */ | |
6396 | if (!memcg) | |
6397 | return; | |
6398 | ||
1f47b61f VD |
6399 | /* |
6400 | * In case the memcg owning these pages has been offlined and doesn't | |
6401 | * have an ID allocated to it anymore, charge the closest online | |
6402 | * ancestor for the swap instead and transfer the memory+swap charge. | |
6403 | */ | |
6404 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
d6810d73 HY |
6405 | nr_entries = hpage_nr_pages(page); |
6406 | /* Get references for the tail pages, too */ | |
6407 | if (nr_entries > 1) | |
6408 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
6409 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
6410 | nr_entries); | |
21afa38e | 6411 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6412 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
6413 | |
6414 | page->mem_cgroup = NULL; | |
6415 | ||
6416 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 6417 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 6418 | |
1f47b61f VD |
6419 | if (memcg != swap_memcg) { |
6420 | if (!mem_cgroup_is_root(swap_memcg)) | |
d6810d73 HY |
6421 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
6422 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
6423 | } |
6424 | ||
ce9ce665 SAS |
6425 | /* |
6426 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 6427 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 6428 | * important here to have the interrupts disabled because it is the |
b93b0163 | 6429 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
6430 | */ |
6431 | VM_BUG_ON(!irqs_disabled()); | |
d6810d73 HY |
6432 | mem_cgroup_charge_statistics(memcg, page, PageTransHuge(page), |
6433 | -nr_entries); | |
21afa38e | 6434 | memcg_check_events(memcg, page); |
73f576c0 JW |
6435 | |
6436 | if (!mem_cgroup_is_root(memcg)) | |
d08afa14 | 6437 | css_put_many(&memcg->css, nr_entries); |
21afa38e JW |
6438 | } |
6439 | ||
38d8b4e6 HY |
6440 | /** |
6441 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
6442 | * @page: page being added to swap |
6443 | * @entry: swap entry to charge | |
6444 | * | |
38d8b4e6 | 6445 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
6446 | * |
6447 | * Returns 0 on success, -ENOMEM on failure. | |
6448 | */ | |
6449 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
6450 | { | |
38d8b4e6 | 6451 | unsigned int nr_pages = hpage_nr_pages(page); |
37e84351 | 6452 | struct page_counter *counter; |
38d8b4e6 | 6453 | struct mem_cgroup *memcg; |
37e84351 VD |
6454 | unsigned short oldid; |
6455 | ||
6456 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
6457 | return 0; | |
6458 | ||
6459 | memcg = page->mem_cgroup; | |
6460 | ||
6461 | /* Readahead page, never charged */ | |
6462 | if (!memcg) | |
6463 | return 0; | |
6464 | ||
f3a53a3a TH |
6465 | if (!entry.val) { |
6466 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 6467 | return 0; |
f3a53a3a | 6468 | } |
bb98f2c5 | 6469 | |
1f47b61f VD |
6470 | memcg = mem_cgroup_id_get_online(memcg); |
6471 | ||
37e84351 | 6472 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 6473 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
6474 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
6475 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 6476 | mem_cgroup_id_put(memcg); |
37e84351 | 6477 | return -ENOMEM; |
1f47b61f | 6478 | } |
37e84351 | 6479 | |
38d8b4e6 HY |
6480 | /* Get references for the tail pages, too */ |
6481 | if (nr_pages > 1) | |
6482 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
6483 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 6484 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6485 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 6486 | |
37e84351 VD |
6487 | return 0; |
6488 | } | |
6489 | ||
21afa38e | 6490 | /** |
38d8b4e6 | 6491 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 6492 | * @entry: swap entry to uncharge |
38d8b4e6 | 6493 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 6494 | */ |
38d8b4e6 | 6495 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
6496 | { |
6497 | struct mem_cgroup *memcg; | |
6498 | unsigned short id; | |
6499 | ||
37e84351 | 6500 | if (!do_swap_account) |
21afa38e JW |
6501 | return; |
6502 | ||
38d8b4e6 | 6503 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 6504 | rcu_read_lock(); |
adbe427b | 6505 | memcg = mem_cgroup_from_id(id); |
21afa38e | 6506 | if (memcg) { |
37e84351 VD |
6507 | if (!mem_cgroup_is_root(memcg)) { |
6508 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
38d8b4e6 | 6509 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 6510 | else |
38d8b4e6 | 6511 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 6512 | } |
c9019e9b | 6513 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 6514 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
6515 | } |
6516 | rcu_read_unlock(); | |
6517 | } | |
6518 | ||
d8b38438 VD |
6519 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
6520 | { | |
6521 | long nr_swap_pages = get_nr_swap_pages(); | |
6522 | ||
6523 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6524 | return nr_swap_pages; | |
6525 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
