]>
Commit | Line | Data |
---|---|---|
c942fddf | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
8cdea7c0 BS |
2 | /* memcontrol.c - Memory Controller |
3 | * | |
4 | * Copyright IBM Corporation, 2007 | |
5 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
6 | * | |
78fb7466 PE |
7 | * Copyright 2007 OpenVZ SWsoft Inc |
8 | * Author: Pavel Emelianov <xemul@openvz.org> | |
9 | * | |
2e72b634 KS |
10 | * Memory thresholds |
11 | * Copyright (C) 2009 Nokia Corporation | |
12 | * Author: Kirill A. Shutemov | |
13 | * | |
7ae1e1d0 GC |
14 | * Kernel Memory Controller |
15 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
16 | * Authors: Glauber Costa and Suleiman Souhlal | |
17 | * | |
1575e68b JW |
18 | * Native page reclaim |
19 | * Charge lifetime sanitation | |
20 | * Lockless page tracking & accounting | |
21 | * Unified hierarchy configuration model | |
22 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
8cdea7c0 BS |
23 | */ |
24 | ||
3e32cb2e | 25 | #include <linux/page_counter.h> |
8cdea7c0 BS |
26 | #include <linux/memcontrol.h> |
27 | #include <linux/cgroup.h> | |
78fb7466 | 28 | #include <linux/mm.h> |
6e84f315 | 29 | #include <linux/sched/mm.h> |
3a4f8a0b | 30 | #include <linux/shmem_fs.h> |
4ffef5fe | 31 | #include <linux/hugetlb.h> |
d13d1443 | 32 | #include <linux/pagemap.h> |
1ff9e6e1 | 33 | #include <linux/vm_event_item.h> |
d52aa412 | 34 | #include <linux/smp.h> |
8a9f3ccd | 35 | #include <linux/page-flags.h> |
66e1707b | 36 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
37 | #include <linux/bit_spinlock.h> |
38 | #include <linux/rcupdate.h> | |
e222432b | 39 | #include <linux/limits.h> |
b9e15baf | 40 | #include <linux/export.h> |
8c7c6e34 | 41 | #include <linux/mutex.h> |
bb4cc1a8 | 42 | #include <linux/rbtree.h> |
b6ac57d5 | 43 | #include <linux/slab.h> |
66e1707b | 44 | #include <linux/swap.h> |
02491447 | 45 | #include <linux/swapops.h> |
66e1707b | 46 | #include <linux/spinlock.h> |
2e72b634 | 47 | #include <linux/eventfd.h> |
79bd9814 | 48 | #include <linux/poll.h> |
2e72b634 | 49 | #include <linux/sort.h> |
66e1707b | 50 | #include <linux/fs.h> |
d2ceb9b7 | 51 | #include <linux/seq_file.h> |
70ddf637 | 52 | #include <linux/vmpressure.h> |
b69408e8 | 53 | #include <linux/mm_inline.h> |
5d1ea48b | 54 | #include <linux/swap_cgroup.h> |
cdec2e42 | 55 | #include <linux/cpu.h> |
158e0a2d | 56 | #include <linux/oom.h> |
0056f4e6 | 57 | #include <linux/lockdep.h> |
79bd9814 | 58 | #include <linux/file.h> |
b23afb93 | 59 | #include <linux/tracehook.h> |
08e552c6 | 60 | #include "internal.h" |
d1a4c0b3 | 61 | #include <net/sock.h> |
4bd2c1ee | 62 | #include <net/ip.h> |
f35c3a8e | 63 | #include "slab.h" |
8cdea7c0 | 64 | |
7c0f6ba6 | 65 | #include <linux/uaccess.h> |
8697d331 | 66 | |
cc8e970c KM |
67 | #include <trace/events/vmscan.h> |
68 | ||
073219e9 TH |
69 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
70 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 71 | |
7d828602 JW |
72 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
73 | ||
a181b0e8 | 74 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 75 | |
f7e1cb6e JW |
76 | /* Socket memory accounting disabled? */ |
77 | static bool cgroup_memory_nosocket; | |
78 | ||
04823c83 VD |
79 | /* Kernel memory accounting disabled? */ |
80 | static bool cgroup_memory_nokmem; | |
81 | ||
21afa38e | 82 | /* Whether the swap controller is active */ |
c255a458 | 83 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 84 | int do_swap_account __read_mostly; |
c077719b | 85 | #else |
a0db00fc | 86 | #define do_swap_account 0 |
c077719b KH |
87 | #endif |
88 | ||
7941d214 JW |
89 | /* Whether legacy memory+swap accounting is active */ |
90 | static bool do_memsw_account(void) | |
91 | { | |
92 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; | |
93 | } | |
94 | ||
71cd3113 | 95 | static const char *const mem_cgroup_lru_names[] = { |
58cf188e SZ |
96 | "inactive_anon", |
97 | "active_anon", | |
98 | "inactive_file", | |
99 | "active_file", | |
100 | "unevictable", | |
101 | }; | |
102 | ||
a0db00fc KS |
103 | #define THRESHOLDS_EVENTS_TARGET 128 |
104 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
105 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 106 | |
bb4cc1a8 AM |
107 | /* |
108 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
109 | * their hierarchy representation | |
110 | */ | |
111 | ||
ef8f2327 | 112 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 113 | struct rb_root rb_root; |
fa90b2fd | 114 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
115 | spinlock_t lock; |
116 | }; | |
117 | ||
bb4cc1a8 AM |
118 | struct mem_cgroup_tree { |
119 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
120 | }; | |
121 | ||
122 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
123 | ||
9490ff27 KH |
124 | /* for OOM */ |
125 | struct mem_cgroup_eventfd_list { | |
126 | struct list_head list; | |
127 | struct eventfd_ctx *eventfd; | |
128 | }; | |
2e72b634 | 129 | |
79bd9814 TH |
130 | /* |
131 | * cgroup_event represents events which userspace want to receive. | |
132 | */ | |
3bc942f3 | 133 | struct mem_cgroup_event { |
79bd9814 | 134 | /* |
59b6f873 | 135 | * memcg which the event belongs to. |
79bd9814 | 136 | */ |
59b6f873 | 137 | struct mem_cgroup *memcg; |
79bd9814 TH |
138 | /* |
139 | * eventfd to signal userspace about the event. | |
140 | */ | |
141 | struct eventfd_ctx *eventfd; | |
142 | /* | |
143 | * Each of these stored in a list by the cgroup. | |
144 | */ | |
145 | struct list_head list; | |
fba94807 TH |
146 | /* |
147 | * register_event() callback will be used to add new userspace | |
148 | * waiter for changes related to this event. Use eventfd_signal() | |
149 | * on eventfd to send notification to userspace. | |
150 | */ | |
59b6f873 | 151 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 152 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
153 | /* |
154 | * unregister_event() callback will be called when userspace closes | |
155 | * the eventfd or on cgroup removing. This callback must be set, | |
156 | * if you want provide notification functionality. | |
157 | */ | |
59b6f873 | 158 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 159 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
160 | /* |
161 | * All fields below needed to unregister event when | |
162 | * userspace closes eventfd. | |
163 | */ | |
164 | poll_table pt; | |
165 | wait_queue_head_t *wqh; | |
ac6424b9 | 166 | wait_queue_entry_t wait; |
79bd9814 TH |
167 | struct work_struct remove; |
168 | }; | |
169 | ||
c0ff4b85 R |
170 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
171 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 172 | |
7dc74be0 DN |
173 | /* Stuffs for move charges at task migration. */ |
174 | /* | |
1dfab5ab | 175 | * Types of charges to be moved. |
7dc74be0 | 176 | */ |
1dfab5ab JW |
177 | #define MOVE_ANON 0x1U |
178 | #define MOVE_FILE 0x2U | |
179 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 180 | |
4ffef5fe DN |
181 | /* "mc" and its members are protected by cgroup_mutex */ |
182 | static struct move_charge_struct { | |
b1dd693e | 183 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 184 | struct mm_struct *mm; |
4ffef5fe DN |
185 | struct mem_cgroup *from; |
186 | struct mem_cgroup *to; | |
1dfab5ab | 187 | unsigned long flags; |
4ffef5fe | 188 | unsigned long precharge; |
854ffa8d | 189 | unsigned long moved_charge; |
483c30b5 | 190 | unsigned long moved_swap; |
8033b97c DN |
191 | struct task_struct *moving_task; /* a task moving charges */ |
192 | wait_queue_head_t waitq; /* a waitq for other context */ | |
193 | } mc = { | |
2bd9bb20 | 194 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
195 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
196 | }; | |
4ffef5fe | 197 | |
4e416953 BS |
198 | /* |
199 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
200 | * limit reclaim to prevent infinite loops, if they ever occur. | |
201 | */ | |
a0db00fc | 202 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 203 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 204 | |
217bc319 KH |
205 | enum charge_type { |
206 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 207 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 208 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 209 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
210 | NR_CHARGE_TYPE, |
211 | }; | |
212 | ||
8c7c6e34 | 213 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
214 | enum res_type { |
215 | _MEM, | |
216 | _MEMSWAP, | |
217 | _OOM_TYPE, | |
510fc4e1 | 218 | _KMEM, |
d55f90bf | 219 | _TCP, |
86ae53e1 GC |
220 | }; |
221 | ||
a0db00fc KS |
222 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
223 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 224 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
225 | /* Used for OOM nofiier */ |
226 | #define OOM_CONTROL (0) | |
8c7c6e34 | 227 | |
b05706f1 KT |
228 | /* |
229 | * Iteration constructs for visiting all cgroups (under a tree). If | |
230 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
231 | * be used for reference counting. | |
232 | */ | |
233 | #define for_each_mem_cgroup_tree(iter, root) \ | |
234 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
235 | iter != NULL; \ | |
236 | iter = mem_cgroup_iter(root, iter, NULL)) | |
237 | ||
238 | #define for_each_mem_cgroup(iter) \ | |
239 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
240 | iter != NULL; \ | |
241 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
242 | ||
7775face TH |
243 | static inline bool should_force_charge(void) |
244 | { | |
245 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
246 | (current->flags & PF_EXITING); | |
247 | } | |
248 | ||
70ddf637 AV |
249 | /* Some nice accessors for the vmpressure. */ |
250 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
251 | { | |
252 | if (!memcg) | |
253 | memcg = root_mem_cgroup; | |
254 | return &memcg->vmpressure; | |
255 | } | |
256 | ||
257 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
258 | { | |
259 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
260 | } | |
261 | ||
84c07d11 | 262 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 263 | /* |
f7ce3190 | 264 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
265 | * The main reason for not using cgroup id for this: |
266 | * this works better in sparse environments, where we have a lot of memcgs, | |
267 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
268 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
269 | * 200 entry array for that. | |
55007d84 | 270 | * |
dbcf73e2 VD |
271 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
272 | * will double each time we have to increase it. | |
55007d84 | 273 | */ |
dbcf73e2 VD |
274 | static DEFINE_IDA(memcg_cache_ida); |
275 | int memcg_nr_cache_ids; | |
749c5415 | 276 | |
05257a1a VD |
277 | /* Protects memcg_nr_cache_ids */ |
278 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
279 | ||
280 | void memcg_get_cache_ids(void) | |
281 | { | |
282 | down_read(&memcg_cache_ids_sem); | |
283 | } | |
284 | ||
285 | void memcg_put_cache_ids(void) | |
286 | { | |
287 | up_read(&memcg_cache_ids_sem); | |
288 | } | |
289 | ||
55007d84 GC |
290 | /* |
291 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
292 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
293 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
294 | * tunable, but that is strictly not necessary. | |
295 | * | |
b8627835 | 296 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
297 | * this constant directly from cgroup, but it is understandable that this is |
298 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 299 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
300 | * increase ours as well if it increases. |
301 | */ | |
302 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 303 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 304 | |
d7f25f8a GC |
305 | /* |
306 | * A lot of the calls to the cache allocation functions are expected to be | |
307 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
308 | * conditional to this static branch, we'll have to allow modules that does | |
309 | * kmem_cache_alloc and the such to see this symbol as well | |
310 | */ | |
ef12947c | 311 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 312 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 313 | |
17cc4dfe TH |
314 | struct workqueue_struct *memcg_kmem_cache_wq; |
315 | ||
0a4465d3 KT |
316 | static int memcg_shrinker_map_size; |
317 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); | |
318 | ||
319 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) | |
320 | { | |
321 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); | |
322 | } | |
323 | ||
324 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, | |
325 | int size, int old_size) | |
326 | { | |
327 | struct memcg_shrinker_map *new, *old; | |
328 | int nid; | |
329 | ||
330 | lockdep_assert_held(&memcg_shrinker_map_mutex); | |
331 | ||
332 | for_each_node(nid) { | |
333 | old = rcu_dereference_protected( | |
334 | mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); | |
335 | /* Not yet online memcg */ | |
336 | if (!old) | |
337 | return 0; | |
338 | ||
339 | new = kvmalloc(sizeof(*new) + size, GFP_KERNEL); | |
340 | if (!new) | |
341 | return -ENOMEM; | |
342 | ||
343 | /* Set all old bits, clear all new bits */ | |
344 | memset(new->map, (int)0xff, old_size); | |
345 | memset((void *)new->map + old_size, 0, size - old_size); | |
346 | ||
347 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); | |
348 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); | |
349 | } | |
350 | ||
351 | return 0; | |
352 | } | |
353 | ||
354 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) | |
355 | { | |
356 | struct mem_cgroup_per_node *pn; | |
357 | struct memcg_shrinker_map *map; | |
358 | int nid; | |
359 | ||
360 | if (mem_cgroup_is_root(memcg)) | |
361 | return; | |
362 | ||
363 | for_each_node(nid) { | |
364 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
365 | map = rcu_dereference_protected(pn->shrinker_map, true); | |
366 | if (map) | |
367 | kvfree(map); | |
368 | rcu_assign_pointer(pn->shrinker_map, NULL); | |
369 | } | |
370 | } | |
371 | ||
372 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
373 | { | |
374 | struct memcg_shrinker_map *map; | |
375 | int nid, size, ret = 0; | |
376 | ||
377 | if (mem_cgroup_is_root(memcg)) | |
378 | return 0; | |
379 | ||
380 | mutex_lock(&memcg_shrinker_map_mutex); | |
381 | size = memcg_shrinker_map_size; | |
382 | for_each_node(nid) { | |
383 | map = kvzalloc(sizeof(*map) + size, GFP_KERNEL); | |
384 | if (!map) { | |
385 | memcg_free_shrinker_maps(memcg); | |
386 | ret = -ENOMEM; | |
387 | break; | |
388 | } | |
389 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); | |
390 | } | |
391 | mutex_unlock(&memcg_shrinker_map_mutex); | |
392 | ||
393 | return ret; | |
394 | } | |
395 | ||
396 | int memcg_expand_shrinker_maps(int new_id) | |
397 | { | |
398 | int size, old_size, ret = 0; | |
399 | struct mem_cgroup *memcg; | |
400 | ||
401 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); | |
402 | old_size = memcg_shrinker_map_size; | |
403 | if (size <= old_size) | |
404 | return 0; | |
405 | ||
406 | mutex_lock(&memcg_shrinker_map_mutex); | |
407 | if (!root_mem_cgroup) | |
408 | goto unlock; | |
409 | ||
410 | for_each_mem_cgroup(memcg) { | |
411 | if (mem_cgroup_is_root(memcg)) | |
412 | continue; | |
413 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); | |
414 | if (ret) | |
415 | goto unlock; | |
416 | } | |
417 | unlock: | |
418 | if (!ret) | |
419 | memcg_shrinker_map_size = size; | |
420 | mutex_unlock(&memcg_shrinker_map_mutex); | |
421 | return ret; | |
422 | } | |
fae91d6d KT |
423 | |
424 | void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id) | |
425 | { | |
426 | if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) { | |
427 | struct memcg_shrinker_map *map; | |
428 | ||
429 | rcu_read_lock(); | |
430 | map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map); | |
f90280d6 KT |
431 | /* Pairs with smp mb in shrink_slab() */ |
432 | smp_mb__before_atomic(); | |
fae91d6d KT |
433 | set_bit(shrinker_id, map->map); |
434 | rcu_read_unlock(); | |
435 | } | |
436 | } | |
437 | ||
0a4465d3 KT |
438 | #else /* CONFIG_MEMCG_KMEM */ |
439 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
440 | { | |
441 | return 0; | |
442 | } | |
443 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) { } | |
84c07d11 | 444 | #endif /* CONFIG_MEMCG_KMEM */ |
a8964b9b | 445 | |
ad7fa852 TH |
446 | /** |
447 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
448 | * @page: page of interest | |
449 | * | |
450 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
451 | * with @page is returned. The returned css remains associated with @page | |
452 | * until it is released. | |
453 | * | |
454 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
455 | * is returned. | |
ad7fa852 TH |
456 | */ |
457 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
458 | { | |
459 | struct mem_cgroup *memcg; | |
460 | ||
ad7fa852 TH |
461 | memcg = page->mem_cgroup; |
462 | ||
9e10a130 | 463 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
464 | memcg = root_mem_cgroup; |
465 | ||
ad7fa852 TH |
466 | return &memcg->css; |
467 | } | |
468 | ||
2fc04524 VD |
469 | /** |
470 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
471 | * @page: the page | |
472 | * | |
473 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
474 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
475 | * is safe to call this function without holding a reference to @page. | |
476 | * | |
477 | * Note, this function is inherently racy, because there is nothing to prevent | |
478 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
479 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
480 | * do not care (such as procfs interfaces). | |
481 | */ | |
482 | ino_t page_cgroup_ino(struct page *page) | |
483 | { | |
484 | struct mem_cgroup *memcg; | |
485 | unsigned long ino = 0; | |
486 | ||
487 | rcu_read_lock(); | |
488 | memcg = READ_ONCE(page->mem_cgroup); | |
489 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
490 | memcg = parent_mem_cgroup(memcg); | |
491 | if (memcg) | |
492 | ino = cgroup_ino(memcg->css.cgroup); | |
493 | rcu_read_unlock(); | |
494 | return ino; | |
495 | } | |
496 | ||
ef8f2327 MG |
497 | static struct mem_cgroup_per_node * |
498 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 499 | { |
97a6c37b | 500 | int nid = page_to_nid(page); |
f64c3f54 | 501 | |
ef8f2327 | 502 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
503 | } |
504 | ||
ef8f2327 MG |
505 | static struct mem_cgroup_tree_per_node * |
506 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 507 | { |
ef8f2327 | 508 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
509 | } |
510 | ||
ef8f2327 | 511 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
512 | soft_limit_tree_from_page(struct page *page) |
513 | { | |
514 | int nid = page_to_nid(page); | |
bb4cc1a8 | 515 | |
ef8f2327 | 516 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
517 | } |
518 | ||
ef8f2327 MG |
519 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
520 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 521 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
522 | { |
523 | struct rb_node **p = &mctz->rb_root.rb_node; | |
524 | struct rb_node *parent = NULL; | |
ef8f2327 | 525 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 526 | bool rightmost = true; |
bb4cc1a8 AM |
527 | |
528 | if (mz->on_tree) | |
529 | return; | |
530 | ||
531 | mz->usage_in_excess = new_usage_in_excess; | |
532 | if (!mz->usage_in_excess) | |
533 | return; | |
534 | while (*p) { | |
535 | parent = *p; | |
ef8f2327 | 536 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 537 | tree_node); |
fa90b2fd | 538 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 539 | p = &(*p)->rb_left; |
fa90b2fd DB |
540 | rightmost = false; |
541 | } | |
542 | ||
bb4cc1a8 AM |
543 | /* |
544 | * We can't avoid mem cgroups that are over their soft | |
545 | * limit by the same amount | |
546 | */ | |
547 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
548 | p = &(*p)->rb_right; | |
549 | } | |
fa90b2fd DB |
550 | |
551 | if (rightmost) | |
552 | mctz->rb_rightmost = &mz->tree_node; | |
553 | ||
bb4cc1a8 AM |
554 | rb_link_node(&mz->tree_node, parent, p); |
555 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
556 | mz->on_tree = true; | |
557 | } | |
558 | ||
ef8f2327 MG |
559 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
560 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
561 | { |
562 | if (!mz->on_tree) | |
563 | return; | |
fa90b2fd DB |
564 | |
565 | if (&mz->tree_node == mctz->rb_rightmost) | |
566 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
567 | ||
bb4cc1a8 AM |
568 | rb_erase(&mz->tree_node, &mctz->rb_root); |
569 | mz->on_tree = false; | |
570 | } | |
571 | ||
ef8f2327 MG |
572 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
573 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 574 | { |
0a31bc97 JW |
575 | unsigned long flags; |
576 | ||
577 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 578 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 579 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
580 | } |
581 | ||
3e32cb2e JW |
582 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
583 | { | |
584 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 585 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
586 | unsigned long excess = 0; |
587 | ||
588 | if (nr_pages > soft_limit) | |
589 | excess = nr_pages - soft_limit; | |
590 | ||
591 | return excess; | |
592 | } | |
bb4cc1a8 AM |
593 | |
594 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
595 | { | |
3e32cb2e | 596 | unsigned long excess; |
ef8f2327 MG |
597 | struct mem_cgroup_per_node *mz; |
598 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 599 | |
e231875b | 600 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
601 | if (!mctz) |
602 | return; | |
bb4cc1a8 AM |
603 | /* |
604 | * Necessary to update all ancestors when hierarchy is used. | |
605 | * because their event counter is not touched. | |
606 | */ | |
607 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 608 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 609 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
610 | /* |
611 | * We have to update the tree if mz is on RB-tree or | |
612 | * mem is over its softlimit. | |
613 | */ | |
614 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
615 | unsigned long flags; |
616 | ||
617 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
618 | /* if on-tree, remove it */ |
619 | if (mz->on_tree) | |
cf2c8127 | 620 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
621 | /* |
622 | * Insert again. mz->usage_in_excess will be updated. | |
623 | * If excess is 0, no tree ops. | |
624 | */ | |
cf2c8127 | 625 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 626 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
627 | } |
628 | } | |
629 | } | |
630 | ||
631 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
632 | { | |
ef8f2327 MG |
633 | struct mem_cgroup_tree_per_node *mctz; |
634 | struct mem_cgroup_per_node *mz; | |
635 | int nid; | |
bb4cc1a8 | 636 | |
e231875b | 637 | for_each_node(nid) { |
ef8f2327 MG |
638 | mz = mem_cgroup_nodeinfo(memcg, nid); |
639 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
640 | if (mctz) |
641 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
642 | } |
643 | } | |
644 | ||
ef8f2327 MG |
645 | static struct mem_cgroup_per_node * |
646 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 647 | { |
ef8f2327 | 648 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
649 | |
650 | retry: | |
651 | mz = NULL; | |
fa90b2fd | 652 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
653 | goto done; /* Nothing to reclaim from */ |
654 | ||
fa90b2fd DB |
655 | mz = rb_entry(mctz->rb_rightmost, |
656 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
657 | /* |
658 | * Remove the node now but someone else can add it back, | |
659 | * we will to add it back at the end of reclaim to its correct | |
660 | * position in the tree. | |
661 | */ | |
cf2c8127 | 662 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 663 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 664 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
665 | goto retry; |
666 | done: | |
667 | return mz; | |
668 | } | |
669 | ||
ef8f2327 MG |
670 | static struct mem_cgroup_per_node * |
671 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 672 | { |
ef8f2327 | 673 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 674 | |
0a31bc97 | 675 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 676 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 677 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
678 | return mz; |
679 | } | |
680 | ||
db9adbcb JW |
681 | /** |
682 | * __mod_memcg_state - update cgroup memory statistics | |
683 | * @memcg: the memory cgroup | |
684 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
685 | * @val: delta to add to the counter, can be negative | |
686 | */ | |
687 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
688 | { | |
689 | long x; | |
690 | ||
691 | if (mem_cgroup_disabled()) | |
692 | return; | |
693 | ||
694 | x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); | |
695 | if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
696 | struct mem_cgroup *mi; |
697 | ||
698 | atomic_long_add(x, &memcg->vmstats_local[idx]); | |
699 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
700 | atomic_long_add(x, &mi->vmstats[idx]); | |
db9adbcb JW |
701 | x = 0; |
702 | } | |
703 | __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); | |
704 | } | |
705 | ||
42a30035 JW |
706 | static struct mem_cgroup_per_node * |
707 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
708 | { | |
709 | struct mem_cgroup *parent; | |
710 | ||
711 | parent = parent_mem_cgroup(pn->memcg); | |
712 | if (!parent) | |
713 | return NULL; | |
714 | return mem_cgroup_nodeinfo(parent, nid); | |
715 | } | |
716 | ||
db9adbcb JW |
717 | /** |
718 | * __mod_lruvec_state - update lruvec memory statistics | |
719 | * @lruvec: the lruvec | |
720 | * @idx: the stat item | |
721 | * @val: delta to add to the counter, can be negative | |
722 | * | |
723 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
724 | * function updates the all three counters that are affected by a | |
725 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
726 | */ | |
727 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
728 | int val) | |
729 | { | |
42a30035 | 730 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
db9adbcb | 731 | struct mem_cgroup_per_node *pn; |
42a30035 | 732 | struct mem_cgroup *memcg; |
db9adbcb JW |
733 | long x; |
734 | ||
735 | /* Update node */ | |
42a30035 | 736 | __mod_node_page_state(pgdat, idx, val); |
db9adbcb JW |
737 | |
738 | if (mem_cgroup_disabled()) | |
739 | return; | |
740 | ||
741 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); | |
42a30035 | 742 | memcg = pn->memcg; |
db9adbcb JW |
743 | |
744 | /* Update memcg */ | |
42a30035 | 745 | __mod_memcg_state(memcg, idx, val); |
db9adbcb JW |
746 | |
747 | /* Update lruvec */ | |
748 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); | |
749 | if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
750 | struct mem_cgroup_per_node *pi; |
751 | ||
752 | atomic_long_add(x, &pn->lruvec_stat_local[idx]); | |
753 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) | |
754 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
755 | x = 0; |
756 | } | |
757 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
758 | } | |
759 | ||
760 | /** | |
761 | * __count_memcg_events - account VM events in a cgroup | |
762 | * @memcg: the memory cgroup | |
763 | * @idx: the event item | |
764 | * @count: the number of events that occured | |
765 | */ | |
766 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
767 | unsigned long count) | |
768 | { | |
769 | unsigned long x; | |
770 | ||
771 | if (mem_cgroup_disabled()) | |
772 | return; | |
773 | ||
774 | x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); | |
775 | if (unlikely(x > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
776 | struct mem_cgroup *mi; |
777 | ||
778 | atomic_long_add(x, &memcg->vmevents_local[idx]); | |
779 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
780 | atomic_long_add(x, &mi->vmevents[idx]); | |
db9adbcb JW |
781 | x = 0; |
782 | } | |
783 | __this_cpu_write(memcg->vmstats_percpu->events[idx], x); | |
784 | } | |
785 | ||
42a30035 | 786 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 787 | { |
871789d4 | 788 | return atomic_long_read(&memcg->vmevents[event]); |
e9f8974f JW |
789 | } |
790 | ||
42a30035 JW |
791 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
792 | { | |
793 | return atomic_long_read(&memcg->vmevents_local[event]); | |
794 | } | |
795 | ||
c0ff4b85 | 796 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 797 | struct page *page, |
f627c2f5 | 798 | bool compound, int nr_pages) |
d52aa412 | 799 | { |
b2402857 KH |
800 | /* |
801 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
802 | * counted as CACHE even if it's on ANON LRU. | |
803 | */ | |
0a31bc97 | 804 | if (PageAnon(page)) |
c9019e9b | 805 | __mod_memcg_state(memcg, MEMCG_RSS, nr_pages); |
9a4caf1e | 806 | else { |
c9019e9b | 807 | __mod_memcg_state(memcg, MEMCG_CACHE, nr_pages); |
9a4caf1e | 808 | if (PageSwapBacked(page)) |
c9019e9b | 809 | __mod_memcg_state(memcg, NR_SHMEM, nr_pages); |
9a4caf1e | 810 | } |
55e462b0 | 811 | |
f627c2f5 KS |
812 | if (compound) { |
813 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
c9019e9b | 814 | __mod_memcg_state(memcg, MEMCG_RSS_HUGE, nr_pages); |
f627c2f5 | 815 | } |
b070e65c | 816 | |
e401f176 KH |
817 | /* pagein of a big page is an event. So, ignore page size */ |
818 | if (nr_pages > 0) | |
c9019e9b | 819 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 820 | else { |
c9019e9b | 821 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
822 | nr_pages = -nr_pages; /* for event */ |
823 | } | |
e401f176 | 824 | |
871789d4 | 825 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
826 | } |
827 | ||
f53d7ce3 JW |
828 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
829 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
830 | { |
831 | unsigned long val, next; | |
832 | ||
871789d4 CD |
833 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
834 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 835 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 836 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
837 | switch (target) { |
838 | case MEM_CGROUP_TARGET_THRESH: | |
839 | next = val + THRESHOLDS_EVENTS_TARGET; | |
840 | break; | |
bb4cc1a8 AM |
841 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
842 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
843 | break; | |
f53d7ce3 JW |
844 | case MEM_CGROUP_TARGET_NUMAINFO: |
845 | next = val + NUMAINFO_EVENTS_TARGET; | |
846 | break; | |
847 | default: | |
848 | break; | |
849 | } | |
871789d4 | 850 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 851 | return true; |
7a159cc9 | 852 | } |
f53d7ce3 | 853 | return false; |
d2265e6f KH |
854 | } |
855 | ||
856 | /* | |
857 | * Check events in order. | |
858 | * | |
859 | */ | |
c0ff4b85 | 860 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
861 | { |
862 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
863 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
864 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 865 | bool do_softlimit; |
82b3f2a7 | 866 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 867 | |
