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 | ||
815744d7 JW |
694 | __this_cpu_add(memcg->vmstats_local->stat[idx], val); |
695 | ||
db9adbcb JW |
696 | x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); |
697 | if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
698 | struct mem_cgroup *mi; |
699 | ||
42a30035 JW |
700 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
701 | atomic_long_add(x, &mi->vmstats[idx]); | |
db9adbcb JW |
702 | x = 0; |
703 | } | |
704 | __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); | |
705 | } | |
706 | ||
42a30035 JW |
707 | static struct mem_cgroup_per_node * |
708 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
709 | { | |
710 | struct mem_cgroup *parent; | |
711 | ||
712 | parent = parent_mem_cgroup(pn->memcg); | |
713 | if (!parent) | |
714 | return NULL; | |
715 | return mem_cgroup_nodeinfo(parent, nid); | |
716 | } | |
717 | ||
db9adbcb JW |
718 | /** |
719 | * __mod_lruvec_state - update lruvec memory statistics | |
720 | * @lruvec: the lruvec | |
721 | * @idx: the stat item | |
722 | * @val: delta to add to the counter, can be negative | |
723 | * | |
724 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
725 | * function updates the all three counters that are affected by a | |
726 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
727 | */ | |
728 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
729 | int val) | |
730 | { | |
42a30035 | 731 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
db9adbcb | 732 | struct mem_cgroup_per_node *pn; |
42a30035 | 733 | struct mem_cgroup *memcg; |
db9adbcb JW |
734 | long x; |
735 | ||
736 | /* Update node */ | |
42a30035 | 737 | __mod_node_page_state(pgdat, idx, val); |
db9adbcb JW |
738 | |
739 | if (mem_cgroup_disabled()) | |
740 | return; | |
741 | ||
742 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); | |
42a30035 | 743 | memcg = pn->memcg; |
db9adbcb JW |
744 | |
745 | /* Update memcg */ | |
42a30035 | 746 | __mod_memcg_state(memcg, idx, val); |
db9adbcb JW |
747 | |
748 | /* Update lruvec */ | |
815744d7 JW |
749 | __this_cpu_add(pn->lruvec_stat_local->count[idx], val); |
750 | ||
db9adbcb JW |
751 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); |
752 | if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
753 | struct mem_cgroup_per_node *pi; |
754 | ||
42a30035 JW |
755 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) |
756 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
757 | x = 0; |
758 | } | |
759 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
760 | } | |
761 | ||
762 | /** | |
763 | * __count_memcg_events - account VM events in a cgroup | |
764 | * @memcg: the memory cgroup | |
765 | * @idx: the event item | |
766 | * @count: the number of events that occured | |
767 | */ | |
768 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
769 | unsigned long count) | |
770 | { | |
771 | unsigned long x; | |
772 | ||
773 | if (mem_cgroup_disabled()) | |
774 | return; | |
775 | ||
815744d7 JW |
776 | __this_cpu_add(memcg->vmstats_local->events[idx], count); |
777 | ||
db9adbcb JW |
778 | x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); |
779 | if (unlikely(x > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
780 | struct mem_cgroup *mi; |
781 | ||
42a30035 JW |
782 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
783 | atomic_long_add(x, &mi->vmevents[idx]); | |
db9adbcb JW |
784 | x = 0; |
785 | } | |
786 | __this_cpu_write(memcg->vmstats_percpu->events[idx], x); | |
787 | } | |
788 | ||
42a30035 | 789 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 790 | { |
871789d4 | 791 | return atomic_long_read(&memcg->vmevents[event]); |
e9f8974f JW |
792 | } |
793 | ||
42a30035 JW |
794 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
795 | { | |
815744d7 JW |
796 | long x = 0; |
797 | int cpu; | |
798 | ||
799 | for_each_possible_cpu(cpu) | |
800 | x += per_cpu(memcg->vmstats_local->events[event], cpu); | |
801 | return x; | |
42a30035 JW |
802 | } |
803 | ||
c0ff4b85 | 804 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 805 | struct page *page, |
f627c2f5 | 806 | bool compound, int nr_pages) |
d52aa412 | 807 | { |
b2402857 KH |
808 | /* |
809 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
810 | * counted as CACHE even if it's on ANON LRU. | |
811 | */ | |
0a31bc97 | 812 | if (PageAnon(page)) |
c9019e9b | 813 | __mod_memcg_state(memcg, MEMCG_RSS, nr_pages); |
9a4caf1e | 814 | else { |
c9019e9b | 815 | __mod_memcg_state(memcg, MEMCG_CACHE, nr_pages); |
9a4caf1e | 816 | if (PageSwapBacked(page)) |
c9019e9b | 817 | __mod_memcg_state(memcg, NR_SHMEM, nr_pages); |
9a4caf1e | 818 | } |
55e462b0 | 819 | |
f627c2f5 KS |
820 | if (compound) { |
821 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
c9019e9b | 822 | __mod_memcg_state(memcg, MEMCG_RSS_HUGE, nr_pages); |
f627c2f5 | 823 | } |
b070e65c | 824 | |
e401f176 KH |
825 | /* pagein of a big page is an event. So, ignore page size */ |
826 | if (nr_pages > 0) | |
c9019e9b | 827 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 828 | else { |
c9019e9b | 829 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
830 | nr_pages = -nr_pages; /* for event */ |
831 | } | |
e401f176 | 832 | |
871789d4 | 833 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
834 | } |
835 | ||
f53d7ce3 JW |
836 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
837 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
838 | { |
839 | unsigned long val, next; | |
840 | ||
871789d4 CD |
841 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
842 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 843 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 844 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
845 | switch (target) { |
846 | case MEM_CGROUP_TARGET_THRESH: | |
847 | next = val + THRESHOLDS_EVENTS_TARGET; | |
848 | break; | |
bb4cc1a8 AM |
849 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
850 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
851 | break; | |
f53d7ce3 JW |
852 | case MEM_CGROUP_TARGET_NUMAINFO: |
853 | next = val + NUMAINFO_EVENTS_TARGET; | |
854 | break; | |
855 | default: | |
856 | break; | |
857 | } | |
871789d4 | 858 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 859 | return true; |
7a159cc9 | 860 | } |
f53d7ce3 | 861 | return false; |
d2265e6f KH |
862 | } |
863 | ||
864 | /* | |
865 | * Check events in order. | |
866 | * | |
867 | */ | |
c0ff4b85 | 868 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
869 | { |
870 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
871 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
872 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 873 | bool do_softlimit; |
82b3f2a7 | 874 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 875 | |
bb4cc1a8 AM |
876 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
877 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
878 | #if MAX_NUMNODES > 1 |
879 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
880 | MEM_CGROUP_TARGET_NUMAINFO); | |
881 | #endif | |
c0ff4b85 | 882 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
883 | if (unlikely(do_softlimit)) |
884 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 885 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 886 | if (unlikely(do_numainfo)) |
c0ff4b85 | 887 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 888 | #endif |
0a31bc97 | 889 | } |
d2265e6f KH |
890 | } |
891 | ||
cf475ad2 | 892 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 893 | { |
31a78f23 BS |
894 | /* |
895 | * mm_update_next_owner() may clear mm->owner to NULL | |
896 | * if it races with swapoff, page migration, etc. | |
897 | * So this can be called with p == NULL. | |
898 | */ | |
899 | if (unlikely(!p)) | |
900 | return NULL; | |
901 | ||
073219e9 | 902 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 903 | } |
33398cf2 | 904 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 905 | |
d46eb14b SB |
906 | /** |
907 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
908 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
909 | * | |
910 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
911 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
912 | * returned. | |
913 | */ | |
914 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 915 | { |
d46eb14b SB |
916 | struct mem_cgroup *memcg; |
917 | ||
918 | if (mem_cgroup_disabled()) | |
919 | return NULL; | |
0b7f569e | 920 | |
54595fe2 KH |
921 | rcu_read_lock(); |
922 | do { | |
6f6acb00 MH |
923 | /* |
924 | * Page cache insertions can happen withou an | |
925 | * actual mm context, e.g. during disk probing | |
926 | * on boot, loopback IO, acct() writes etc. | |
927 | */ | |
928 | if (unlikely(!mm)) | |
df381975 | 929 | memcg = root_mem_cgroup; |
6f6acb00 MH |
930 | else { |
931 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
932 | if (unlikely(!memcg)) | |
933 | memcg = root_mem_cgroup; | |
934 | } | |
ec903c0c | 935 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 936 | rcu_read_unlock(); |
c0ff4b85 | 937 | return memcg; |
54595fe2 | 938 | } |
d46eb14b SB |
939 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
940 | ||
f745c6f5 SB |
941 | /** |
942 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
943 | * @page: page from which memcg should be extracted. | |
944 | * | |
945 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
946 | * root_mem_cgroup is returned. | |
947 | */ | |
948 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
949 | { | |
950 | struct mem_cgroup *memcg = page->mem_cgroup; | |
951 | ||
952 | if (mem_cgroup_disabled()) | |
953 | return NULL; | |
954 | ||
955 | rcu_read_lock(); | |
956 | if (!memcg || !css_tryget_online(&memcg->css)) | |
957 | memcg = root_mem_cgroup; | |
958 | rcu_read_unlock(); | |
959 | return memcg; | |
960 | } | |
961 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
962 | ||
d46eb14b SB |
963 | /** |
964 | * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg. | |
965 | */ | |
966 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
967 | { | |
968 | if (unlikely(current->active_memcg)) { | |
969 | struct mem_cgroup *memcg = root_mem_cgroup; | |
970 | ||
971 | rcu_read_lock(); | |
972 | if (css_tryget_online(¤t->active_memcg->css)) | |
973 | memcg = current->active_memcg; | |
974 | rcu_read_unlock(); | |
975 | return memcg; | |
976 | } | |
977 | return get_mem_cgroup_from_mm(current->mm); | |
978 | } | |
54595fe2 | 979 | |
5660048c JW |
980 | /** |
981 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
982 | * @root: hierarchy root | |
983 | * @prev: previously returned memcg, NULL on first invocation | |
984 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
985 | * | |
986 | * Returns references to children of the hierarchy below @root, or | |
987 | * @root itself, or %NULL after a full round-trip. | |
988 | * | |
989 | * Caller must pass the return value in @prev on subsequent | |
990 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
991 | * to cancel a hierarchy walk before the round-trip is complete. | |
992 | * | |
b213b54f | 993 | * Reclaimers can specify a node and a priority level in @reclaim to |