6526 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 6527 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
6528 | page_counter_read(&memcg->swap)); |
6529 | return nr_swap_pages; | |
6530 | } | |
6531 | ||
5ccc5aba VD |
6532 | bool mem_cgroup_swap_full(struct page *page) |
6533 | { | |
6534 | struct mem_cgroup *memcg; | |
6535 | ||
6536 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
6537 | ||
6538 | if (vm_swap_full()) | |
6539 | return true; | |
6540 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6541 | return false; | |
6542 | ||
6543 | memcg = page->mem_cgroup; | |
6544 | if (!memcg) | |
6545 | return false; | |
6546 | ||
6547 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
bbec2e15 | 6548 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.max) |
5ccc5aba VD |
6549 | return true; |
6550 | ||
6551 | return false; | |
6552 | } | |
6553 | ||
21afa38e JW |
6554 | /* for remember boot option*/ |
6555 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
6556 | static int really_do_swap_account __initdata = 1; | |
6557 | #else | |
6558 | static int really_do_swap_account __initdata; | |
6559 | #endif | |
6560 | ||
6561 | static int __init enable_swap_account(char *s) | |
6562 | { | |
6563 | if (!strcmp(s, "1")) | |
6564 | really_do_swap_account = 1; | |
6565 | else if (!strcmp(s, "0")) | |
6566 | really_do_swap_account = 0; | |
6567 | return 1; | |
6568 | } | |
6569 | __setup("swapaccount=", enable_swap_account); | |
6570 | ||
37e84351 VD |
6571 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
6572 | struct cftype *cft) | |
6573 | { | |
6574 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6575 | ||
6576 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
6577 | } | |
6578 | ||
6579 | static int swap_max_show(struct seq_file *m, void *v) | |
6580 | { | |
6581 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
bbec2e15 | 6582 | unsigned long max = READ_ONCE(memcg->swap.max); |
37e84351 VD |
6583 | |
6584 | if (max == PAGE_COUNTER_MAX) | |
6585 | seq_puts(m, "max\n"); | |
6586 | else | |
6587 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
6588 | ||
6589 | return 0; | |
6590 | } | |
6591 | ||
6592 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
6593 | char *buf, size_t nbytes, loff_t off) | |
6594 | { | |
6595 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6596 | unsigned long max; | |
6597 | int err; | |
6598 | ||
6599 | buf = strstrip(buf); | |
6600 | err = page_counter_memparse(buf, "max", &max); | |
6601 | if (err) | |
6602 | return err; | |
6603 | ||
be09102b | 6604 | xchg(&memcg->swap.max, max); |
37e84351 VD |
6605 | |
6606 | return nbytes; | |
6607 | } | |
6608 | ||
f3a53a3a TH |
6609 | static int swap_events_show(struct seq_file *m, void *v) |
6610 | { | |
6611 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
6612 | ||
6613 | seq_printf(m, "max %lu\n", | |
6614 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
6615 | seq_printf(m, "fail %lu\n", | |
6616 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
6617 | ||
6618 | return 0; | |
6619 | } | |
6620 | ||
37e84351 VD |
6621 | static struct cftype swap_files[] = { |
6622 | { | |
6623 | .name = "swap.current", | |
6624 | .flags = CFTYPE_NOT_ON_ROOT, | |
6625 | .read_u64 = swap_current_read, | |
6626 | }, | |
6627 | { | |
6628 | .name = "swap.max", | |
6629 | .flags = CFTYPE_NOT_ON_ROOT, | |
6630 | .seq_show = swap_max_show, | |
6631 | .write = swap_max_write, | |
6632 | }, | |
f3a53a3a TH |
6633 | { |
6634 | .name = "swap.events", | |
6635 | .flags = CFTYPE_NOT_ON_ROOT, | |
6636 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
6637 | .seq_show = swap_events_show, | |
6638 | }, | |
37e84351 VD |
6639 | { } /* terminate */ |
6640 | }; | |
6641 | ||
21afa38e JW |
6642 | static struct cftype memsw_cgroup_files[] = { |
6643 | { | |
6644 | .name = "memsw.usage_in_bytes", | |
6645 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6646 | .read_u64 = mem_cgroup_read_u64, | |
6647 | }, | |
6648 | { | |
6649 | .name = "memsw.max_usage_in_bytes", | |
6650 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6651 | .write = mem_cgroup_reset, | |
6652 | .read_u64 = mem_cgroup_read_u64, | |
6653 | }, | |
6654 | { | |
6655 | .name = "memsw.limit_in_bytes", | |
6656 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6657 | .write = mem_cgroup_write, | |
6658 | .read_u64 = mem_cgroup_read_u64, | |
6659 | }, | |
6660 | { | |
6661 | .name = "memsw.failcnt", | |
6662 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6663 | .write = mem_cgroup_reset, | |
6664 | .read_u64 = mem_cgroup_read_u64, | |
6665 | }, | |
6666 | { }, /* terminate */ | |
6667 | }; | |
6668 | ||
6669 | static int __init mem_cgroup_swap_init(void) | |
6670 | { | |
6671 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6672 | do_swap_account = 1; | |
37e84351 VD |
6673 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
6674 | swap_files)); | |
21afa38e JW |
6675 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6676 | memsw_cgroup_files)); | |
6677 | } | |
6678 | return 0; | |
6679 | } | |
6680 | subsys_initcall(mem_cgroup_swap_init); | |
6681 | ||
6682 | #endif /* CONFIG_MEMCG_SWAP */ |