bb4cc1a8 AM |
868 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
869 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
870 | #if MAX_NUMNODES > 1 |
871 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
872 | MEM_CGROUP_TARGET_NUMAINFO); | |
873 | #endif | |
c0ff4b85 | 874 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
875 | if (unlikely(do_softlimit)) |
876 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 877 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 878 | if (unlikely(do_numainfo)) |
c0ff4b85 | 879 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 880 | #endif |
0a31bc97 | 881 | } |
d2265e6f KH |
882 | } |
883 | ||
cf475ad2 | 884 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 885 | { |
31a78f23 BS |
886 | /* |
887 | * mm_update_next_owner() may clear mm->owner to NULL | |
888 | * if it races with swapoff, page migration, etc. | |
889 | * So this can be called with p == NULL. | |
890 | */ | |
891 | if (unlikely(!p)) | |
892 | return NULL; | |
893 | ||
073219e9 | 894 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 895 | } |
33398cf2 | 896 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 897 | |
d46eb14b SB |
898 | /** |
899 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
900 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
901 | * | |
902 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
903 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
904 | * returned. | |
905 | */ | |
906 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 907 | { |
d46eb14b SB |
908 | struct mem_cgroup *memcg; |
909 | ||
910 | if (mem_cgroup_disabled()) | |
911 | return NULL; | |
0b7f569e | 912 | |
54595fe2 KH |
913 | rcu_read_lock(); |
914 | do { | |
6f6acb00 MH |
915 | /* |
916 | * Page cache insertions can happen withou an | |
917 | * actual mm context, e.g. during disk probing | |
918 | * on boot, loopback IO, acct() writes etc. | |
919 | */ | |
920 | if (unlikely(!mm)) | |
df381975 | 921 | memcg = root_mem_cgroup; |
6f6acb00 MH |
922 | else { |
923 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
924 | if (unlikely(!memcg)) | |
925 | memcg = root_mem_cgroup; | |
926 | } | |
ec903c0c | 927 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 928 | rcu_read_unlock(); |
c0ff4b85 | 929 | return memcg; |
54595fe2 | 930 | } |
d46eb14b SB |
931 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
932 | ||
f745c6f5 SB |
933 | /** |
934 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
935 | * @page: page from which memcg should be extracted. | |
936 | * | |
937 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
938 | * root_mem_cgroup is returned. | |
939 | */ | |
940 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
941 | { | |
942 | struct mem_cgroup *memcg = page->mem_cgroup; | |
943 | ||
944 | if (mem_cgroup_disabled()) | |
945 | return NULL; | |
946 | ||
947 | rcu_read_lock(); | |
948 | if (!memcg || !css_tryget_online(&memcg->css)) | |
949 | memcg = root_mem_cgroup; | |
950 | rcu_read_unlock(); | |
951 | return memcg; | |
952 | } | |
953 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
954 | ||
d46eb14b SB |
955 | /** |
956 | * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg. | |
957 | */ | |
958 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
959 | { | |
960 | if (unlikely(current->active_memcg)) { | |
961 | struct mem_cgroup *memcg = root_mem_cgroup; | |
962 | ||
963 | rcu_read_lock(); | |
964 | if (css_tryget_online(¤t->active_memcg->css)) | |
965 | memcg = current->active_memcg; | |
966 | rcu_read_unlock(); | |
967 | return memcg; | |
968 | } | |
969 | return get_mem_cgroup_from_mm(current->mm); | |
970 | } | |
54595fe2 | 971 | |
5660048c JW |
972 | /** |
973 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
974 | * @root: hierarchy root | |
975 | * @prev: previously returned memcg, NULL on first invocation | |
976 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
977 | * | |
978 | * Returns references to children of the hierarchy below @root, or | |
979 | * @root itself, or %NULL after a full round-trip. | |
980 | * | |
981 | * Caller must pass the return value in @prev on subsequent | |
982 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
983 | * to cancel a hierarchy walk before the round-trip is complete. | |
984 | * | |
b213b54f | 985 | * Reclaimers can specify a node and a priority level in @reclaim to |
5660048c | 986 | * divide up the memcgs in the hierarchy among all concurrent |
b213b54f | 987 | * reclaimers operating on the same node and priority. |
5660048c | 988 | */ |
694fbc0f | 989 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 990 | struct mem_cgroup *prev, |
694fbc0f | 991 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 992 | { |
33398cf2 | 993 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 994 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 995 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 996 | struct mem_cgroup *pos = NULL; |
711d3d2c | 997 | |
694fbc0f AM |
998 | if (mem_cgroup_disabled()) |
999 | return NULL; | |
5660048c | 1000 | |
9f3a0d09 JW |
1001 | if (!root) |
1002 | root = root_mem_cgroup; | |
7d74b06f | 1003 | |
9f3a0d09 | 1004 | if (prev && !reclaim) |
5ac8fb31 | 1005 | pos = prev; |
14067bb3 | 1006 | |
9f3a0d09 JW |
1007 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1008 | if (prev) | |
5ac8fb31 | 1009 | goto out; |
694fbc0f | 1010 | return root; |
9f3a0d09 | 1011 | } |
14067bb3 | 1012 | |
542f85f9 | 1013 | rcu_read_lock(); |
5f578161 | 1014 | |
5ac8fb31 | 1015 | if (reclaim) { |
ef8f2327 | 1016 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1017 | |
ef8f2327 | 1018 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
5ac8fb31 JW |
1019 | iter = &mz->iter[reclaim->priority]; |
1020 | ||
1021 | if (prev && reclaim->generation != iter->generation) | |
1022 | goto out_unlock; | |
1023 | ||
6df38689 | 1024 | while (1) { |
4db0c3c2 | 1025 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1026 | if (!pos || css_tryget(&pos->css)) |
1027 | break; | |
5ac8fb31 | 1028 | /* |
6df38689 VD |
1029 | * css reference reached zero, so iter->position will |
1030 | * be cleared by ->css_released. However, we should not | |
1031 | * rely on this happening soon, because ->css_released | |
1032 | * is called from a work queue, and by busy-waiting we | |
1033 | * might block it. So we clear iter->position right | |
1034 | * away. | |
5ac8fb31 | 1035 | */ |
6df38689 VD |
1036 | (void)cmpxchg(&iter->position, pos, NULL); |
1037 | } | |
5ac8fb31 JW |
1038 | } |
1039 | ||
1040 | if (pos) | |
1041 | css = &pos->css; | |
1042 | ||
1043 | for (;;) { | |
1044 | css = css_next_descendant_pre(css, &root->css); | |
1045 | if (!css) { | |
1046 | /* | |
1047 | * Reclaimers share the hierarchy walk, and a | |
1048 | * new one might jump in right at the end of | |
1049 | * the hierarchy - make sure they see at least | |
1050 | * one group and restart from the beginning. | |
1051 | */ | |
1052 | if (!prev) | |
1053 | continue; | |
1054 | break; | |
527a5ec9 | 1055 | } |
7d74b06f | 1056 | |
5ac8fb31 JW |
1057 | /* |
1058 | * Verify the css and acquire a reference. The root | |
1059 | * is provided by the caller, so we know it's alive | |
1060 | * and kicking, and don't take an extra reference. | |
1061 | */ | |
1062 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1063 | |
5ac8fb31 JW |
1064 | if (css == &root->css) |
1065 | break; | |
14067bb3 | 1066 | |
0b8f73e1 JW |
1067 | if (css_tryget(css)) |
1068 | break; | |
9f3a0d09 | 1069 | |
5ac8fb31 | 1070 | memcg = NULL; |
9f3a0d09 | 1071 | } |
5ac8fb31 JW |
1072 | |
1073 | if (reclaim) { | |
5ac8fb31 | 1074 | /* |
6df38689 VD |
1075 | * The position could have already been updated by a competing |
1076 | * thread, so check that the value hasn't changed since we read | |
1077 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1078 | */ |
6df38689 VD |
1079 | (void)cmpxchg(&iter->position, pos, memcg); |
1080 | ||
5ac8fb31 JW |
1081 | if (pos) |
1082 | css_put(&pos->css); | |
1083 | ||
1084 | if (!memcg) | |
1085 | iter->generation++; | |
1086 | else if (!prev) | |
1087 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1088 | } |
5ac8fb31 | 1089 | |
542f85f9 MH |
1090 | out_unlock: |
1091 | rcu_read_unlock(); | |
5ac8fb31 | 1092 | out: |
c40046f3 MH |
1093 | if (prev && prev != root) |
1094 | css_put(&prev->css); | |
1095 | ||
9f3a0d09 | 1096 | return memcg; |
14067bb3 | 1097 | } |
7d74b06f | 1098 | |
5660048c JW |
1099 | /** |
1100 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1101 | * @root: hierarchy root | |
1102 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1103 | */ | |
1104 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1105 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1106 | { |
1107 | if (!root) | |
1108 | root = root_mem_cgroup; | |
1109 | if (prev && prev != root) | |
1110 | css_put(&prev->css); | |
1111 | } | |
7d74b06f | 1112 | |
6df38689 VD |
1113 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1114 | { | |
1115 | struct mem_cgroup *memcg = dead_memcg; | |
1116 | struct mem_cgroup_reclaim_iter *iter; | |
ef8f2327 MG |
1117 | struct mem_cgroup_per_node *mz; |
1118 | int nid; | |
6df38689 VD |
1119 | int i; |
1120 | ||
9f15bde6 | 1121 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
6df38689 | 1122 | for_each_node(nid) { |
ef8f2327 MG |
1123 | mz = mem_cgroup_nodeinfo(memcg, nid); |
1124 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
1125 | iter = &mz->iter[i]; | |
1126 | cmpxchg(&iter->position, | |
1127 | dead_memcg, NULL); | |
6df38689 VD |
1128 | } |
1129 | } | |
1130 | } | |
1131 | } | |
1132 | ||
7c5f64f8 VD |
1133 | /** |
1134 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1135 | * @memcg: hierarchy root | |
1136 | * @fn: function to call for each task | |
1137 | * @arg: argument passed to @fn | |
1138 | * | |
1139 | * This function iterates over tasks attached to @memcg or to any of its | |
1140 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1141 | * value, the function breaks the iteration loop and returns the value. | |
1142 | * Otherwise, it will iterate over all tasks and return 0. | |
1143 | * | |
1144 | * This function must not be called for the root memory cgroup. | |
1145 | */ | |
1146 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1147 | int (*fn)(struct task_struct *, void *), void *arg) | |
1148 | { | |
1149 | struct mem_cgroup *iter; | |
1150 | int ret = 0; | |
1151 | ||
1152 | BUG_ON(memcg == root_mem_cgroup); | |
1153 | ||
1154 | for_each_mem_cgroup_tree(iter, memcg) { | |
1155 | struct css_task_iter it; | |
1156 | struct task_struct *task; | |
1157 | ||
bc2fb7ed | 1158 | css_task_iter_start(&iter->css, 0, &it); |
7c5f64f8 VD |
1159 | while (!ret && (task = css_task_iter_next(&it))) |
1160 | ret = fn(task, arg); | |
1161 | css_task_iter_end(&it); | |
1162 | if (ret) { | |
1163 | mem_cgroup_iter_break(memcg, iter); | |
1164 | break; | |
1165 | } | |
1166 | } | |
1167 | return ret; | |
1168 | } | |
1169 | ||
925b7673 | 1170 | /** |
dfe0e773 | 1171 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1172 | * @page: the page |
f144c390 | 1173 | * @pgdat: pgdat of the page |
dfe0e773 JW |
1174 | * |
1175 | * This function is only safe when following the LRU page isolation | |
1176 | * and putback protocol: the LRU lock must be held, and the page must | |
1177 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1178 | */ |
599d0c95 | 1179 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1180 | { |
ef8f2327 | 1181 | struct mem_cgroup_per_node *mz; |
925b7673 | 1182 | struct mem_cgroup *memcg; |
bea8c150 | 1183 | struct lruvec *lruvec; |
6d12e2d8 | 1184 | |
bea8c150 | 1185 | if (mem_cgroup_disabled()) { |
599d0c95 | 1186 | lruvec = &pgdat->lruvec; |
bea8c150 HD |
1187 | goto out; |
1188 | } | |
925b7673 | 1189 | |
1306a85a | 1190 | memcg = page->mem_cgroup; |
7512102c | 1191 | /* |
dfe0e773 | 1192 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1193 | * possibly migrated - before they are charged. |
7512102c | 1194 | */ |
29833315 JW |
1195 | if (!memcg) |
1196 | memcg = root_mem_cgroup; | |
7512102c | 1197 | |
ef8f2327 | 1198 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1199 | lruvec = &mz->lruvec; |
1200 | out: | |
1201 | /* | |
1202 | * Since a node can be onlined after the mem_cgroup was created, | |
1203 | * we have to be prepared to initialize lruvec->zone here; | |
1204 | * and if offlined then reonlined, we need to reinitialize it. | |
1205 | */ | |
599d0c95 MG |
1206 | if (unlikely(lruvec->pgdat != pgdat)) |
1207 | lruvec->pgdat = pgdat; | |
bea8c150 | 1208 | return lruvec; |
08e552c6 | 1209 | } |
b69408e8 | 1210 | |
925b7673 | 1211 | /** |
fa9add64 HD |
1212 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1213 | * @lruvec: mem_cgroup per zone lru vector | |
1214 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1215 | * @zid: zone id of the accounted pages |
fa9add64 | 1216 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1217 | * |
ca707239 HD |
1218 | * This function must be called under lru_lock, just before a page is added |
1219 | * to or just after a page is removed from an lru list (that ordering being | |
1220 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1221 | */ |
fa9add64 | 1222 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1223 | int zid, int nr_pages) |
3f58a829 | 1224 | { |
ef8f2327 | 1225 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1226 | unsigned long *lru_size; |
ca707239 | 1227 | long size; |
3f58a829 MK |
1228 | |
1229 | if (mem_cgroup_disabled()) | |
1230 | return; | |
1231 | ||
ef8f2327 | 1232 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1233 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1234 | |
1235 | if (nr_pages < 0) | |
1236 | *lru_size += nr_pages; | |
1237 | ||
1238 | size = *lru_size; | |
b4536f0c MH |
1239 | if (WARN_ONCE(size < 0, |
1240 | "%s(%p, %d, %d): lru_size %ld\n", | |
1241 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1242 | VM_BUG_ON(1); |
1243 | *lru_size = 0; | |
1244 | } | |
1245 | ||
1246 | if (nr_pages > 0) | |
1247 | *lru_size += nr_pages; | |
08e552c6 | 1248 | } |
544122e5 | 1249 | |
2314b42d | 1250 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1251 | { |
2314b42d | 1252 | struct mem_cgroup *task_memcg; |
158e0a2d | 1253 | struct task_struct *p; |
ffbdccf5 | 1254 | bool ret; |
4c4a2214 | 1255 | |
158e0a2d | 1256 | p = find_lock_task_mm(task); |
de077d22 | 1257 | if (p) { |
2314b42d | 1258 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1259 | task_unlock(p); |
1260 | } else { | |
1261 | /* | |
1262 | * All threads may have already detached their mm's, but the oom | |
1263 | * killer still needs to detect if they have already been oom | |
1264 | * killed to prevent needlessly killing additional tasks. | |
1265 | */ | |
ffbdccf5 | 1266 | rcu_read_lock(); |
2314b42d JW |
1267 | task_memcg = mem_cgroup_from_task(task); |
1268 | css_get(&task_memcg->css); | |
ffbdccf5 | 1269 | rcu_read_unlock(); |
de077d22 | 1270 | } |
2314b42d JW |
1271 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1272 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1273 | return ret; |
1274 | } | |
1275 | ||
19942822 | 1276 | /** |
9d11ea9f | 1277 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1278 | * @memcg: the memory cgroup |
19942822 | 1279 | * |
9d11ea9f | 1280 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1281 | * pages. |
19942822 | 1282 | */ |
c0ff4b85 | 1283 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1284 | { |
3e32cb2e JW |
1285 | unsigned long margin = 0; |
1286 | unsigned long count; | |
1287 | unsigned long limit; | |
9d11ea9f | 1288 | |
3e32cb2e | 1289 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1290 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1291 | if (count < limit) |
1292 | margin = limit - count; | |
1293 | ||
7941d214 | 1294 | if (do_memsw_account()) { |
3e32cb2e | 1295 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1296 | limit = READ_ONCE(memcg->memsw.max); |
3e32cb2e JW |
1297 | if (count <= limit) |
1298 | margin = min(margin, limit - count); | |
cbedbac3 LR |
1299 | else |
1300 | margin = 0; | |
3e32cb2e JW |
1301 | } |
1302 | ||
1303 | return margin; | |
19942822 JW |
1304 | } |
1305 | ||
32047e2a | 1306 | /* |
bdcbb659 | 1307 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1308 | * |
bdcbb659 QH |
1309 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1310 | * moving cgroups. This is for waiting at high-memory pressure | |
1311 | * caused by "move". | |
32047e2a | 1312 | */ |
c0ff4b85 | 1313 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1314 | { |
2bd9bb20 KH |
1315 | struct mem_cgroup *from; |
1316 | struct mem_cgroup *to; | |
4b534334 | 1317 | bool ret = false; |
2bd9bb20 KH |
1318 | /* |
1319 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1320 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1321 | */ | |
1322 | spin_lock(&mc.lock); | |
1323 | from = mc.from; | |
1324 | to = mc.to; | |
1325 | if (!from) | |
1326 | goto unlock; | |
3e92041d | 1327 | |
2314b42d JW |
1328 | ret = mem_cgroup_is_descendant(from, memcg) || |
1329 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1330 | unlock: |
1331 | spin_unlock(&mc.lock); | |
4b534334 KH |
1332 | return ret; |
1333 | } | |
1334 | ||
c0ff4b85 | 1335 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1336 | { |
1337 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1338 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1339 | DEFINE_WAIT(wait); |
1340 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1341 | /* moving charge context might have finished. */ | |
1342 | if (mc.moving_task) | |
1343 | schedule(); | |
1344 | finish_wait(&mc.waitq, &wait); | |
1345 | return true; | |
1346 | } | |
1347 | } | |
1348 | return false; | |
1349 | } | |
1350 | ||
8ad6e404 | 1351 | static const unsigned int memcg1_stats[] = { |
71cd3113 JW |
1352 | MEMCG_CACHE, |
1353 | MEMCG_RSS, | |
1354 | MEMCG_RSS_HUGE, | |
1355 | NR_SHMEM, | |
1356 | NR_FILE_MAPPED, | |
1357 | NR_FILE_DIRTY, | |
1358 | NR_WRITEBACK, | |
1359 | MEMCG_SWAP, | |
1360 | }; | |
1361 | ||
1362 | static const char *const memcg1_stat_names[] = { | |
1363 | "cache", | |
1364 | "rss", | |
1365 | "rss_huge", | |
1366 | "shmem", | |
1367 | "mapped_file", | |
1368 | "dirty", | |
1369 | "writeback", | |
1370 | "swap", | |
1371 | }; | |
1372 | ||
58cf188e | 1373 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1374 | /** |
f0c867d9 | 1375 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1376 | * memory controller. | |
e222432b BS |
1377 | * @memcg: The memory cgroup that went over limit |
1378 | * @p: Task that is going to be killed | |
1379 | * | |
1380 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1381 | * enabled | |
1382 | */ | |
f0c867d9 | 1383 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1384 | { |
e222432b BS |
1385 | rcu_read_lock(); |
1386 | ||
f0c867d9 | 1387 | if (memcg) { |
1388 | pr_cont(",oom_memcg="); | |
1389 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1390 | } else | |
1391 | pr_cont(",global_oom"); | |
2415b9f5 | 1392 | if (p) { |
f0c867d9 | 1393 | pr_cont(",task_memcg="); |
2415b9f5 | 1394 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1395 | } |
e222432b | 1396 | rcu_read_unlock(); |
f0c867d9 | 1397 | } |
1398 | ||
1399 | /** | |
1400 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1401 | * memory controller. | |
1402 | * @memcg: The memory cgroup that went over limit | |
1403 | */ | |
1404 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1405 | { | |
1406 | struct mem_cgroup *iter; | |
1407 | unsigned int i; | |
e222432b | 1408 | |
3e32cb2e JW |
1409 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1410 | K((u64)page_counter_read(&memcg->memory)), | |
bbec2e15 | 1411 | K((u64)memcg->memory.max), memcg->memory.failcnt); |
3e32cb2e JW |
1412 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", |
1413 | K((u64)page_counter_read(&memcg->memsw)), | |
bbec2e15 | 1414 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); |
3e32cb2e JW |
1415 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", |
1416 | K((u64)page_counter_read(&memcg->kmem)), | |
bbec2e15 | 1417 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); |
58cf188e SZ |
1418 | |
1419 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1420 | pr_info("Memory cgroup stats for "); |
1421 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1422 | pr_cont(":"); |
1423 | ||
71cd3113 JW |
1424 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
1425 | if (memcg1_stats[i] == MEMCG_SWAP && !do_swap_account) | |
58cf188e | 1426 | continue; |
71cd3113 | 1427 | pr_cont(" %s:%luKB", memcg1_stat_names[i], |
205b20cc JW |
1428 | K(memcg_page_state_local(iter, |
1429 | memcg1_stats[i]))); | |
58cf188e SZ |
1430 | } |
1431 | ||
1432 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1433 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
205b20cc JW |
1434 | K(memcg_page_state_local(iter, |
1435 | NR_LRU_BASE + i))); | |
58cf188e SZ |
1436 | |
1437 | pr_cont("\n"); | |
1438 | } | |
e222432b BS |
1439 | } |
1440 | ||
a63d83f4 DR |
1441 | /* |
1442 | * Return the memory (and swap, if configured) limit for a memcg. | |
1443 | */ | |
bbec2e15 | 1444 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1445 | { |
bbec2e15 | 1446 | unsigned long max; |
f3e8eb70 | 1447 | |
bbec2e15 | 1448 | max = memcg->memory.max; |
9a5a8f19 | 1449 | if (mem_cgroup_swappiness(memcg)) { |
bbec2e15 RG |
1450 | unsigned long memsw_max; |
1451 | unsigned long swap_max; | |
9a5a8f19 | 1452 | |
bbec2e15 RG |
1453 | memsw_max = memcg->memsw.max; |
1454 | swap_max = memcg->swap.max; | |
1455 | swap_max = min(swap_max, (unsigned long)total_swap_pages); | |
1456 | max = min(max + swap_max, memsw_max); | |
9a5a8f19 | 1457 | } |
bbec2e15 | 1458 | return max; |
a63d83f4 DR |
1459 | } |
1460 | ||
b6e6edcf | 1461 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1462 | int order) |
9cbb78bb | 1463 | { |
6e0fc46d DR |
1464 | struct oom_control oc = { |
1465 | .zonelist = NULL, | |
1466 | .nodemask = NULL, | |
2a966b77 | 1467 | .memcg = memcg, |
6e0fc46d DR |
1468 | .gfp_mask = gfp_mask, |
1469 | .order = order, | |
6e0fc46d | 1470 | }; |
7c5f64f8 | 1471 | bool ret; |
9cbb78bb | 1472 | |
7775face TH |
1473 | if (mutex_lock_killable(&oom_lock)) |
1474 | return true; | |
1475 | /* | |
1476 | * A few threads which were not waiting at mutex_lock_killable() can | |
1477 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1478 | */ | |
1479 | ret = should_force_charge() || out_of_memory(&oc); | |
dc56401f | 1480 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1481 | return ret; |
9cbb78bb DR |
1482 | } |
1483 | ||
ae6e71d3 MC |
1484 | #if MAX_NUMNODES > 1 |
1485 | ||
4d0c066d KH |
1486 | /** |
1487 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1488 | * @memcg: the target memcg |
4d0c066d KH |
1489 | * @nid: the node ID to be checked. |
1490 | * @noswap : specify true here if the user wants flle only information. | |
1491 | * | |
1492 | * This function returns whether the specified memcg contains any | |
1493 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1494 | * pages in the node. | |
1495 | */ | |
c0ff4b85 | 1496 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1497 | int nid, bool noswap) |
1498 | { | |
2b487e59 JW |
1499 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); |
1500 | ||
def0fdae JW |
1501 | if (lruvec_page_state(lruvec, NR_INACTIVE_FILE) || |
1502 | lruvec_page_state(lruvec, NR_ACTIVE_FILE)) | |
4d0c066d KH |
1503 | return true; |
1504 | if (noswap || !total_swap_pages) | |
1505 | return false; | |
def0fdae JW |
1506 | if (lruvec_page_state(lruvec, NR_INACTIVE_ANON) || |
1507 | lruvec_page_state(lruvec, NR_ACTIVE_ANON)) | |
4d0c066d KH |
1508 | return true; |
1509 | return false; | |
1510 | ||
1511 | } | |
889976db YH |
1512 | |
1513 | /* | |
1514 | * Always updating the nodemask is not very good - even if we have an empty | |
1515 | * list or the wrong list here, we can start from some node and traverse all | |
1516 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1517 | * | |
1518 | */ | |
c0ff4b85 | 1519 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1520 | { |
1521 | int nid; | |
453a9bf3 KH |
1522 | /* |
1523 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1524 | * pagein/pageout changes since the last update. | |
1525 | */ | |
c0ff4b85 | 1526 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1527 | return; |
c0ff4b85 | 1528 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1529 | return; |
1530 | ||
889976db | 1531 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1532 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1533 | |
31aaea4a | 1534 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1535 | |
c0ff4b85 R |
1536 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1537 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1538 | } |
453a9bf3 | 1539 | |
c0ff4b85 R |
1540 | atomic_set(&memcg->numainfo_events, 0); |
1541 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1542 | } |
1543 | ||
1544 | /* | |
1545 | * Selecting a node where we start reclaim from. Because what we need is just | |
1546 | * reducing usage counter, start from anywhere is O,K. Considering | |
1547 | * memory reclaim from current node, there are pros. and cons. | |
1548 | * | |
1549 | * Freeing memory from current node means freeing memory from a node which | |
1550 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1551 | * hit limits, it will see a contention on a node. But freeing from remote | |
1552 | * node means more costs for memory reclaim because of memory latency. | |
1553 | * | |
1554 | * Now, we use round-robin. Better algorithm is welcomed. | |
1555 | */ | |
c0ff4b85 | 1556 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1557 | { |
1558 | int node; | |
1559 | ||
c0ff4b85 R |
1560 | mem_cgroup_may_update_nodemask(memcg); |
1561 | node = memcg->last_scanned_node; | |
889976db | 1562 | |
0edaf86c | 1563 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1564 | /* |
fda3d69b MH |
1565 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1566 | * last time it really checked all the LRUs due to rate limiting. | |
1567 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1568 | */ |
1569 | if (unlikely(node == MAX_NUMNODES)) | |
1570 | node = numa_node_id(); | |
1571 | ||
c0ff4b85 | 1572 | memcg->last_scanned_node = node; |
889976db YH |
1573 | return node; |
1574 | } | |
889976db | 1575 | #else |
c0ff4b85 | 1576 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1577 | { |
1578 | return 0; | |
1579 | } | |
1580 | #endif | |
1581 | ||
0608f43d | 1582 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1583 | pg_data_t *pgdat, |
0608f43d AM |
1584 | gfp_t gfp_mask, |
1585 | unsigned long *total_scanned) | |
1586 | { | |
1587 | struct mem_cgroup *victim = NULL; | |
1588 | int total = 0; | |
1589 | int loop = 0; | |
1590 | unsigned long excess; | |
1591 | unsigned long nr_scanned; | |
1592 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1593 | .pgdat = pgdat, |
0608f43d AM |
1594 | .priority = 0, |
1595 | }; | |
1596 | ||
3e32cb2e | 1597 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1598 | |
1599 | while (1) { | |
1600 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1601 | if (!victim) { | |
1602 | loop++; | |
1603 | if (loop >= 2) { | |
1604 | /* | |
1605 | * If we have not been able to reclaim | |
1606 | * anything, it might because there are | |
1607 | * no reclaimable pages under this hierarchy | |
1608 | */ | |
1609 | if (!total) | |
1610 | break; | |
1611 | /* | |
1612 | * We want to do more targeted reclaim. | |
1613 | * excess >> 2 is not to excessive so as to | |
1614 | * reclaim too much, nor too less that we keep | |
1615 | * coming back to reclaim from this cgroup | |
1616 | */ | |
1617 | if (total >= (excess >> 2) || | |
1618 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1619 | break; | |
1620 | } | |
1621 | continue; | |
1622 | } | |
a9dd0a83 | 1623 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1624 | pgdat, &nr_scanned); |
0608f43d | 1625 | *total_scanned += nr_scanned; |
3e32cb2e | 1626 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1627 | break; |
6d61ef40 | 1628 | } |
0608f43d AM |
1629 | mem_cgroup_iter_break(root_memcg, victim); |
1630 | return total; | |
6d61ef40 BS |
1631 | } |
1632 | ||
0056f4e6 JW |
1633 | #ifdef CONFIG_LOCKDEP |
1634 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1635 | .name = "memcg_oom_lock", | |
1636 | }; | |
1637 | #endif | |
1638 | ||
fb2a6fc5 JW |
1639 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1640 | ||
867578cb KH |
1641 | /* |
1642 | * Check OOM-Killer is already running under our hierarchy. | |
1643 | * If someone is running, return false. | |
1644 | */ | |
fb2a6fc5 | 1645 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1646 | { |
79dfdacc | 1647 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1648 | |
fb2a6fc5 JW |
1649 | spin_lock(&memcg_oom_lock); |
1650 | ||
9f3a0d09 | 1651 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1652 | if (iter->oom_lock) { |
79dfdacc MH |
1653 | /* |
1654 | * this subtree of our hierarchy is already locked | |
1655 | * so we cannot give a lock. | |
1656 | */ | |
79dfdacc | 1657 | failed = iter; |
9f3a0d09 JW |
1658 | mem_cgroup_iter_break(memcg, iter); |
1659 | break; | |
23751be0 JW |
1660 | } else |
1661 | iter->oom_lock = true; | |
7d74b06f | 1662 | } |
867578cb | 1663 | |
fb2a6fc5 JW |
1664 | if (failed) { |
1665 | /* | |
1666 | * OK, we failed to lock the whole subtree so we have | |
1667 | * to clean up what we set up to the failing subtree | |
1668 | */ | |
1669 | for_each_mem_cgroup_tree(iter, memcg) { | |
1670 | if (iter == failed) { | |
1671 | mem_cgroup_iter_break(memcg, iter); | |
1672 | break; | |
1673 | } | |
1674 | iter->oom_lock = false; | |
79dfdacc | 1675 | } |
0056f4e6 JW |
1676 | } else |
1677 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1678 | |
1679 | spin_unlock(&memcg_oom_lock); | |
1680 | ||
1681 | return !failed; | |
a636b327 | 1682 | } |
0b7f569e | 1683 | |
fb2a6fc5 | 1684 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1685 | { |
7d74b06f KH |
1686 | struct mem_cgroup *iter; |
1687 | ||
fb2a6fc5 | 1688 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1689 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1690 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1691 | iter->oom_lock = false; |
fb2a6fc5 | 1692 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1693 | } |
1694 | ||
c0ff4b85 | 1695 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1696 | { |
1697 | struct mem_cgroup *iter; | |
1698 | ||
c2b42d3c | 1699 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1700 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1701 | iter->under_oom++; |
1702 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1703 | } |
1704 | ||
c0ff4b85 | 1705 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1706 | { |
1707 | struct mem_cgroup *iter; | |
1708 | ||
867578cb KH |
1709 | /* |
1710 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1711 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1712 | */ |
c2b42d3c | 1713 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1714 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1715 | if (iter->under_oom > 0) |
1716 | iter->under_oom--; | |
1717 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1718 | } |
1719 | ||
867578cb KH |
1720 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1721 | ||
dc98df5a | 1722 | struct oom_wait_info { |
d79154bb | 1723 | struct mem_cgroup *memcg; |
ac6424b9 | 1724 | wait_queue_entry_t wait; |
dc98df5a KH |
1725 | }; |
1726 | ||
ac6424b9 | 1727 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1728 | unsigned mode, int sync, void *arg) |
1729 | { | |
d79154bb HD |