5660048c | 994 | * divide up the memcgs in the hierarchy among all concurrent |
b213b54f | 995 | * reclaimers operating on the same node and priority. |
5660048c | 996 | */ |
694fbc0f | 997 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 998 | struct mem_cgroup *prev, |
694fbc0f | 999 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1000 | { |
33398cf2 | 1001 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 1002 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1003 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1004 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1005 | |
694fbc0f AM |
1006 | if (mem_cgroup_disabled()) |
1007 | return NULL; | |
5660048c | 1008 | |
9f3a0d09 JW |
1009 | if (!root) |
1010 | root = root_mem_cgroup; | |
7d74b06f | 1011 | |
9f3a0d09 | 1012 | if (prev && !reclaim) |
5ac8fb31 | 1013 | pos = prev; |
14067bb3 | 1014 | |
9f3a0d09 JW |
1015 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1016 | if (prev) | |
5ac8fb31 | 1017 | goto out; |
694fbc0f | 1018 | return root; |
9f3a0d09 | 1019 | } |
14067bb3 | 1020 | |
542f85f9 | 1021 | rcu_read_lock(); |
5f578161 | 1022 | |
5ac8fb31 | 1023 | if (reclaim) { |
ef8f2327 | 1024 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1025 | |
ef8f2327 | 1026 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
5ac8fb31 JW |
1027 | iter = &mz->iter[reclaim->priority]; |
1028 | ||
1029 | if (prev && reclaim->generation != iter->generation) | |
1030 | goto out_unlock; | |
1031 | ||
6df38689 | 1032 | while (1) { |
4db0c3c2 | 1033 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1034 | if (!pos || css_tryget(&pos->css)) |
1035 | break; | |
5ac8fb31 | 1036 | /* |
6df38689 VD |
1037 | * css reference reached zero, so iter->position will |
1038 | * be cleared by ->css_released. However, we should not | |
1039 | * rely on this happening soon, because ->css_released | |
1040 | * is called from a work queue, and by busy-waiting we | |
1041 | * might block it. So we clear iter->position right | |
1042 | * away. | |
5ac8fb31 | 1043 | */ |
6df38689 VD |
1044 | (void)cmpxchg(&iter->position, pos, NULL); |
1045 | } | |
5ac8fb31 JW |
1046 | } |
1047 | ||
1048 | if (pos) | |
1049 | css = &pos->css; | |
1050 | ||
1051 | for (;;) { | |
1052 | css = css_next_descendant_pre(css, &root->css); | |
1053 | if (!css) { | |
1054 | /* | |
1055 | * Reclaimers share the hierarchy walk, and a | |
1056 | * new one might jump in right at the end of | |
1057 | * the hierarchy - make sure they see at least | |
1058 | * one group and restart from the beginning. | |
1059 | */ | |
1060 | if (!prev) | |
1061 | continue; | |
1062 | break; | |
527a5ec9 | 1063 | } |
7d74b06f | 1064 | |
5ac8fb31 JW |
1065 | /* |
1066 | * Verify the css and acquire a reference. The root | |
1067 | * is provided by the caller, so we know it's alive | |
1068 | * and kicking, and don't take an extra reference. | |
1069 | */ | |
1070 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1071 | |
5ac8fb31 JW |
1072 | if (css == &root->css) |
1073 | break; | |
14067bb3 | 1074 | |
0b8f73e1 JW |
1075 | if (css_tryget(css)) |
1076 | break; | |
9f3a0d09 | 1077 | |
5ac8fb31 | 1078 | memcg = NULL; |
9f3a0d09 | 1079 | } |
5ac8fb31 JW |
1080 | |
1081 | if (reclaim) { | |
5ac8fb31 | 1082 | /* |
6df38689 VD |
1083 | * The position could have already been updated by a competing |
1084 | * thread, so check that the value hasn't changed since we read | |
1085 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1086 | */ |
6df38689 VD |
1087 | (void)cmpxchg(&iter->position, pos, memcg); |
1088 | ||
5ac8fb31 JW |
1089 | if (pos) |
1090 | css_put(&pos->css); | |
1091 | ||
1092 | if (!memcg) | |
1093 | iter->generation++; | |
1094 | else if (!prev) | |
1095 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1096 | } |
5ac8fb31 | 1097 | |
542f85f9 MH |
1098 | out_unlock: |
1099 | rcu_read_unlock(); | |
5ac8fb31 | 1100 | out: |
c40046f3 MH |
1101 | if (prev && prev != root) |
1102 | css_put(&prev->css); | |
1103 | ||
9f3a0d09 | 1104 | return memcg; |
14067bb3 | 1105 | } |
7d74b06f | 1106 | |
5660048c JW |
1107 | /** |
1108 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1109 | * @root: hierarchy root | |
1110 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1111 | */ | |
1112 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1113 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1114 | { |
1115 | if (!root) | |
1116 | root = root_mem_cgroup; | |
1117 | if (prev && prev != root) | |
1118 | css_put(&prev->css); | |
1119 | } | |
7d74b06f | 1120 | |
6df38689 VD |
1121 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1122 | { | |
1123 | struct mem_cgroup *memcg = dead_memcg; | |
1124 | struct mem_cgroup_reclaim_iter *iter; | |
ef8f2327 MG |
1125 | struct mem_cgroup_per_node *mz; |
1126 | int nid; | |
6df38689 VD |
1127 | int i; |
1128 | ||
9f15bde6 | 1129 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
6df38689 | 1130 | for_each_node(nid) { |
ef8f2327 MG |
1131 | mz = mem_cgroup_nodeinfo(memcg, nid); |
1132 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
1133 | iter = &mz->iter[i]; | |
1134 | cmpxchg(&iter->position, | |
1135 | dead_memcg, NULL); | |
6df38689 VD |
1136 | } |
1137 | } | |
1138 | } | |
1139 | } | |
1140 | ||
7c5f64f8 VD |
1141 | /** |
1142 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1143 | * @memcg: hierarchy root | |
1144 | * @fn: function to call for each task | |
1145 | * @arg: argument passed to @fn | |
1146 | * | |
1147 | * This function iterates over tasks attached to @memcg or to any of its | |
1148 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1149 | * value, the function breaks the iteration loop and returns the value. | |
1150 | * Otherwise, it will iterate over all tasks and return 0. | |
1151 | * | |
1152 | * This function must not be called for the root memory cgroup. | |
1153 | */ | |
1154 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1155 | int (*fn)(struct task_struct *, void *), void *arg) | |
1156 | { | |
1157 | struct mem_cgroup *iter; | |
1158 | int ret = 0; | |
1159 | ||
1160 | BUG_ON(memcg == root_mem_cgroup); | |
1161 | ||
1162 | for_each_mem_cgroup_tree(iter, memcg) { | |
1163 | struct css_task_iter it; | |
1164 | struct task_struct *task; | |
1165 | ||
bc2fb7ed | 1166 | css_task_iter_start(&iter->css, 0, &it); |
7c5f64f8 VD |
1167 | while (!ret && (task = css_task_iter_next(&it))) |
1168 | ret = fn(task, arg); | |
1169 | css_task_iter_end(&it); | |
1170 | if (ret) { | |
1171 | mem_cgroup_iter_break(memcg, iter); | |
1172 | break; | |
1173 | } | |
1174 | } | |
1175 | return ret; | |
1176 | } | |
1177 | ||
925b7673 | 1178 | /** |
dfe0e773 | 1179 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1180 | * @page: the page |
f144c390 | 1181 | * @pgdat: pgdat of the page |
dfe0e773 JW |
1182 | * |
1183 | * This function is only safe when following the LRU page isolation | |
1184 | * and putback protocol: the LRU lock must be held, and the page must | |
1185 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1186 | */ |
599d0c95 | 1187 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1188 | { |
ef8f2327 | 1189 | struct mem_cgroup_per_node *mz; |
925b7673 | 1190 | struct mem_cgroup *memcg; |
bea8c150 | 1191 | struct lruvec *lruvec; |
6d12e2d8 | 1192 | |
bea8c150 | 1193 | if (mem_cgroup_disabled()) { |
599d0c95 | 1194 | lruvec = &pgdat->lruvec; |
bea8c150 HD |
1195 | goto out; |
1196 | } | |
925b7673 | 1197 | |
1306a85a | 1198 | memcg = page->mem_cgroup; |
7512102c | 1199 | /* |
dfe0e773 | 1200 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1201 | * possibly migrated - before they are charged. |
7512102c | 1202 | */ |
29833315 JW |
1203 | if (!memcg) |
1204 | memcg = root_mem_cgroup; | |
7512102c | 1205 | |
ef8f2327 | 1206 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1207 | lruvec = &mz->lruvec; |
1208 | out: | |
1209 | /* | |
1210 | * Since a node can be onlined after the mem_cgroup was created, | |
1211 | * we have to be prepared to initialize lruvec->zone here; | |
1212 | * and if offlined then reonlined, we need to reinitialize it. | |
1213 | */ | |
599d0c95 MG |
1214 | if (unlikely(lruvec->pgdat != pgdat)) |
1215 | lruvec->pgdat = pgdat; | |
bea8c150 | 1216 | return lruvec; |
08e552c6 | 1217 | } |
b69408e8 | 1218 | |
925b7673 | 1219 | /** |
fa9add64 HD |
1220 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1221 | * @lruvec: mem_cgroup per zone lru vector | |
1222 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1223 | * @zid: zone id of the accounted pages |
fa9add64 | 1224 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1225 | * |
ca707239 HD |
1226 | * This function must be called under lru_lock, just before a page is added |
1227 | * to or just after a page is removed from an lru list (that ordering being | |
1228 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1229 | */ |
fa9add64 | 1230 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1231 | int zid, int nr_pages) |
3f58a829 | 1232 | { |
ef8f2327 | 1233 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1234 | unsigned long *lru_size; |
ca707239 | 1235 | long size; |
3f58a829 MK |
1236 | |
1237 | if (mem_cgroup_disabled()) | |
1238 | return; | |
1239 | ||
ef8f2327 | 1240 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1241 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1242 | |
1243 | if (nr_pages < 0) | |
1244 | *lru_size += nr_pages; | |
1245 | ||
1246 | size = *lru_size; | |
b4536f0c MH |
1247 | if (WARN_ONCE(size < 0, |
1248 | "%s(%p, %d, %d): lru_size %ld\n", | |
1249 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1250 | VM_BUG_ON(1); |
1251 | *lru_size = 0; | |
1252 | } | |
1253 | ||
1254 | if (nr_pages > 0) | |
1255 | *lru_size += nr_pages; | |
08e552c6 | 1256 | } |
544122e5 | 1257 | |
2314b42d | 1258 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1259 | { |
2314b42d | 1260 | struct mem_cgroup *task_memcg; |
158e0a2d | 1261 | struct task_struct *p; |
ffbdccf5 | 1262 | bool ret; |
4c4a2214 | 1263 | |
158e0a2d | 1264 | p = find_lock_task_mm(task); |
de077d22 | 1265 | if (p) { |
2314b42d | 1266 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1267 | task_unlock(p); |
1268 | } else { | |
1269 | /* | |
1270 | * All threads may have already detached their mm's, but the oom | |
1271 | * killer still needs to detect if they have already been oom | |
1272 | * killed to prevent needlessly killing additional tasks. | |
1273 | */ | |
ffbdccf5 | 1274 | rcu_read_lock(); |
2314b42d JW |
1275 | task_memcg = mem_cgroup_from_task(task); |
1276 | css_get(&task_memcg->css); | |
ffbdccf5 | 1277 | rcu_read_unlock(); |
de077d22 | 1278 | } |
2314b42d JW |
1279 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1280 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1281 | return ret; |
1282 | } | |
1283 | ||
19942822 | 1284 | /** |
9d11ea9f | 1285 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1286 | * @memcg: the memory cgroup |
19942822 | 1287 | * |
9d11ea9f | 1288 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1289 | * pages. |
19942822 | 1290 | */ |
c0ff4b85 | 1291 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1292 | { |