1730 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1731 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1732 | struct oom_wait_info *oom_wait_info; |
1733 | ||
1734 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1735 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1736 | |
2314b42d JW |
1737 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1738 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1739 | return 0; |
dc98df5a KH |
1740 | return autoremove_wake_function(wait, mode, sync, arg); |
1741 | } | |
1742 | ||
c0ff4b85 | 1743 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1744 | { |
c2b42d3c TH |
1745 | /* |
1746 | * For the following lockless ->under_oom test, the only required | |
1747 | * guarantee is that it must see the state asserted by an OOM when | |
1748 | * this function is called as a result of userland actions | |
1749 | * triggered by the notification of the OOM. This is trivially | |
1750 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1751 | * triggering notification. | |
1752 | */ | |
1753 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1754 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1755 | } |
1756 | ||
29ef680a MH |
1757 | enum oom_status { |
1758 | OOM_SUCCESS, | |
1759 | OOM_FAILED, | |
1760 | OOM_ASYNC, | |
1761 | OOM_SKIPPED | |
1762 | }; | |
1763 | ||
1764 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1765 | { |
7056d3a3 MH |
1766 | enum oom_status ret; |
1767 | bool locked; | |
1768 | ||
29ef680a MH |
1769 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1770 | return OOM_SKIPPED; | |
1771 | ||
7a1adfdd RG |
1772 | memcg_memory_event(memcg, MEMCG_OOM); |
1773 | ||
867578cb | 1774 | /* |
49426420 JW |
1775 | * We are in the middle of the charge context here, so we |
1776 | * don't want to block when potentially sitting on a callstack | |
1777 | * that holds all kinds of filesystem and mm locks. | |
1778 | * | |
29ef680a MH |
1779 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1780 | * handling until the charge can succeed; remember the context and put | |
1781 | * the task to sleep at the end of the page fault when all locks are | |
1782 | * released. | |
49426420 | 1783 | * |
29ef680a MH |
1784 | * On the other hand, in-kernel OOM killer allows for an async victim |
1785 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1786 | * relying on the oom victim to make a forward progress and we can | |
1787 | * invoke the oom killer here. | |
1788 | * | |
1789 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1790 | * victim and then we have to bail out from the charge path. | |
867578cb | 1791 | */ |
29ef680a MH |
1792 | if (memcg->oom_kill_disable) { |
1793 | if (!current->in_user_fault) | |
1794 | return OOM_SKIPPED; | |
1795 | css_get(&memcg->css); | |
1796 | current->memcg_in_oom = memcg; | |
1797 | current->memcg_oom_gfp_mask = mask; | |
1798 | current->memcg_oom_order = order; | |
1799 | ||
1800 | return OOM_ASYNC; | |
1801 | } | |
1802 | ||
7056d3a3 MH |
1803 | mem_cgroup_mark_under_oom(memcg); |
1804 | ||
1805 | locked = mem_cgroup_oom_trylock(memcg); | |
1806 | ||
1807 | if (locked) | |
1808 | mem_cgroup_oom_notify(memcg); | |
1809 | ||
1810 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1811 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1812 | ret = OOM_SUCCESS; |
1813 | else | |
1814 | ret = OOM_FAILED; | |
1815 | ||
1816 | if (locked) | |
1817 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1818 | |
7056d3a3 | 1819 | return ret; |
3812c8c8 JW |
1820 | } |
1821 | ||
1822 | /** | |
1823 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1824 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1825 | * |
49426420 JW |
1826 | * This has to be called at the end of a page fault if the memcg OOM |
1827 | * handler was enabled. | |
3812c8c8 | 1828 | * |
49426420 | 1829 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1830 | * sleep on a waitqueue until the userspace task resolves the |
1831 | * situation. Sleeping directly in the charge context with all kinds | |
1832 | * of locks held is not a good idea, instead we remember an OOM state | |
1833 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1834 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1835 | * |
1836 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1837 | * completed, %false otherwise. |
3812c8c8 | 1838 | */ |
49426420 | 1839 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1840 | { |
626ebc41 | 1841 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1842 | struct oom_wait_info owait; |
49426420 | 1843 | bool locked; |
3812c8c8 JW |
1844 | |
1845 | /* OOM is global, do not handle */ | |
3812c8c8 | 1846 | if (!memcg) |
49426420 | 1847 | return false; |
3812c8c8 | 1848 | |
7c5f64f8 | 1849 | if (!handle) |
49426420 | 1850 | goto cleanup; |
3812c8c8 JW |
1851 | |
1852 | owait.memcg = memcg; | |
1853 | owait.wait.flags = 0; | |
1854 | owait.wait.func = memcg_oom_wake_function; | |
1855 | owait.wait.private = current; | |
2055da97 | 1856 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1857 | |
3812c8c8 | 1858 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1859 | mem_cgroup_mark_under_oom(memcg); |
1860 | ||
1861 | locked = mem_cgroup_oom_trylock(memcg); | |
1862 | ||
1863 | if (locked) | |
1864 | mem_cgroup_oom_notify(memcg); | |
1865 | ||
1866 | if (locked && !memcg->oom_kill_disable) { | |
1867 | mem_cgroup_unmark_under_oom(memcg); | |
1868 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1869 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1870 | current->memcg_oom_order); | |
49426420 | 1871 | } else { |
3812c8c8 | 1872 | schedule(); |
49426420 JW |
1873 | mem_cgroup_unmark_under_oom(memcg); |
1874 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1875 | } | |
1876 | ||
1877 | if (locked) { | |
fb2a6fc5 JW |
1878 | mem_cgroup_oom_unlock(memcg); |
1879 | /* | |
1880 | * There is no guarantee that an OOM-lock contender | |
1881 | * sees the wakeups triggered by the OOM kill | |
1882 | * uncharges. Wake any sleepers explicitely. | |
1883 | */ | |
1884 | memcg_oom_recover(memcg); | |
1885 | } | |
49426420 | 1886 | cleanup: |
626ebc41 | 1887 | current->memcg_in_oom = NULL; |
3812c8c8 | 1888 | css_put(&memcg->css); |
867578cb | 1889 | return true; |
0b7f569e KH |
1890 | } |
1891 | ||
3d8b38eb RG |
1892 | /** |
1893 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
1894 | * @victim: task to be killed by the OOM killer | |
1895 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
1896 | * | |
1897 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
1898 | * by killing all belonging OOM-killable tasks. | |
1899 | * | |
1900 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
1901 | */ | |
1902 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
1903 | struct mem_cgroup *oom_domain) | |
1904 | { | |
1905 | struct mem_cgroup *oom_group = NULL; | |
1906 | struct mem_cgroup *memcg; | |
1907 | ||
1908 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
1909 | return NULL; | |
1910 | ||
1911 | if (!oom_domain) | |
1912 | oom_domain = root_mem_cgroup; | |
1913 | ||
1914 | rcu_read_lock(); | |
1915 | ||
1916 | memcg = mem_cgroup_from_task(victim); | |
1917 | if (memcg == root_mem_cgroup) | |
1918 | goto out; | |
1919 | ||
1920 | /* | |
1921 | * Traverse the memory cgroup hierarchy from the victim task's | |
1922 | * cgroup up to the OOMing cgroup (or root) to find the | |
1923 | * highest-level memory cgroup with oom.group set. | |
1924 | */ | |
1925 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
1926 | if (memcg->oom_group) | |
1927 | oom_group = memcg; | |
1928 | ||
1929 | if (memcg == oom_domain) | |
1930 | break; | |
1931 | } | |
1932 | ||
1933 | if (oom_group) | |
1934 | css_get(&oom_group->css); | |
1935 | out: | |
1936 | rcu_read_unlock(); | |
1937 | ||
1938 | return oom_group; | |
1939 | } | |
1940 | ||
1941 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
1942 | { | |
1943 | pr_info("Tasks in "); | |
1944 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1945 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
1946 | } | |
1947 | ||
d7365e78 | 1948 | /** |
81f8c3a4 JW |
1949 | * lock_page_memcg - lock a page->mem_cgroup binding |
1950 | * @page: the page | |
32047e2a | 1951 | * |
81f8c3a4 | 1952 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
1953 | * another cgroup. |
1954 | * | |
1955 | * It ensures lifetime of the returned memcg. Caller is responsible | |
1956 | * for the lifetime of the page; __unlock_page_memcg() is available | |
1957 | * when @page might get freed inside the locked section. | |
d69b042f | 1958 | */ |
739f79fc | 1959 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 KH |
1960 | { |
1961 | struct mem_cgroup *memcg; | |
6de22619 | 1962 | unsigned long flags; |
89c06bd5 | 1963 | |
6de22619 JW |
1964 | /* |
1965 | * The RCU lock is held throughout the transaction. The fast | |
1966 | * path can get away without acquiring the memcg->move_lock | |
1967 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
1968 | * |
1969 | * The RCU lock also protects the memcg from being freed when | |
1970 | * the page state that is going to change is the only thing | |
1971 | * preventing the page itself from being freed. E.g. writeback | |
1972 | * doesn't hold a page reference and relies on PG_writeback to | |
1973 | * keep off truncation, migration and so forth. | |
1974 | */ | |
d7365e78 JW |
1975 | rcu_read_lock(); |
1976 | ||
1977 | if (mem_cgroup_disabled()) | |
739f79fc | 1978 | return NULL; |
89c06bd5 | 1979 | again: |
1306a85a | 1980 | memcg = page->mem_cgroup; |
29833315 | 1981 | if (unlikely(!memcg)) |
739f79fc | 1982 | return NULL; |
d7365e78 | 1983 | |
bdcbb659 | 1984 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 1985 | return memcg; |
89c06bd5 | 1986 | |
6de22619 | 1987 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1988 | if (memcg != page->mem_cgroup) { |
6de22619 | 1989 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1990 | goto again; |
1991 | } | |
6de22619 JW |
1992 | |
1993 | /* | |
1994 | * When charge migration first begins, we can have locked and | |
1995 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 1996 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
1997 | */ |
1998 | memcg->move_lock_task = current; | |
1999 | memcg->move_lock_flags = flags; | |
d7365e78 | 2000 | |
739f79fc | 2001 | return memcg; |
89c06bd5 | 2002 | } |
81f8c3a4 | 2003 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2004 | |
d7365e78 | 2005 | /** |
739f79fc JW |
2006 | * __unlock_page_memcg - unlock and unpin a memcg |
2007 | * @memcg: the memcg | |
2008 | * | |
2009 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2010 | */ |
739f79fc | 2011 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2012 | { |
6de22619 JW |
2013 | if (memcg && memcg->move_lock_task == current) { |
2014 | unsigned long flags = memcg->move_lock_flags; | |
2015 | ||
2016 | memcg->move_lock_task = NULL; | |
2017 | memcg->move_lock_flags = 0; | |
2018 | ||
2019 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2020 | } | |
89c06bd5 | 2021 | |
d7365e78 | 2022 | rcu_read_unlock(); |
89c06bd5 | 2023 | } |
739f79fc JW |
2024 | |
2025 | /** | |
2026 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
2027 | * @page: the page | |
2028 | */ | |
2029 | void unlock_page_memcg(struct page *page) | |
2030 | { | |
2031 | __unlock_page_memcg(page->mem_cgroup); | |
2032 | } | |
81f8c3a4 | 2033 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2034 | |
cdec2e42 KH |
2035 | struct memcg_stock_pcp { |
2036 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2037 | unsigned int nr_pages; |
cdec2e42 | 2038 | struct work_struct work; |
26fe6168 | 2039 | unsigned long flags; |
a0db00fc | 2040 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2041 | }; |
2042 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2043 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2044 | |
a0956d54 SS |
2045 | /** |
2046 | * consume_stock: Try to consume stocked charge on this cpu. | |
2047 | * @memcg: memcg to consume from. | |
2048 | * @nr_pages: how many pages to charge. | |
2049 | * | |
2050 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2051 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2052 | * service an allocation will refill the stock. | |
2053 | * | |
2054 | * returns true if successful, false otherwise. | |
cdec2e42 | 2055 | */ |
a0956d54 | 2056 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2057 | { |
2058 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2059 | unsigned long flags; |
3e32cb2e | 2060 | bool ret = false; |
cdec2e42 | 2061 | |
a983b5eb | 2062 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2063 | return ret; |
a0956d54 | 2064 | |
db2ba40c JW |
2065 | local_irq_save(flags); |
2066 | ||
2067 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2068 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2069 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2070 | ret = true; |
2071 | } | |
db2ba40c JW |
2072 | |
2073 | local_irq_restore(flags); | |
2074 | ||
cdec2e42 KH |
2075 | return ret; |
2076 | } | |
2077 | ||
2078 | /* | |
3e32cb2e | 2079 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2080 | */ |
2081 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2082 | { | |
2083 | struct mem_cgroup *old = stock->cached; | |
2084 | ||
11c9ea4e | 2085 | if (stock->nr_pages) { |
3e32cb2e | 2086 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2087 | if (do_memsw_account()) |
3e32cb2e | 2088 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2089 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2090 | stock->nr_pages = 0; |
cdec2e42 KH |
2091 | } |
2092 | stock->cached = NULL; | |
cdec2e42 KH |
2093 | } |
2094 | ||
cdec2e42 KH |
2095 | static void drain_local_stock(struct work_struct *dummy) |
2096 | { | |
db2ba40c JW |
2097 | struct memcg_stock_pcp *stock; |
2098 | unsigned long flags; | |
2099 | ||
72f0184c MH |
2100 | /* |
2101 | * The only protection from memory hotplug vs. drain_stock races is | |
2102 | * that we always operate on local CPU stock here with IRQ disabled | |
2103 | */ | |
db2ba40c JW |
2104 | local_irq_save(flags); |
2105 | ||
2106 | stock = this_cpu_ptr(&memcg_stock); | |
cdec2e42 | 2107 | drain_stock(stock); |
26fe6168 | 2108 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2109 | |
2110 | local_irq_restore(flags); | |
cdec2e42 KH |
2111 | } |
2112 | ||
2113 | /* | |
3e32cb2e | 2114 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2115 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2116 | */ |
c0ff4b85 | 2117 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2118 | { |
db2ba40c JW |
2119 | struct memcg_stock_pcp *stock; |
2120 | unsigned long flags; | |
2121 | ||
2122 | local_irq_save(flags); | |
cdec2e42 | 2123 | |
db2ba40c | 2124 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2125 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2126 | drain_stock(stock); |
c0ff4b85 | 2127 | stock->cached = memcg; |
cdec2e42 | 2128 | } |
11c9ea4e | 2129 | stock->nr_pages += nr_pages; |
db2ba40c | 2130 | |
a983b5eb | 2131 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2132 | drain_stock(stock); |
2133 | ||
db2ba40c | 2134 | local_irq_restore(flags); |
cdec2e42 KH |
2135 | } |
2136 | ||
2137 | /* | |
c0ff4b85 | 2138 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2139 | * of the hierarchy under it. |
cdec2e42 | 2140 | */ |
6d3d6aa2 | 2141 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2142 | { |
26fe6168 | 2143 | int cpu, curcpu; |
d38144b7 | 2144 | |
6d3d6aa2 JW |
2145 | /* If someone's already draining, avoid adding running more workers. */ |
2146 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2147 | return; | |
72f0184c MH |
2148 | /* |
2149 | * Notify other cpus that system-wide "drain" is running | |
2150 | * We do not care about races with the cpu hotplug because cpu down | |
2151 | * as well as workers from this path always operate on the local | |
2152 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2153 | */ | |
5af12d0e | 2154 | curcpu = get_cpu(); |
cdec2e42 KH |
2155 | for_each_online_cpu(cpu) { |
2156 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2157 | struct mem_cgroup *memcg; |
26fe6168 | 2158 | |
c0ff4b85 | 2159 | memcg = stock->cached; |
72f0184c | 2160 | if (!memcg || !stock->nr_pages || !css_tryget(&memcg->css)) |
26fe6168 | 2161 | continue; |
72f0184c MH |
2162 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) { |
2163 | css_put(&memcg->css); | |
3e92041d | 2164 | continue; |
72f0184c | 2165 | } |
d1a05b69 MH |
2166 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2167 | if (cpu == curcpu) | |
2168 | drain_local_stock(&stock->work); | |
2169 | else | |
2170 | schedule_work_on(cpu, &stock->work); | |
2171 | } | |
72f0184c | 2172 | css_put(&memcg->css); |
cdec2e42 | 2173 | } |
5af12d0e | 2174 | put_cpu(); |
9f50fad6 | 2175 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2176 | } |
2177 | ||
308167fc | 2178 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2179 | { |
cdec2e42 | 2180 | struct memcg_stock_pcp *stock; |
42a30035 | 2181 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2182 | |
cdec2e42 KH |
2183 | stock = &per_cpu(memcg_stock, cpu); |
2184 | drain_stock(stock); | |
a983b5eb JW |
2185 | |
2186 | for_each_mem_cgroup(memcg) { | |
2187 | int i; | |
2188 | ||
2189 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2190 | int nid; | |
2191 | long x; | |
2192 | ||
871789d4 | 2193 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
42a30035 JW |
2194 | if (x) { |
2195 | atomic_long_add(x, &memcg->vmstats_local[i]); | |
2196 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
2197 | atomic_long_add(x, &memcg->vmstats[i]); | |
2198 | } | |
a983b5eb JW |
2199 | |
2200 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2201 | continue; | |
2202 | ||
2203 | for_each_node(nid) { | |
2204 | struct mem_cgroup_per_node *pn; | |
2205 | ||
2206 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2207 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
42a30035 JW |
2208 | if (x) { |
2209 | atomic_long_add(x, &pn->lruvec_stat_local[i]); | |
2210 | do { | |
2211 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2212 | } while ((pn = parent_nodeinfo(pn, nid))); | |
2213 | } | |
a983b5eb JW |
2214 | } |
2215 | } | |
2216 | ||
e27be240 | 2217 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2218 | long x; |
2219 | ||
871789d4 | 2220 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
42a30035 JW |
2221 | if (x) { |
2222 | atomic_long_add(x, &memcg->vmevents_local[i]); | |
2223 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
2224 | atomic_long_add(x, &memcg->vmevents[i]); | |
2225 | } | |
a983b5eb JW |
2226 | } |
2227 | } | |
2228 | ||
308167fc | 2229 | return 0; |
cdec2e42 KH |
2230 | } |
2231 | ||
f7e1cb6e JW |
2232 | static void reclaim_high(struct mem_cgroup *memcg, |
2233 | unsigned int nr_pages, | |
2234 | gfp_t gfp_mask) | |
2235 | { | |
2236 | do { | |
2237 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
2238 | continue; | |
e27be240 | 2239 | memcg_memory_event(memcg, MEMCG_HIGH); |
f7e1cb6e JW |
2240 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
2241 | } while ((memcg = parent_mem_cgroup(memcg))); | |
2242 | } | |
2243 | ||
2244 | static void high_work_func(struct work_struct *work) | |
2245 | { | |
2246 | struct mem_cgroup *memcg; | |
2247 | ||
2248 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2249 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2250 | } |
2251 | ||
b23afb93 TH |
2252 | /* |
2253 | * Scheduled by try_charge() to be executed from the userland return path | |
2254 | * and reclaims memory over the high limit. | |
2255 | */ | |
2256 | void mem_cgroup_handle_over_high(void) | |
2257 | { | |
2258 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 2259 | struct mem_cgroup *memcg; |
b23afb93 TH |
2260 | |
2261 | if (likely(!nr_pages)) | |
2262 | return; | |
2263 | ||
f7e1cb6e JW |
2264 | memcg = get_mem_cgroup_from_mm(current->mm); |
2265 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
2266 | css_put(&memcg->css); |
2267 | current->memcg_nr_pages_over_high = 0; | |
2268 | } | |
2269 | ||
00501b53 JW |
2270 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2271 | unsigned int nr_pages) | |
8a9f3ccd | 2272 | { |
a983b5eb | 2273 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
9b130619 | 2274 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2275 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2276 | struct page_counter *counter; |
6539cc05 | 2277 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2278 | bool may_swap = true; |
2279 | bool drained = false; | |
29ef680a MH |
2280 | bool oomed = false; |
2281 | enum oom_status oom_status; | |
a636b327 | 2282 | |
ce00a967 | 2283 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2284 | return 0; |
6539cc05 | 2285 | retry: |
b6b6cc72 | 2286 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2287 | return 0; |
8a9f3ccd | 2288 | |
7941d214 | 2289 | if (!do_memsw_account() || |
6071ca52 JW |
2290 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2291 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2292 | goto done_restock; |
7941d214 | 2293 | if (do_memsw_account()) |
3e32cb2e JW |
2294 | page_counter_uncharge(&memcg->memsw, batch); |
2295 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2296 | } else { |
3e32cb2e | 2297 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2298 | may_swap = false; |
3fbe7244 | 2299 | } |
7a81b88c | 2300 | |
6539cc05 JW |
2301 | if (batch > nr_pages) { |
2302 | batch = nr_pages; | |
2303 | goto retry; | |
2304 | } | |
6d61ef40 | 2305 | |
06b078fc JW |
2306 | /* |
2307 | * Unlike in global OOM situations, memcg is not in a physical | |
2308 | * memory shortage. Allow dying and OOM-killed tasks to | |
2309 | * bypass the last charges so that they can exit quickly and | |
2310 | * free their memory. | |
2311 | */ | |
7775face | 2312 | if (unlikely(should_force_charge())) |
10d53c74 | 2313 | goto force; |
06b078fc | 2314 | |
89a28483 JW |
2315 | /* |
2316 | * Prevent unbounded recursion when reclaim operations need to | |
2317 | * allocate memory. This might exceed the limits temporarily, | |
2318 | * but we prefer facilitating memory reclaim and getting back | |
2319 | * under the limit over triggering OOM kills in these cases. | |
2320 | */ | |
2321 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2322 | goto force; | |
2323 | ||
06b078fc JW |
2324 | if (unlikely(task_in_memcg_oom(current))) |
2325 | goto nomem; | |
2326 | ||
d0164adc | 2327 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2328 | goto nomem; |
4b534334 | 2329 | |
e27be240 | 2330 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2331 | |
b70a2a21 JW |
2332 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2333 | gfp_mask, may_swap); | |
6539cc05 | 2334 | |
61e02c74 | 2335 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2336 | goto retry; |
28c34c29 | 2337 | |
b70a2a21 | 2338 | if (!drained) { |
6d3d6aa2 | 2339 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2340 | drained = true; |
2341 | goto retry; | |
2342 | } | |
2343 | ||
28c34c29 JW |
2344 | if (gfp_mask & __GFP_NORETRY) |
2345 | goto nomem; | |
6539cc05 JW |
2346 | /* |
2347 | * Even though the limit is exceeded at this point, reclaim | |
2348 | * may have been able to free some pages. Retry the charge | |
2349 | * before killing the task. | |
2350 | * | |
2351 | * Only for regular pages, though: huge pages are rather | |
2352 | * unlikely to succeed so close to the limit, and we fall back | |
2353 | * to regular pages anyway in case of failure. | |
2354 | */ | |
61e02c74 | 2355 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2356 | goto retry; |
2357 | /* | |
2358 | * At task move, charge accounts can be doubly counted. So, it's | |
2359 | * better to wait until the end of task_move if something is going on. | |
2360 | */ | |
2361 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2362 | goto retry; | |
2363 | ||
9b130619 JW |
2364 | if (nr_retries--) |
2365 | goto retry; | |
2366 | ||
29ef680a MH |
2367 | if (gfp_mask & __GFP_RETRY_MAYFAIL && oomed) |
2368 | goto nomem; | |
2369 | ||
06b078fc | 2370 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2371 | goto force; |
06b078fc | 2372 | |
6539cc05 | 2373 | if (fatal_signal_pending(current)) |
10d53c74 | 2374 | goto force; |
6539cc05 | 2375 | |
29ef680a MH |
2376 | /* |
2377 | * keep retrying as long as the memcg oom killer is able to make | |
2378 | * a forward progress or bypass the charge if the oom killer | |
2379 | * couldn't make any progress. | |
2380 | */ | |
2381 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2382 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2383 | switch (oom_status) { |
2384 | case OOM_SUCCESS: | |
2385 | nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
2386 | oomed = true; | |
2387 | goto retry; | |
2388 | case OOM_FAILED: | |
2389 | goto force; | |
2390 | default: | |
2391 | goto nomem; | |
2392 | } | |
7a81b88c | 2393 | nomem: |
6d1fdc48 | 2394 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2395 | return -ENOMEM; |
10d53c74 TH |
2396 | force: |
2397 | /* | |
2398 | * The allocation either can't fail or will lead to more memory | |
2399 | * being freed very soon. Allow memory usage go over the limit | |
2400 | * temporarily by force charging it. | |
2401 | */ | |
2402 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2403 | if (do_memsw_account()) |
10d53c74 TH |
2404 | page_counter_charge(&memcg->memsw, nr_pages); |
2405 | css_get_many(&memcg->css, nr_pages); | |
2406 | ||
2407 | return 0; | |
6539cc05 JW |
2408 | |
2409 | done_restock: | |
e8ea14cc | 2410 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2411 | if (batch > nr_pages) |
2412 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2413 | |
241994ed | 2414 | /* |
b23afb93 TH |
2415 | * If the hierarchy is above the normal consumption range, schedule |
2416 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2417 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2418 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2419 | * not recorded as it most likely matches current's and won't | |
2420 | * change in the meantime. As high limit is checked again before | |
2421 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2422 | */ |
2423 | do { | |
b23afb93 | 2424 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2425 | /* Don't bother a random interrupted task */ |
2426 | if (in_interrupt()) { | |
2427 | schedule_work(&memcg->high_work); | |
2428 | break; | |
2429 | } | |
9516a18a | 2430 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2431 | set_notify_resume(current); |
2432 | break; | |
2433 | } | |
241994ed | 2434 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2435 | |
2436 | return 0; | |
7a81b88c | 2437 | } |
8a9f3ccd | 2438 | |
00501b53 | 2439 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2440 | { |
ce00a967 JW |
2441 | if (mem_cgroup_is_root(memcg)) |
2442 | return; | |
2443 | ||
3e32cb2e | 2444 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2445 | if (do_memsw_account()) |
3e32cb2e | 2446 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2447 | |
e8ea14cc | 2448 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2449 | } |
2450 | ||
0a31bc97 JW |
2451 | static void lock_page_lru(struct page *page, int *isolated) |
2452 | { | |
f4b7e272 | 2453 | pg_data_t *pgdat = page_pgdat(page); |
0a31bc97 | 2454 | |
f4b7e272 | 2455 | spin_lock_irq(&pgdat->lru_lock); |
0a31bc97 JW |
2456 | if (PageLRU(page)) { |
2457 | struct lruvec *lruvec; | |
2458 | ||
f4b7e272 | 2459 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
0a31bc97 JW |
2460 | ClearPageLRU(page); |
2461 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2462 | *isolated = 1; | |
2463 | } else | |
2464 | *isolated = 0; | |
2465 | } | |
2466 | ||
2467 | static void unlock_page_lru(struct page *page, int isolated) | |
2468 | { | |
f4b7e272 | 2469 | pg_data_t *pgdat = page_pgdat(page); |
0a31bc97 JW |
2470 | |
2471 | if (isolated) { | |
2472 | struct lruvec *lruvec; | |
2473 | ||
f4b7e272 | 2474 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
0a31bc97 JW |
2475 | VM_BUG_ON_PAGE(PageLRU(page), page); |
2476 | SetPageLRU(page); | |
2477 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2478 | } | |
f4b7e272 | 2479 | spin_unlock_irq(&pgdat->lru_lock); |
0a31bc97 JW |
2480 | } |
2481 | ||
00501b53 | 2482 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2483 | bool lrucare) |
7a81b88c | 2484 | { |
0a31bc97 | 2485 | int isolated; |
9ce70c02 | 2486 | |
1306a85a | 2487 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2488 | |
2489 | /* | |
2490 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2491 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2492 | */ | |
0a31bc97 JW |
2493 | if (lrucare) |
2494 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2495 | |
0a31bc97 JW |
2496 | /* |
2497 | * Nobody should be changing or seriously looking at | |
1306a85a | 2498 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2499 | * |
2500 | * - the page is uncharged | |
2501 | * | |
2502 | * - the page is off-LRU | |
2503 | * | |
2504 | * - an anonymous fault has exclusive page access, except for | |
2505 | * a locked page table | |
2506 | * | |
2507 | * - a page cache insertion, a swapin fault, or a migration | |
2508 | * have the page locked | |
2509 | */ | |
1306a85a | 2510 | page->mem_cgroup = memcg; |
9ce70c02 | 2511 | |
0a31bc97 JW |
2512 | if (lrucare) |
2513 | unlock_page_lru(page, isolated); | |
7a81b88c | 2514 | } |
66e1707b | 2515 | |
84c07d11 | 2516 | #ifdef CONFIG_MEMCG_KMEM |
f3bb3043 | 2517 | static int memcg_alloc_cache_id(void) |
55007d84 | 2518 | { |
f3bb3043 VD |
2519 | int id, size; |
2520 | int err; | |
2521 | ||
dbcf73e2 | 2522 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2523 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2524 | if (id < 0) | |
2525 | return id; | |
55007d84 | 2526 | |
dbcf73e2 | 2527 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2528 | return id; |
2529 | ||
2530 | /* | |
2531 | * There's no space for the new id in memcg_caches arrays, | |
2532 | * so we have to grow them. | |
2533 | */ | |
05257a1a | 2534 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2535 | |
2536 | size = 2 * (id + 1); | |
55007d84 GC |
2537 | if (size < MEMCG_CACHES_MIN_SIZE) |
2538 | size = MEMCG_CACHES_MIN_SIZE; | |
2539 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2540 | size = MEMCG_CACHES_MAX_SIZE; | |
2541 | ||
f3bb3043 | 2542 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2543 | if (!err) |
2544 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2545 | if (!err) |
2546 | memcg_nr_cache_ids = size; | |
2547 | ||
2548 | up_write(&memcg_cache_ids_sem); | |
2549 | ||
f3bb3043 | 2550 | if (err) { |
dbcf73e2 | 2551 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2552 | return err; |
2553 | } | |
2554 | return id; | |
2555 | } | |
2556 | ||
2557 | static void memcg_free_cache_id(int id) | |
2558 | { | |
dbcf73e2 | 2559 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2560 | } |
2561 | ||
d5b3cf71 | 2562 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2563 | struct mem_cgroup *memcg; |
2564 | struct kmem_cache *cachep; | |
2565 | struct work_struct work; | |
2566 | }; | |
2567 | ||
d5b3cf71 | 2568 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2569 | { |
d5b3cf71 VD |
2570 | struct memcg_kmem_cache_create_work *cw = |
2571 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2572 | struct mem_cgroup *memcg = cw->memcg; |
2573 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2574 | |
d5b3cf71 | 2575 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2576 | |
5722d094 | 2577 | css_put(&memcg->css); |
d7f25f8a GC |
2578 | kfree(cw); |
2579 | } | |
2580 | ||
2581 | /* | |
2582 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2583 | */ |
85cfb245 | 2584 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
d5b3cf71 | 2585 | struct kmem_cache *cachep) |
d7f25f8a | 2586 | { |
d5b3cf71 | 2587 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2588 | |
c892fd82 | 2589 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT | __GFP_NOWARN); |
8135be5a | 2590 | if (!cw) |
d7f25f8a | 2591 | return; |
8135be5a VD |
2592 | |
2593 | css_get(&memcg->css); | |
d7f25f8a GC |
2594 | |
2595 | cw->memcg = memcg; | |
2596 | cw->cachep = cachep; | |
d5b3cf71 | 2597 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2598 | |
17cc4dfe | 2599 | queue_work(memcg_kmem_cache_wq, &cw->work); |
d7f25f8a GC |
2600 | } |
2601 | ||
45264778 VD |
2602 | static inline bool memcg_kmem_bypass(void) |
2603 | { | |
2604 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
2605 | return true; | |
2606 | return false; | |
2607 | } | |
2608 | ||
2609 | /** | |
2610 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation | |
2611 | * @cachep: the original global kmem cache | |
2612 | * | |
d7f25f8a GC |
2613 | * Return the kmem_cache we're supposed to use for a slab allocation. |
2614 | * We try to use the current memcg's version of the cache. | |
2615 | * | |
45264778 VD |
2616 | * If the cache does not exist yet, if we are the first user of it, we |
2617 | * create it asynchronously in a workqueue and let the current allocation | |
2618 | * go through with the original cache. | |
d7f25f8a | 2619 | * |
45264778 VD |
2620 | * This function takes a reference to the cache it returns to assure it |
2621 | * won't get destroyed while we are working with it. Once the caller is | |
2622 | * done with it, memcg_kmem_put_cache() must be called to release the | |
2623 | * reference. | |
d7f25f8a | 2624 | */ |
45264778 | 2625 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2626 | { |
2627 | struct mem_cgroup *memcg; | |
959c8963 | 2628 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2629 | int kmemcg_id; |
d7f25f8a | 2630 | |
f7ce3190 | 2631 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2632 | |
45264778 | 2633 | if (memcg_kmem_bypass()) |
230e9fc2 VD |
2634 | return cachep; |
2635 | ||
d46eb14b | 2636 | memcg = get_mem_cgroup_from_current(); |
4db0c3c2 | 2637 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2638 | if (kmemcg_id < 0) |
ca0dde97 | 2639 | goto out; |
d7f25f8a | 2640 | |
2a4db7eb | 2641 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2642 | if (likely(memcg_cachep)) |
2643 | return memcg_cachep; | |
ca0dde97 LZ |
2644 | |
2645 | /* | |
2646 | * If we are in a safe context (can wait, and not in interrupt | |
2647 | * context), we could be be predictable and return right away. | |
2648 | * This would guarantee that the allocation being performed | |
2649 | * already belongs in the new cache. | |
2650 | * | |
2651 | * However, there are some clashes that can arrive from locking. | |
2652 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2653 | * memcg_create_kmem_cache, this means no further allocation |
2654 | * could happen with the slab_mutex held. So it's better to | |
2655 | * defer everything. | |
ca0dde97 | 2656 | */ |
d5b3cf71 | 2657 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2658 | out: |
8135be5a | 2659 | css_put(&memcg->css); |
ca0dde97 | 2660 | return cachep; |
d7f25f8a | 2661 | } |
d7f25f8a | 2662 | |
45264778 VD |
2663 | /** |
2664 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache | |
2665 | * @cachep: the cache returned by memcg_kmem_get_cache | |
2666 | */ | |
2667 | void memcg_kmem_put_cache(struct kmem_cache *cachep) | |
8135be5a VD |
2668 | { |
2669 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2670 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2671 | } |
2672 | ||
45264778 | 2673 | /** |
60cd4bcd | 2674 | * __memcg_kmem_charge_memcg: charge a kmem page |
45264778 VD |
2675 | * @page: page to charge |
2676 | * @gfp: reclaim mode | |
2677 | * @order: allocation order | |
2678 | * @memcg: memory cgroup to charge | |
2679 | * | |
2680 | * Returns 0 on success, an error code on failure. | |
2681 | */ | |
60cd4bcd | 2682 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
45264778 | 2683 | struct mem_cgroup *memcg) |
7ae1e1d0 | 2684 | { |
f3ccb2c4 VD |
2685 | unsigned int nr_pages = 1 << order; |
2686 | struct page_counter *counter; | |
7ae1e1d0 GC |
2687 | int ret; |
2688 | ||
f3ccb2c4 | 2689 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2690 | if (ret) |
f3ccb2c4 | 2691 | return ret; |
52c29b04 JW |
2692 | |
2693 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2694 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2695 | cancel_charge(memcg, nr_pages); | |
2696 | return -ENOMEM; | |
7ae1e1d0 GC |
2697 | } |
2698 | ||
f3ccb2c4 | 2699 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2700 | |
f3ccb2c4 | 2701 | return 0; |
7ae1e1d0 GC |
2702 | } |
2703 | ||
45264778 | 2704 | /** |
60cd4bcd | 2705 | * __memcg_kmem_charge: charge a kmem page to the current memory cgroup |
45264778 VD |
2706 | * @page: page to charge |
2707 | * @gfp: reclaim mode | |
2708 | * @order: allocation order | |
2709 | * | |
2710 | * Returns 0 on success, an error code on failure. | |
2711 | */ | |
60cd4bcd | 2712 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2713 | { |
f3ccb2c4 | 2714 | struct mem_cgroup *memcg; |
fcff7d7e | 2715 | int ret = 0; |
7ae1e1d0 | 2716 | |
60cd4bcd | 2717 | if (memcg_kmem_bypass()) |
45264778 VD |
2718 | return 0; |
2719 | ||
d46eb14b | 2720 | memcg = get_mem_cgroup_from_current(); |
c4159a75 | 2721 | if (!mem_cgroup_is_root(memcg)) { |
60cd4bcd | 2722 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); |
c4159a75 VD |
2723 | if (!ret) |
2724 | __SetPageKmemcg(page); | |
2725 | } | |
7ae1e1d0 | 2726 | css_put(&memcg->css); |
d05e83a6 | 2727 | return ret; |
7ae1e1d0 | 2728 | } |
45264778 | 2729 | /** |
60cd4bcd | 2730 | * __memcg_kmem_uncharge: uncharge a kmem page |
45264778 VD |
2731 | * @page: page to uncharge |
2732 | * @order: allocation order | |
2733 | */ | |
60cd4bcd | 2734 | void __memcg_kmem_uncharge(struct page *page, int order) |
7ae1e1d0 | 2735 | { |
1306a85a | 2736 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2737 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2738 | |
7ae1e1d0 GC |
2739 | if (!memcg) |
2740 | return; | |
2741 | ||
309381fe | 2742 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2743 | |
52c29b04 JW |
2744 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2745 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2746 | ||
f3ccb2c4 | 2747 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2748 | if (do_memsw_account()) |
f3ccb2c4 | 2749 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2750 | |
1306a85a | 2751 | page->mem_cgroup = NULL; |
c4159a75 VD |
2752 | |
2753 | /* slab pages do not have PageKmemcg flag set */ | |
2754 | if (PageKmemcg(page)) | |
2755 | __ClearPageKmemcg(page); | |
2756 | ||
f3ccb2c4 | 2757 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2758 | } |
84c07d11 | 2759 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 2760 | |
ca3e0214 KH |
2761 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2762 | ||
ca3e0214 KH |
2763 | /* |
2764 | * Because tail pages are not marked as "used", set it. We're under | |
f4b7e272 | 2765 | * pgdat->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2766 | */ |
e94c8a9c | 2767 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2768 | { |
e94c8a9c | 2769 | int i; |
ca3e0214 | 2770 | |
3d37c4a9 KH |
2771 | if (mem_cgroup_disabled()) |
2772 | return; | |
b070e65c | 2773 | |
29833315 | 2774 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2775 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2776 | |
c9019e9b | 2777 | __mod_memcg_state(head->mem_cgroup, MEMCG_RSS_HUGE, -HPAGE_PMD_NR); |
ca3e0214 | 2778 | } |
12d27107 | 2779 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2780 | |
c255a458 | 2781 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
2782 | /** |
2783 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2784 | * @entry: swap entry to be moved | |
2785 | * @from: mem_cgroup which the entry is moved from | |
2786 | * @to: mem_cgroup which the entry is moved to | |
2787 | * | |
2788 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2789 | * as the mem_cgroup's id of @from. | |
2790 | * | |
2791 | * Returns 0 on success, -EINVAL on failure. | |
2792 | * | |
3e32cb2e | 2793 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2794 | * both res and memsw, and called css_get(). |
2795 | */ | |
2796 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2797 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2798 | { |
2799 | unsigned short old_id, new_id; | |
2800 | ||
34c00c31 LZ |
2801 | old_id = mem_cgroup_id(from); |
2802 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2803 | |
2804 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
2805 | mod_memcg_state(from, MEMCG_SWAP, -1); |
2806 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
2807 | return 0; |
2808 | } | |
2809 | return -EINVAL; | |
2810 | } | |
2811 | #else | |
2812 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2813 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2814 | { |
2815 | return -EINVAL; | |
2816 | } | |
8c7c6e34 | 2817 | #endif |
d13d1443 | 2818 | |
bbec2e15 | 2819 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 2820 | |
bbec2e15 RG |
2821 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
2822 | unsigned long max, bool memsw) | |
628f4235 | 2823 | { |
3e32cb2e | 2824 | bool enlarge = false; |
bb4a7ea2 | 2825 | bool drained = false; |
3e32cb2e | 2826 | int ret; |
c054a78c YZ |
2827 | bool limits_invariant; |
2828 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 2829 | |
3e32cb2e | 2830 | do { |
628f4235 KH |
2831 | if (signal_pending(current)) { |
2832 | ret = -EINTR; | |
2833 | break; | |
2834 | } | |
3e32cb2e | 2835 | |
bbec2e15 | 2836 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
2837 | /* |
2838 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 2839 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 2840 | */ |
bbec2e15 RG |
2841 | limits_invariant = memsw ? max >= memcg->memory.max : |
2842 | max <= memcg->memsw.max; | |
c054a78c | 2843 | if (!limits_invariant) { |
bbec2e15 | 2844 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 2845 | ret = -EINVAL; |
8c7c6e34 KH |
2846 | break; |
2847 | } | |
bbec2e15 | 2848 | if (max > counter->max) |
3e32cb2e | 2849 | enlarge = true; |
bbec2e15 RG |
2850 | ret = page_counter_set_max(counter, max); |
2851 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
2852 | |
2853 | if (!ret) | |
2854 | break; | |
2855 | ||
bb4a7ea2 SB |
2856 | if (!drained) { |
2857 | drain_all_stock(memcg); | |
2858 | drained = true; | |
2859 | continue; | |
2860 | } | |
2861 | ||
1ab5c056 AR |
2862 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
2863 | GFP_KERNEL, !memsw)) { | |
2864 | ret = -EBUSY; | |
2865 | break; | |
2866 | } | |
2867 | } while (true); | |
3e32cb2e | 2868 | |
3c11ecf4 KH |
2869 | if (!ret && enlarge) |
2870 | memcg_oom_recover(memcg); | |
3e32cb2e | 2871 | |
628f4235 KH |
2872 | return ret; |
2873 | } | |
2874 | ||
ef8f2327 | 2875 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
2876 | gfp_t gfp_mask, |
2877 | unsigned long *total_scanned) | |
2878 | { | |
2879 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 2880 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
2881 | unsigned long reclaimed; |
2882 | int loop = 0; | |
ef8f2327 | 2883 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 2884 | unsigned long excess; |
0608f43d AM |
2885 | unsigned long nr_scanned; |
2886 | ||
2887 | if (order > 0) | |
2888 | return 0; | |
2889 | ||
ef8f2327 | 2890 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
2891 | |
2892 | /* | |
2893 | * Do not even bother to check the largest node if the root | |
2894 | * is empty. Do it lockless to prevent lock bouncing. Races | |
2895 | * are acceptable as soft limit is best effort anyway. | |
2896 | */ | |
bfc7228b | 2897 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
2898 | return 0; |
2899 | ||
0608f43d AM |
2900 | /* |
2901 | * This loop can run a while, specially if mem_cgroup's continuously | |
2902 | * keep exceeding their soft limit and putting the system under | |
2903 | * pressure | |
2904 | */ | |
2905 | do { | |
2906 | if (next_mz) | |
2907 | mz = next_mz; | |
2908 | else | |
2909 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2910 | if (!mz) | |
2911 | break; | |
2912 | ||
2913 | nr_scanned = 0; | |
ef8f2327 | 2914 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
2915 | gfp_mask, &nr_scanned); |
2916 | nr_reclaimed += reclaimed; | |
2917 | *total_scanned += nr_scanned; | |
0a31bc97 | 2918 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2919 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2920 | |
2921 | /* | |
2922 | * If we failed to reclaim anything from this memory cgroup | |
2923 | * it is time to move on to the next cgroup | |
2924 | */ | |
2925 | next_mz = NULL; | |
bc2f2e7f VD |
2926 | if (!reclaimed) |
2927 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2928 | ||
3e32cb2e | 2929 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2930 | /* |
2931 | * One school of thought says that we should not add | |
2932 | * back the node to the tree if reclaim returns 0. | |
2933 | * But our reclaim could return 0, simply because due | |
2934 | * to priority we are exposing a smaller subset of | |
2935 | * memory to reclaim from. Consider this as a longer | |
2936 | * term TODO. | |
2937 | */ | |
2938 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2939 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2940 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2941 | css_put(&mz->memcg->css); |
2942 | loop++; | |
2943 | /* | |
2944 | * Could not reclaim anything and there are no more | |
2945 | * mem cgroups to try or we seem to be looping without | |
2946 | * reclaiming anything. | |
2947 | */ | |
2948 | if (!nr_reclaimed && | |
2949 | (next_mz == NULL || | |
2950 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2951 | break; | |
2952 | } while (!nr_reclaimed); | |
2953 | if (next_mz) | |
2954 | css_put(&next_mz->memcg->css); | |
2955 | return nr_reclaimed; | |
2956 | } | |
2957 | ||
ea280e7b TH |
2958 | /* |
2959 | * Test whether @memcg has children, dead or alive. Note that this | |
2960 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2961 | * returns %true if there are child csses according to the cgroup | |
2962 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2963 | */ | |
b5f99b53 GC |
2964 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2965 | { | |
ea280e7b TH |
2966 | bool ret; |
2967 | ||
ea280e7b TH |
2968 | rcu_read_lock(); |
2969 | ret = css_next_child(NULL, &memcg->css); | |
2970 | rcu_read_unlock(); | |
2971 | return ret; | |
b5f99b53 GC |
2972 | } |
2973 | ||
c26251f9 | 2974 | /* |
51038171 | 2975 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
2976 | * |
2977 | * Caller is responsible for holding css reference for memcg. | |
2978 | */ | |
2979 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2980 | { | |
2981 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2982 | |
c1e862c1 KH |
2983 | /* we call try-to-free pages for make this cgroup empty */ |
2984 | lru_add_drain_all(); | |
d12c60f6 JS |
2985 | |
2986 | drain_all_stock(memcg); | |
2987 | ||
f817ed48 | 2988 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2989 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2990 | int progress; |
c1e862c1 | 2991 | |
c26251f9 MH |
2992 | if (signal_pending(current)) |
2993 | return -EINTR; | |
2994 | ||
b70a2a21 JW |
2995 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2996 | GFP_KERNEL, true); | |
c1e862c1 | 2997 | if (!progress) { |
f817ed48 | 2998 | nr_retries--; |
c1e862c1 | 2999 | /* maybe some writeback is necessary */ |
8aa7e847 | 3000 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3001 | } |
f817ed48 KH |
3002 | |
3003 | } | |
ab5196c2 MH |
3004 | |
3005 | return 0; | |
cc847582 KH |
3006 | } |
3007 | ||
6770c64e TH |
3008 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3009 | char *buf, size_t nbytes, | |
3010 | loff_t off) | |
c1e862c1 | 3011 | { |
6770c64e | 3012 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3013 | |
d8423011 MH |
3014 | if (mem_cgroup_is_root(memcg)) |
3015 | return -EINVAL; | |
6770c64e | 3016 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3017 | } |
3018 | ||
182446d0 TH |
3019 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3020 | struct cftype *cft) | |
18f59ea7 | 3021 | { |
182446d0 | 3022 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3023 | } |
3024 | ||
182446d0 TH |
3025 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3026 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3027 | { |
3028 | int retval = 0; | |
182446d0 | 3029 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3030 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3031 | |
567fb435 | 3032 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 3033 | return 0; |
567fb435 | 3034 | |
18f59ea7 | 3035 | /* |
af901ca1 | 3036 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3037 | * in the child subtrees. If it is unset, then the change can |
3038 | * occur, provided the current cgroup has no children. | |
3039 | * | |
3040 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3041 | * set if there are no children. | |
3042 | */ | |
c0ff4b85 | 3043 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3044 | (val == 1 || val == 0)) { |
ea280e7b | 3045 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3046 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3047 | else |
3048 | retval = -EBUSY; | |
3049 | } else | |
3050 | retval = -EINVAL; | |
567fb435 | 3051 | |
18f59ea7 BS |
3052 | return retval; |
3053 | } | |
3054 | ||
6f646156 | 3055 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3056 | { |
42a30035 | 3057 | unsigned long val; |
ce00a967 | 3058 | |
3e32cb2e | 3059 | if (mem_cgroup_is_root(memcg)) { |
42a30035 JW |
3060 | val = memcg_page_state(memcg, MEMCG_CACHE) + |
3061 | memcg_page_state(memcg, MEMCG_RSS); | |
3062 | if (swap) | |
3063 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3064 | } else { |
ce00a967 | 3065 | if (!swap) |
3e32cb2e | 3066 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3067 | else |
3e32cb2e | 3068 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3069 | } |
c12176d3 | 3070 | return val; |
ce00a967 JW |
3071 | } |
3072 | ||
3e32cb2e JW |
3073 | enum { |
3074 | RES_USAGE, | |
3075 | RES_LIMIT, | |
3076 | RES_MAX_USAGE, | |
3077 | RES_FAILCNT, | |
3078 | RES_SOFT_LIMIT, | |
3079 | }; | |
ce00a967 | 3080 | |
791badbd | 3081 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3082 | struct cftype *cft) |
8cdea7c0 | 3083 | { |
182446d0 | 3084 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3085 | struct page_counter *counter; |
af36f906 | 3086 | |
3e32cb2e | 3087 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3088 | case _MEM: |
3e32cb2e JW |
3089 | counter = &memcg->memory; |
3090 | break; | |
8c7c6e34 | 3091 | case _MEMSWAP: |
3e32cb2e JW |
3092 | counter = &memcg->memsw; |
3093 | break; | |
510fc4e1 | 3094 | case _KMEM: |
3e32cb2e | 3095 | counter = &memcg->kmem; |
510fc4e1 | 3096 | break; |
d55f90bf | 3097 | case _TCP: |
0db15298 | 3098 | counter = &memcg->tcpmem; |
d55f90bf | 3099 | break; |
8c7c6e34 KH |
3100 | default: |
3101 | BUG(); | |
8c7c6e34 | 3102 | } |
3e32cb2e JW |
3103 | |
3104 | switch (MEMFILE_ATTR(cft->private)) { | |
3105 | case RES_USAGE: | |
3106 | if (counter == &memcg->memory) | |
c12176d3 | 3107 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3108 | if (counter == &memcg->memsw) |
c12176d3 | 3109 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3110 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3111 | case RES_LIMIT: | |
bbec2e15 | 3112 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3113 | case RES_MAX_USAGE: |
3114 | return (u64)counter->watermark * PAGE_SIZE; | |
3115 | case RES_FAILCNT: | |
3116 | return counter->failcnt; | |
3117 | case RES_SOFT_LIMIT: | |
3118 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3119 | default: | |
3120 | BUG(); | |
3121 | } | |
8cdea7c0 | 3122 | } |
510fc4e1 | 3123 | |
84c07d11 | 3124 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3125 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3126 | { |
d6441637 VD |
3127 | int memcg_id; |
3128 | ||
b313aeee VD |
3129 | if (cgroup_memory_nokmem) |
3130 | return 0; | |
3131 | ||
2a4db7eb | 3132 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3133 | BUG_ON(memcg->kmem_state); |
d6441637 | 3134 | |
f3bb3043 | 3135 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3136 | if (memcg_id < 0) |
3137 | return memcg_id; | |
d6441637 | 3138 | |
ef12947c | 3139 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 3140 | /* |
567e9ab2 | 3141 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 3142 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
3143 | * guarantee no one starts accounting before all call sites are |
3144 | * patched. | |
3145 | */ | |
900a38f0 | 3146 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3147 | memcg->kmem_state = KMEM_ONLINE; |
bc2791f8 | 3148 | INIT_LIST_HEAD(&memcg->kmem_caches); |
0b8f73e1 JW |
3149 | |
3150 | return 0; | |
d6441637 VD |
3151 | } |
3152 | ||
8e0a8912 JW |
3153 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3154 | { | |
3155 | struct cgroup_subsys_state *css; | |
3156 | struct mem_cgroup *parent, *child; | |
3157 | int kmemcg_id; | |
3158 | ||
3159 | if (memcg->kmem_state != KMEM_ONLINE) | |
3160 | return; | |
3161 | /* | |
3162 | * Clear the online state before clearing memcg_caches array | |
3163 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
3164 | * guarantees that no cache will be created for this cgroup | |
3165 | * after we are done (see memcg_create_kmem_cache()). | |
3166 | */ | |
3167 | memcg->kmem_state = KMEM_ALLOCATED; | |
3168 | ||
3169 | memcg_deactivate_kmem_caches(memcg); | |
3170 | ||
3171 | kmemcg_id = memcg->kmemcg_id; | |
3172 | BUG_ON(kmemcg_id < 0); | |
3173 | ||
3174 | parent = parent_mem_cgroup(memcg); | |
3175 | if (!parent) | |
3176 | parent = root_mem_cgroup; | |
3177 | ||
3178 | /* | |
3179 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3180 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3181 | * to ones of the parent. After we have finished, all list_lrus | |
3182 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3183 | * ordering is imposed by list_lru_node->lock taken by | |
3184 | * memcg_drain_all_list_lrus(). | |
3185 | */ | |
3a06bb78 | 3186 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3187 | css_for_each_descendant_pre(css, &memcg->css) { |
3188 | child = mem_cgroup_from_css(css); | |
3189 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3190 | child->kmemcg_id = parent->kmemcg_id; | |
3191 | if (!memcg->use_hierarchy) | |
3192 | break; | |
3193 | } | |
3a06bb78 TH |
3194 | rcu_read_unlock(); |
3195 | ||
9bec5c35 | 3196 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3197 | |
3198 | memcg_free_cache_id(kmemcg_id); | |
3199 | } | |
3200 | ||
3201 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3202 | { | |
0b8f73e1 JW |
3203 | /* css_alloc() failed, offlining didn't happen */ |
3204 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3205 | memcg_offline_kmem(memcg); | |
3206 | ||
8e0a8912 JW |
3207 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
3208 | memcg_destroy_kmem_caches(memcg); | |
3209 | static_branch_dec(&memcg_kmem_enabled_key); | |
3210 | WARN_ON(page_counter_read(&memcg->kmem)); | |
3211 | } | |
8e0a8912 | 3212 | } |
d6441637 | 3213 | #else |
0b8f73e1 | 3214 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3215 | { |
3216 | return 0; | |
3217 | } | |
3218 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3219 | { | |
3220 | } | |
3221 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3222 | { | |
3223 | } | |
84c07d11 | 3224 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3225 | |
bbec2e15 RG |
3226 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3227 | unsigned long max) | |
d6441637 | 3228 | { |
b313aeee | 3229 | int ret; |
127424c8 | 3230 | |
bbec2e15 RG |
3231 | mutex_lock(&memcg_max_mutex); |
3232 | ret = page_counter_set_max(&memcg->kmem, max); | |
3233 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3234 | return ret; |
d6441637 | 3235 | } |
510fc4e1 | 3236 | |
bbec2e15 | 3237 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3238 | { |
3239 | int ret; | |
3240 | ||
bbec2e15 | 3241 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3242 | |
bbec2e15 | 3243 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3244 | if (ret) |
3245 | goto out; | |
3246 | ||
0db15298 | 3247 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3248 | /* |
3249 | * The active flag needs to be written after the static_key | |
3250 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3251 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3252 | * for details, and note that we don't mark any socket as | |
3253 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3254 | * |
3255 | * We need to do this, because static_keys will span multiple | |
3256 | * sites, but we can't control their order. If we mark a socket | |
3257 | * as accounted, but the accounting functions are not patched in | |
3258 | * yet, we'll lose accounting. | |
3259 | * | |
2d758073 | 3260 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3261 | * because when this value change, the code to process it is not |
3262 | * patched in yet. | |
3263 | */ | |
3264 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3265 | memcg->tcpmem_active = true; |
d55f90bf VD |
3266 | } |
3267 | out: | |
bbec2e15 | 3268 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3269 | return ret; |
3270 | } | |
d55f90bf | 3271 | |
628f4235 KH |
3272 | /* |
3273 | * The user of this function is... | |
3274 | * RES_LIMIT. | |
3275 | */ | |
451af504 TH |
3276 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3277 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3278 | { |
451af504 | 3279 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3280 | unsigned long nr_pages; |
628f4235 KH |
3281 | int ret; |
3282 | ||
451af504 | 3283 | buf = strstrip(buf); |
650c5e56 | 3284 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3285 | if (ret) |
3286 | return ret; | |
af36f906 | 3287 | |
3e32cb2e | 3288 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3289 | case RES_LIMIT: |
4b3bde4c BS |
3290 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3291 | ret = -EINVAL; | |
3292 | break; | |
3293 | } | |
3e32cb2e JW |
3294 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3295 | case _MEM: | |
bbec2e15 | 3296 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3297 | break; |
3e32cb2e | 3298 | case _MEMSWAP: |
bbec2e15 | 3299 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3300 | break; |
3e32cb2e | 3301 | case _KMEM: |
bbec2e15 | 3302 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3303 | break; |
d55f90bf | 3304 | case _TCP: |
bbec2e15 | 3305 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3306 | break; |
3e32cb2e | 3307 | } |
296c81d8 | 3308 | break; |
3e32cb2e JW |
3309 | case RES_SOFT_LIMIT: |
3310 | memcg->soft_limit = nr_pages; | |
3311 | ret = 0; | |
628f4235 KH |
3312 | break; |
3313 | } | |
451af504 | 3314 | return ret ?: nbytes; |
8cdea7c0 BS |
3315 | } |
3316 | ||
6770c64e TH |
3317 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3318 | size_t nbytes, loff_t off) | |
c84872e1 | 3319 | { |
6770c64e | 3320 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3321 | struct page_counter *counter; |
c84872e1 | 3322 | |
3e32cb2e JW |
3323 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3324 | case _MEM: | |
3325 | counter = &memcg->memory; | |
3326 | break; | |
3327 | case _MEMSWAP: | |
3328 | counter = &memcg->memsw; | |
3329 | break; | |
3330 | case _KMEM: | |
3331 | counter = &memcg->kmem; | |
3332 | break; | |
d55f90bf | 3333 | case _TCP: |
0db15298 | 3334 | counter = &memcg->tcpmem; |
d55f90bf | 3335 | break; |
3e32cb2e JW |
3336 | default: |
3337 | BUG(); | |
3338 | } | |
af36f906 | 3339 | |
3e32cb2e | 3340 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3341 | case RES_MAX_USAGE: |
3e32cb2e | 3342 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3343 | break; |
3344 | case RES_FAILCNT: | |
3e32cb2e | 3345 | counter->failcnt = 0; |
29f2a4da | 3346 | break; |
3e32cb2e JW |
3347 | default: |
3348 | BUG(); | |
29f2a4da | 3349 | } |
f64c3f54 | 3350 | |
6770c64e | 3351 | return nbytes; |
c84872e1 PE |
3352 | } |
3353 | ||
182446d0 | 3354 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3355 | struct cftype *cft) |
3356 | { | |
182446d0 | 3357 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3358 | } |
3359 | ||
02491447 | 3360 | #ifdef CONFIG_MMU |
182446d0 | 3361 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3362 | struct cftype *cft, u64 val) |
3363 | { | |
182446d0 | 3364 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3365 | |
1dfab5ab | 3366 | if (val & ~MOVE_MASK) |
7dc74be0 | 3367 | return -EINVAL; |
ee5e8472 | 3368 | |
7dc74be0 | 3369 | /* |
ee5e8472 GC |
3370 | * No kind of locking is needed in here, because ->can_attach() will |
3371 | * check this value once in the beginning of the process, and then carry | |
3372 | * on with stale data. This means that changes to this value will only | |
3373 | * affect task migrations starting after the change. | |
7dc74be0 | 3374 | */ |
c0ff4b85 | 3375 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3376 | return 0; |
3377 | } | |
02491447 | 3378 | #else |
182446d0 | 3379 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3380 | struct cftype *cft, u64 val) |
3381 | { | |
3382 | return -ENOSYS; | |
3383 | } | |
3384 | #endif | |
7dc74be0 | 3385 | |
406eb0c9 | 3386 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3387 | |
3388 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3389 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3390 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3391 | ||
3392 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
3393 | int nid, unsigned int lru_mask) | |
3394 | { | |
3395 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); | |
3396 | unsigned long nr = 0; | |
3397 | enum lru_list lru; | |
3398 | ||
3399 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3400 | ||
3401 | for_each_lru(lru) { | |
3402 | if (!(BIT(lru) & lru_mask)) | |
3403 | continue; | |
205b20cc | 3404 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); |
113b7dfd JW |
3405 | } |
3406 | return nr; | |
3407 | } | |
3408 | ||
3409 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
3410 | unsigned int lru_mask) | |
3411 | { | |
3412 | unsigned long nr = 0; | |
3413 | enum lru_list lru; | |
3414 | ||
3415 | for_each_lru(lru) { | |
3416 | if (!(BIT(lru) & lru_mask)) | |
3417 | continue; | |
205b20cc | 3418 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); |
113b7dfd JW |
3419 | } |
3420 | return nr; | |
3421 | } | |
3422 | ||
2da8ca82 | 3423 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3424 | { |
25485de6 GT |
3425 | struct numa_stat { |
3426 | const char *name; | |
3427 | unsigned int lru_mask; | |
3428 | }; | |
3429 | ||
3430 | static const struct numa_stat stats[] = { | |
3431 | { "total", LRU_ALL }, | |
3432 | { "file", LRU_ALL_FILE }, | |
3433 | { "anon", LRU_ALL_ANON }, | |
3434 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3435 | }; | |
3436 | const struct numa_stat *stat; | |
406eb0c9 | 3437 | int nid; |
25485de6 | 3438 | unsigned long nr; |
aa9694bb | 3439 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3440 | |
25485de6 GT |