3e32cb2e JW |
1293 | unsigned long margin = 0; |
1294 | unsigned long count; | |
1295 | unsigned long limit; | |
9d11ea9f | 1296 | |
3e32cb2e | 1297 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1298 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1299 | if (count < limit) |
1300 | margin = limit - count; | |
1301 | ||
7941d214 | 1302 | if (do_memsw_account()) { |
3e32cb2e | 1303 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1304 | limit = READ_ONCE(memcg->memsw.max); |
3e32cb2e JW |
1305 | if (count <= limit) |
1306 | margin = min(margin, limit - count); | |
cbedbac3 LR |
1307 | else |
1308 | margin = 0; | |
3e32cb2e JW |
1309 | } |
1310 | ||
1311 | return margin; | |
19942822 JW |
1312 | } |
1313 | ||
32047e2a | 1314 | /* |
bdcbb659 | 1315 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1316 | * |
bdcbb659 QH |
1317 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1318 | * moving cgroups. This is for waiting at high-memory pressure | |
1319 | * caused by "move". | |
32047e2a | 1320 | */ |
c0ff4b85 | 1321 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1322 | { |
2bd9bb20 KH |
1323 | struct mem_cgroup *from; |
1324 | struct mem_cgroup *to; | |
4b534334 | 1325 | bool ret = false; |
2bd9bb20 KH |
1326 | /* |
1327 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1328 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1329 | */ | |
1330 | spin_lock(&mc.lock); | |
1331 | from = mc.from; | |
1332 | to = mc.to; | |
1333 | if (!from) | |
1334 | goto unlock; | |
3e92041d | 1335 | |
2314b42d JW |
1336 | ret = mem_cgroup_is_descendant(from, memcg) || |
1337 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1338 | unlock: |
1339 | spin_unlock(&mc.lock); | |
4b534334 KH |
1340 | return ret; |
1341 | } | |
1342 | ||
c0ff4b85 | 1343 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1344 | { |
1345 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1346 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1347 | DEFINE_WAIT(wait); |
1348 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1349 | /* moving charge context might have finished. */ | |
1350 | if (mc.moving_task) | |
1351 | schedule(); | |
1352 | finish_wait(&mc.waitq, &wait); | |
1353 | return true; | |
1354 | } | |
1355 | } | |
1356 | return false; | |
1357 | } | |
1358 | ||
8ad6e404 | 1359 | static const unsigned int memcg1_stats[] = { |
71cd3113 JW |
1360 | MEMCG_CACHE, |
1361 | MEMCG_RSS, | |
1362 | MEMCG_RSS_HUGE, | |
1363 | NR_SHMEM, | |
1364 | NR_FILE_MAPPED, | |
1365 | NR_FILE_DIRTY, | |
1366 | NR_WRITEBACK, | |
1367 | MEMCG_SWAP, | |
1368 | }; | |
1369 | ||
1370 | static const char *const memcg1_stat_names[] = { | |
1371 | "cache", | |
1372 | "rss", | |
1373 | "rss_huge", | |
1374 | "shmem", | |
1375 | "mapped_file", | |
1376 | "dirty", | |
1377 | "writeback", | |
1378 | "swap", | |
1379 | }; | |
1380 | ||
58cf188e | 1381 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1382 | /** |
f0c867d9 | 1383 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1384 | * memory controller. | |
e222432b BS |
1385 | * @memcg: The memory cgroup that went over limit |
1386 | * @p: Task that is going to be killed | |
1387 | * | |
1388 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1389 | * enabled | |
1390 | */ | |
f0c867d9 | 1391 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1392 | { |
e222432b BS |
1393 | rcu_read_lock(); |
1394 | ||
f0c867d9 | 1395 | if (memcg) { |
1396 | pr_cont(",oom_memcg="); | |
1397 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1398 | } else | |
1399 | pr_cont(",global_oom"); | |
2415b9f5 | 1400 | if (p) { |
f0c867d9 | 1401 | pr_cont(",task_memcg="); |
2415b9f5 | 1402 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1403 | } |
e222432b | 1404 | rcu_read_unlock(); |
f0c867d9 | 1405 | } |
1406 | ||
1407 | /** | |
1408 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1409 | * memory controller. | |
1410 | * @memcg: The memory cgroup that went over limit | |
1411 | */ | |
1412 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1413 | { | |
1414 | struct mem_cgroup *iter; | |
1415 | unsigned int i; | |
e222432b | 1416 | |
3e32cb2e JW |
1417 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1418 | K((u64)page_counter_read(&memcg->memory)), | |
bbec2e15 | 1419 | K((u64)memcg->memory.max), memcg->memory.failcnt); |
3e32cb2e JW |
1420 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", |
1421 | K((u64)page_counter_read(&memcg->memsw)), | |
bbec2e15 | 1422 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); |
3e32cb2e JW |
1423 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", |
1424 | K((u64)page_counter_read(&memcg->kmem)), | |
bbec2e15 | 1425 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); |
58cf188e SZ |
1426 | |
1427 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1428 | pr_info("Memory cgroup stats for "); |
1429 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1430 | pr_cont(":"); |
1431 | ||
71cd3113 JW |
1432 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
1433 | if (memcg1_stats[i] == MEMCG_SWAP && !do_swap_account) | |
58cf188e | 1434 | continue; |
71cd3113 | 1435 | pr_cont(" %s:%luKB", memcg1_stat_names[i], |
205b20cc JW |
1436 | K(memcg_page_state_local(iter, |
1437 | memcg1_stats[i]))); | |
58cf188e SZ |
1438 | } |
1439 | ||
1440 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1441 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
205b20cc JW |
1442 | K(memcg_page_state_local(iter, |
1443 | NR_LRU_BASE + i))); | |
58cf188e SZ |
1444 | |
1445 | pr_cont("\n"); | |
1446 | } | |
e222432b BS |
1447 | } |
1448 | ||
a63d83f4 DR |
1449 | /* |
1450 | * Return the memory (and swap, if configured) limit for a memcg. | |
1451 | */ | |
bbec2e15 | 1452 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1453 | { |
bbec2e15 | 1454 | unsigned long max; |
f3e8eb70 | 1455 | |
bbec2e15 | 1456 | max = memcg->memory.max; |
9a5a8f19 | 1457 | if (mem_cgroup_swappiness(memcg)) { |
bbec2e15 RG |
1458 | unsigned long memsw_max; |
1459 | unsigned long swap_max; | |
9a5a8f19 | 1460 | |
bbec2e15 RG |
1461 | memsw_max = memcg->memsw.max; |
1462 | swap_max = memcg->swap.max; | |
1463 | swap_max = min(swap_max, (unsigned long)total_swap_pages); | |
1464 | max = min(max + swap_max, memsw_max); | |
9a5a8f19 | 1465 | } |
bbec2e15 | 1466 | return max; |
a63d83f4 DR |
1467 | } |
1468 | ||
b6e6edcf | 1469 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1470 | int order) |
9cbb78bb | 1471 | { |
6e0fc46d DR |
1472 | struct oom_control oc = { |
1473 | .zonelist = NULL, | |
1474 | .nodemask = NULL, | |
2a966b77 | 1475 | .memcg = memcg, |
6e0fc46d DR |
1476 | .gfp_mask = gfp_mask, |
1477 | .order = order, | |
6e0fc46d | 1478 | }; |
7c5f64f8 | 1479 | bool ret; |
9cbb78bb | 1480 | |
7775face TH |
1481 | if (mutex_lock_killable(&oom_lock)) |
1482 | return true; | |
1483 | /* | |
1484 | * A few threads which were not waiting at mutex_lock_killable() can | |
1485 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1486 | */ | |
1487 | ret = should_force_charge() || out_of_memory(&oc); | |
dc56401f | 1488 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1489 | return ret; |
9cbb78bb DR |
1490 | } |
1491 | ||
ae6e71d3 MC |
1492 | #if MAX_NUMNODES > 1 |
1493 | ||
4d0c066d KH |
1494 | /** |
1495 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1496 | * @memcg: the target memcg |
4d0c066d KH |
1497 | * @nid: the node ID to be checked. |
1498 | * @noswap : specify true here if the user wants flle only information. | |
1499 | * | |
1500 | * This function returns whether the specified memcg contains any | |
1501 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1502 | * pages in the node. | |
1503 | */ | |
c0ff4b85 | 1504 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1505 | int nid, bool noswap) |
1506 | { | |
2b487e59 JW |
1507 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); |
1508 | ||
def0fdae JW |
1509 | if (lruvec_page_state(lruvec, NR_INACTIVE_FILE) || |
1510 | lruvec_page_state(lruvec, NR_ACTIVE_FILE)) | |
4d0c066d KH |
1511 | return true; |
1512 | if (noswap || !total_swap_pages) | |
1513 | return false; | |
def0fdae JW |
1514 | if (lruvec_page_state(lruvec, NR_INACTIVE_ANON) || |
1515 | lruvec_page_state(lruvec, NR_ACTIVE_ANON)) | |
4d0c066d KH |
1516 | return true; |
1517 | return false; | |
1518 | ||
1519 | } | |
889976db YH |
1520 | |
1521 | /* | |
1522 | * Always updating the nodemask is not very good - even if we have an empty | |
1523 | * list or the wrong list here, we can start from some node and traverse all | |
1524 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1525 | * | |
1526 | */ | |
c0ff4b85 | 1527 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1528 | { |
1529 | int nid; | |
453a9bf3 KH |
1530 | /* |
1531 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1532 | * pagein/pageout changes since the last update. | |
1533 | */ | |
c0ff4b85 | 1534 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1535 | return; |
c0ff4b85 | 1536 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1537 | return; |
1538 | ||
889976db | 1539 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1540 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1541 | |
31aaea4a | 1542 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1543 | |
c0ff4b85 R |
1544 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1545 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1546 | } |
453a9bf3 | 1547 | |
c0ff4b85 R |
1548 | atomic_set(&memcg->numainfo_events, 0); |
1549 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1550 | } |
1551 | ||
1552 | /* | |
1553 | * Selecting a node where we start reclaim from. Because what we need is just | |
1554 | * reducing usage counter, start from anywhere is O,K. Considering | |
1555 | * memory reclaim from current node, there are pros. and cons. | |
1556 | * | |
1557 | * Freeing memory from current node means freeing memory from a node which | |
1558 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1559 | * hit limits, it will see a contention on a node. But freeing from remote | |
1560 | * node means more costs for memory reclaim because of memory latency. | |
1561 | * | |
1562 | * Now, we use round-robin. Better algorithm is welcomed. | |
1563 | */ | |
c0ff4b85 | 1564 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1565 | { |
1566 | int node; | |
1567 | ||
c0ff4b85 R |
1568 | mem_cgroup_may_update_nodemask(memcg); |
1569 | node = memcg->last_scanned_node; | |
889976db | 1570 | |
0edaf86c | 1571 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1572 | /* |
fda3d69b MH |
1573 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1574 | * last time it really checked all the LRUs due to rate limiting. | |
1575 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1576 | */ |
1577 | if (unlikely(node == MAX_NUMNODES)) | |
1578 | node = numa_node_id(); | |
1579 | ||
c0ff4b85 | 1580 | memcg->last_scanned_node = node; |
889976db YH |
1581 | return node; |
1582 | } | |
889976db | 1583 | #else |
c0ff4b85 | 1584 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1585 | { |
1586 | return 0; | |