3441 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3442 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3443 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3444 | for_each_node_state(nid, N_MEMORY) { | |
3445 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3446 | stat->lru_mask); | |
3447 | seq_printf(m, " N%d=%lu", nid, nr); | |
3448 | } | |
3449 | seq_putc(m, '\n'); | |
406eb0c9 | 3450 | } |
406eb0c9 | 3451 | |
071aee13 YH |
3452 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3453 | struct mem_cgroup *iter; | |
3454 | ||
3455 | nr = 0; | |
3456 | for_each_mem_cgroup_tree(iter, memcg) | |
3457 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3458 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3459 | for_each_node_state(nid, N_MEMORY) { | |
3460 | nr = 0; | |
3461 | for_each_mem_cgroup_tree(iter, memcg) | |
3462 | nr += mem_cgroup_node_nr_lru_pages( | |
3463 | iter, nid, stat->lru_mask); | |
3464 | seq_printf(m, " N%d=%lu", nid, nr); | |
3465 | } | |
3466 | seq_putc(m, '\n'); | |
406eb0c9 | 3467 | } |
406eb0c9 | 3468 | |
406eb0c9 YH |
3469 | return 0; |
3470 | } | |
3471 | #endif /* CONFIG_NUMA */ | |
3472 | ||
df0e53d0 | 3473 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3474 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
3475 | PGPGIN, |
3476 | PGPGOUT, | |
3477 | PGFAULT, | |
3478 | PGMAJFAULT, | |
3479 | }; | |
3480 | ||
3481 | static const char *const memcg1_event_names[] = { | |
3482 | "pgpgin", | |
3483 | "pgpgout", | |
3484 | "pgfault", | |
3485 | "pgmajfault", | |
3486 | }; | |
3487 | ||
2da8ca82 | 3488 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3489 | { |
aa9694bb | 3490 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 3491 | unsigned long memory, memsw; |
af7c4b0e JW |
3492 | struct mem_cgroup *mi; |
3493 | unsigned int i; | |
406eb0c9 | 3494 | |
71cd3113 | 3495 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c RS |
3496 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3497 | ||
71cd3113 JW |
3498 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
3499 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) | |
1dd3a273 | 3500 | continue; |
71cd3113 | 3501 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], |
205b20cc | 3502 | memcg_page_state_local(memcg, memcg1_stats[i]) * |
71cd3113 | 3503 | PAGE_SIZE); |
1dd3a273 | 3504 | } |
7b854121 | 3505 | |
df0e53d0 JW |
3506 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3507 | seq_printf(m, "%s %lu\n", memcg1_event_names[i], | |
205b20cc | 3508 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
3509 | |
3510 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3511 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
205b20cc | 3512 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 3513 | PAGE_SIZE); |
af7c4b0e | 3514 | |
14067bb3 | 3515 | /* Hierarchical information */ |
3e32cb2e JW |
3516 | memory = memsw = PAGE_COUNTER_MAX; |
3517 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
bbec2e15 RG |
3518 | memory = min(memory, mi->memory.max); |
3519 | memsw = min(memsw, mi->memsw.max); | |
fee7b548 | 3520 | } |
3e32cb2e JW |
3521 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3522 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3523 | if (do_memsw_account()) |
3e32cb2e JW |
3524 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3525 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3526 | |
8de7ecc6 | 3527 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
71cd3113 | 3528 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3529 | continue; |
8de7ecc6 | 3530 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
42a30035 | 3531 | (u64)memcg_page_state(memcg, i) * PAGE_SIZE); |
af7c4b0e JW |
3532 | } |
3533 | ||
8de7ecc6 SB |
3534 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3535 | seq_printf(m, "total_%s %llu\n", memcg1_event_names[i], | |
42a30035 | 3536 | (u64)memcg_events(memcg, i)); |
af7c4b0e | 3537 | |
8de7ecc6 SB |
3538 | for (i = 0; i < NR_LRU_LISTS; i++) |
3539 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], | |
42a30035 JW |
3540 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
3541 | PAGE_SIZE); | |
14067bb3 | 3542 | |
7f016ee8 | 3543 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 3544 | { |
ef8f2327 MG |
3545 | pg_data_t *pgdat; |
3546 | struct mem_cgroup_per_node *mz; | |
89abfab1 | 3547 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3548 | unsigned long recent_rotated[2] = {0, 0}; |
3549 | unsigned long recent_scanned[2] = {0, 0}; | |
3550 | ||
ef8f2327 MG |
3551 | for_each_online_pgdat(pgdat) { |
3552 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
3553 | rstat = &mz->lruvec.reclaim_stat; | |
7f016ee8 | 3554 | |
ef8f2327 MG |
3555 | recent_rotated[0] += rstat->recent_rotated[0]; |
3556 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3557 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3558 | recent_scanned[1] += rstat->recent_scanned[1]; | |
3559 | } | |
78ccf5b5 JW |
3560 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3561 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3562 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3563 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3564 | } |
3565 | #endif | |
3566 | ||
d2ceb9b7 KH |
3567 | return 0; |
3568 | } | |
3569 | ||
182446d0 TH |
3570 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3571 | struct cftype *cft) | |
a7885eb8 | 3572 | { |
182446d0 | 3573 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3574 | |
1f4c025b | 3575 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3576 | } |
3577 | ||
182446d0 TH |
3578 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3579 | struct cftype *cft, u64 val) | |
a7885eb8 | 3580 | { |
182446d0 | 3581 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3582 | |
3dae7fec | 3583 | if (val > 100) |
a7885eb8 KM |
3584 | return -EINVAL; |
3585 | ||
14208b0e | 3586 | if (css->parent) |
3dae7fec JW |
3587 | memcg->swappiness = val; |
3588 | else | |
3589 | vm_swappiness = val; | |
068b38c1 | 3590 | |
a7885eb8 KM |
3591 | return 0; |
3592 | } | |
3593 | ||
2e72b634 KS |
3594 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3595 | { | |
3596 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3597 | unsigned long usage; |
2e72b634 KS |
3598 | int i; |
3599 | ||
3600 | rcu_read_lock(); | |
3601 | if (!swap) | |
2c488db2 | 3602 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3603 | else |
2c488db2 | 3604 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3605 | |
3606 | if (!t) | |
3607 | goto unlock; | |
3608 | ||
ce00a967 | 3609 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3610 | |
3611 | /* | |
748dad36 | 3612 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3613 | * If it's not true, a threshold was crossed after last |
3614 | * call of __mem_cgroup_threshold(). | |
3615 | */ | |
5407a562 | 3616 | i = t->current_threshold; |
2e72b634 KS |
3617 | |
3618 | /* | |
3619 | * Iterate backward over array of thresholds starting from | |
3620 | * current_threshold and check if a threshold is crossed. | |
3621 | * If none of thresholds below usage is crossed, we read | |
3622 | * only one element of the array here. | |
3623 | */ | |
3624 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3625 | eventfd_signal(t->entries[i].eventfd, 1); | |
3626 | ||
3627 | /* i = current_threshold + 1 */ | |
3628 | i++; | |
3629 | ||
3630 | /* | |
3631 | * Iterate forward over array of thresholds starting from | |
3632 | * current_threshold+1 and check if a threshold is crossed. | |
3633 | * If none of thresholds above usage is crossed, we read | |
3634 | * only one element of the array here. | |
3635 | */ | |
3636 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3637 | eventfd_signal(t->entries[i].eventfd, 1); | |
3638 | ||
3639 | /* Update current_threshold */ | |
5407a562 | 3640 | t->current_threshold = i - 1; |
2e72b634 KS |
3641 | unlock: |
3642 | rcu_read_unlock(); | |
3643 | } | |
3644 | ||
3645 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3646 | { | |
ad4ca5f4 KS |
3647 | while (memcg) { |
3648 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3649 | if (do_memsw_account()) |
ad4ca5f4 KS |
3650 | __mem_cgroup_threshold(memcg, true); |
3651 | ||
3652 | memcg = parent_mem_cgroup(memcg); | |
3653 | } | |
2e72b634 KS |
3654 | } |
3655 | ||
3656 | static int compare_thresholds(const void *a, const void *b) | |
3657 | { | |
3658 | const struct mem_cgroup_threshold *_a = a; | |
3659 | const struct mem_cgroup_threshold *_b = b; | |
3660 | ||
2bff24a3 GT |
3661 | if (_a->threshold > _b->threshold) |
3662 | return 1; | |
3663 | ||
3664 | if (_a->threshold < _b->threshold) | |
3665 | return -1; | |
3666 | ||
3667 | return 0; | |
2e72b634 KS |
3668 | } |
3669 | ||
c0ff4b85 | 3670 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3671 | { |
3672 | struct mem_cgroup_eventfd_list *ev; | |
3673 | ||
2bcf2e92 MH |
3674 | spin_lock(&memcg_oom_lock); |
3675 | ||
c0ff4b85 | 3676 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3677 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3678 | |
3679 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3680 | return 0; |
3681 | } | |
3682 | ||
c0ff4b85 | 3683 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3684 | { |
7d74b06f KH |
3685 | struct mem_cgroup *iter; |
3686 | ||
c0ff4b85 | 3687 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3688 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3689 | } |
3690 | ||
59b6f873 | 3691 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3692 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3693 | { |
2c488db2 KS |
3694 | struct mem_cgroup_thresholds *thresholds; |
3695 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3696 | unsigned long threshold; |
3697 | unsigned long usage; | |
2c488db2 | 3698 | int i, size, ret; |
2e72b634 | 3699 | |
650c5e56 | 3700 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3701 | if (ret) |
3702 | return ret; | |
3703 | ||
3704 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3705 | |
05b84301 | 3706 | if (type == _MEM) { |
2c488db2 | 3707 | thresholds = &memcg->thresholds; |
ce00a967 | 3708 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3709 | } else if (type == _MEMSWAP) { |
2c488db2 | 3710 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3711 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3712 | } else |
2e72b634 KS |
3713 | BUG(); |
3714 | ||
2e72b634 | 3715 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3716 | if (thresholds->primary) |
2e72b634 KS |
3717 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3718 | ||
2c488db2 | 3719 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3720 | |
3721 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 3722 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 3723 | if (!new) { |
2e72b634 KS |
3724 | ret = -ENOMEM; |
3725 | goto unlock; | |
3726 | } | |
2c488db2 | 3727 | new->size = size; |
2e72b634 KS |
3728 | |
3729 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3730 | if (thresholds->primary) { |
3731 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3732 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3733 | } |
3734 | ||
2e72b634 | 3735 | /* Add new threshold */ |
2c488db2 KS |
3736 | new->entries[size - 1].eventfd = eventfd; |
3737 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3738 | |
3739 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3740 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3741 | compare_thresholds, NULL); |
3742 | ||
3743 | /* Find current threshold */ | |
2c488db2 | 3744 | new->current_threshold = -1; |
2e72b634 | 3745 | for (i = 0; i < size; i++) { |
748dad36 | 3746 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3747 | /* |
2c488db2 KS |
3748 | * new->current_threshold will not be used until |
3749 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3750 | * it here. |
3751 | */ | |
2c488db2 | 3752 | ++new->current_threshold; |
748dad36 SZ |
3753 | } else |
3754 | break; | |
2e72b634 KS |
3755 | } |
3756 | ||
2c488db2 KS |
3757 | /* Free old spare buffer and save old primary buffer as spare */ |
3758 | kfree(thresholds->spare); | |
3759 | thresholds->spare = thresholds->primary; | |
3760 | ||
3761 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3762 | |
907860ed | 3763 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3764 | synchronize_rcu(); |
3765 | ||
2e72b634 KS |
3766 | unlock: |
3767 | mutex_unlock(&memcg->thresholds_lock); | |
3768 | ||
3769 | return ret; | |
3770 | } | |
3771 | ||
59b6f873 | 3772 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3773 | struct eventfd_ctx *eventfd, const char *args) |
3774 | { | |
59b6f873 | 3775 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3776 | } |
3777 | ||
59b6f873 | 3778 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3779 | struct eventfd_ctx *eventfd, const char *args) |
3780 | { | |
59b6f873 | 3781 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3782 | } |
3783 | ||
59b6f873 | 3784 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3785 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3786 | { |
2c488db2 KS |
3787 | struct mem_cgroup_thresholds *thresholds; |
3788 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3789 | unsigned long usage; |
2c488db2 | 3790 | int i, j, size; |
2e72b634 KS |
3791 | |
3792 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3793 | |
3794 | if (type == _MEM) { | |
2c488db2 | 3795 | thresholds = &memcg->thresholds; |
ce00a967 | 3796 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3797 | } else if (type == _MEMSWAP) { |
2c488db2 | 3798 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3799 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3800 | } else |
2e72b634 KS |
3801 | BUG(); |
3802 | ||
371528ca AV |
3803 | if (!thresholds->primary) |
3804 | goto unlock; | |
3805 | ||
2e72b634 KS |
3806 | /* Check if a threshold crossed before removing */ |
3807 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3808 | ||
3809 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3810 | size = 0; |
3811 | for (i = 0; i < thresholds->primary->size; i++) { | |
3812 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3813 | size++; |
3814 | } | |
3815 | ||
2c488db2 | 3816 | new = thresholds->spare; |
907860ed | 3817 | |
2e72b634 KS |
3818 | /* Set thresholds array to NULL if we don't have thresholds */ |
3819 | if (!size) { | |
2c488db2 KS |
3820 | kfree(new); |
3821 | new = NULL; | |
907860ed | 3822 | goto swap_buffers; |
2e72b634 KS |
3823 | } |
3824 | ||
2c488db2 | 3825 | new->size = size; |
2e72b634 KS |
3826 | |
3827 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3828 | new->current_threshold = -1; |
3829 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3830 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3831 | continue; |
3832 | ||
2c488db2 | 3833 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3834 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3835 | /* |
2c488db2 | 3836 | * new->current_threshold will not be used |
2e72b634 KS |
3837 | * until rcu_assign_pointer(), so it's safe to increment |
3838 | * it here. | |
3839 | */ | |
2c488db2 | 3840 | ++new->current_threshold; |
2e72b634 KS |
3841 | } |
3842 | j++; | |
3843 | } | |
3844 | ||
907860ed | 3845 | swap_buffers: |
2c488db2 KS |
3846 | /* Swap primary and spare array */ |
3847 | thresholds->spare = thresholds->primary; | |
8c757763 | 3848 | |
2c488db2 | 3849 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3850 | |
907860ed | 3851 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3852 | synchronize_rcu(); |
6611d8d7 MC |
3853 | |
3854 | /* If all events are unregistered, free the spare array */ | |
3855 | if (!new) { | |
3856 | kfree(thresholds->spare); | |
3857 | thresholds->spare = NULL; | |
3858 | } | |
371528ca | 3859 | unlock: |
2e72b634 | 3860 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3861 | } |
c1e862c1 | 3862 | |
59b6f873 | 3863 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3864 | struct eventfd_ctx *eventfd) |
3865 | { | |
59b6f873 | 3866 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3867 | } |
3868 | ||
59b6f873 | 3869 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3870 | struct eventfd_ctx *eventfd) |
3871 | { | |
59b6f873 | 3872 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3873 | } |
3874 | ||
59b6f873 | 3875 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3876 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3877 | { |
9490ff27 | 3878 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3879 | |
9490ff27 KH |
3880 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3881 | if (!event) | |
3882 | return -ENOMEM; | |
3883 | ||
1af8efe9 | 3884 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3885 | |
3886 | event->eventfd = eventfd; | |
3887 | list_add(&event->list, &memcg->oom_notify); | |
3888 | ||
3889 | /* already in OOM ? */ | |
c2b42d3c | 3890 | if (memcg->under_oom) |
9490ff27 | 3891 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3892 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3893 | |
3894 | return 0; | |
3895 | } | |
3896 | ||
59b6f873 | 3897 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3898 | struct eventfd_ctx *eventfd) |
9490ff27 | 3899 | { |
9490ff27 | 3900 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3901 | |
1af8efe9 | 3902 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3903 | |
c0ff4b85 | 3904 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3905 | if (ev->eventfd == eventfd) { |
3906 | list_del(&ev->list); | |
3907 | kfree(ev); | |
3908 | } | |
3909 | } | |
3910 | ||
1af8efe9 | 3911 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3912 | } |
3913 | ||
2da8ca82 | 3914 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3915 | { |
aa9694bb | 3916 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 3917 | |
791badbd | 3918 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3919 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
3920 | seq_printf(sf, "oom_kill %lu\n", |
3921 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
3922 | return 0; |
3923 | } | |
3924 | ||
182446d0 | 3925 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3926 | struct cftype *cft, u64 val) |
3927 | { | |
182446d0 | 3928 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3929 | |
3930 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3931 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3932 | return -EINVAL; |
3933 | ||
c0ff4b85 | 3934 | memcg->oom_kill_disable = val; |
4d845ebf | 3935 | if (!val) |
c0ff4b85 | 3936 | memcg_oom_recover(memcg); |
3dae7fec | 3937 | |
3c11ecf4 KH |
3938 | return 0; |
3939 | } | |
3940 | ||
52ebea74 TH |
3941 | #ifdef CONFIG_CGROUP_WRITEBACK |
3942 | ||
841710aa TH |
3943 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3944 | { | |
3945 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3946 | } | |
3947 | ||
3948 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3949 | { | |
3950 | wb_domain_exit(&memcg->cgwb_domain); | |
3951 | } | |
3952 | ||
2529bb3a TH |
3953 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3954 | { | |
3955 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3956 | } | |
3957 | ||
841710aa TH |
3958 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3959 | { | |
3960 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3961 | ||
3962 | if (!memcg->css.parent) | |
3963 | return NULL; | |
3964 | ||
3965 | return &memcg->cgwb_domain; | |
3966 | } | |
3967 | ||
0b3d6e6f GT |
3968 | /* |
3969 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
3970 | * Keep in sync with memcg_exact_page(). | |
3971 | */ | |
3972 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
3973 | { | |
871789d4 | 3974 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
3975 | int cpu; |
3976 | ||
3977 | for_each_online_cpu(cpu) | |
871789d4 | 3978 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
3979 | if (x < 0) |
3980 | x = 0; | |
3981 | return x; | |
3982 | } | |
3983 | ||
c2aa723a TH |
3984 | /** |
3985 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3986 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3987 | * @pfilepages: out parameter for number of file pages |
3988 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3989 | * @pdirty: out parameter for number of dirty pages |
3990 | * @pwriteback: out parameter for number of pages under writeback | |
3991 | * | |
c5edf9cd TH |
3992 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3993 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3994 | * is a bit more involved. | |
c2aa723a | 3995 | * |
c5edf9cd TH |
3996 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3997 | * headroom is calculated as the lowest headroom of itself and the | |
3998 | * ancestors. Note that this doesn't consider the actual amount of | |
3999 | * available memory in the system. The caller should further cap | |
4000 | * *@pheadroom accordingly. | |
c2aa723a | 4001 | */ |
c5edf9cd TH |
4002 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4003 | unsigned long *pheadroom, unsigned long *pdirty, | |
4004 | unsigned long *pwriteback) | |
c2aa723a TH |
4005 | { |
4006 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4007 | struct mem_cgroup *parent; | |
c2aa723a | 4008 | |
0b3d6e6f | 4009 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a TH |
4010 | |
4011 | /* this should eventually include NR_UNSTABLE_NFS */ | |
0b3d6e6f | 4012 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4013 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4014 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4015 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4016 | |
c2aa723a | 4017 | while ((parent = parent_mem_cgroup(memcg))) { |
bbec2e15 | 4018 | unsigned long ceiling = min(memcg->memory.max, memcg->high); |
c2aa723a TH |
4019 | unsigned long used = page_counter_read(&memcg->memory); |
4020 | ||
c5edf9cd | 4021 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4022 | memcg = parent; |
4023 | } | |
c2aa723a TH |
4024 | } |
4025 | ||
841710aa TH |
4026 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4027 | ||
4028 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4029 | { | |
4030 | return 0; | |
4031 | } | |
4032 | ||
4033 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4034 | { | |
4035 | } | |
4036 | ||
2529bb3a TH |
4037 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4038 | { | |
4039 | } | |
4040 | ||
52ebea74 TH |
4041 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4042 | ||
3bc942f3 TH |
4043 | /* |
4044 | * DO NOT USE IN NEW FILES. | |
4045 | * | |
4046 | * "cgroup.event_control" implementation. | |
4047 | * | |
4048 | * This is way over-engineered. It tries to support fully configurable | |
4049 | * events for each user. Such level of flexibility is completely | |
4050 | * unnecessary especially in the light of the planned unified hierarchy. | |
4051 | * | |
4052 | * Please deprecate this and replace with something simpler if at all | |
4053 | * possible. | |
4054 | */ | |
4055 | ||
79bd9814 TH |
4056 | /* |
4057 | * Unregister event and free resources. | |
4058 | * | |
4059 | * Gets called from workqueue. | |
4060 | */ | |
3bc942f3 | 4061 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4062 | { |
3bc942f3 TH |
4063 | struct mem_cgroup_event *event = |
4064 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4065 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4066 | |
4067 | remove_wait_queue(event->wqh, &event->wait); | |
4068 | ||
59b6f873 | 4069 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4070 | |
4071 | /* Notify userspace the event is going away. */ | |
4072 | eventfd_signal(event->eventfd, 1); | |
4073 | ||
4074 | eventfd_ctx_put(event->eventfd); | |
4075 | kfree(event); | |
59b6f873 | 4076 | css_put(&memcg->css); |
79bd9814 TH |
4077 | } |
4078 | ||
4079 | /* | |
a9a08845 | 4080 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4081 | * |
4082 | * Called with wqh->lock held and interrupts disabled. | |
4083 | */ | |
ac6424b9 | 4084 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4085 | int sync, void *key) |
79bd9814 | 4086 | { |
3bc942f3 TH |
4087 | struct mem_cgroup_event *event = |
4088 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4089 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4090 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4091 | |
a9a08845 | 4092 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4093 | /* |
4094 | * If the event has been detached at cgroup removal, we | |
4095 | * can simply return knowing the other side will cleanup | |
4096 | * for us. | |
4097 | * | |
4098 | * We can't race against event freeing since the other | |
4099 | * side will require wqh->lock via remove_wait_queue(), | |
4100 | * which we hold. | |
4101 | */ | |
fba94807 | 4102 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4103 | if (!list_empty(&event->list)) { |
4104 | list_del_init(&event->list); | |
4105 | /* | |
4106 | * We are in atomic context, but cgroup_event_remove() | |
4107 | * may sleep, so we have to call it in workqueue. | |
4108 | */ | |
4109 | schedule_work(&event->remove); | |
4110 | } | |
fba94807 | 4111 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4112 | } |
4113 | ||
4114 | return 0; | |
4115 | } | |
4116 | ||
3bc942f3 | 4117 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4118 | wait_queue_head_t *wqh, poll_table *pt) |
4119 | { | |
3bc942f3 TH |
4120 | struct mem_cgroup_event *event = |
4121 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4122 | |
4123 | event->wqh = wqh; | |
4124 | add_wait_queue(wqh, &event->wait); | |
4125 | } | |
4126 | ||
4127 | /* | |
3bc942f3 TH |
4128 | * DO NOT USE IN NEW FILES. |
4129 | * | |
79bd9814 TH |
4130 | * Parse input and register new cgroup event handler. |
4131 | * | |
4132 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4133 | * Interpretation of args is defined by control file implementation. | |
4134 | */ | |
451af504 TH |
4135 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4136 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4137 | { |
451af504 | 4138 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4139 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4140 | struct mem_cgroup_event *event; |
79bd9814 TH |
4141 | struct cgroup_subsys_state *cfile_css; |
4142 | unsigned int efd, cfd; | |
4143 | struct fd efile; | |
4144 | struct fd cfile; | |
fba94807 | 4145 | const char *name; |
79bd9814 TH |
4146 | char *endp; |
4147 | int ret; | |
4148 | ||
451af504 TH |
4149 | buf = strstrip(buf); |
4150 | ||
4151 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4152 | if (*endp != ' ') |
4153 | return -EINVAL; | |
451af504 | 4154 | buf = endp + 1; |
79bd9814 | 4155 | |
451af504 | 4156 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4157 | if ((*endp != ' ') && (*endp != '\0')) |
4158 | return -EINVAL; | |
451af504 | 4159 | buf = endp + 1; |
79bd9814 TH |
4160 | |
4161 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4162 | if (!event) | |
4163 | return -ENOMEM; | |
4164 | ||
59b6f873 | 4165 | event->memcg = memcg; |
79bd9814 | 4166 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4167 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4168 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4169 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4170 | |
4171 | efile = fdget(efd); | |
4172 | if (!efile.file) { | |
4173 | ret = -EBADF; | |
4174 | goto out_kfree; | |
4175 | } | |
4176 | ||
4177 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4178 | if (IS_ERR(event->eventfd)) { | |
4179 | ret = PTR_ERR(event->eventfd); | |
4180 | goto out_put_efile; | |
4181 | } | |
4182 | ||
4183 | cfile = fdget(cfd); | |
4184 | if (!cfile.file) { | |
4185 | ret = -EBADF; | |
4186 | goto out_put_eventfd; | |
4187 | } | |
4188 | ||
4189 | /* the process need read permission on control file */ | |
4190 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4191 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4192 | if (ret < 0) | |
4193 | goto out_put_cfile; | |
4194 | ||
fba94807 TH |
4195 | /* |
4196 | * Determine the event callbacks and set them in @event. This used | |
4197 | * to be done via struct cftype but cgroup core no longer knows | |
4198 | * about these events. The following is crude but the whole thing | |
4199 | * is for compatibility anyway. | |
3bc942f3 TH |
4200 | * |
4201 | * DO NOT ADD NEW FILES. | |
fba94807 | 4202 | */ |
b583043e | 4203 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4204 | |
4205 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4206 | event->register_event = mem_cgroup_usage_register_event; | |
4207 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4208 | } else if (!strcmp(name, "memory.oom_control")) { | |
4209 | event->register_event = mem_cgroup_oom_register_event; | |
4210 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4211 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4212 | event->register_event = vmpressure_register_event; | |
4213 | event->unregister_event = vmpressure_unregister_event; | |
4214 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4215 | event->register_event = memsw_cgroup_usage_register_event; |
4216 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4217 | } else { |
4218 | ret = -EINVAL; | |
4219 | goto out_put_cfile; | |
4220 | } | |
4221 | ||
79bd9814 | 4222 | /* |
b5557c4c TH |
4223 | * Verify @cfile should belong to @css. Also, remaining events are |
4224 | * automatically removed on cgroup destruction but the removal is | |
4225 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4226 | */ |
b583043e | 4227 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4228 | &memory_cgrp_subsys); |
79bd9814 | 4229 | ret = -EINVAL; |
5a17f543 | 4230 | if (IS_ERR(cfile_css)) |
79bd9814 | 4231 | goto out_put_cfile; |
5a17f543 TH |
4232 | if (cfile_css != css) { |
4233 | css_put(cfile_css); | |
79bd9814 | 4234 | goto out_put_cfile; |
5a17f543 | 4235 | } |
79bd9814 | 4236 | |
451af504 | 4237 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4238 | if (ret) |
4239 | goto out_put_css; | |
4240 | ||
9965ed17 | 4241 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4242 | |
fba94807 TH |
4243 | spin_lock(&memcg->event_list_lock); |
4244 | list_add(&event->list, &memcg->event_list); | |
4245 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4246 | |
4247 | fdput(cfile); | |
4248 | fdput(efile); | |
4249 | ||
451af504 | 4250 | return nbytes; |
79bd9814 TH |
4251 | |
4252 | out_put_css: | |
b5557c4c | 4253 | css_put(css); |
79bd9814 TH |
4254 | out_put_cfile: |
4255 | fdput(cfile); | |
4256 | out_put_eventfd: | |
4257 | eventfd_ctx_put(event->eventfd); | |
4258 | out_put_efile: | |
4259 | fdput(efile); | |
4260 | out_kfree: | |
4261 | kfree(event); | |
4262 | ||
4263 | return ret; | |
4264 | } | |
4265 | ||
241994ed | 4266 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4267 | { |
0eea1030 | 4268 | .name = "usage_in_bytes", |
8c7c6e34 | 4269 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4270 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4271 | }, |
c84872e1 PE |
4272 | { |
4273 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4274 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4275 | .write = mem_cgroup_reset, |
791badbd | 4276 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4277 | }, |
8cdea7c0 | 4278 | { |
0eea1030 | 4279 | .name = "limit_in_bytes", |
8c7c6e34 | 4280 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4281 | .write = mem_cgroup_write, |
791badbd | 4282 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4283 | }, |
296c81d8 BS |
4284 | { |
4285 | .name = "soft_limit_in_bytes", | |
4286 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4287 | .write = mem_cgroup_write, |
791badbd | 4288 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4289 | }, |