1587 | } | |
1588 | #endif | |
1589 | ||
0608f43d | 1590 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1591 | pg_data_t *pgdat, |
0608f43d AM |
1592 | gfp_t gfp_mask, |
1593 | unsigned long *total_scanned) | |
1594 | { | |
1595 | struct mem_cgroup *victim = NULL; | |
1596 | int total = 0; | |
1597 | int loop = 0; | |
1598 | unsigned long excess; | |
1599 | unsigned long nr_scanned; | |
1600 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1601 | .pgdat = pgdat, |
0608f43d AM |
1602 | .priority = 0, |
1603 | }; | |
1604 | ||
3e32cb2e | 1605 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1606 | |
1607 | while (1) { | |
1608 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1609 | if (!victim) { | |
1610 | loop++; | |
1611 | if (loop >= 2) { | |
1612 | /* | |
1613 | * If we have not been able to reclaim | |
1614 | * anything, it might because there are | |
1615 | * no reclaimable pages under this hierarchy | |
1616 | */ | |
1617 | if (!total) | |
1618 | break; | |
1619 | /* | |
1620 | * We want to do more targeted reclaim. | |
1621 | * excess >> 2 is not to excessive so as to | |
1622 | * reclaim too much, nor too less that we keep | |
1623 | * coming back to reclaim from this cgroup | |
1624 | */ | |
1625 | if (total >= (excess >> 2) || | |
1626 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1627 | break; | |
1628 | } | |
1629 | continue; | |
1630 | } | |
a9dd0a83 | 1631 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1632 | pgdat, &nr_scanned); |
0608f43d | 1633 | *total_scanned += nr_scanned; |
3e32cb2e | 1634 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1635 | break; |
6d61ef40 | 1636 | } |
0608f43d AM |
1637 | mem_cgroup_iter_break(root_memcg, victim); |
1638 | return total; | |
6d61ef40 BS |
1639 | } |
1640 | ||
0056f4e6 JW |
1641 | #ifdef CONFIG_LOCKDEP |
1642 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1643 | .name = "memcg_oom_lock", | |
1644 | }; | |
1645 | #endif | |
1646 | ||
fb2a6fc5 JW |
1647 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1648 | ||
867578cb KH |
1649 | /* |
1650 | * Check OOM-Killer is already running under our hierarchy. | |
1651 | * If someone is running, return false. | |
1652 | */ | |
fb2a6fc5 | 1653 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1654 | { |
79dfdacc | 1655 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1656 | |
fb2a6fc5 JW |
1657 | spin_lock(&memcg_oom_lock); |
1658 | ||
9f3a0d09 | 1659 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1660 | if (iter->oom_lock) { |
79dfdacc MH |
1661 | /* |
1662 | * this subtree of our hierarchy is already locked | |
1663 | * so we cannot give a lock. | |
1664 | */ | |
79dfdacc | 1665 | failed = iter; |
9f3a0d09 JW |
1666 | mem_cgroup_iter_break(memcg, iter); |
1667 | break; | |
23751be0 JW |
1668 | } else |
1669 | iter->oom_lock = true; | |
7d74b06f | 1670 | } |
867578cb | 1671 | |
fb2a6fc5 JW |
1672 | if (failed) { |
1673 | /* | |
1674 | * OK, we failed to lock the whole subtree so we have | |
1675 | * to clean up what we set up to the failing subtree | |
1676 | */ | |
1677 | for_each_mem_cgroup_tree(iter, memcg) { | |
1678 | if (iter == failed) { | |
1679 | mem_cgroup_iter_break(memcg, iter); | |
1680 | break; | |
1681 | } | |
1682 | iter->oom_lock = false; | |
79dfdacc | 1683 | } |
0056f4e6 JW |
1684 | } else |
1685 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1686 | |
1687 | spin_unlock(&memcg_oom_lock); | |
1688 | ||
1689 | return !failed; | |
a636b327 | 1690 | } |
0b7f569e | 1691 | |
fb2a6fc5 | 1692 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1693 | { |
7d74b06f KH |
1694 | struct mem_cgroup *iter; |
1695 | ||
fb2a6fc5 | 1696 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1697 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1698 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1699 | iter->oom_lock = false; |
fb2a6fc5 | 1700 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1701 | } |
1702 | ||
c0ff4b85 | 1703 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1704 | { |
1705 | struct mem_cgroup *iter; | |
1706 | ||
c2b42d3c | 1707 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1708 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1709 | iter->under_oom++; |
1710 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1711 | } |
1712 | ||
c0ff4b85 | 1713 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1714 | { |
1715 | struct mem_cgroup *iter; | |
1716 | ||
867578cb KH |
1717 | /* |
1718 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1719 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1720 | */ |
c2b42d3c | 1721 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1722 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1723 | if (iter->under_oom > 0) |
1724 | iter->under_oom--; | |
1725 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1726 | } |
1727 | ||
867578cb KH |
1728 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1729 | ||
dc98df5a | 1730 | struct oom_wait_info { |
d79154bb | 1731 | struct mem_cgroup *memcg; |
ac6424b9 | 1732 | wait_queue_entry_t wait; |
dc98df5a KH |
1733 | }; |
1734 | ||
ac6424b9 | 1735 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1736 | unsigned mode, int sync, void *arg) |
1737 | { | |
d79154bb HD |
1738 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1739 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1740 | struct oom_wait_info *oom_wait_info; |
1741 | ||
1742 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1743 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1744 | |
2314b42d JW |
1745 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1746 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1747 | return 0; |
dc98df5a KH |
1748 | return autoremove_wake_function(wait, mode, sync, arg); |
1749 | } | |
1750 | ||
c0ff4b85 | 1751 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1752 | { |
c2b42d3c TH |
1753 | /* |
1754 | * For the following lockless ->under_oom test, the only required | |
1755 | * guarantee is that it must see the state asserted by an OOM when | |
1756 | * this function is called as a result of userland actions | |
1757 | * triggered by the notification of the OOM. This is trivially | |
1758 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1759 | * triggering notification. | |
1760 | */ | |
1761 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1762 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1763 | } |
1764 | ||
29ef680a MH |
1765 | enum oom_status { |
1766 | OOM_SUCCESS, | |
1767 | OOM_FAILED, | |
1768 | OOM_ASYNC, | |
1769 | OOM_SKIPPED | |
1770 | }; | |
1771 | ||
1772 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1773 | { |
7056d3a3 MH |
1774 | enum oom_status ret; |
1775 | bool locked; | |
1776 | ||
29ef680a MH |
1777 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1778 | return OOM_SKIPPED; | |
1779 | ||
7a1adfdd RG |
1780 | memcg_memory_event(memcg, MEMCG_OOM); |
1781 | ||
867578cb | 1782 | /* |
49426420 JW |
1783 | * We are in the middle of the charge context here, so we |
1784 | * don't want to block when potentially sitting on a callstack | |
1785 | * that holds all kinds of filesystem and mm locks. | |
1786 | * | |
29ef680a MH |
1787 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1788 | * handling until the charge can succeed; remember the context and put | |
1789 | * the task to sleep at the end of the page fault when all locks are | |
1790 | * released. | |
49426420 | 1791 | * |
29ef680a MH |
1792 | * On the other hand, in-kernel OOM killer allows for an async victim |
1793 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1794 | * relying on the oom victim to make a forward progress and we can | |
1795 | * invoke the oom killer here. | |
1796 | * | |
1797 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1798 | * victim and then we have to bail out from the charge path. | |
867578cb | 1799 | */ |
29ef680a MH |
1800 | if (memcg->oom_kill_disable) { |
1801 | if (!current->in_user_fault) | |
1802 | return OOM_SKIPPED; | |
1803 | css_get(&memcg->css); | |
1804 | current->memcg_in_oom = memcg; | |
1805 | current->memcg_oom_gfp_mask = mask; | |
1806 | current->memcg_oom_order = order; | |
1807 | ||
1808 | return OOM_ASYNC; | |
1809 | } | |
1810 | ||
7056d3a3 MH |
1811 | mem_cgroup_mark_under_oom(memcg); |
1812 | ||
1813 | locked = mem_cgroup_oom_trylock(memcg); | |
1814 | ||
1815 | if (locked) | |
1816 | mem_cgroup_oom_notify(memcg); | |
1817 | ||
1818 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1819 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1820 | ret = OOM_SUCCESS; |
1821 | else | |
1822 | ret = OOM_FAILED; | |
1823 | ||
1824 | if (locked) | |
1825 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1826 | |
7056d3a3 | 1827 | return ret; |
3812c8c8 JW |
1828 | } |
1829 | ||
1830 | /** | |
1831 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1832 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1833 | * |
49426420 JW |
1834 | * This has to be called at the end of a page fault if the memcg OOM |
1835 | * handler was enabled. | |
3812c8c8 | 1836 | * |
49426420 | 1837 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1838 | * sleep on a waitqueue until the userspace task resolves the |
1839 | * situation. Sleeping directly in the charge context with all kinds | |
1840 | * of locks held is not a good idea, instead we remember an OOM state | |
1841 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1842 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1843 | * |
1844 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1845 | * completed, %false otherwise. |
3812c8c8 | 1846 | */ |
49426420 | 1847 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1848 | { |
626ebc41 | 1849 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1850 | struct oom_wait_info owait; |
49426420 | 1851 | bool locked; |
3812c8c8 JW |
1852 | |
1853 | /* OOM is global, do not handle */ | |
3812c8c8 | 1854 | if (!memcg) |
49426420 | 1855 | return false; |
3812c8c8 | 1856 | |
7c5f64f8 | 1857 | if (!handle) |
49426420 | 1858 | goto cleanup; |
3812c8c8 JW |
1859 | |
1860 | owait.memcg = memcg; | |
1861 | owait.wait.flags = 0; | |
1862 | owait.wait.func = memcg_oom_wake_function; | |
1863 | owait.wait.private = current; | |
2055da97 | 1864 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1865 | |
3812c8c8 | 1866 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1867 | mem_cgroup_mark_under_oom(memcg); |
1868 | ||
1869 | locked = mem_cgroup_oom_trylock(memcg); | |
1870 | ||
1871 | if (locked) | |
1872 | mem_cgroup_oom_notify(memcg); | |
1873 | ||
1874 | if (locked && !memcg->oom_kill_disable) { | |
1875 | mem_cgroup_unmark_under_oom(memcg); | |
1876 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1877 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1878 | current->memcg_oom_order); | |
49426420 | 1879 | } else { |
3812c8c8 | 1880 | schedule(); |
49426420 JW |
1881 | mem_cgroup_unmark_under_oom(memcg); |
1882 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1883 | } | |
1884 | ||
1885 | if (locked) { | |
fb2a6fc5 JW |
1886 | mem_cgroup_oom_unlock(memcg); |
1887 | /* | |