8cdea7c0 BS |
4290 | { |
4291 | .name = "failcnt", | |
8c7c6e34 | 4292 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4293 | .write = mem_cgroup_reset, |
791badbd | 4294 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4295 | }, |
d2ceb9b7 KH |
4296 | { |
4297 | .name = "stat", | |
2da8ca82 | 4298 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4299 | }, |
c1e862c1 KH |
4300 | { |
4301 | .name = "force_empty", | |
6770c64e | 4302 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4303 | }, |
18f59ea7 BS |
4304 | { |
4305 | .name = "use_hierarchy", | |
4306 | .write_u64 = mem_cgroup_hierarchy_write, | |
4307 | .read_u64 = mem_cgroup_hierarchy_read, | |
4308 | }, | |
79bd9814 | 4309 | { |
3bc942f3 | 4310 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4311 | .write = memcg_write_event_control, |
7dbdb199 | 4312 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4313 | }, |
a7885eb8 KM |
4314 | { |
4315 | .name = "swappiness", | |
4316 | .read_u64 = mem_cgroup_swappiness_read, | |
4317 | .write_u64 = mem_cgroup_swappiness_write, | |
4318 | }, | |
7dc74be0 DN |
4319 | { |
4320 | .name = "move_charge_at_immigrate", | |
4321 | .read_u64 = mem_cgroup_move_charge_read, | |
4322 | .write_u64 = mem_cgroup_move_charge_write, | |
4323 | }, | |
9490ff27 KH |
4324 | { |
4325 | .name = "oom_control", | |
2da8ca82 | 4326 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4327 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4328 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4329 | }, | |
70ddf637 AV |
4330 | { |
4331 | .name = "pressure_level", | |
70ddf637 | 4332 | }, |
406eb0c9 YH |
4333 | #ifdef CONFIG_NUMA |
4334 | { | |
4335 | .name = "numa_stat", | |
2da8ca82 | 4336 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4337 | }, |
4338 | #endif | |
510fc4e1 GC |
4339 | { |
4340 | .name = "kmem.limit_in_bytes", | |
4341 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4342 | .write = mem_cgroup_write, |
791badbd | 4343 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4344 | }, |
4345 | { | |
4346 | .name = "kmem.usage_in_bytes", | |
4347 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4348 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4349 | }, |
4350 | { | |
4351 | .name = "kmem.failcnt", | |
4352 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4353 | .write = mem_cgroup_reset, |
791badbd | 4354 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4355 | }, |
4356 | { | |
4357 | .name = "kmem.max_usage_in_bytes", | |
4358 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4359 | .write = mem_cgroup_reset, |
791badbd | 4360 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4361 | }, |
5b365771 | 4362 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
749c5415 GC |
4363 | { |
4364 | .name = "kmem.slabinfo", | |
bc2791f8 TH |
4365 | .seq_start = memcg_slab_start, |
4366 | .seq_next = memcg_slab_next, | |
4367 | .seq_stop = memcg_slab_stop, | |
b047501c | 4368 | .seq_show = memcg_slab_show, |
749c5415 GC |
4369 | }, |
4370 | #endif | |
d55f90bf VD |
4371 | { |
4372 | .name = "kmem.tcp.limit_in_bytes", | |
4373 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4374 | .write = mem_cgroup_write, | |
4375 | .read_u64 = mem_cgroup_read_u64, | |
4376 | }, | |
4377 | { | |
4378 | .name = "kmem.tcp.usage_in_bytes", | |
4379 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4380 | .read_u64 = mem_cgroup_read_u64, | |
4381 | }, | |
4382 | { | |
4383 | .name = "kmem.tcp.failcnt", | |
4384 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4385 | .write = mem_cgroup_reset, | |
4386 | .read_u64 = mem_cgroup_read_u64, | |
4387 | }, | |
4388 | { | |
4389 | .name = "kmem.tcp.max_usage_in_bytes", | |
4390 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4391 | .write = mem_cgroup_reset, | |
4392 | .read_u64 = mem_cgroup_read_u64, | |
4393 | }, | |
6bc10349 | 4394 | { }, /* terminate */ |
af36f906 | 4395 | }; |
8c7c6e34 | 4396 | |
73f576c0 JW |
4397 | /* |
4398 | * Private memory cgroup IDR | |
4399 | * | |
4400 | * Swap-out records and page cache shadow entries need to store memcg | |
4401 | * references in constrained space, so we maintain an ID space that is | |
4402 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4403 | * memory-controlled cgroups to 64k. | |
4404 | * | |
4405 | * However, there usually are many references to the oflline CSS after | |
4406 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4407 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4408 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4409 | * relatively small ID space and prevent the creation of new cgroups | |
4410 | * even when there are much fewer than 64k cgroups - possibly none. | |
4411 | * | |
4412 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4413 | * be freed and recycled when it's no longer needed, which is usually | |
4414 | * when the CSS is offlined. | |
4415 | * | |
4416 | * The only exception to that are records of swapped out tmpfs/shmem | |
4417 | * pages that need to be attributed to live ancestors on swapin. But | |
4418 | * those references are manageable from userspace. | |
4419 | */ | |
4420 | ||
4421 | static DEFINE_IDR(mem_cgroup_idr); | |
4422 | ||
7e97de0b KT |
4423 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
4424 | { | |
4425 | if (memcg->id.id > 0) { | |
4426 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
4427 | memcg->id.id = 0; | |
4428 | } | |
4429 | } | |
4430 | ||
615d66c3 | 4431 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4432 | { |
1c2d479a | 4433 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
4434 | } |
4435 | ||
615d66c3 | 4436 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4437 | { |
1c2d479a | 4438 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 4439 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
4440 | |
4441 | /* Memcg ID pins CSS */ | |
4442 | css_put(&memcg->css); | |
4443 | } | |
4444 | } | |
4445 | ||
615d66c3 VD |
4446 | static inline void mem_cgroup_id_get(struct mem_cgroup *memcg) |
4447 | { | |
4448 | mem_cgroup_id_get_many(memcg, 1); | |
4449 | } | |
4450 | ||
4451 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) | |
4452 | { | |
4453 | mem_cgroup_id_put_many(memcg, 1); | |
4454 | } | |
4455 | ||
73f576c0 JW |
4456 | /** |
4457 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4458 | * @id: the memcg id to look up | |
4459 | * | |
4460 | * Caller must hold rcu_read_lock(). | |
4461 | */ | |
4462 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4463 | { | |
4464 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4465 | return idr_find(&mem_cgroup_idr, id); | |
4466 | } | |
4467 | ||
ef8f2327 | 4468 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4469 | { |
4470 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 4471 | int tmp = node; |
1ecaab2b KH |
4472 | /* |
4473 | * This routine is called against possible nodes. | |
4474 | * But it's BUG to call kmalloc() against offline node. | |
4475 | * | |
4476 | * TODO: this routine can waste much memory for nodes which will | |
4477 | * never be onlined. It's better to use memory hotplug callback | |
4478 | * function. | |
4479 | */ | |
41e3355d KH |
4480 | if (!node_state(node, N_NORMAL_MEMORY)) |
4481 | tmp = -1; | |
17295c88 | 4482 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4483 | if (!pn) |
4484 | return 1; | |
1ecaab2b | 4485 | |
a983b5eb JW |
4486 | pn->lruvec_stat_cpu = alloc_percpu(struct lruvec_stat); |
4487 | if (!pn->lruvec_stat_cpu) { | |
00f3ca2c JW |
4488 | kfree(pn); |
4489 | return 1; | |
4490 | } | |
4491 | ||
ef8f2327 MG |
4492 | lruvec_init(&pn->lruvec); |
4493 | pn->usage_in_excess = 0; | |
4494 | pn->on_tree = false; | |
4495 | pn->memcg = memcg; | |
4496 | ||
54f72fe0 | 4497 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4498 | return 0; |
4499 | } | |
4500 | ||
ef8f2327 | 4501 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4502 | { |
00f3ca2c JW |
4503 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
4504 | ||
4eaf431f MH |
4505 | if (!pn) |
4506 | return; | |
4507 | ||
a983b5eb | 4508 | free_percpu(pn->lruvec_stat_cpu); |
00f3ca2c | 4509 | kfree(pn); |
1ecaab2b KH |
4510 | } |
4511 | ||
40e952f9 | 4512 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4513 | { |
c8b2a36f | 4514 | int node; |
59927fb9 | 4515 | |
c8b2a36f | 4516 | for_each_node(node) |
ef8f2327 | 4517 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 4518 | free_percpu(memcg->vmstats_percpu); |
8ff69e2c | 4519 | kfree(memcg); |
59927fb9 | 4520 | } |
3afe36b1 | 4521 | |
40e952f9 TE |
4522 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
4523 | { | |
4524 | memcg_wb_domain_exit(memcg); | |
4525 | __mem_cgroup_free(memcg); | |
4526 | } | |
4527 | ||
0b8f73e1 | 4528 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4529 | { |
d142e3e6 | 4530 | struct mem_cgroup *memcg; |
b9726c26 | 4531 | unsigned int size; |
6d12e2d8 | 4532 | int node; |
8cdea7c0 | 4533 | |
0b8f73e1 JW |
4534 | size = sizeof(struct mem_cgroup); |
4535 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4536 | ||
4537 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4538 | if (!memcg) |
0b8f73e1 JW |
4539 | return NULL; |
4540 | ||
73f576c0 JW |
4541 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
4542 | 1, MEM_CGROUP_ID_MAX, | |
4543 | GFP_KERNEL); | |
4544 | if (memcg->id.id < 0) | |
4545 | goto fail; | |
4546 | ||
871789d4 CD |
4547 | memcg->vmstats_percpu = alloc_percpu(struct memcg_vmstats_percpu); |
4548 | if (!memcg->vmstats_percpu) | |
0b8f73e1 | 4549 | goto fail; |
78fb7466 | 4550 | |
3ed28fa1 | 4551 | for_each_node(node) |
ef8f2327 | 4552 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 4553 | goto fail; |
f64c3f54 | 4554 | |
0b8f73e1 JW |
4555 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4556 | goto fail; | |
28dbc4b6 | 4557 | |
f7e1cb6e | 4558 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4559 | memcg->last_scanned_node = MAX_NUMNODES; |
4560 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4561 | mutex_init(&memcg->thresholds_lock); |
4562 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4563 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4564 | INIT_LIST_HEAD(&memcg->event_list); |
4565 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4566 | memcg->socket_pressure = jiffies; |
84c07d11 | 4567 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 4568 | memcg->kmemcg_id = -1; |
900a38f0 | 4569 | #endif |
52ebea74 TH |
4570 | #ifdef CONFIG_CGROUP_WRITEBACK |
4571 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4572 | #endif | |
73f576c0 | 4573 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
4574 | return memcg; |
4575 | fail: | |
7e97de0b | 4576 | mem_cgroup_id_remove(memcg); |
40e952f9 | 4577 | __mem_cgroup_free(memcg); |
0b8f73e1 | 4578 | return NULL; |
d142e3e6 GC |
4579 | } |
4580 | ||
0b8f73e1 JW |
4581 | static struct cgroup_subsys_state * __ref |
4582 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4583 | { |
0b8f73e1 JW |
4584 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4585 | struct mem_cgroup *memcg; | |
4586 | long error = -ENOMEM; | |
d142e3e6 | 4587 | |
0b8f73e1 JW |
4588 | memcg = mem_cgroup_alloc(); |
4589 | if (!memcg) | |
4590 | return ERR_PTR(error); | |
d142e3e6 | 4591 | |
0b8f73e1 JW |
4592 | memcg->high = PAGE_COUNTER_MAX; |
4593 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4594 | if (parent) { | |
4595 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4596 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4597 | } | |
4598 | if (parent && parent->use_hierarchy) { | |
4599 | memcg->use_hierarchy = true; | |
3e32cb2e | 4600 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4601 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4602 | page_counter_init(&memcg->memsw, &parent->memsw); |
4603 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4604 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4605 | } else { |
3e32cb2e | 4606 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4607 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4608 | page_counter_init(&memcg->memsw, NULL); |
4609 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4610 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4611 | /* |
4612 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4613 | * much sense so let cgroup subsystem know about this | |
4614 | * unfortunate state in our controller. | |
4615 | */ | |
d142e3e6 | 4616 | if (parent != root_mem_cgroup) |
073219e9 | 4617 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4618 | } |
d6441637 | 4619 | |
0b8f73e1 JW |
4620 | /* The following stuff does not apply to the root */ |
4621 | if (!parent) { | |
4622 | root_mem_cgroup = memcg; | |
4623 | return &memcg->css; | |
4624 | } | |
4625 | ||
b313aeee | 4626 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
4627 | if (error) |
4628 | goto fail; | |
127424c8 | 4629 | |
f7e1cb6e | 4630 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4631 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4632 | |
0b8f73e1 JW |
4633 | return &memcg->css; |
4634 | fail: | |
7e97de0b | 4635 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 4636 | mem_cgroup_free(memcg); |
ea3a9645 | 4637 | return ERR_PTR(-ENOMEM); |
0b8f73e1 JW |
4638 | } |
4639 | ||
73f576c0 | 4640 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 4641 | { |
58fa2a55 VD |
4642 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
4643 | ||
0a4465d3 KT |
4644 | /* |
4645 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
4646 | * by the time the maps are allocated. So, we allocate maps | |
4647 | * here, when for_each_mem_cgroup() can't skip it. | |
4648 | */ | |
4649 | if (memcg_alloc_shrinker_maps(memcg)) { | |
4650 | mem_cgroup_id_remove(memcg); | |
4651 | return -ENOMEM; | |
4652 | } | |
4653 | ||
73f576c0 | 4654 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 4655 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 4656 | css_get(css); |
2f7dd7a4 | 4657 | return 0; |
8cdea7c0 BS |
4658 | } |
4659 | ||
eb95419b | 4660 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4661 | { |
eb95419b | 4662 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4663 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4664 | |
4665 | /* | |
4666 | * Unregister events and notify userspace. | |
4667 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4668 | * directory to avoid race between userspace and kernelspace. | |
4669 | */ | |
fba94807 TH |
4670 | spin_lock(&memcg->event_list_lock); |
4671 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4672 | list_del_init(&event->list); |
4673 | schedule_work(&event->remove); | |
4674 | } | |
fba94807 | 4675 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4676 | |
bf8d5d52 | 4677 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4678 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 4679 | |
567e9ab2 | 4680 | memcg_offline_kmem(memcg); |
52ebea74 | 4681 | wb_memcg_offline(memcg); |
73f576c0 | 4682 | |
591edfb1 RG |
4683 | drain_all_stock(memcg); |
4684 | ||
73f576c0 | 4685 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
4686 | } |
4687 | ||
6df38689 VD |
4688 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4689 | { | |
4690 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4691 | ||
4692 | invalidate_reclaim_iterators(memcg); | |
4693 | } | |
4694 | ||
eb95419b | 4695 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4696 | { |
eb95419b | 4697 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4698 | |
f7e1cb6e | 4699 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4700 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4701 | |
0db15298 | 4702 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4703 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4704 | |
0b8f73e1 JW |
4705 | vmpressure_cleanup(&memcg->vmpressure); |
4706 | cancel_work_sync(&memcg->high_work); | |
4707 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 4708 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 4709 | memcg_free_kmem(memcg); |
0b8f73e1 | 4710 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4711 | } |
4712 | ||
1ced953b TH |
4713 | /** |
4714 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4715 | * @css: the target css | |
4716 | * | |
4717 | * Reset the states of the mem_cgroup associated with @css. This is | |
4718 | * invoked when the userland requests disabling on the default hierarchy | |
4719 | * but the memcg is pinned through dependency. The memcg should stop | |
4720 | * applying policies and should revert to the vanilla state as it may be | |
4721 | * made visible again. | |
4722 | * | |
4723 | * The current implementation only resets the essential configurations. | |
4724 | * This needs to be expanded to cover all the visible parts. | |
4725 | */ | |
4726 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4727 | { | |
4728 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4729 | ||
bbec2e15 RG |
4730 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
4731 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
4732 | page_counter_set_max(&memcg->memsw, PAGE_COUNTER_MAX); | |
4733 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); | |
4734 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 4735 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4736 | page_counter_set_low(&memcg->memory, 0); |
241994ed | 4737 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4738 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4739 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4740 | } |
4741 | ||
02491447 | 4742 | #ifdef CONFIG_MMU |
7dc74be0 | 4743 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4744 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4745 | { |
05b84301 | 4746 | int ret; |
9476db97 | 4747 | |
d0164adc MG |
4748 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4749 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4750 | if (!ret) { |
854ffa8d | 4751 | mc.precharge += count; |
854ffa8d DN |
4752 | return ret; |
4753 | } | |
9476db97 | 4754 | |
3674534b | 4755 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 4756 | while (count--) { |
3674534b | 4757 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 4758 | if (ret) |
38c5d72f | 4759 | return ret; |
854ffa8d | 4760 | mc.precharge++; |
9476db97 | 4761 | cond_resched(); |
854ffa8d | 4762 | } |
9476db97 | 4763 | return 0; |
4ffef5fe DN |
4764 | } |
4765 | ||
4ffef5fe DN |
4766 | union mc_target { |
4767 | struct page *page; | |
02491447 | 4768 | swp_entry_t ent; |
4ffef5fe DN |
4769 | }; |
4770 | ||
4ffef5fe | 4771 | enum mc_target_type { |
8d32ff84 | 4772 | MC_TARGET_NONE = 0, |
4ffef5fe | 4773 | MC_TARGET_PAGE, |
02491447 | 4774 | MC_TARGET_SWAP, |
c733a828 | 4775 | MC_TARGET_DEVICE, |
4ffef5fe DN |
4776 | }; |
4777 | ||
90254a65 DN |
4778 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4779 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4780 | { |
c733a828 | 4781 | struct page *page = _vm_normal_page(vma, addr, ptent, true); |
4ffef5fe | 4782 | |
90254a65 DN |
4783 | if (!page || !page_mapped(page)) |
4784 | return NULL; | |
4785 | if (PageAnon(page)) { | |
1dfab5ab | 4786 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4787 | return NULL; |
1dfab5ab JW |
4788 | } else { |
4789 | if (!(mc.flags & MOVE_FILE)) | |
4790 | return NULL; | |
4791 | } | |
90254a65 DN |
4792 | if (!get_page_unless_zero(page)) |
4793 | return NULL; | |
4794 | ||
4795 | return page; | |
4796 | } | |
4797 | ||
c733a828 | 4798 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 4799 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 4800 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 4801 | { |
90254a65 DN |
4802 | struct page *page = NULL; |
4803 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4804 | ||
1dfab5ab | 4805 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4806 | return NULL; |
c733a828 JG |
4807 | |
4808 | /* | |
4809 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
4810 | * a device and because they are not accessible by CPU they are store | |
4811 | * as special swap entry in the CPU page table. | |
4812 | */ | |
4813 | if (is_device_private_entry(ent)) { | |
4814 | page = device_private_entry_to_page(ent); | |
4815 | /* | |
4816 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
4817 | * a refcount of 1 when free (unlike normal page) | |
4818 | */ | |
4819 | if (!page_ref_add_unless(page, 1, 1)) | |
4820 | return NULL; | |
4821 | return page; | |
4822 | } | |
4823 | ||
4b91355e KH |
4824 | /* |
4825 | * Because lookup_swap_cache() updates some statistics counter, | |
4826 | * we call find_get_page() with swapper_space directly. | |
4827 | */ | |
f6ab1f7f | 4828 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
7941d214 | 4829 | if (do_memsw_account()) |
90254a65 DN |
4830 | entry->val = ent.val; |
4831 | ||
4832 | return page; | |
4833 | } | |
4b91355e KH |
4834 | #else |
4835 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 4836 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
4837 | { |
4838 | return NULL; | |
4839 | } | |
4840 | #endif | |
90254a65 | 4841 | |
87946a72 DN |
4842 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4843 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4844 | { | |
4845 | struct page *page = NULL; | |
87946a72 DN |
4846 | struct address_space *mapping; |
4847 | pgoff_t pgoff; | |
4848 | ||
4849 | if (!vma->vm_file) /* anonymous vma */ | |
4850 | return NULL; | |
1dfab5ab | 4851 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4852 | return NULL; |
4853 | ||
87946a72 | 4854 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4855 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4856 | |
4857 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4858 | #ifdef CONFIG_SWAP |
4859 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4860 | if (shmem_mapping(mapping)) { |
4861 | page = find_get_entry(mapping, pgoff); | |
3159f943 | 4862 | if (xa_is_value(page)) { |
139b6a6f | 4863 | swp_entry_t swp = radix_to_swp_entry(page); |
7941d214 | 4864 | if (do_memsw_account()) |
139b6a6f | 4865 | *entry = swp; |
f6ab1f7f HY |
4866 | page = find_get_page(swap_address_space(swp), |
4867 | swp_offset(swp)); | |
139b6a6f JW |
4868 | } |
4869 | } else | |
4870 | page = find_get_page(mapping, pgoff); | |
4871 | #else | |
4872 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4873 | #endif |
87946a72 DN |
4874 | return page; |
4875 | } | |
4876 | ||
b1b0deab CG |
4877 | /** |
4878 | * mem_cgroup_move_account - move account of the page | |
4879 | * @page: the page | |
25843c2b | 4880 | * @compound: charge the page as compound or small page |
b1b0deab CG |
4881 | * @from: mem_cgroup which the page is moved from. |
4882 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4883 | * | |
3ac808fd | 4884 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
4885 | * |
4886 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4887 | * from old cgroup. | |
4888 | */ | |
4889 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 4890 | bool compound, |
b1b0deab CG |
4891 | struct mem_cgroup *from, |
4892 | struct mem_cgroup *to) | |
4893 | { | |
4894 | unsigned long flags; | |
f627c2f5 | 4895 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 4896 | int ret; |
c4843a75 | 4897 | bool anon; |
b1b0deab CG |
4898 | |
4899 | VM_BUG_ON(from == to); | |
4900 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 4901 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
4902 | |
4903 | /* | |
6a93ca8f | 4904 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 4905 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 4906 | */ |
f627c2f5 | 4907 | ret = -EBUSY; |
b1b0deab CG |
4908 | if (!trylock_page(page)) |
4909 | goto out; | |
4910 | ||
4911 | ret = -EINVAL; | |
4912 | if (page->mem_cgroup != from) | |
4913 | goto out_unlock; | |
4914 | ||
c4843a75 GT |
4915 | anon = PageAnon(page); |
4916 | ||
b1b0deab CG |
4917 | spin_lock_irqsave(&from->move_lock, flags); |
4918 | ||
c4843a75 | 4919 | if (!anon && page_mapped(page)) { |
c9019e9b JW |
4920 | __mod_memcg_state(from, NR_FILE_MAPPED, -nr_pages); |
4921 | __mod_memcg_state(to, NR_FILE_MAPPED, nr_pages); | |
b1b0deab CG |
4922 | } |
4923 | ||
c4843a75 GT |
4924 | /* |
4925 | * move_lock grabbed above and caller set from->moving_account, so | |
ccda7f43 | 4926 | * mod_memcg_page_state will serialize updates to PageDirty. |
c4843a75 GT |
4927 | * So mapping should be stable for dirty pages. |
4928 | */ | |
4929 | if (!anon && PageDirty(page)) { | |
4930 | struct address_space *mapping = page_mapping(page); | |
4931 | ||
4932 | if (mapping_cap_account_dirty(mapping)) { | |
c9019e9b JW |
4933 | __mod_memcg_state(from, NR_FILE_DIRTY, -nr_pages); |
4934 | __mod_memcg_state(to, NR_FILE_DIRTY, nr_pages); | |
c4843a75 GT |
4935 | } |
4936 | } | |
4937 | ||
b1b0deab | 4938 | if (PageWriteback(page)) { |
c9019e9b JW |
4939 | __mod_memcg_state(from, NR_WRITEBACK, -nr_pages); |
4940 | __mod_memcg_state(to, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
4941 | } |
4942 | ||
4943 | /* | |
4944 | * It is safe to change page->mem_cgroup here because the page | |
4945 | * is referenced, charged, and isolated - we can't race with | |
4946 | * uncharging, charging, migration, or LRU putback. | |
4947 | */ | |
4948 | ||
4949 | /* caller should have done css_get */ | |
4950 | page->mem_cgroup = to; | |
4951 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4952 | ||
4953 | ret = 0; | |
4954 | ||
4955 | local_irq_disable(); | |
f627c2f5 | 4956 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 4957 | memcg_check_events(to, page); |
f627c2f5 | 4958 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
4959 | memcg_check_events(from, page); |
4960 | local_irq_enable(); | |
4961 | out_unlock: | |
4962 | unlock_page(page); | |
4963 | out: | |
4964 | return ret; | |
4965 | } | |
4966 | ||
7cf7806c LR |
4967 | /** |
4968 | * get_mctgt_type - get target type of moving charge | |
4969 | * @vma: the vma the pte to be checked belongs | |
4970 | * @addr: the address corresponding to the pte to be checked | |
4971 | * @ptent: the pte to be checked | |
4972 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
4973 | * | |
4974 | * Returns | |
4975 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4976 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4977 | * move charge. if @target is not NULL, the page is stored in target->page | |
4978 | * with extra refcnt got(Callers should handle it). | |
4979 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
4980 | * target for charge migration. if @target is not NULL, the entry is stored | |
4981 | * in target->ent. | |
df6ad698 JG |
4982 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PUBLIC |
4983 | * or MEMORY_DEVICE_PRIVATE (so ZONE_DEVICE page and thus not on the lru). | |
4984 | * For now we such page is charge like a regular page would be as for all | |
4985 | * intent and purposes it is just special memory taking the place of a | |
4986 | * regular page. | |
c733a828 JG |
4987 | * |
4988 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
4989 | * |
4990 | * Called with pte lock held. | |
4991 | */ | |
4992 | ||
8d32ff84 | 4993 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4994 | unsigned long addr, pte_t ptent, union mc_target *target) |
4995 | { | |
4996 | struct page *page = NULL; | |
8d32ff84 | 4997 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4998 | swp_entry_t ent = { .val = 0 }; |
4999 | ||
5000 | if (pte_present(ptent)) | |
5001 | page = mc_handle_present_pte(vma, addr, ptent); | |
5002 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5003 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5004 | else if (pte_none(ptent)) |
87946a72 | 5005 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5006 | |
5007 | if (!page && !ent.val) | |
8d32ff84 | 5008 | return ret; |
02491447 | 5009 | if (page) { |
02491447 | 5010 | /* |
0a31bc97 | 5011 | * Do only loose check w/o serialization. |
1306a85a | 5012 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5013 | * not under LRU exclusion. |
02491447 | 5014 | */ |
1306a85a | 5015 | if (page->mem_cgroup == mc.from) { |
02491447 | 5016 | ret = MC_TARGET_PAGE; |
df6ad698 JG |
5017 | if (is_device_private_page(page) || |
5018 | is_device_public_page(page)) | |
c733a828 | 5019 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5020 | if (target) |
5021 | target->page = page; | |
5022 | } | |
5023 | if (!ret || !target) | |
5024 | put_page(page); | |
5025 | } | |
3e14a57b HY |
5026 | /* |
5027 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5028 | * But we cannot move a tail-page in a THP. | |
5029 | */ | |
5030 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5031 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5032 | ret = MC_TARGET_SWAP; |
5033 | if (target) | |
5034 | target->ent = ent; | |
4ffef5fe | 5035 | } |
4ffef5fe DN |
5036 | return ret; |
5037 | } | |
5038 | ||
12724850 NH |
5039 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5040 | /* | |
d6810d73 HY |
5041 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5042 | * not support them for now. | |
12724850 NH |
5043 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5044 | */ | |
5045 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5046 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5047 | { | |
5048 | struct page *page = NULL; | |
12724850 NH |
5049 | enum mc_target_type ret = MC_TARGET_NONE; |
5050 | ||
84c3fc4e ZY |
5051 | if (unlikely(is_swap_pmd(pmd))) { |
5052 | VM_BUG_ON(thp_migration_supported() && | |
5053 | !is_pmd_migration_entry(pmd)); | |
5054 | return ret; | |
5055 | } | |
12724850 | 5056 | page = pmd_page(pmd); |
309381fe | 5057 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5058 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5059 | return ret; |
1306a85a | 5060 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
5061 | ret = MC_TARGET_PAGE; |
5062 | if (target) { | |
5063 | get_page(page); | |
5064 | target->page = page; | |
5065 | } | |
5066 | } | |
5067 | return ret; | |
5068 | } | |
5069 | #else | |
5070 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5071 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5072 | { | |
5073 | return MC_TARGET_NONE; | |
5074 | } | |
5075 | #endif | |
5076 | ||
4ffef5fe DN |
5077 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5078 | unsigned long addr, unsigned long end, | |
5079 | struct mm_walk *walk) | |
5080 | { | |
26bcd64a | 5081 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5082 | pte_t *pte; |
5083 | spinlock_t *ptl; | |
5084 | ||
b6ec57f4 KS |
5085 | ptl = pmd_trans_huge_lock(pmd, vma); |
5086 | if (ptl) { | |
c733a828 JG |
5087 | /* |
5088 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
5089 | * support transparent huge page with MEMORY_DEVICE_PUBLIC or | |
5090 | * MEMORY_DEVICE_PRIVATE but this might change. | |
5091 | */ | |
12724850 NH |
5092 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5093 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5094 | spin_unlock(ptl); |
1a5a9906 | 5095 | return 0; |
12724850 | 5096 | } |
03319327 | 5097 | |
45f83cef AA |
5098 | if (pmd_trans_unstable(pmd)) |
5099 | return 0; | |
4ffef5fe DN |
5100 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5101 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5102 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5103 | mc.precharge++; /* increment precharge temporarily */ |
5104 | pte_unmap_unlock(pte - 1, ptl); | |
5105 | cond_resched(); | |
5106 | ||
7dc74be0 DN |
5107 | return 0; |
5108 | } | |
5109 | ||
4ffef5fe DN |
5110 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5111 | { | |
5112 | unsigned long precharge; | |
4ffef5fe | 5113 | |
26bcd64a NH |
5114 | struct mm_walk mem_cgroup_count_precharge_walk = { |