1888 | * There is no guarantee that an OOM-lock contender | |
1889 | * sees the wakeups triggered by the OOM kill | |
1890 | * uncharges. Wake any sleepers explicitely. | |
1891 | */ | |
1892 | memcg_oom_recover(memcg); | |
1893 | } | |
49426420 | 1894 | cleanup: |
626ebc41 | 1895 | current->memcg_in_oom = NULL; |
3812c8c8 | 1896 | css_put(&memcg->css); |
867578cb | 1897 | return true; |
0b7f569e KH |
1898 | } |
1899 | ||
3d8b38eb RG |
1900 | /** |
1901 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
1902 | * @victim: task to be killed by the OOM killer | |
1903 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
1904 | * | |
1905 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
1906 | * by killing all belonging OOM-killable tasks. | |
1907 | * | |
1908 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
1909 | */ | |
1910 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
1911 | struct mem_cgroup *oom_domain) | |
1912 | { | |
1913 | struct mem_cgroup *oom_group = NULL; | |
1914 | struct mem_cgroup *memcg; | |
1915 | ||
1916 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
1917 | return NULL; | |
1918 | ||
1919 | if (!oom_domain) | |
1920 | oom_domain = root_mem_cgroup; | |
1921 | ||
1922 | rcu_read_lock(); | |
1923 | ||
1924 | memcg = mem_cgroup_from_task(victim); | |
1925 | if (memcg == root_mem_cgroup) | |
1926 | goto out; | |
1927 | ||
1928 | /* | |
1929 | * Traverse the memory cgroup hierarchy from the victim task's | |
1930 | * cgroup up to the OOMing cgroup (or root) to find the | |
1931 | * highest-level memory cgroup with oom.group set. | |
1932 | */ | |
1933 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
1934 | if (memcg->oom_group) | |
1935 | oom_group = memcg; | |
1936 | ||
1937 | if (memcg == oom_domain) | |
1938 | break; | |
1939 | } | |
1940 | ||
1941 | if (oom_group) | |
1942 | css_get(&oom_group->css); | |
1943 | out: | |
1944 | rcu_read_unlock(); | |
1945 | ||
1946 | return oom_group; | |
1947 | } | |
1948 | ||
1949 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
1950 | { | |
1951 | pr_info("Tasks in "); | |
1952 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1953 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
1954 | } | |
1955 | ||
d7365e78 | 1956 | /** |
81f8c3a4 JW |
1957 | * lock_page_memcg - lock a page->mem_cgroup binding |
1958 | * @page: the page | |
32047e2a | 1959 | * |
81f8c3a4 | 1960 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
1961 | * another cgroup. |
1962 | * | |
1963 | * It ensures lifetime of the returned memcg. Caller is responsible | |
1964 | * for the lifetime of the page; __unlock_page_memcg() is available | |
1965 | * when @page might get freed inside the locked section. | |
d69b042f | 1966 | */ |
739f79fc | 1967 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 KH |
1968 | { |
1969 | struct mem_cgroup *memcg; | |
6de22619 | 1970 | unsigned long flags; |
89c06bd5 | 1971 | |
6de22619 JW |
1972 | /* |
1973 | * The RCU lock is held throughout the transaction. The fast | |
1974 | * path can get away without acquiring the memcg->move_lock | |
1975 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
1976 | * |
1977 | * The RCU lock also protects the memcg from being freed when | |
1978 | * the page state that is going to change is the only thing | |
1979 | * preventing the page itself from being freed. E.g. writeback | |
1980 | * doesn't hold a page reference and relies on PG_writeback to | |
1981 | * keep off truncation, migration and so forth. | |
1982 | */ | |
d7365e78 JW |
1983 | rcu_read_lock(); |
1984 | ||
1985 | if (mem_cgroup_disabled()) | |
739f79fc | 1986 | return NULL; |
89c06bd5 | 1987 | again: |
1306a85a | 1988 | memcg = page->mem_cgroup; |
29833315 | 1989 | if (unlikely(!memcg)) |
739f79fc | 1990 | return NULL; |
d7365e78 | 1991 | |
bdcbb659 | 1992 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 1993 | return memcg; |
89c06bd5 | 1994 | |
6de22619 | 1995 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1996 | if (memcg != page->mem_cgroup) { |
6de22619 | 1997 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1998 | goto again; |
1999 | } | |
6de22619 JW |
2000 | |
2001 | /* | |
2002 | * When charge migration first begins, we can have locked and | |
2003 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 2004 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
2005 | */ |
2006 | memcg->move_lock_task = current; | |
2007 | memcg->move_lock_flags = flags; | |
d7365e78 | 2008 | |
739f79fc | 2009 | return memcg; |
89c06bd5 | 2010 | } |
81f8c3a4 | 2011 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2012 | |
d7365e78 | 2013 | /** |
739f79fc JW |
2014 | * __unlock_page_memcg - unlock and unpin a memcg |
2015 | * @memcg: the memcg | |
2016 | * | |
2017 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2018 | */ |
739f79fc | 2019 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2020 | { |
6de22619 JW |
2021 | if (memcg && memcg->move_lock_task == current) { |
2022 | unsigned long flags = memcg->move_lock_flags; | |
2023 | ||
2024 | memcg->move_lock_task = NULL; | |
2025 | memcg->move_lock_flags = 0; | |
2026 | ||
2027 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2028 | } | |
89c06bd5 | 2029 | |
d7365e78 | 2030 | rcu_read_unlock(); |
89c06bd5 | 2031 | } |
739f79fc JW |
2032 | |
2033 | /** | |
2034 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
2035 | * @page: the page | |
2036 | */ | |
2037 | void unlock_page_memcg(struct page *page) | |
2038 | { | |
2039 | __unlock_page_memcg(page->mem_cgroup); | |
2040 | } | |
81f8c3a4 | 2041 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2042 | |
cdec2e42 KH |
2043 | struct memcg_stock_pcp { |
2044 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2045 | unsigned int nr_pages; |
cdec2e42 | 2046 | struct work_struct work; |
26fe6168 | 2047 | unsigned long flags; |
a0db00fc | 2048 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2049 | }; |
2050 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2051 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2052 | |
a0956d54 SS |
2053 | /** |
2054 | * consume_stock: Try to consume stocked charge on this cpu. | |
2055 | * @memcg: memcg to consume from. | |
2056 | * @nr_pages: how many pages to charge. | |
2057 | * | |
2058 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2059 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2060 | * service an allocation will refill the stock. | |
2061 | * | |
2062 | * returns true if successful, false otherwise. | |
cdec2e42 | 2063 | */ |
a0956d54 | 2064 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2065 | { |
2066 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2067 | unsigned long flags; |
3e32cb2e | 2068 | bool ret = false; |
cdec2e42 | 2069 | |
a983b5eb | 2070 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2071 | return ret; |
a0956d54 | 2072 | |
db2ba40c JW |
2073 | local_irq_save(flags); |
2074 | ||
2075 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2076 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2077 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2078 | ret = true; |
2079 | } | |
db2ba40c JW |
2080 | |
2081 | local_irq_restore(flags); | |
2082 | ||
cdec2e42 KH |
2083 | return ret; |
2084 | } | |
2085 | ||
2086 | /* | |
3e32cb2e | 2087 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2088 | */ |
2089 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2090 | { | |
2091 | struct mem_cgroup *old = stock->cached; | |
2092 | ||
11c9ea4e | 2093 | if (stock->nr_pages) { |
3e32cb2e | 2094 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2095 | if (do_memsw_account()) |
3e32cb2e | 2096 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2097 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2098 | stock->nr_pages = 0; |
cdec2e42 KH |
2099 | } |
2100 | stock->cached = NULL; | |
cdec2e42 KH |
2101 | } |
2102 | ||
cdec2e42 KH |
2103 | static void drain_local_stock(struct work_struct *dummy) |
2104 | { | |
db2ba40c JW |
2105 | struct memcg_stock_pcp *stock; |
2106 | unsigned long flags; | |
2107 | ||
72f0184c MH |
2108 | /* |
2109 | * The only protection from memory hotplug vs. drain_stock races is | |
2110 | * that we always operate on local CPU stock here with IRQ disabled | |
2111 | */ | |
db2ba40c JW |
2112 | local_irq_save(flags); |
2113 | ||
2114 | stock = this_cpu_ptr(&memcg_stock); | |
cdec2e42 | 2115 | drain_stock(stock); |
26fe6168 | 2116 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2117 | |
2118 | local_irq_restore(flags); | |
cdec2e42 KH |
2119 | } |
2120 | ||
2121 | /* | |
3e32cb2e | 2122 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2123 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2124 | */ |
c0ff4b85 | 2125 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2126 | { |
db2ba40c JW |
2127 | struct memcg_stock_pcp *stock; |
2128 | unsigned long flags; | |
2129 | ||
2130 | local_irq_save(flags); | |
cdec2e42 | 2131 | |
db2ba40c | 2132 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2133 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2134 | drain_stock(stock); |
c0ff4b85 | 2135 | stock->cached = memcg; |
cdec2e42 | 2136 | } |
11c9ea4e | 2137 | stock->nr_pages += nr_pages; |
db2ba40c | 2138 | |
a983b5eb | 2139 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2140 | drain_stock(stock); |
2141 | ||
db2ba40c | 2142 | local_irq_restore(flags); |
cdec2e42 KH |
2143 | } |
2144 | ||
2145 | /* | |
c0ff4b85 | 2146 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2147 | * of the hierarchy under it. |
cdec2e42 | 2148 | */ |
6d3d6aa2 | 2149 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2150 | { |
26fe6168 | 2151 | int cpu, curcpu; |
d38144b7 | 2152 | |
6d3d6aa2 JW |
2153 | /* If someone's already draining, avoid adding running more workers. */ |
2154 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2155 | return; | |
72f0184c MH |
2156 | /* |
2157 | * Notify other cpus that system-wide "drain" is running | |
2158 | * We do not care about races with the cpu hotplug because cpu down | |
2159 | * as well as workers from this path always operate on the local | |
2160 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2161 | */ | |
5af12d0e | 2162 | curcpu = get_cpu(); |
cdec2e42 KH |
2163 | for_each_online_cpu(cpu) { |
2164 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2165 | struct mem_cgroup *memcg; |
26fe6168 | 2166 | |
c0ff4b85 | 2167 | memcg = stock->cached; |
72f0184c | 2168 | if (!memcg || !stock->nr_pages || !css_tryget(&memcg->css)) |
26fe6168 | 2169 | continue; |
72f0184c MH |
2170 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) { |
2171 | css_put(&memcg->css); | |
3e92041d | 2172 | continue; |
72f0184c | 2173 | } |
d1a05b69 MH |
2174 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2175 | if (cpu == curcpu) | |
2176 | drain_local_stock(&stock->work); | |
2177 | else | |
2178 | schedule_work_on(cpu, &stock->work); | |
2179 | } | |
72f0184c | 2180 | css_put(&memcg->css); |
cdec2e42 | 2181 | } |
5af12d0e | 2182 | put_cpu(); |
9f50fad6 | 2183 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2184 | } |
2185 | ||
308167fc | 2186 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2187 | { |