5115 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5116 | .mm = mm, | |
5117 | }; | |
dfe076b0 | 5118 | down_read(&mm->mmap_sem); |
0247f3f4 JM |
5119 | walk_page_range(0, mm->highest_vm_end, |
5120 | &mem_cgroup_count_precharge_walk); | |
dfe076b0 | 5121 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5122 | |
5123 | precharge = mc.precharge; | |
5124 | mc.precharge = 0; | |
5125 | ||
5126 | return precharge; | |
5127 | } | |
5128 | ||
4ffef5fe DN |
5129 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5130 | { | |
dfe076b0 DN |
5131 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5132 | ||
5133 | VM_BUG_ON(mc.moving_task); | |
5134 | mc.moving_task = current; | |
5135 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5136 | } |
5137 | ||
dfe076b0 DN |
5138 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5139 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5140 | { |
2bd9bb20 KH |
5141 | struct mem_cgroup *from = mc.from; |
5142 | struct mem_cgroup *to = mc.to; | |
5143 | ||
4ffef5fe | 5144 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5145 | if (mc.precharge) { |
00501b53 | 5146 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5147 | mc.precharge = 0; |
5148 | } | |
5149 | /* | |
5150 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5151 | * we must uncharge here. | |
5152 | */ | |
5153 | if (mc.moved_charge) { | |
00501b53 | 5154 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5155 | mc.moved_charge = 0; |
4ffef5fe | 5156 | } |
483c30b5 DN |
5157 | /* we must fixup refcnts and charges */ |
5158 | if (mc.moved_swap) { | |
483c30b5 | 5159 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5160 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5161 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5162 | |
615d66c3 VD |
5163 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5164 | ||
05b84301 | 5165 | /* |
3e32cb2e JW |
5166 | * we charged both to->memory and to->memsw, so we |
5167 | * should uncharge to->memory. | |
05b84301 | 5168 | */ |
ce00a967 | 5169 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5170 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5171 | ||
615d66c3 VD |
5172 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
5173 | css_put_many(&mc.to->css, mc.moved_swap); | |
3e32cb2e | 5174 | |
483c30b5 DN |
5175 | mc.moved_swap = 0; |
5176 | } | |
dfe076b0 DN |
5177 | memcg_oom_recover(from); |
5178 | memcg_oom_recover(to); | |
5179 | wake_up_all(&mc.waitq); | |
5180 | } | |
5181 | ||
5182 | static void mem_cgroup_clear_mc(void) | |
5183 | { | |
264a0ae1 TH |
5184 | struct mm_struct *mm = mc.mm; |
5185 | ||
dfe076b0 DN |
5186 | /* |
5187 | * we must clear moving_task before waking up waiters at the end of | |
5188 | * task migration. | |
5189 | */ | |
5190 | mc.moving_task = NULL; | |
5191 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5192 | spin_lock(&mc.lock); |
4ffef5fe DN |
5193 | mc.from = NULL; |
5194 | mc.to = NULL; | |
264a0ae1 | 5195 | mc.mm = NULL; |
2bd9bb20 | 5196 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5197 | |
5198 | mmput(mm); | |
4ffef5fe DN |
5199 | } |
5200 | ||
1f7dd3e5 | 5201 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5202 | { |
1f7dd3e5 | 5203 | struct cgroup_subsys_state *css; |
eed67d75 | 5204 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5205 | struct mem_cgroup *from; |
4530eddb | 5206 | struct task_struct *leader, *p; |
9f2115f9 | 5207 | struct mm_struct *mm; |
1dfab5ab | 5208 | unsigned long move_flags; |
9f2115f9 | 5209 | int ret = 0; |
7dc74be0 | 5210 | |
1f7dd3e5 TH |
5211 | /* charge immigration isn't supported on the default hierarchy */ |
5212 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5213 | return 0; |
5214 | ||
4530eddb TH |
5215 | /* |
5216 | * Multi-process migrations only happen on the default hierarchy | |
5217 | * where charge immigration is not used. Perform charge | |
5218 | * immigration if @tset contains a leader and whine if there are | |
5219 | * multiple. | |
5220 | */ | |
5221 | p = NULL; | |
1f7dd3e5 | 5222 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5223 | WARN_ON_ONCE(p); |
5224 | p = leader; | |
1f7dd3e5 | 5225 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5226 | } |
5227 | if (!p) | |
5228 | return 0; | |
5229 | ||
1f7dd3e5 TH |
5230 | /* |
5231 | * We are now commited to this value whatever it is. Changes in this | |
5232 | * tunable will only affect upcoming migrations, not the current one. | |
5233 | * So we need to save it, and keep it going. | |
5234 | */ | |
5235 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5236 | if (!move_flags) | |
5237 | return 0; | |
5238 | ||
9f2115f9 TH |
5239 | from = mem_cgroup_from_task(p); |
5240 | ||
5241 | VM_BUG_ON(from == memcg); | |
5242 | ||
5243 | mm = get_task_mm(p); | |
5244 | if (!mm) | |
5245 | return 0; | |
5246 | /* We move charges only when we move a owner of the mm */ | |
5247 | if (mm->owner == p) { | |
5248 | VM_BUG_ON(mc.from); | |
5249 | VM_BUG_ON(mc.to); | |
5250 | VM_BUG_ON(mc.precharge); | |
5251 | VM_BUG_ON(mc.moved_charge); | |
5252 | VM_BUG_ON(mc.moved_swap); | |
5253 | ||
5254 | spin_lock(&mc.lock); | |
264a0ae1 | 5255 | mc.mm = mm; |
9f2115f9 TH |
5256 | mc.from = from; |
5257 | mc.to = memcg; | |
5258 | mc.flags = move_flags; | |
5259 | spin_unlock(&mc.lock); | |
5260 | /* We set mc.moving_task later */ | |
5261 | ||
5262 | ret = mem_cgroup_precharge_mc(mm); | |
5263 | if (ret) | |
5264 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5265 | } else { |
5266 | mmput(mm); | |
7dc74be0 DN |
5267 | } |
5268 | return ret; | |
5269 | } | |
5270 | ||
1f7dd3e5 | 5271 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5272 | { |
4e2f245d JW |
5273 | if (mc.to) |
5274 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5275 | } |
5276 | ||
4ffef5fe DN |
5277 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5278 | unsigned long addr, unsigned long end, | |
5279 | struct mm_walk *walk) | |
7dc74be0 | 5280 | { |
4ffef5fe | 5281 | int ret = 0; |
26bcd64a | 5282 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5283 | pte_t *pte; |
5284 | spinlock_t *ptl; | |
12724850 NH |
5285 | enum mc_target_type target_type; |
5286 | union mc_target target; | |
5287 | struct page *page; | |
4ffef5fe | 5288 | |
b6ec57f4 KS |
5289 | ptl = pmd_trans_huge_lock(pmd, vma); |
5290 | if (ptl) { | |
62ade86a | 5291 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5292 | spin_unlock(ptl); |
12724850 NH |
5293 | return 0; |
5294 | } | |
5295 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5296 | if (target_type == MC_TARGET_PAGE) { | |
5297 | page = target.page; | |
5298 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5299 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5300 | mc.from, mc.to)) { |
12724850 NH |
5301 | mc.precharge -= HPAGE_PMD_NR; |
5302 | mc.moved_charge += HPAGE_PMD_NR; | |
5303 | } | |
5304 | putback_lru_page(page); | |
5305 | } | |
5306 | put_page(page); | |
c733a828 JG |
5307 | } else if (target_type == MC_TARGET_DEVICE) { |
5308 | page = target.page; | |
5309 | if (!mem_cgroup_move_account(page, true, | |
5310 | mc.from, mc.to)) { | |
5311 | mc.precharge -= HPAGE_PMD_NR; | |
5312 | mc.moved_charge += HPAGE_PMD_NR; | |
5313 | } | |
5314 | put_page(page); | |
12724850 | 5315 | } |
bf929152 | 5316 | spin_unlock(ptl); |
1a5a9906 | 5317 | return 0; |
12724850 NH |
5318 | } |
5319 | ||
45f83cef AA |
5320 | if (pmd_trans_unstable(pmd)) |
5321 | return 0; | |
4ffef5fe DN |
5322 | retry: |
5323 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5324 | for (; addr != end; addr += PAGE_SIZE) { | |
5325 | pte_t ptent = *(pte++); | |
c733a828 | 5326 | bool device = false; |
02491447 | 5327 | swp_entry_t ent; |
4ffef5fe DN |
5328 | |
5329 | if (!mc.precharge) | |
5330 | break; | |
5331 | ||
8d32ff84 | 5332 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
5333 | case MC_TARGET_DEVICE: |
5334 | device = true; | |
5335 | /* fall through */ | |
4ffef5fe DN |
5336 | case MC_TARGET_PAGE: |
5337 | page = target.page; | |
53f9263b KS |
5338 | /* |
5339 | * We can have a part of the split pmd here. Moving it | |
5340 | * can be done but it would be too convoluted so simply | |
5341 | * ignore such a partial THP and keep it in original | |
5342 | * memcg. There should be somebody mapping the head. | |
5343 | */ | |
5344 | if (PageTransCompound(page)) | |
5345 | goto put; | |
c733a828 | 5346 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 5347 | goto put; |
f627c2f5 KS |
5348 | if (!mem_cgroup_move_account(page, false, |
5349 | mc.from, mc.to)) { | |
4ffef5fe | 5350 | mc.precharge--; |
854ffa8d DN |
5351 | /* we uncharge from mc.from later. */ |
5352 | mc.moved_charge++; | |
4ffef5fe | 5353 | } |
c733a828 JG |
5354 | if (!device) |
5355 | putback_lru_page(page); | |
8d32ff84 | 5356 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5357 | put_page(page); |
5358 | break; | |
02491447 DN |
5359 | case MC_TARGET_SWAP: |
5360 | ent = target.ent; | |
e91cbb42 | 5361 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5362 | mc.precharge--; |
483c30b5 DN |
5363 | /* we fixup refcnts and charges later. */ |
5364 | mc.moved_swap++; | |
5365 | } | |
02491447 | 5366 | break; |
4ffef5fe DN |
5367 | default: |
5368 | break; | |
5369 | } | |
5370 | } | |
5371 | pte_unmap_unlock(pte - 1, ptl); | |
5372 | cond_resched(); | |
5373 | ||
5374 | if (addr != end) { | |
5375 | /* | |
5376 | * We have consumed all precharges we got in can_attach(). | |
5377 | * We try charge one by one, but don't do any additional | |
5378 | * charges to mc.to if we have failed in charge once in attach() | |
5379 | * phase. | |
5380 | */ | |
854ffa8d | 5381 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5382 | if (!ret) |
5383 | goto retry; | |
5384 | } | |
5385 | ||
5386 | return ret; | |
5387 | } | |
5388 | ||
264a0ae1 | 5389 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 5390 | { |
26bcd64a NH |
5391 | struct mm_walk mem_cgroup_move_charge_walk = { |
5392 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
264a0ae1 | 5393 | .mm = mc.mm, |
26bcd64a | 5394 | }; |
4ffef5fe DN |
5395 | |
5396 | lru_add_drain_all(); | |
312722cb | 5397 | /* |
81f8c3a4 JW |
5398 | * Signal lock_page_memcg() to take the memcg's move_lock |
5399 | * while we're moving its pages to another memcg. Then wait | |
5400 | * for already started RCU-only updates to finish. | |
312722cb JW |
5401 | */ |
5402 | atomic_inc(&mc.from->moving_account); | |
5403 | synchronize_rcu(); | |
dfe076b0 | 5404 | retry: |
264a0ae1 | 5405 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
5406 | /* |
5407 | * Someone who are holding the mmap_sem might be waiting in | |
5408 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5409 | * and retry. Because we cancel precharges, we might not be able | |
5410 | * to move enough charges, but moving charge is a best-effort | |
5411 | * feature anyway, so it wouldn't be a big problem. | |
5412 | */ | |
5413 | __mem_cgroup_clear_mc(); | |
5414 | cond_resched(); | |
5415 | goto retry; | |
5416 | } | |
26bcd64a NH |
5417 | /* |
5418 | * When we have consumed all precharges and failed in doing | |
5419 | * additional charge, the page walk just aborts. | |
5420 | */ | |
0247f3f4 JM |
5421 | walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk); |
5422 | ||
264a0ae1 | 5423 | up_read(&mc.mm->mmap_sem); |
312722cb | 5424 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5425 | } |
5426 | ||
264a0ae1 | 5427 | static void mem_cgroup_move_task(void) |
67e465a7 | 5428 | { |
264a0ae1 TH |
5429 | if (mc.to) { |
5430 | mem_cgroup_move_charge(); | |
a433658c | 5431 | mem_cgroup_clear_mc(); |
264a0ae1 | 5432 | } |
67e465a7 | 5433 | } |
5cfb80a7 | 5434 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5435 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5436 | { |
5437 | return 0; | |
5438 | } | |
1f7dd3e5 | 5439 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5440 | { |
5441 | } | |
264a0ae1 | 5442 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
5443 | { |
5444 | } | |
5445 | #endif | |
67e465a7 | 5446 | |
f00baae7 TH |
5447 | /* |
5448 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5449 | * to verify whether we're attached to the default hierarchy on each mount |
5450 | * attempt. | |
f00baae7 | 5451 | */ |
eb95419b | 5452 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5453 | { |
5454 | /* | |
aa6ec29b | 5455 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5456 | * guarantees that @root doesn't have any children, so turning it |
5457 | * on for the root memcg is enough. | |
5458 | */ | |
9e10a130 | 5459 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5460 | root_mem_cgroup->use_hierarchy = true; |
5461 | else | |
5462 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5463 | } |
5464 | ||
677dc973 CD |
5465 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
5466 | { | |
5467 | if (value == PAGE_COUNTER_MAX) | |
5468 | seq_puts(m, "max\n"); | |
5469 | else | |
5470 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
5471 | ||
5472 | return 0; | |
5473 | } | |
5474 | ||
241994ed JW |
5475 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5476 | struct cftype *cft) | |
5477 | { | |
f5fc3c5d JW |
5478 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5479 | ||
5480 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5481 | } |
5482 | ||
bf8d5d52 RG |
5483 | static int memory_min_show(struct seq_file *m, void *v) |
5484 | { | |
677dc973 CD |
5485 | return seq_puts_memcg_tunable(m, |
5486 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
5487 | } |
5488 | ||
5489 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
5490 | char *buf, size_t nbytes, loff_t off) | |
5491 | { | |
5492 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5493 | unsigned long min; | |
5494 | int err; | |
5495 | ||
5496 | buf = strstrip(buf); | |
5497 | err = page_counter_memparse(buf, "max", &min); | |
5498 | if (err) | |
5499 | return err; | |
5500 | ||
5501 | page_counter_set_min(&memcg->memory, min); | |
5502 | ||
5503 | return nbytes; | |
5504 | } | |
5505 | ||
241994ed JW |
5506 | static int memory_low_show(struct seq_file *m, void *v) |
5507 | { | |
677dc973 CD |
5508 | return seq_puts_memcg_tunable(m, |
5509 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
5510 | } |
5511 | ||
5512 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5513 | char *buf, size_t nbytes, loff_t off) | |
5514 | { | |
5515 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5516 | unsigned long low; | |
5517 | int err; | |
5518 | ||
5519 | buf = strstrip(buf); | |
d2973697 | 5520 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5521 | if (err) |
5522 | return err; | |
5523 | ||
23067153 | 5524 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
5525 | |
5526 | return nbytes; | |
5527 | } | |
5528 | ||
5529 | static int memory_high_show(struct seq_file *m, void *v) | |
5530 | { | |
677dc973 | 5531 | return seq_puts_memcg_tunable(m, READ_ONCE(mem_cgroup_from_seq(m)->high)); |
241994ed JW |
5532 | } |
5533 | ||
5534 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5535 | char *buf, size_t nbytes, loff_t off) | |
5536 | { | |
5537 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 5538 | unsigned long nr_pages; |
241994ed JW |
5539 | unsigned long high; |
5540 | int err; | |
5541 | ||
5542 | buf = strstrip(buf); | |
d2973697 | 5543 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5544 | if (err) |
5545 | return err; | |
5546 | ||
5547 | memcg->high = high; | |
5548 | ||
588083bb JW |
5549 | nr_pages = page_counter_read(&memcg->memory); |
5550 | if (nr_pages > high) | |
5551 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5552 | GFP_KERNEL, true); | |
5553 | ||
2529bb3a | 5554 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5555 | return nbytes; |
5556 | } | |
5557 | ||
5558 | static int memory_max_show(struct seq_file *m, void *v) | |
5559 | { | |
677dc973 CD |
5560 | return seq_puts_memcg_tunable(m, |
5561 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
5562 | } |
5563 | ||
5564 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5565 | char *buf, size_t nbytes, loff_t off) | |
5566 | { | |
5567 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
5568 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5569 | bool drained = false; | |
241994ed JW |
5570 | unsigned long max; |
5571 | int err; | |
5572 | ||
5573 | buf = strstrip(buf); | |
d2973697 | 5574 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5575 | if (err) |
5576 | return err; | |
5577 | ||
bbec2e15 | 5578 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
5579 | |
5580 | for (;;) { | |
5581 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5582 | ||
5583 | if (nr_pages <= max) | |
5584 | break; | |
5585 | ||
5586 | if (signal_pending(current)) { | |
5587 | err = -EINTR; | |
5588 | break; | |
5589 | } | |
5590 | ||
5591 | if (!drained) { | |
5592 | drain_all_stock(memcg); | |
5593 | drained = true; | |
5594 | continue; | |
5595 | } | |
5596 | ||
5597 | if (nr_reclaims) { | |
5598 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5599 | GFP_KERNEL, true)) | |
5600 | nr_reclaims--; | |
5601 | continue; | |
5602 | } | |
5603 | ||
e27be240 | 5604 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
5605 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
5606 | break; | |
5607 | } | |
241994ed | 5608 | |
2529bb3a | 5609 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5610 | return nbytes; |
5611 | } | |
5612 | ||
5613 | static int memory_events_show(struct seq_file *m, void *v) | |
5614 | { | |
aa9694bb | 5615 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 5616 | |
e27be240 JW |
5617 | seq_printf(m, "low %lu\n", |
5618 | atomic_long_read(&memcg->memory_events[MEMCG_LOW])); | |
5619 | seq_printf(m, "high %lu\n", | |
5620 | atomic_long_read(&memcg->memory_events[MEMCG_HIGH])); | |
5621 | seq_printf(m, "max %lu\n", | |
5622 | atomic_long_read(&memcg->memory_events[MEMCG_MAX])); | |
5623 | seq_printf(m, "oom %lu\n", | |
5624 | atomic_long_read(&memcg->memory_events[MEMCG_OOM])); | |
fe6bdfc8 RG |
5625 | seq_printf(m, "oom_kill %lu\n", |
5626 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
241994ed JW |
5627 | |
5628 | return 0; | |
5629 | } | |
5630 | ||
587d9f72 JW |
5631 | static int memory_stat_show(struct seq_file *m, void *v) |
5632 | { | |
aa9694bb | 5633 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
587d9f72 JW |
5634 | int i; |
5635 | ||
5636 | /* | |
5637 | * Provide statistics on the state of the memory subsystem as | |
5638 | * well as cumulative event counters that show past behavior. | |
5639 | * | |
5640 | * This list is ordered following a combination of these gradients: | |
5641 | * 1) generic big picture -> specifics and details | |
5642 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
5643 | * | |
5644 | * Current memory state: | |
5645 | */ | |
5646 | ||
5647 | seq_printf(m, "anon %llu\n", | |
42a30035 | 5648 | (u64)memcg_page_state(memcg, MEMCG_RSS) * PAGE_SIZE); |
587d9f72 | 5649 | seq_printf(m, "file %llu\n", |
42a30035 | 5650 | (u64)memcg_page_state(memcg, MEMCG_CACHE) * PAGE_SIZE); |
12580e4b | 5651 | seq_printf(m, "kernel_stack %llu\n", |
42a30035 | 5652 | (u64)memcg_page_state(memcg, MEMCG_KERNEL_STACK_KB) * 1024); |
27ee57c9 | 5653 | seq_printf(m, "slab %llu\n", |
42a30035 JW |
5654 | (u64)(memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) + |
5655 | memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE)) * | |
5656 | PAGE_SIZE); | |
b2807f07 | 5657 | seq_printf(m, "sock %llu\n", |
42a30035 | 5658 | (u64)memcg_page_state(memcg, MEMCG_SOCK) * PAGE_SIZE); |
587d9f72 | 5659 | |
9a4caf1e | 5660 | seq_printf(m, "shmem %llu\n", |
42a30035 | 5661 | (u64)memcg_page_state(memcg, NR_SHMEM) * PAGE_SIZE); |
587d9f72 | 5662 | seq_printf(m, "file_mapped %llu\n", |
42a30035 | 5663 | (u64)memcg_page_state(memcg, NR_FILE_MAPPED) * PAGE_SIZE); |
587d9f72 | 5664 | seq_printf(m, "file_dirty %llu\n", |
42a30035 | 5665 | (u64)memcg_page_state(memcg, NR_FILE_DIRTY) * PAGE_SIZE); |
587d9f72 | 5666 | seq_printf(m, "file_writeback %llu\n", |
42a30035 | 5667 | (u64)memcg_page_state(memcg, NR_WRITEBACK) * PAGE_SIZE); |
587d9f72 | 5668 | |
1ff9e6e1 CD |
5669 | /* |
5670 | * TODO: We should eventually replace our own MEMCG_RSS_HUGE counter | |
5671 | * with the NR_ANON_THP vm counter, but right now it's a pain in the | |
5672 | * arse because it requires migrating the work out of rmap to a place | |
5673 | * where the page->mem_cgroup is set up and stable. | |
5674 | */ | |
5675 | seq_printf(m, "anon_thp %llu\n", | |
42a30035 | 5676 | (u64)memcg_page_state(memcg, MEMCG_RSS_HUGE) * PAGE_SIZE); |
1ff9e6e1 | 5677 | |
8de7ecc6 SB |
5678 | for (i = 0; i < NR_LRU_LISTS; i++) |
5679 | seq_printf(m, "%s %llu\n", mem_cgroup_lru_names[i], | |
42a30035 JW |
5680 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
5681 | PAGE_SIZE); | |
587d9f72 | 5682 | |
27ee57c9 | 5683 | seq_printf(m, "slab_reclaimable %llu\n", |
42a30035 JW |
5684 | (u64)memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) * |
5685 | PAGE_SIZE); | |
27ee57c9 | 5686 | seq_printf(m, "slab_unreclaimable %llu\n", |
42a30035 JW |
5687 | (u64)memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE) * |
5688 | PAGE_SIZE); | |
27ee57c9 | 5689 | |
587d9f72 JW |
5690 | /* Accumulated memory events */ |
5691 | ||
42a30035 JW |
5692 | seq_printf(m, "pgfault %lu\n", memcg_events(memcg, PGFAULT)); |
5693 | seq_printf(m, "pgmajfault %lu\n", memcg_events(memcg, PGMAJFAULT)); | |
587d9f72 | 5694 | |
e9b257ed | 5695 | seq_printf(m, "workingset_refault %lu\n", |
42a30035 | 5696 | memcg_page_state(memcg, WORKINGSET_REFAULT)); |
e9b257ed | 5697 | seq_printf(m, "workingset_activate %lu\n", |
42a30035 | 5698 | memcg_page_state(memcg, WORKINGSET_ACTIVATE)); |
e9b257ed | 5699 | seq_printf(m, "workingset_nodereclaim %lu\n", |
42a30035 JW |
5700 | memcg_page_state(memcg, WORKINGSET_NODERECLAIM)); |
5701 | ||
5702 | seq_printf(m, "pgrefill %lu\n", memcg_events(memcg, PGREFILL)); | |
5703 | seq_printf(m, "pgscan %lu\n", memcg_events(memcg, PGSCAN_KSWAPD) + | |
5704 | memcg_events(memcg, PGSCAN_DIRECT)); | |
5705 | seq_printf(m, "pgsteal %lu\n", memcg_events(memcg, PGSTEAL_KSWAPD) + | |
5706 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
5707 | seq_printf(m, "pgactivate %lu\n", memcg_events(memcg, PGACTIVATE)); | |
5708 | seq_printf(m, "pgdeactivate %lu\n", memcg_events(memcg, PGDEACTIVATE)); | |
5709 | seq_printf(m, "pglazyfree %lu\n", memcg_events(memcg, PGLAZYFREE)); | |
5710 | seq_printf(m, "pglazyfreed %lu\n", memcg_events(memcg, PGLAZYFREED)); | |
2262185c | 5711 | |
1ff9e6e1 | 5712 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
42a30035 JW |
5713 | seq_printf(m, "thp_fault_alloc %lu\n", |
5714 | memcg_events(memcg, THP_FAULT_ALLOC)); | |
1ff9e6e1 | 5715 | seq_printf(m, "thp_collapse_alloc %lu\n", |
42a30035 | 5716 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1ff9e6e1 CD |
5717 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
5718 | ||
587d9f72 JW |
5719 | return 0; |
5720 | } | |
5721 | ||
3d8b38eb RG |
5722 | static int memory_oom_group_show(struct seq_file *m, void *v) |
5723 | { | |
aa9694bb | 5724 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
5725 | |
5726 | seq_printf(m, "%d\n", memcg->oom_group); | |
5727 | ||
5728 | return 0; | |
5729 | } | |
5730 | ||
5731 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
5732 | char *buf, size_t nbytes, loff_t off) | |
5733 | { | |
5734 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5735 | int ret, oom_group; | |
5736 | ||
5737 | buf = strstrip(buf); | |
5738 | if (!buf) | |
5739 | return -EINVAL; | |
5740 | ||
5741 | ret = kstrtoint(buf, 0, &oom_group); | |
5742 | if (ret) | |
5743 | return ret; | |
5744 | ||
5745 | if (oom_group != 0 && oom_group != 1) | |
5746 | return -EINVAL; | |
5747 | ||
5748 | memcg->oom_group = oom_group; | |
5749 | ||
5750 | return nbytes; | |
5751 | } | |
5752 | ||
241994ed JW |
5753 | static struct cftype memory_files[] = { |
5754 | { | |
5755 | .name = "current", | |
f5fc3c5d | 5756 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5757 | .read_u64 = memory_current_read, |
5758 | }, | |
bf8d5d52 RG |
5759 | { |
5760 | .name = "min", | |
5761 | .flags = CFTYPE_NOT_ON_ROOT, | |
5762 | .seq_show = memory_min_show, | |
5763 | .write = memory_min_write, | |
5764 | }, | |
241994ed JW |
5765 | { |
5766 | .name = "low", | |
5767 | .flags = CFTYPE_NOT_ON_ROOT, | |
5768 | .seq_show = memory_low_show, | |
5769 | .write = memory_low_write, | |
5770 | }, | |
5771 | { | |
5772 | .name = "high", | |
5773 | .flags = CFTYPE_NOT_ON_ROOT, | |
5774 | .seq_show = memory_high_show, | |
5775 | .write = memory_high_write, | |
5776 | }, | |
5777 | { | |
5778 | .name = "max", | |
5779 | .flags = CFTYPE_NOT_ON_ROOT, | |
5780 | .seq_show = memory_max_show, | |
5781 | .write = memory_max_write, | |
5782 | }, | |
5783 | { | |
5784 | .name = "events", | |
5785 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5786 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5787 | .seq_show = memory_events_show, |
5788 | }, | |
587d9f72 JW |
5789 | { |
5790 | .name = "stat", | |
5791 | .flags = CFTYPE_NOT_ON_ROOT, | |
5792 | .seq_show = memory_stat_show, | |
5793 | }, | |
3d8b38eb RG |
5794 | { |
5795 | .name = "oom.group", | |
5796 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
5797 | .seq_show = memory_oom_group_show, | |
5798 | .write = memory_oom_group_write, | |
5799 | }, | |
241994ed JW |
5800 | { } /* terminate */ |
5801 | }; | |
5802 | ||
073219e9 | 5803 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5804 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5805 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5806 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5807 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5808 | .css_free = mem_cgroup_css_free, |
1ced953b | 5809 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5810 | .can_attach = mem_cgroup_can_attach, |
5811 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 5812 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5813 | .bind = mem_cgroup_bind, |
241994ed JW |
5814 | .dfl_cftypes = memory_files, |
5815 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5816 | .early_init = 0, |
8cdea7c0 | 5817 | }; |
c077719b | 5818 | |
241994ed | 5819 | /** |
bf8d5d52 | 5820 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 5821 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
5822 | * @memcg: the memory cgroup to check |
5823 | * | |
23067153 RG |
5824 | * WARNING: This function is not stateless! It can only be used as part |
5825 | * of a top-down tree iteration, not for isolated queries. | |
34c81057 | 5826 | * |
bf8d5d52 RG |
5827 | * Returns one of the following: |
5828 | * MEMCG_PROT_NONE: cgroup memory is not protected | |
5829 | * MEMCG_PROT_LOW: cgroup memory is protected as long there is | |
5830 | * an unprotected supply of reclaimable memory from other cgroups. | |
5831 | * MEMCG_PROT_MIN: cgroup memory is protected | |
34c81057 | 5832 | * |
bf8d5d52 | 5833 | * @root is exclusive; it is never protected when looked at directly |
34c81057 | 5834 | * |
bf8d5d52 RG |
5835 | * To provide a proper hierarchical behavior, effective memory.min/low values |
5836 | * are used. Below is the description of how effective memory.low is calculated. | |
5837 | * Effective memory.min values is calculated in the same way. | |
34c81057 | 5838 | * |
23067153 RG |
5839 | * Effective memory.low is always equal or less than the original memory.low. |
5840 | * If there is no memory.low overcommittment (which is always true for | |
5841 | * top-level memory cgroups), these two values are equal. | |
5842 | * Otherwise, it's a part of parent's effective memory.low, | |
5843 | * calculated as a cgroup's memory.low usage divided by sum of sibling's | |
5844 | * memory.low usages, where memory.low usage is the size of actually | |
5845 | * protected memory. | |
34c81057 | 5846 | * |
23067153 RG |
5847 | * low_usage |
5848 | * elow = min( memory.low, parent->elow * ------------------ ), | |
5849 | * siblings_low_usage | |
34c81057 | 5850 | * |
23067153 RG |
5851 | * | memory.current, if memory.current < memory.low |
5852 | * low_usage = | | |
82ede7ee | 5853 | * | 0, otherwise. |
34c81057 | 5854 | * |
23067153 RG |
5855 | * |
5856 | * Such definition of the effective memory.low provides the expected | |
5857 | * hierarchical behavior: parent's memory.low value is limiting | |
5858 | * children, unprotected memory is reclaimed first and cgroups, | |
5859 | * which are not using their guarantee do not affect actual memory | |
5860 | * distribution. | |
5861 | * | |
5862 | * For example, if there are memcgs A, A/B, A/C, A/D and A/E: | |
5863 | * | |
5864 | * A A/memory.low = 2G, A/memory.current = 6G | |
5865 | * //\\ | |
5866 | * BC DE B/memory.low = 3G B/memory.current = 2G | |
5867 | * C/memory.low = 1G C/memory.current = 2G | |
5868 | * D/memory.low = 0 D/memory.current = 2G | |
5869 | * E/memory.low = 10G E/memory.current = 0 | |
5870 | * | |
5871 | * and the memory pressure is applied, the following memory distribution | |
5872 | * is expected (approximately): | |
5873 | * | |
5874 | * A/memory.current = 2G | |
5875 | * | |
5876 | * B/memory.current = 1.3G | |
5877 | * C/memory.current = 0.6G | |
5878 | * D/memory.current = 0 | |
5879 | * E/memory.current = 0 | |
5880 | * | |
5881 | * These calculations require constant tracking of the actual low usages | |
bf8d5d52 RG |
5882 | * (see propagate_protected_usage()), as well as recursive calculation of |
5883 | * effective memory.low values. But as we do call mem_cgroup_protected() | |
23067153 RG |
5884 | * path for each memory cgroup top-down from the reclaim, |
5885 | * it's possible to optimize this part, and save calculated elow | |
5886 | * for next usage. This part is intentionally racy, but it's ok, | |
5887 | * as memory.low is a best-effort mechanism. | |
241994ed | 5888 | */ |
bf8d5d52 RG |
5889 | enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root, |
5890 | struct mem_cgroup *memcg) | |
241994ed | 5891 | { |
23067153 | 5892 | struct mem_cgroup *parent; |
bf8d5d52 RG |
5893 | unsigned long emin, parent_emin; |
5894 | unsigned long elow, parent_elow; | |
5895 | unsigned long usage; | |
23067153 | 5896 | |
241994ed | 5897 | if (mem_cgroup_disabled()) |
bf8d5d52 | 5898 | return MEMCG_PROT_NONE; |
241994ed | 5899 | |
34c81057 SC |
5900 | if (!root) |
5901 | root = root_mem_cgroup; | |
5902 | if (memcg == root) | |
bf8d5d52 | 5903 | return MEMCG_PROT_NONE; |
241994ed | 5904 | |
23067153 | 5905 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 RG |
5906 | if (!usage) |
5907 | return MEMCG_PROT_NONE; | |
5908 | ||
5909 | emin = memcg->memory.min; | |
5910 | elow = memcg->memory.low; | |
34c81057 | 5911 | |
bf8d5d52 | 5912 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
5913 | /* No parent means a non-hierarchical mode on v1 memcg */ |
5914 | if (!parent) | |
5915 | return MEMCG_PROT_NONE; | |
5916 | ||
23067153 RG |
5917 | if (parent == root) |
5918 | goto exit; | |
5919 | ||
bf8d5d52 RG |
5920 | parent_emin = READ_ONCE(parent->memory.emin); |
5921 | emin = min(emin, parent_emin); | |
5922 | if (emin && parent_emin) { | |
5923 | unsigned long min_usage, siblings_min_usage; | |
5924 | ||
5925 | min_usage = min(usage, memcg->memory.min); | |
5926 | siblings_min_usage = atomic_long_read( | |
5927 | &parent->memory.children_min_usage); | |
5928 | ||
5929 | if (min_usage && siblings_min_usage) | |
5930 | emin = min(emin, parent_emin * min_usage / | |
5931 | siblings_min_usage); | |
5932 | } | |
5933 | ||
23067153 RG |
5934 | parent_elow = READ_ONCE(parent->memory.elow); |
5935 | elow = min(elow, parent_elow); | |
bf8d5d52 RG |
5936 | if (elow && parent_elow) { |
5937 | unsigned long low_usage, siblings_low_usage; | |
23067153 | 5938 | |
bf8d5d52 RG |
5939 | low_usage = min(usage, memcg->memory.low); |