cdec2e42 | 2188 | struct memcg_stock_pcp *stock; |
42a30035 | 2189 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2190 | |
cdec2e42 KH |
2191 | stock = &per_cpu(memcg_stock, cpu); |
2192 | drain_stock(stock); | |
a983b5eb JW |
2193 | |
2194 | for_each_mem_cgroup(memcg) { | |
2195 | int i; | |
2196 | ||
2197 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2198 | int nid; | |
2199 | long x; | |
2200 | ||
871789d4 | 2201 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
815744d7 | 2202 | if (x) |
42a30035 JW |
2203 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2204 | atomic_long_add(x, &memcg->vmstats[i]); | |
a983b5eb JW |
2205 | |
2206 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2207 | continue; | |
2208 | ||
2209 | for_each_node(nid) { | |
2210 | struct mem_cgroup_per_node *pn; | |
2211 | ||
2212 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2213 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
815744d7 | 2214 | if (x) |
42a30035 JW |
2215 | do { |
2216 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2217 | } while ((pn = parent_nodeinfo(pn, nid))); | |
a983b5eb JW |
2218 | } |
2219 | } | |
2220 | ||
e27be240 | 2221 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2222 | long x; |
2223 | ||
871789d4 | 2224 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
815744d7 | 2225 | if (x) |
42a30035 JW |
2226 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2227 | atomic_long_add(x, &memcg->vmevents[i]); | |
a983b5eb JW |
2228 | } |
2229 | } | |
2230 | ||
308167fc | 2231 | return 0; |
cdec2e42 KH |
2232 | } |
2233 | ||
f7e1cb6e JW |
2234 | static void reclaim_high(struct mem_cgroup *memcg, |
2235 | unsigned int nr_pages, | |
2236 | gfp_t gfp_mask) | |
2237 | { | |
2238 | do { | |
2239 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
2240 | continue; | |
e27be240 | 2241 | memcg_memory_event(memcg, MEMCG_HIGH); |
f7e1cb6e JW |
2242 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
2243 | } while ((memcg = parent_mem_cgroup(memcg))); | |
2244 | } | |
2245 | ||
2246 | static void high_work_func(struct work_struct *work) | |
2247 | { | |
2248 | struct mem_cgroup *memcg; | |
2249 | ||
2250 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2251 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2252 | } |
2253 | ||
b23afb93 TH |
2254 | /* |
2255 | * Scheduled by try_charge() to be executed from the userland return path | |
2256 | * and reclaims memory over the high limit. | |
2257 | */ | |
2258 | void mem_cgroup_handle_over_high(void) | |
2259 | { | |
2260 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 2261 | struct mem_cgroup *memcg; |
b23afb93 TH |
2262 | |
2263 | if (likely(!nr_pages)) | |
2264 | return; | |
2265 | ||
f7e1cb6e JW |
2266 | memcg = get_mem_cgroup_from_mm(current->mm); |
2267 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
2268 | css_put(&memcg->css); |
2269 | current->memcg_nr_pages_over_high = 0; | |
2270 | } | |
2271 | ||
00501b53 JW |
2272 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2273 | unsigned int nr_pages) | |
8a9f3ccd | 2274 | { |
a983b5eb | 2275 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
9b130619 | 2276 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2277 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2278 | struct page_counter *counter; |
6539cc05 | 2279 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2280 | bool may_swap = true; |
2281 | bool drained = false; | |
29ef680a MH |
2282 | bool oomed = false; |
2283 | enum oom_status oom_status; | |
a636b327 | 2284 | |
ce00a967 | 2285 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2286 | return 0; |
6539cc05 | 2287 | retry: |
b6b6cc72 | 2288 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2289 | return 0; |
8a9f3ccd | 2290 | |
7941d214 | 2291 | if (!do_memsw_account() || |
6071ca52 JW |
2292 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2293 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2294 | goto done_restock; |
7941d214 | 2295 | if (do_memsw_account()) |
3e32cb2e JW |
2296 | page_counter_uncharge(&memcg->memsw, batch); |
2297 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2298 | } else { |
3e32cb2e | 2299 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2300 | may_swap = false; |
3fbe7244 | 2301 | } |
7a81b88c | 2302 | |
6539cc05 JW |
2303 | if (batch > nr_pages) { |
2304 | batch = nr_pages; | |
2305 | goto retry; | |
2306 | } | |
6d61ef40 | 2307 | |
06b078fc JW |
2308 | /* |
2309 | * Unlike in global OOM situations, memcg is not in a physical | |
2310 | * memory shortage. Allow dying and OOM-killed tasks to | |
2311 | * bypass the last charges so that they can exit quickly and | |
2312 | * free their memory. | |
2313 | */ | |
7775face | 2314 | if (unlikely(should_force_charge())) |
10d53c74 | 2315 | goto force; |
06b078fc | 2316 | |
89a28483 JW |
2317 | /* |
2318 | * Prevent unbounded recursion when reclaim operations need to | |
2319 | * allocate memory. This might exceed the limits temporarily, | |
2320 | * but we prefer facilitating memory reclaim and getting back | |
2321 | * under the limit over triggering OOM kills in these cases. | |
2322 | */ | |
2323 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2324 | goto force; | |
2325 | ||
06b078fc JW |
2326 | if (unlikely(task_in_memcg_oom(current))) |
2327 | goto nomem; | |
2328 | ||
d0164adc | 2329 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2330 | goto nomem; |
4b534334 | 2331 | |
e27be240 | 2332 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2333 | |
b70a2a21 JW |
2334 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2335 | gfp_mask, may_swap); | |
6539cc05 | 2336 | |
61e02c74 | 2337 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2338 | goto retry; |
28c34c29 | 2339 | |
b70a2a21 | 2340 | if (!drained) { |
6d3d6aa2 | 2341 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2342 | drained = true; |
2343 | goto retry; | |
2344 | } | |
2345 | ||
28c34c29 JW |
2346 | if (gfp_mask & __GFP_NORETRY) |
2347 | goto nomem; | |
6539cc05 JW |
2348 | /* |
2349 | * Even though the limit is exceeded at this point, reclaim | |
2350 | * may have been able to free some pages. Retry the charge | |
2351 | * before killing the task. | |
2352 | * | |
2353 | * Only for regular pages, though: huge pages are rather | |
2354 | * unlikely to succeed so close to the limit, and we fall back | |
2355 | * to regular pages anyway in case of failure. | |
2356 | */ | |
61e02c74 | 2357 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2358 | goto retry; |
2359 | /* | |
2360 | * At task move, charge accounts can be doubly counted. So, it's | |
2361 | * better to wait until the end of task_move if something is going on. | |
2362 | */ | |
2363 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2364 | goto retry; | |
2365 | ||
9b130619 JW |
2366 | if (nr_retries--) |
2367 | goto retry; | |
2368 | ||
29ef680a MH |
2369 | if (gfp_mask & __GFP_RETRY_MAYFAIL && oomed) |
2370 | goto nomem; | |
2371 | ||
06b078fc | 2372 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2373 | goto force; |
06b078fc | 2374 | |
6539cc05 | 2375 | if (fatal_signal_pending(current)) |
10d53c74 | 2376 | goto force; |
6539cc05 | 2377 | |
29ef680a MH |
2378 | /* |
2379 | * keep retrying as long as the memcg oom killer is able to make | |
2380 | * a forward progress or bypass the charge if the oom killer | |
2381 | * couldn't make any progress. | |
2382 | */ | |
2383 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2384 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2385 | switch (oom_status) { |
2386 | case OOM_SUCCESS: | |
2387 | nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
2388 | oomed = true; | |
2389 | goto retry; | |
2390 | case OOM_FAILED: | |
2391 | goto force; | |
2392 | default: | |
2393 | goto nomem; | |
2394 | } | |
7a81b88c | 2395 | nomem: |
6d1fdc48 | 2396 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2397 | return -ENOMEM; |
10d53c74 TH |
2398 | force: |
2399 | /* | |
2400 | * The allocation either can't fail or will lead to more memory | |
2401 | * being freed very soon. Allow memory usage go over the limit | |
2402 | * temporarily by force charging it. | |
2403 | */ | |
2404 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2405 | if (do_memsw_account()) |
10d53c74 TH |
2406 | page_counter_charge(&memcg->memsw, nr_pages); |
2407 | css_get_many(&memcg->css, nr_pages); | |
2408 | ||
2409 | return 0; | |
6539cc05 JW |
2410 | |
2411 | done_restock: | |
e8ea14cc | 2412 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2413 | if (batch > nr_pages) |
2414 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2415 | |
241994ed | 2416 | /* |
b23afb93 TH |
2417 | * If the hierarchy is above the normal consumption range, schedule |
2418 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2419 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2420 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2421 | * not recorded as it most likely matches current's and won't | |
2422 | * change in the meantime. As high limit is checked again before | |
2423 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2424 | */ |
2425 | do { | |
b23afb93 | 2426 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2427 | /* Don't bother a random interrupted task */ |
2428 | if (in_interrupt()) { | |
2429 | schedule_work(&memcg->high_work); | |
2430 | break; | |
2431 | } | |
9516a18a | 2432 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2433 | set_notify_resume(current); |
2434 | break; | |
2435 | } | |
241994ed | 2436 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2437 | |
2438 | return 0; | |
7a81b88c | 2439 | } |
8a9f3ccd | 2440 | |
00501b53 | 2441 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2442 | { |
ce00a967 JW |
2443 | if (mem_cgroup_is_root(memcg)) |
2444 | return; | |
2445 | ||
3e32cb2e | 2446 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2447 | if (do_memsw_account()) |
3e32cb2e | 2448 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2449 | |
e8ea14cc | 2450 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2451 | } |
2452 | ||
0a31bc97 JW |
2453 | static void lock_page_lru(struct page *page, int *isolated) |
2454 | { | |
f4b7e272 | 2455 | pg_data_t *pgdat = page_pgdat(page); |
0a31bc97 | 2456 | |
f4b7e272 | 2457 | spin_lock_irq(&pgdat->lru_lock); |
0a31bc97 JW |
2458 | if (PageLRU(page)) { |
2459 | struct lruvec *lruvec; | |
2460 | ||
f4b7e272 | 2461 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
0a31bc97 JW |
2462 | ClearPageLRU(page); |
2463 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2464 | *isolated = 1; | |
2465 | } else | |
2466 | *isolated = 0; | |
2467 | } | |
2468 | ||
2469 | static void unlock_page_lru(struct page *page, int isolated) | |
2470 | { | |
f4b7e272 | 2471 | pg_data_t *pgdat = page_pgdat(page); |
0a31bc97 JW |
2472 | |
2473 | if (isolated) { | |
2474 | struct lruvec *lruvec; | |
2475 | ||
f4b7e272 | 2476 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
0a31bc97 JW |
2477 | VM_BUG_ON_PAGE(PageLRU(page), page); |
2478 | SetPageLRU(page); | |
2479 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2480 | } | |
f4b7e272 | 2481 | spin_unlock_irq(&pgdat->lru_lock); |
0a31bc97 JW |
2482 | } |
2483 | ||
00501b53 | 2484 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2485 | bool lrucare) |
7a81b88c | 2486 | { |
0a31bc97 | 2487 | int isolated; |
9ce70c02 | 2488 | |
1306a85a | 2489 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2490 | |
2491 | /* | |
2492 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2493 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2494 | */ | |
0a31bc97 JW |
2495 | if (lrucare) |
2496 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2497 | |
0a31bc97 JW |
2498 | /* |
2499 | * Nobody should be changing or seriously looking at | |
1306a85a | 2500 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2501 | * |
2502 | * - the page is uncharged | |
2503 | * | |
2504 | * - the page is off-LRU | |
2505 | * | |
2506 | * - an anonymous fault has exclusive page access, except for | |
2507 | * a locked page table | |
2508 | * | |
2509 | * - a page cache insertion, a swapin fault, or a migration | |
2510 | * have the page locked | |
2511 | */ | |
1306a85a | 2512 | page->mem_cgroup = memcg; |
9ce70c02 | 2513 | |
0a31bc97 JW |
2514 | if (lrucare) |
2515 | unlock_page_lru(page, isolated); | |
7a81b88c | 2516 | } |
66e1707b | 2517 | |
84c07d11 | 2518 | #ifdef CONFIG_MEMCG_KMEM |
f3bb3043 | 2519 | static int memcg_alloc_cache_id(void) |
55007d84 | 2520 | { |
f3bb3043 VD |
2521 | int id, size; |
2522 | int err; | |
2523 | ||
dbcf73e2 | 2524 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2525 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2526 | if (id < 0) | |
2527 | return id; | |
55007d84 | 2528 | |
dbcf73e2 | 2529 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2530 | return id; |
2531 | ||
2532 | /* | |
2533 | * There's no space for the new id in memcg_caches arrays, | |
2534 | * so we have to grow them. | |
2535 | */ | |
05257a1a | 2536 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2537 | |
2538 | size = 2 * (id + 1); | |
55007d84 GC |
2539 | if (size < MEMCG_CACHES_MIN_SIZE) |
2540 | size = MEMCG_CACHES_MIN_SIZE; | |
2541 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2542 | size = MEMCG_CACHES_MAX_SIZE; | |
2543 | ||
f3bb3043 | 2544 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2545 | if (!err) |
2546 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2547 | if (!err) |
2548 | memcg_nr_cache_ids = size; | |
2549 | ||
2550 | up_write(&memcg_cache_ids_sem); | |
2551 | ||
f3bb3043 | 2552 | if (err) { |
dbcf73e2 | 2553 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2554 | return err; |
2555 | } | |
2556 | return id; | |
2557 | } | |
2558 | ||
2559 | static void memcg_free_cache_id(int id) | |
2560 | { | |
dbcf73e2 | 2561 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2562 | } |
2563 | ||
d5b3cf71 | 2564 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2565 | struct mem_cgroup *memcg; |
2566 | struct kmem_cache *cachep; | |
2567 | struct work_struct work; | |
2568 | }; | |
2569 | ||
d5b3cf71 | 2570 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2571 | { |
d5b3cf71 VD |
2572 | struct memcg_kmem_cache_create_work *cw = |
2573 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2574 | struct mem_cgroup *memcg = cw->memcg; |
2575 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2576 | |
d5b3cf71 | 2577 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2578 | |
5722d094 | 2579 | css_put(&memcg->css); |
d7f25f8a GC |
2580 | kfree(cw); |
2581 | } | |
2582 | ||
2583 | /* | |
2584 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2585 | */ |
85cfb245 | 2586 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
d5b3cf71 | 2587 | struct kmem_cache *cachep) |
d7f25f8a | 2588 | { |
d5b3cf71 | 2589 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2590 | |
c892fd82 | 2591 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT | __GFP_NOWARN); |
8135be5a | 2592 | if (!cw) |
d7f25f8a | 2593 | return; |
8135be5a VD |
2594 | |
2595 | css_get(&memcg->css); | |
d7f25f8a GC |
2596 | |
2597 | cw->memcg = memcg; | |
2598 | cw->cachep = cachep; | |
d5b3cf71 | 2599 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2600 | |
17cc4dfe | 2601 | queue_work(memcg_kmem_cache_wq, &cw->work); |
d7f25f8a GC |
2602 | } |
2603 | ||
45264778 VD |
2604 | static inline bool memcg_kmem_bypass(void) |
2605 | { | |
2606 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
2607 | return true; | |
2608 | return false; | |
2609 | } | |
2610 | ||
2611 | /** | |
2612 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation | |
2613 | * @cachep: the original global kmem cache | |
2614 | * | |
d7f25f8a GC |
2615 | * Return the kmem_cache we're supposed to use for a slab allocation. |
2616 | * We try to use the current memcg's version of the cache. | |
2617 | * | |
45264778 VD |
2618 | * If the cache does not exist yet, if we are the first user of it, we |
2619 | * create it asynchronously in a workqueue and let the current allocation | |
2620 | * go through with the original cache. | |
d7f25f8a | 2621 | * |
45264778 VD |
2622 | * This function takes a reference to the cache it returns to assure it |
2623 | * won't get destroyed while we are working with it. Once the caller is | |
2624 | * done with it, memcg_kmem_put_cache() must be called to release the | |
2625 | * reference. | |
d7f25f8a | 2626 | */ |
45264778 | 2627 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2628 | { |
2629 | struct mem_cgroup *memcg; | |
959c8963 | 2630 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2631 | int kmemcg_id; |
d7f25f8a | 2632 | |
f7ce3190 | 2633 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2634 | |
45264778 | 2635 | if (memcg_kmem_bypass()) |
230e9fc2 VD |
2636 | return cachep; |
2637 | ||
d46eb14b | 2638 | memcg = get_mem_cgroup_from_current(); |
4db0c3c2 | 2639 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2640 | if (kmemcg_id < 0) |
ca0dde97 | 2641 | goto out; |
d7f25f8a | 2642 | |
2a4db7eb | 2643 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2644 | if (likely(memcg_cachep)) |
2645 | return memcg_cachep; | |
ca0dde97 LZ |
2646 | |
2647 | /* | |
2648 | * If we are in a safe context (can wait, and not in interrupt | |
2649 | * context), we could be be predictable and return right away. | |
2650 | * This would guarantee that the allocation being performed | |
2651 | * already belongs in the new cache. | |
2652 | * | |
2653 | * However, there are some clashes that can arrive from locking. | |
2654 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2655 | * memcg_create_kmem_cache, this means no further allocation |
2656 | * could happen with the slab_mutex held. So it's better to | |
2657 | * defer everything. | |
ca0dde97 | 2658 | */ |
d5b3cf71 | 2659 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2660 | out: |
8135be5a | 2661 | css_put(&memcg->css); |
ca0dde97 | 2662 | return cachep; |
d7f25f8a | 2663 | } |
d7f25f8a | 2664 | |
45264778 VD |
2665 | /** |
2666 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache | |
2667 | * @cachep: the cache returned by memcg_kmem_get_cache | |
2668 | */ | |
2669 | void memcg_kmem_put_cache(struct kmem_cache *cachep) | |
8135be5a VD |
2670 | { |
2671 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2672 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2673 | } |
2674 | ||
45264778 | 2675 | /** |
60cd4bcd | 2676 | * __memcg_kmem_charge_memcg: charge a kmem page |
45264778 VD |
2677 | * @page: page to charge |
2678 | * @gfp: reclaim mode | |
2679 | * @order: allocation order | |
2680 | * @memcg: memory cgroup to charge | |
2681 | * | |
2682 | * Returns 0 on success, an error code on failure. | |
2683 | */ | |
60cd4bcd | 2684 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
45264778 | 2685 | struct mem_cgroup *memcg) |
7ae1e1d0 | 2686 | { |
f3ccb2c4 VD |
2687 | unsigned int nr_pages = 1 << order; |
2688 | struct page_counter *counter; | |
7ae1e1d0 GC |
2689 | int ret; |
2690 | ||
f3ccb2c4 | 2691 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2692 | if (ret) |
f3ccb2c4 | 2693 | return ret; |
52c29b04 JW |
2694 | |
2695 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2696 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2697 | cancel_charge(memcg, nr_pages); | |
2698 | return -ENOMEM; | |
7ae1e1d0 GC |
2699 | } |
2700 | ||
f3ccb2c4 | 2701 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2702 | |
f3ccb2c4 | 2703 | return 0; |
7ae1e1d0 GC |
2704 | } |
2705 | ||
45264778 | 2706 | /** |
60cd4bcd | 2707 | * __memcg_kmem_charge: charge a kmem page to the current memory cgroup |
45264778 VD |
2708 | * @page: page to charge |
2709 | * @gfp: reclaim mode | |
2710 | * @order: allocation order | |
2711 | * | |
2712 | * Returns 0 on success, an error code on failure. | |
2713 | */ | |
60cd4bcd | 2714 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2715 | { |
f3ccb2c4 | 2716 | struct mem_cgroup *memcg; |
fcff7d7e | 2717 | int ret = 0; |
7ae1e1d0 | 2718 | |
60cd4bcd | 2719 | if (memcg_kmem_bypass()) |
45264778 VD |
2720 | return 0; |
2721 | ||
d46eb14b | 2722 | memcg = get_mem_cgroup_from_current(); |
c4159a75 | 2723 | if (!mem_cgroup_is_root(memcg)) { |
60cd4bcd | 2724 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); |
c4159a75 VD |
2725 | if (!ret) |
2726 | __SetPageKmemcg(page); | |
2727 | } | |
7ae1e1d0 | 2728 | css_put(&memcg->css); |
d05e83a6 | 2729 | return ret; |
7ae1e1d0 | 2730 | } |
45264778 | 2731 | /** |
60cd4bcd | 2732 | * __memcg_kmem_uncharge: uncharge a kmem page |
45264778 VD |
2733 | * @page: page to uncharge |
2734 | * @order: allocation order | |
2735 | */ | |
60cd4bcd | 2736 | void __memcg_kmem_uncharge(struct page *page, int order) |
7ae1e1d0 | 2737 | { |
1306a85a | 2738 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2739 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2740 | |
7ae1e1d0 GC |
2741 | if (!memcg) |
2742 | return; | |
2743 | ||
309381fe | 2744 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2745 | |
52c29b04 JW |
2746 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2747 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2748 | ||
f3ccb2c4 | 2749 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2750 | if (do_memsw_account()) |
f3ccb2c4 | 2751 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2752 | |
1306a85a | 2753 | page->mem_cgroup = NULL; |
c4159a75 VD |
2754 | |
2755 | /* slab pages do not have PageKmemcg flag set */ | |
2756 | if (PageKmemcg(page)) | |
2757 | __ClearPageKmemcg(page); | |
2758 | ||
f3ccb2c4 | 2759 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2760 | } |
84c07d11 | 2761 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 2762 | |
ca3e0214 KH |
2763 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2764 | ||
ca3e0214 KH |
2765 | /* |
2766 | * Because tail pages are not marked as "used", set it. We're under | |
f4b7e272 | 2767 | * pgdat->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2768 | */ |
e94c8a9c | 2769 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2770 | { |
e94c8a9c | 2771 | int i; |
ca3e0214 | 2772 | |
3d37c4a9 KH |
2773 | if (mem_cgroup_disabled()) |
2774 | return; | |