5940 | siblings_low_usage = atomic_long_read( | |
5941 | &parent->memory.children_low_usage); | |
23067153 | 5942 | |
bf8d5d52 RG |
5943 | if (low_usage && siblings_low_usage) |
5944 | elow = min(elow, parent_elow * low_usage / | |
5945 | siblings_low_usage); | |
5946 | } | |
23067153 | 5947 | |
23067153 | 5948 | exit: |
bf8d5d52 | 5949 | memcg->memory.emin = emin; |
23067153 | 5950 | memcg->memory.elow = elow; |
bf8d5d52 RG |
5951 | |
5952 | if (usage <= emin) | |
5953 | return MEMCG_PROT_MIN; | |
5954 | else if (usage <= elow) | |
5955 | return MEMCG_PROT_LOW; | |
5956 | else | |
5957 | return MEMCG_PROT_NONE; | |
241994ed JW |
5958 | } |
5959 | ||
00501b53 JW |
5960 | /** |
5961 | * mem_cgroup_try_charge - try charging a page | |
5962 | * @page: page to charge | |
5963 | * @mm: mm context of the victim | |
5964 | * @gfp_mask: reclaim mode | |
5965 | * @memcgp: charged memcg return | |
25843c2b | 5966 | * @compound: charge the page as compound or small page |
00501b53 JW |
5967 | * |
5968 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5969 | * pages according to @gfp_mask if necessary. | |
5970 | * | |
5971 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5972 | * Otherwise, an error code is returned. | |
5973 | * | |
5974 | * After page->mapping has been set up, the caller must finalize the | |
5975 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5976 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5977 | */ | |
5978 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
5979 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
5980 | bool compound) | |
00501b53 JW |
5981 | { |
5982 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 5983 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5984 | int ret = 0; |
5985 | ||
5986 | if (mem_cgroup_disabled()) | |
5987 | goto out; | |
5988 | ||
5989 | if (PageSwapCache(page)) { | |
00501b53 JW |
5990 | /* |
5991 | * Every swap fault against a single page tries to charge the | |
5992 | * page, bail as early as possible. shmem_unuse() encounters | |
5993 | * already charged pages, too. The USED bit is protected by | |
5994 | * the page lock, which serializes swap cache removal, which | |
5995 | * in turn serializes uncharging. | |
5996 | */ | |
e993d905 | 5997 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 5998 | if (compound_head(page)->mem_cgroup) |
00501b53 | 5999 | goto out; |
e993d905 | 6000 | |
37e84351 | 6001 | if (do_swap_account) { |
e993d905 VD |
6002 | swp_entry_t ent = { .val = page_private(page), }; |
6003 | unsigned short id = lookup_swap_cgroup_id(ent); | |
6004 | ||
6005 | rcu_read_lock(); | |
6006 | memcg = mem_cgroup_from_id(id); | |
6007 | if (memcg && !css_tryget_online(&memcg->css)) | |
6008 | memcg = NULL; | |
6009 | rcu_read_unlock(); | |
6010 | } | |
00501b53 JW |
6011 | } |
6012 | ||
00501b53 JW |
6013 | if (!memcg) |
6014 | memcg = get_mem_cgroup_from_mm(mm); | |
6015 | ||
6016 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
6017 | ||
6018 | css_put(&memcg->css); | |
00501b53 JW |
6019 | out: |
6020 | *memcgp = memcg; | |
6021 | return ret; | |
6022 | } | |
6023 | ||
2cf85583 TH |
6024 | int mem_cgroup_try_charge_delay(struct page *page, struct mm_struct *mm, |
6025 | gfp_t gfp_mask, struct mem_cgroup **memcgp, | |
6026 | bool compound) | |
6027 | { | |
6028 | struct mem_cgroup *memcg; | |
6029 | int ret; | |
6030 | ||
6031 | ret = mem_cgroup_try_charge(page, mm, gfp_mask, memcgp, compound); | |
6032 | memcg = *memcgp; | |
6033 | mem_cgroup_throttle_swaprate(memcg, page_to_nid(page), gfp_mask); | |
6034 | return ret; | |
6035 | } | |
6036 | ||
00501b53 JW |
6037 | /** |
6038 | * mem_cgroup_commit_charge - commit a page charge | |
6039 | * @page: page to charge | |
6040 | * @memcg: memcg to charge the page to | |
6041 | * @lrucare: page might be on LRU already | |
25843c2b | 6042 | * @compound: charge the page as compound or small page |
00501b53 JW |
6043 | * |
6044 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
6045 | * after page->mapping has been set up. This must happen atomically | |
6046 | * as part of the page instantiation, i.e. under the page table lock | |
6047 | * for anonymous pages, under the page lock for page and swap cache. | |
6048 | * | |
6049 | * In addition, the page must not be on the LRU during the commit, to | |
6050 | * prevent racing with task migration. If it might be, use @lrucare. | |
6051 | * | |
6052 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
6053 | */ | |
6054 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 6055 | bool lrucare, bool compound) |
00501b53 | 6056 | { |
f627c2f5 | 6057 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6058 | |
6059 | VM_BUG_ON_PAGE(!page->mapping, page); | |
6060 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
6061 | ||
6062 | if (mem_cgroup_disabled()) | |
6063 | return; | |
6064 | /* | |
6065 | * Swap faults will attempt to charge the same page multiple | |
6066 | * times. But reuse_swap_page() might have removed the page | |
6067 | * from swapcache already, so we can't check PageSwapCache(). | |
6068 | */ | |
6069 | if (!memcg) | |
6070 | return; | |
6071 | ||
6abb5a86 JW |
6072 | commit_charge(page, memcg, lrucare); |
6073 | ||
6abb5a86 | 6074 | local_irq_disable(); |
f627c2f5 | 6075 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
6076 | memcg_check_events(memcg, page); |
6077 | local_irq_enable(); | |
00501b53 | 6078 | |
7941d214 | 6079 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
6080 | swp_entry_t entry = { .val = page_private(page) }; |
6081 | /* | |
6082 | * The swap entry might not get freed for a long time, | |
6083 | * let's not wait for it. The page already received a | |
6084 | * memory+swap charge, drop the swap entry duplicate. | |
6085 | */ | |
38d8b4e6 | 6086 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 JW |
6087 | } |
6088 | } | |
6089 | ||
6090 | /** | |
6091 | * mem_cgroup_cancel_charge - cancel a page charge | |
6092 | * @page: page to charge | |
6093 | * @memcg: memcg to charge the page to | |
25843c2b | 6094 | * @compound: charge the page as compound or small page |
00501b53 JW |
6095 | * |
6096 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
6097 | */ | |
f627c2f5 KS |
6098 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
6099 | bool compound) | |
00501b53 | 6100 | { |
f627c2f5 | 6101 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6102 | |
6103 | if (mem_cgroup_disabled()) | |
6104 | return; | |
6105 | /* | |
6106 | * Swap faults will attempt to charge the same page multiple | |
6107 | * times. But reuse_swap_page() might have removed the page | |
6108 | * from swapcache already, so we can't check PageSwapCache(). | |
6109 | */ | |
6110 | if (!memcg) | |
6111 | return; | |
6112 | ||
00501b53 JW |
6113 | cancel_charge(memcg, nr_pages); |
6114 | } | |
6115 | ||
a9d5adee JG |
6116 | struct uncharge_gather { |
6117 | struct mem_cgroup *memcg; | |
6118 | unsigned long pgpgout; | |
6119 | unsigned long nr_anon; | |
6120 | unsigned long nr_file; | |
6121 | unsigned long nr_kmem; | |
6122 | unsigned long nr_huge; | |
6123 | unsigned long nr_shmem; | |
6124 | struct page *dummy_page; | |
6125 | }; | |
6126 | ||
6127 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6128 | { |
a9d5adee JG |
6129 | memset(ug, 0, sizeof(*ug)); |
6130 | } | |
6131 | ||
6132 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6133 | { | |
6134 | unsigned long nr_pages = ug->nr_anon + ug->nr_file + ug->nr_kmem; | |
747db954 JW |
6135 | unsigned long flags; |
6136 | ||
a9d5adee JG |
6137 | if (!mem_cgroup_is_root(ug->memcg)) { |
6138 | page_counter_uncharge(&ug->memcg->memory, nr_pages); | |
7941d214 | 6139 | if (do_memsw_account()) |
a9d5adee JG |
6140 | page_counter_uncharge(&ug->memcg->memsw, nr_pages); |
6141 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) | |
6142 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6143 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6144 | } |
747db954 JW |
6145 | |
6146 | local_irq_save(flags); | |
c9019e9b JW |
6147 | __mod_memcg_state(ug->memcg, MEMCG_RSS, -ug->nr_anon); |
6148 | __mod_memcg_state(ug->memcg, MEMCG_CACHE, -ug->nr_file); | |
6149 | __mod_memcg_state(ug->memcg, MEMCG_RSS_HUGE, -ug->nr_huge); | |
6150 | __mod_memcg_state(ug->memcg, NR_SHMEM, -ug->nr_shmem); | |
6151 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); | |
871789d4 | 6152 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, nr_pages); |
a9d5adee | 6153 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6154 | local_irq_restore(flags); |
e8ea14cc | 6155 | |
a9d5adee JG |
6156 | if (!mem_cgroup_is_root(ug->memcg)) |
6157 | css_put_many(&ug->memcg->css, nr_pages); | |
6158 | } | |
6159 | ||
6160 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6161 | { | |
6162 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
3f2eb028 JG |
6163 | VM_BUG_ON_PAGE(page_count(page) && !is_zone_device_page(page) && |
6164 | !PageHWPoison(page) , page); | |
a9d5adee JG |
6165 | |
6166 | if (!page->mem_cgroup) | |
6167 | return; | |
6168 | ||
6169 | /* | |
6170 | * Nobody should be changing or seriously looking at | |
6171 | * page->mem_cgroup at this point, we have fully | |
6172 | * exclusive access to the page. | |
6173 | */ | |
6174 | ||
6175 | if (ug->memcg != page->mem_cgroup) { | |
6176 | if (ug->memcg) { | |
6177 | uncharge_batch(ug); | |
6178 | uncharge_gather_clear(ug); | |
6179 | } | |
6180 | ug->memcg = page->mem_cgroup; | |
6181 | } | |
6182 | ||
6183 | if (!PageKmemcg(page)) { | |
6184 | unsigned int nr_pages = 1; | |
6185 | ||
6186 | if (PageTransHuge(page)) { | |
6187 | nr_pages <<= compound_order(page); | |
6188 | ug->nr_huge += nr_pages; | |
6189 | } | |
6190 | if (PageAnon(page)) | |
6191 | ug->nr_anon += nr_pages; | |
6192 | else { | |
6193 | ug->nr_file += nr_pages; | |
6194 | if (PageSwapBacked(page)) | |
6195 | ug->nr_shmem += nr_pages; | |
6196 | } | |
6197 | ug->pgpgout++; | |
6198 | } else { | |
6199 | ug->nr_kmem += 1 << compound_order(page); | |
6200 | __ClearPageKmemcg(page); | |
6201 | } | |
6202 | ||
6203 | ug->dummy_page = page; | |
6204 | page->mem_cgroup = NULL; | |
747db954 JW |
6205 | } |
6206 | ||
6207 | static void uncharge_list(struct list_head *page_list) | |
6208 | { | |
a9d5adee | 6209 | struct uncharge_gather ug; |
747db954 | 6210 | struct list_head *next; |
a9d5adee JG |
6211 | |
6212 | uncharge_gather_clear(&ug); | |
747db954 | 6213 | |
8b592656 JW |
6214 | /* |
6215 | * Note that the list can be a single page->lru; hence the | |
6216 | * do-while loop instead of a simple list_for_each_entry(). | |
6217 | */ | |
747db954 JW |
6218 | next = page_list->next; |
6219 | do { | |
a9d5adee JG |
6220 | struct page *page; |
6221 | ||
747db954 JW |
6222 | page = list_entry(next, struct page, lru); |
6223 | next = page->lru.next; | |
6224 | ||
a9d5adee | 6225 | uncharge_page(page, &ug); |
747db954 JW |
6226 | } while (next != page_list); |
6227 | ||
a9d5adee JG |
6228 | if (ug.memcg) |
6229 | uncharge_batch(&ug); | |
747db954 JW |
6230 | } |
6231 | ||
0a31bc97 JW |
6232 | /** |
6233 | * mem_cgroup_uncharge - uncharge a page | |
6234 | * @page: page to uncharge | |
6235 | * | |
6236 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
6237 | * mem_cgroup_commit_charge(). | |
6238 | */ | |
6239 | void mem_cgroup_uncharge(struct page *page) | |
6240 | { | |
a9d5adee JG |
6241 | struct uncharge_gather ug; |
6242 | ||
0a31bc97 JW |
6243 | if (mem_cgroup_disabled()) |
6244 | return; | |
6245 | ||
747db954 | 6246 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6247 | if (!page->mem_cgroup) |
0a31bc97 JW |
6248 | return; |
6249 | ||
a9d5adee JG |
6250 | uncharge_gather_clear(&ug); |
6251 | uncharge_page(page, &ug); | |
6252 | uncharge_batch(&ug); | |
747db954 | 6253 | } |
0a31bc97 | 6254 | |
747db954 JW |
6255 | /** |
6256 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6257 | * @page_list: list of pages to uncharge | |
6258 | * | |
6259 | * Uncharge a list of pages previously charged with | |
6260 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
6261 | */ | |
6262 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6263 | { | |
6264 | if (mem_cgroup_disabled()) | |
6265 | return; | |
0a31bc97 | 6266 | |
747db954 JW |
6267 | if (!list_empty(page_list)) |
6268 | uncharge_list(page_list); | |
0a31bc97 JW |
6269 | } |
6270 | ||
6271 | /** | |
6a93ca8f JW |
6272 | * mem_cgroup_migrate - charge a page's replacement |
6273 | * @oldpage: currently circulating page | |
6274 | * @newpage: replacement page | |
0a31bc97 | 6275 | * |
6a93ca8f JW |
6276 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6277 | * be uncharged upon free. | |
0a31bc97 JW |
6278 | * |
6279 | * Both pages must be locked, @newpage->mapping must be set up. | |
6280 | */ | |
6a93ca8f | 6281 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6282 | { |
29833315 | 6283 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
6284 | unsigned int nr_pages; |
6285 | bool compound; | |
d93c4130 | 6286 | unsigned long flags; |
0a31bc97 JW |
6287 | |
6288 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6289 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6290 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6291 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6292 | newpage); | |
0a31bc97 JW |
6293 | |
6294 | if (mem_cgroup_disabled()) | |
6295 | return; | |
6296 | ||
6297 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6298 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6299 | return; |
6300 | ||
45637bab | 6301 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6302 | memcg = oldpage->mem_cgroup; |
29833315 | 6303 | if (!memcg) |
0a31bc97 JW |
6304 | return; |
6305 | ||
44b7a8d3 JW |
6306 | /* Force-charge the new page. The old one will be freed soon */ |
6307 | compound = PageTransHuge(newpage); | |
6308 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
6309 | ||
6310 | page_counter_charge(&memcg->memory, nr_pages); | |
6311 | if (do_memsw_account()) | |
6312 | page_counter_charge(&memcg->memsw, nr_pages); | |
6313 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 6314 | |
9cf7666a | 6315 | commit_charge(newpage, memcg, false); |
44b7a8d3 | 6316 | |
d93c4130 | 6317 | local_irq_save(flags); |
44b7a8d3 JW |
6318 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
6319 | memcg_check_events(memcg, newpage); | |
d93c4130 | 6320 | local_irq_restore(flags); |
0a31bc97 JW |
6321 | } |
6322 | ||
ef12947c | 6323 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6324 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6325 | ||
2d758073 | 6326 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6327 | { |
6328 | struct mem_cgroup *memcg; | |
6329 | ||
2d758073 JW |
6330 | if (!mem_cgroup_sockets_enabled) |
6331 | return; | |
6332 | ||
edbe69ef RG |
6333 | /* |
6334 | * Socket cloning can throw us here with sk_memcg already | |
6335 | * filled. It won't however, necessarily happen from | |
6336 | * process context. So the test for root memcg given | |
6337 | * the current task's memcg won't help us in this case. | |
6338 | * | |
6339 | * Respecting the original socket's memcg is a better | |
6340 | * decision in this case. | |
6341 | */ | |
6342 | if (sk->sk_memcg) { | |
6343 | css_get(&sk->sk_memcg->css); | |
6344 | return; | |
6345 | } | |
6346 | ||
11092087 JW |
6347 | rcu_read_lock(); |
6348 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6349 | if (memcg == root_mem_cgroup) |
6350 | goto out; | |
0db15298 | 6351 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6352 | goto out; |
f7e1cb6e | 6353 | if (css_tryget_online(&memcg->css)) |
11092087 | 6354 | sk->sk_memcg = memcg; |
f7e1cb6e | 6355 | out: |
11092087 JW |
6356 | rcu_read_unlock(); |
6357 | } | |
11092087 | 6358 | |
2d758073 | 6359 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6360 | { |
2d758073 JW |
6361 | if (sk->sk_memcg) |
6362 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6363 | } |
6364 | ||
6365 | /** | |
6366 | * mem_cgroup_charge_skmem - charge socket memory | |
6367 | * @memcg: memcg to charge | |
6368 | * @nr_pages: number of pages to charge | |
6369 | * | |
6370 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6371 | * @memcg's configured limit, %false if the charge had to be forced. | |
6372 | */ | |
6373 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6374 | { | |
f7e1cb6e | 6375 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6376 | |
f7e1cb6e | 6377 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6378 | struct page_counter *fail; |
f7e1cb6e | 6379 | |
0db15298 JW |
6380 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6381 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6382 | return true; |
6383 | } | |
0db15298 JW |
6384 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6385 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6386 | return false; |
11092087 | 6387 | } |
d886f4e4 | 6388 | |
f7e1cb6e JW |
6389 | /* Don't block in the packet receive path */ |
6390 | if (in_softirq()) | |
6391 | gfp_mask = GFP_NOWAIT; | |
6392 | ||
c9019e9b | 6393 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6394 | |
f7e1cb6e JW |
6395 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6396 | return true; | |
6397 | ||
6398 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6399 | return false; |
6400 | } | |
6401 | ||
6402 | /** | |
6403 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
6404 | * @memcg: memcg to uncharge |
6405 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
6406 | */ |
6407 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6408 | { | |
f7e1cb6e | 6409 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6410 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
6411 | return; |
6412 | } | |
d886f4e4 | 6413 | |
c9019e9b | 6414 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 6415 | |
475d0487 | 6416 | refill_stock(memcg, nr_pages); |
11092087 JW |
6417 | } |
6418 | ||
f7e1cb6e JW |
6419 | static int __init cgroup_memory(char *s) |
6420 | { | |
6421 | char *token; | |
6422 | ||
6423 | while ((token = strsep(&s, ",")) != NULL) { | |
6424 | if (!*token) | |
6425 | continue; | |
6426 | if (!strcmp(token, "nosocket")) | |
6427 | cgroup_memory_nosocket = true; | |
04823c83 VD |
6428 | if (!strcmp(token, "nokmem")) |
6429 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
6430 | } |
6431 | return 0; | |
6432 | } | |
6433 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 6434 | |
2d11085e | 6435 | /* |
1081312f MH |
6436 | * subsys_initcall() for memory controller. |
6437 | * | |
308167fc SAS |
6438 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
6439 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
6440 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
6441 | * should be initialized from here. | |
2d11085e MH |
6442 | */ |
6443 | static int __init mem_cgroup_init(void) | |
6444 | { | |
95a045f6 JW |
6445 | int cpu, node; |
6446 | ||
84c07d11 | 6447 | #ifdef CONFIG_MEMCG_KMEM |
13583c3d VD |
6448 | /* |
6449 | * Kmem cache creation is mostly done with the slab_mutex held, | |
17cc4dfe TH |
6450 | * so use a workqueue with limited concurrency to avoid stalling |
6451 | * all worker threads in case lots of cgroups are created and | |
6452 | * destroyed simultaneously. | |
13583c3d | 6453 | */ |
17cc4dfe TH |
6454 | memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1); |
6455 | BUG_ON(!memcg_kmem_cache_wq); | |
13583c3d VD |
6456 | #endif |
6457 | ||
308167fc SAS |
6458 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
6459 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
6460 | |
6461 | for_each_possible_cpu(cpu) | |
6462 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
6463 | drain_local_stock); | |
6464 | ||
6465 | for_each_node(node) { | |
6466 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
6467 | |
6468 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
6469 | node_online(node) ? node : NUMA_NO_NODE); | |
6470 | ||
ef8f2327 | 6471 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 6472 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 6473 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
6474 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
6475 | } | |
6476 | ||
2d11085e MH |
6477 | return 0; |
6478 | } | |
6479 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
6480 | |
6481 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
6482 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
6483 | { | |
1c2d479a | 6484 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
6485 | /* |
6486 | * The root cgroup cannot be destroyed, so it's refcount must | |
6487 | * always be >= 1. | |
6488 | */ | |
6489 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
6490 | VM_BUG_ON(1); | |
6491 | break; | |
6492 | } | |
6493 | memcg = parent_mem_cgroup(memcg); | |
6494 | if (!memcg) | |
6495 | memcg = root_mem_cgroup; | |
6496 | } | |
6497 | return memcg; | |
6498 | } | |
6499 | ||
21afa38e JW |
6500 | /** |
6501 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
6502 | * @page: page whose memsw charge to transfer | |
6503 | * @entry: swap entry to move the charge to | |
6504 | * | |
6505 | * Transfer the memsw charge of @page to @entry. | |
6506 | */ | |
6507 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
6508 | { | |
1f47b61f | 6509 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 6510 | unsigned int nr_entries; |
21afa38e JW |
6511 | unsigned short oldid; |
6512 | ||
6513 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6514 | VM_BUG_ON_PAGE(page_count(page), page); | |
6515 | ||
7941d214 | 6516 | if (!do_memsw_account()) |
21afa38e JW |
6517 | return; |
6518 | ||
6519 | memcg = page->mem_cgroup; | |
6520 | ||
6521 | /* Readahead page, never charged */ | |
6522 | if (!memcg) | |
6523 | return; | |
6524 | ||
1f47b61f VD |
6525 | /* |
6526 | * In case the memcg owning these pages has been offlined and doesn't | |
6527 | * have an ID allocated to it anymore, charge the closest online | |
6528 | * ancestor for the swap instead and transfer the memory+swap charge. | |
6529 | */ | |
6530 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
d6810d73 HY |
6531 | nr_entries = hpage_nr_pages(page); |
6532 | /* Get references for the tail pages, too */ | |
6533 | if (nr_entries > 1) | |
6534 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
6535 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
6536 | nr_entries); | |
21afa38e | 6537 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6538 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
6539 | |
6540 | page->mem_cgroup = NULL; | |
6541 | ||
6542 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 6543 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 6544 | |
1f47b61f VD |
6545 | if (memcg != swap_memcg) { |
6546 | if (!mem_cgroup_is_root(swap_memcg)) | |
d6810d73 HY |
6547 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
6548 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
6549 | } |
6550 | ||
ce9ce665 SAS |
6551 | /* |
6552 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 6553 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 6554 | * important here to have the interrupts disabled because it is the |
b93b0163 | 6555 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
6556 | */ |
6557 | VM_BUG_ON(!irqs_disabled()); | |
d6810d73 HY |
6558 | mem_cgroup_charge_statistics(memcg, page, PageTransHuge(page), |
6559 | -nr_entries); | |
21afa38e | 6560 | memcg_check_events(memcg, page); |
73f576c0 JW |
6561 | |
6562 | if (!mem_cgroup_is_root(memcg)) | |
d08afa14 | 6563 | css_put_many(&memcg->css, nr_entries); |
21afa38e JW |
6564 | } |
6565 | ||
38d8b4e6 HY |
6566 | /** |
6567 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
6568 | * @page: page being added to swap |
6569 | * @entry: swap entry to charge | |
6570 | * | |
38d8b4e6 | 6571 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
6572 | * |
6573 | * Returns 0 on success, -ENOMEM on failure. | |
6574 | */ | |
6575 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
6576 | { | |
38d8b4e6 | 6577 | unsigned int nr_pages = hpage_nr_pages(page); |
37e84351 | 6578 | struct page_counter *counter; |
38d8b4e6 | 6579 | struct mem_cgroup *memcg; |
37e84351 VD |
6580 | unsigned short oldid; |
6581 | ||
6582 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
6583 | return 0; | |
6584 | ||
6585 | memcg = page->mem_cgroup; | |
6586 | ||
6587 | /* Readahead page, never charged */ | |
6588 | if (!memcg) | |
6589 | return 0; | |
6590 | ||
f3a53a3a TH |
6591 | if (!entry.val) { |
6592 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 6593 | return 0; |
f3a53a3a | 6594 | } |
bb98f2c5 | 6595 | |
1f47b61f VD |
6596 | memcg = mem_cgroup_id_get_online(memcg); |
6597 | ||
37e84351 | 6598 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 6599 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
6600 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
6601 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 6602 | mem_cgroup_id_put(memcg); |
37e84351 | 6603 | return -ENOMEM; |
1f47b61f | 6604 | } |
37e84351 | 6605 | |
38d8b4e6 HY |
6606 | /* Get references for the tail pages, too */ |
6607 | if (nr_pages > 1) | |
6608 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
6609 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 6610 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6611 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 6612 | |
37e84351 VD |
6613 | return 0; |
6614 | } | |
6615 | ||
21afa38e | 6616 | /** |
38d8b4e6 | 6617 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 6618 | * @entry: swap entry to uncharge |
38d8b4e6 | 6619 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 6620 | */ |
38d8b4e6 | 6621 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
6622 | { |
6623 | struct mem_cgroup *memcg; | |
6624 | unsigned short id; | |
6625 | ||
37e84351 | 6626 | if (!do_swap_account) |
21afa38e JW |
6627 | return; |
6628 | ||
38d8b4e6 | 6629 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 6630 | rcu_read_lock(); |
adbe427b | 6631 | memcg = mem_cgroup_from_id(id); |
21afa38e | 6632 | if (memcg) { |
37e84351 VD |
6633 | if (!mem_cgroup_is_root(memcg)) { |
6634 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
38d8b4e6 | 6635 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 6636 | else |
38d8b4e6 | 6637 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 6638 | } |
c9019e9b | 6639 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 6640 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
6641 | } |
6642 | rcu_read_unlock(); | |
6643 | } | |
6644 | ||
d8b38438 VD |
6645 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
6646 | { | |
6647 | long nr_swap_pages = get_nr_swap_pages(); | |
6648 | ||
6649 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6650 | return nr_swap_pages; | |
6651 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
6652 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 6653 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
6654 | page_counter_read(&memcg->swap)); |
6655 | return nr_swap_pages; | |
6656 | } | |
6657 | ||
5ccc5aba VD |
6658 | bool mem_cgroup_swap_full(struct page *page) |
6659 | { | |
6660 | struct mem_cgroup *memcg; | |
6661 | ||
6662 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
6663 | ||
6664 | if (vm_swap_full()) | |
6665 | return true; | |
6666 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6667 | return false; | |
6668 | ||
6669 | memcg = page->mem_cgroup; | |
6670 | if (!memcg) | |
6671 | return false; | |
6672 | ||
6673 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
bbec2e15 | 6674 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.max) |
5ccc5aba VD |
6675 | return true; |
6676 | ||
6677 | return false; | |
6678 | } | |
6679 | ||
21afa38e JW |
6680 | /* for remember boot option*/ |
6681 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
6682 | static int really_do_swap_account __initdata = 1; | |
6683 | #else | |
6684 | static int really_do_swap_account __initdata; | |
6685 | #endif | |
6686 | ||
6687 | static int __init enable_swap_account(char *s) | |
6688 | { | |
6689 | if (!strcmp(s, "1")) | |
6690 | really_do_swap_account = 1; | |
6691 | else if (!strcmp(s, "0")) | |
6692 | really_do_swap_account = 0; | |
6693 | return 1; | |
6694 | } | |
6695 | __setup("swapaccount=", enable_swap_account); | |
6696 | ||
37e84351 VD |
6697 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
6698 | struct cftype *cft) | |
6699 | { | |
6700 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6701 | ||
6702 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
6703 | } | |
6704 | ||
6705 | static int swap_max_show(struct seq_file *m, void *v) | |
6706 | { | |
677dc973 CD |
6707 | return seq_puts_memcg_tunable(m, |
6708 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
6709 | } |
6710 | ||
6711 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
6712 | char *buf, size_t nbytes, loff_t off) | |
6713 | { | |
6714 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6715 | unsigned long max; | |
6716 | int err; | |
6717 | ||
6718 | buf = strstrip(buf); | |
6719 | err = page_counter_memparse(buf, "max", &max); | |
6720 | if (err) | |
6721 | return err; | |
6722 | ||
be09102b | 6723 | xchg(&memcg->swap.max, max); |
37e84351 VD |
6724 | |
6725 | return nbytes; | |
6726 | } | |
6727 | ||
f3a53a3a TH |
6728 | static int swap_events_show(struct seq_file *m, void *v) |
6729 | { | |
aa9694bb | 6730 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a TH |
6731 | |
6732 | seq_printf(m, "max %lu\n", | |
6733 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
6734 | seq_printf(m, "fail %lu\n", | |
6735 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
6736 | ||
6737 | return 0; | |
6738 | } | |
6739 | ||
37e84351 VD |
6740 | static struct cftype swap_files[] = { |
6741 | { | |
6742 | .name = "swap.current", | |
6743 | .flags = CFTYPE_NOT_ON_ROOT, | |
6744 | .read_u64 = swap_current_read, | |
6745 | }, | |
6746 | { | |
6747 | .name = "swap.max", | |
6748 | .flags = CFTYPE_NOT_ON_ROOT, | |
6749 | .seq_show = swap_max_show, | |
6750 | .write = swap_max_write, | |
6751 | }, | |
f3a53a3a TH |
6752 | { |
6753 | .name = "swap.events", | |
6754 | .flags = CFTYPE_NOT_ON_ROOT, | |
6755 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
6756 | .seq_show = swap_events_show, | |
6757 | }, | |
37e84351 VD |
6758 | { } /* terminate */ |
6759 | }; | |
6760 | ||
21afa38e JW |
6761 | static struct cftype memsw_cgroup_files[] = { |
6762 | { | |
6763 | .name = "memsw.usage_in_bytes", | |
6764 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6765 | .read_u64 = mem_cgroup_read_u64, | |
6766 | }, | |
6767 | { | |
6768 | .name = "memsw.max_usage_in_bytes", | |
6769 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6770 | .write = mem_cgroup_reset, | |
6771 | .read_u64 = mem_cgroup_read_u64, | |
6772 | }, | |
6773 | { | |
6774 | .name = "memsw.limit_in_bytes", | |
6775 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6776 | .write = mem_cgroup_write, | |
6777 | .read_u64 = mem_cgroup_read_u64, | |
6778 | }, | |
6779 | { | |
6780 | .name = "memsw.failcnt", | |
6781 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6782 | .write = mem_cgroup_reset, | |
6783 | .read_u64 = mem_cgroup_read_u64, | |
6784 | }, | |
6785 | { }, /* terminate */ | |
6786 | }; | |
6787 | ||
6788 | static int __init mem_cgroup_swap_init(void) | |
6789 | { | |
6790 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6791 | do_swap_account = 1; | |
37e84351 VD |
6792 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
6793 | swap_files)); | |
21afa38e JW |
6794 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6795 | memsw_cgroup_files)); | |
6796 | } | |
6797 | return 0; | |
6798 | } | |
6799 | subsys_initcall(mem_cgroup_swap_init); | |
6800 | ||
6801 | #endif /* CONFIG_MEMCG_SWAP */ |