b070e65c | 2775 | |
29833315 | 2776 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2777 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2778 | |
c9019e9b | 2779 | __mod_memcg_state(head->mem_cgroup, MEMCG_RSS_HUGE, -HPAGE_PMD_NR); |
ca3e0214 | 2780 | } |
12d27107 | 2781 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2782 | |
c255a458 | 2783 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
2784 | /** |
2785 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2786 | * @entry: swap entry to be moved | |
2787 | * @from: mem_cgroup which the entry is moved from | |
2788 | * @to: mem_cgroup which the entry is moved to | |
2789 | * | |
2790 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2791 | * as the mem_cgroup's id of @from. | |
2792 | * | |
2793 | * Returns 0 on success, -EINVAL on failure. | |
2794 | * | |
3e32cb2e | 2795 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2796 | * both res and memsw, and called css_get(). |
2797 | */ | |
2798 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2799 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2800 | { |
2801 | unsigned short old_id, new_id; | |
2802 | ||
34c00c31 LZ |
2803 | old_id = mem_cgroup_id(from); |
2804 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2805 | |
2806 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
2807 | mod_memcg_state(from, MEMCG_SWAP, -1); |
2808 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
2809 | return 0; |
2810 | } | |
2811 | return -EINVAL; | |
2812 | } | |
2813 | #else | |
2814 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2815 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2816 | { |
2817 | return -EINVAL; | |
2818 | } | |
8c7c6e34 | 2819 | #endif |
d13d1443 | 2820 | |
bbec2e15 | 2821 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 2822 | |
bbec2e15 RG |
2823 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
2824 | unsigned long max, bool memsw) | |
628f4235 | 2825 | { |
3e32cb2e | 2826 | bool enlarge = false; |
bb4a7ea2 | 2827 | bool drained = false; |
3e32cb2e | 2828 | int ret; |
c054a78c YZ |
2829 | bool limits_invariant; |
2830 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 2831 | |
3e32cb2e | 2832 | do { |
628f4235 KH |
2833 | if (signal_pending(current)) { |
2834 | ret = -EINTR; | |
2835 | break; | |
2836 | } | |
3e32cb2e | 2837 | |
bbec2e15 | 2838 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
2839 | /* |
2840 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 2841 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 2842 | */ |
bbec2e15 RG |
2843 | limits_invariant = memsw ? max >= memcg->memory.max : |
2844 | max <= memcg->memsw.max; | |
c054a78c | 2845 | if (!limits_invariant) { |
bbec2e15 | 2846 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 2847 | ret = -EINVAL; |
8c7c6e34 KH |
2848 | break; |
2849 | } | |
bbec2e15 | 2850 | if (max > counter->max) |
3e32cb2e | 2851 | enlarge = true; |
bbec2e15 RG |
2852 | ret = page_counter_set_max(counter, max); |
2853 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
2854 | |
2855 | if (!ret) | |
2856 | break; | |
2857 | ||
bb4a7ea2 SB |
2858 | if (!drained) { |
2859 | drain_all_stock(memcg); | |
2860 | drained = true; | |
2861 | continue; | |
2862 | } | |
2863 | ||
1ab5c056 AR |
2864 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
2865 | GFP_KERNEL, !memsw)) { | |
2866 | ret = -EBUSY; | |
2867 | break; | |
2868 | } | |
2869 | } while (true); | |
3e32cb2e | 2870 | |
3c11ecf4 KH |
2871 | if (!ret && enlarge) |
2872 | memcg_oom_recover(memcg); | |
3e32cb2e | 2873 | |
628f4235 KH |
2874 | return ret; |
2875 | } | |
2876 | ||
ef8f2327 | 2877 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
2878 | gfp_t gfp_mask, |
2879 | unsigned long *total_scanned) | |
2880 | { | |
2881 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 2882 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
2883 | unsigned long reclaimed; |
2884 | int loop = 0; | |
ef8f2327 | 2885 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 2886 | unsigned long excess; |
0608f43d AM |
2887 | unsigned long nr_scanned; |
2888 | ||
2889 | if (order > 0) | |
2890 | return 0; | |
2891 | ||
ef8f2327 | 2892 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
2893 | |
2894 | /* | |
2895 | * Do not even bother to check the largest node if the root | |
2896 | * is empty. Do it lockless to prevent lock bouncing. Races | |
2897 | * are acceptable as soft limit is best effort anyway. | |
2898 | */ | |
bfc7228b | 2899 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
2900 | return 0; |
2901 | ||
0608f43d AM |
2902 | /* |
2903 | * This loop can run a while, specially if mem_cgroup's continuously | |
2904 | * keep exceeding their soft limit and putting the system under | |
2905 | * pressure | |
2906 | */ | |
2907 | do { | |
2908 | if (next_mz) | |
2909 | mz = next_mz; | |
2910 | else | |
2911 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2912 | if (!mz) | |
2913 | break; | |
2914 | ||
2915 | nr_scanned = 0; | |
ef8f2327 | 2916 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
2917 | gfp_mask, &nr_scanned); |
2918 | nr_reclaimed += reclaimed; | |
2919 | *total_scanned += nr_scanned; | |
0a31bc97 | 2920 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2921 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2922 | |
2923 | /* | |
2924 | * If we failed to reclaim anything from this memory cgroup | |
2925 | * it is time to move on to the next cgroup | |
2926 | */ | |
2927 | next_mz = NULL; | |
bc2f2e7f VD |
2928 | if (!reclaimed) |
2929 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2930 | ||
3e32cb2e | 2931 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2932 | /* |
2933 | * One school of thought says that we should not add | |
2934 | * back the node to the tree if reclaim returns 0. | |
2935 | * But our reclaim could return 0, simply because due | |
2936 | * to priority we are exposing a smaller subset of | |
2937 | * memory to reclaim from. Consider this as a longer | |
2938 | * term TODO. | |
2939 | */ | |
2940 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2941 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2942 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2943 | css_put(&mz->memcg->css); |
2944 | loop++; | |
2945 | /* | |
2946 | * Could not reclaim anything and there are no more | |
2947 | * mem cgroups to try or we seem to be looping without | |
2948 | * reclaiming anything. | |
2949 | */ | |
2950 | if (!nr_reclaimed && | |
2951 | (next_mz == NULL || | |
2952 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2953 | break; | |
2954 | } while (!nr_reclaimed); | |
2955 | if (next_mz) | |
2956 | css_put(&next_mz->memcg->css); | |
2957 | return nr_reclaimed; | |
2958 | } | |
2959 | ||
ea280e7b TH |
2960 | /* |
2961 | * Test whether @memcg has children, dead or alive. Note that this | |
2962 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2963 | * returns %true if there are child csses according to the cgroup | |
2964 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2965 | */ | |
b5f99b53 GC |
2966 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2967 | { | |
ea280e7b TH |
2968 | bool ret; |
2969 | ||
ea280e7b TH |
2970 | rcu_read_lock(); |
2971 | ret = css_next_child(NULL, &memcg->css); | |
2972 | rcu_read_unlock(); | |
2973 | return ret; | |
b5f99b53 GC |
2974 | } |
2975 | ||
c26251f9 | 2976 | /* |
51038171 | 2977 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
2978 | * |
2979 | * Caller is responsible for holding css reference for memcg. | |
2980 | */ | |
2981 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2982 | { | |
2983 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2984 | |
c1e862c1 KH |
2985 | /* we call try-to-free pages for make this cgroup empty */ |
2986 | lru_add_drain_all(); | |
d12c60f6 JS |
2987 | |
2988 | drain_all_stock(memcg); | |
2989 | ||
f817ed48 | 2990 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2991 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2992 | int progress; |
c1e862c1 | 2993 | |
c26251f9 MH |
2994 | if (signal_pending(current)) |
2995 | return -EINTR; | |
2996 | ||
b70a2a21 JW |
2997 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2998 | GFP_KERNEL, true); | |
c1e862c1 | 2999 | if (!progress) { |
f817ed48 | 3000 | nr_retries--; |
c1e862c1 | 3001 | /* maybe some writeback is necessary */ |
8aa7e847 | 3002 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3003 | } |
f817ed48 KH |
3004 | |
3005 | } | |
ab5196c2 MH |
3006 | |
3007 | return 0; | |
cc847582 KH |
3008 | } |
3009 | ||
6770c64e TH |
3010 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3011 | char *buf, size_t nbytes, | |
3012 | loff_t off) | |
c1e862c1 | 3013 | { |
6770c64e | 3014 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3015 | |
d8423011 MH |
3016 | if (mem_cgroup_is_root(memcg)) |
3017 | return -EINVAL; | |
6770c64e | 3018 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3019 | } |
3020 | ||
182446d0 TH |
3021 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3022 | struct cftype *cft) | |
18f59ea7 | 3023 | { |
182446d0 | 3024 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3025 | } |
3026 | ||
182446d0 TH |
3027 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3028 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3029 | { |
3030 | int retval = 0; | |
182446d0 | 3031 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3032 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3033 | |
567fb435 | 3034 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 3035 | return 0; |
567fb435 | 3036 | |
18f59ea7 | 3037 | /* |
af901ca1 | 3038 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3039 | * in the child subtrees. If it is unset, then the change can |
3040 | * occur, provided the current cgroup has no children. | |
3041 | * | |
3042 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3043 | * set if there are no children. | |
3044 | */ | |
c0ff4b85 | 3045 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3046 | (val == 1 || val == 0)) { |
ea280e7b | 3047 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3048 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3049 | else |
3050 | retval = -EBUSY; | |
3051 | } else | |
3052 | retval = -EINVAL; | |
567fb435 | 3053 | |
18f59ea7 BS |
3054 | return retval; |
3055 | } | |
3056 | ||
6f646156 | 3057 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3058 | { |
42a30035 | 3059 | unsigned long val; |
ce00a967 | 3060 | |
3e32cb2e | 3061 | if (mem_cgroup_is_root(memcg)) { |
42a30035 JW |
3062 | val = memcg_page_state(memcg, MEMCG_CACHE) + |
3063 | memcg_page_state(memcg, MEMCG_RSS); | |
3064 | if (swap) | |
3065 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3066 | } else { |
ce00a967 | 3067 | if (!swap) |
3e32cb2e | 3068 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3069 | else |
3e32cb2e | 3070 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3071 | } |
c12176d3 | 3072 | return val; |
ce00a967 JW |
3073 | } |
3074 | ||
3e32cb2e JW |
3075 | enum { |
3076 | RES_USAGE, | |
3077 | RES_LIMIT, | |
3078 | RES_MAX_USAGE, | |
3079 | RES_FAILCNT, | |
3080 | RES_SOFT_LIMIT, | |
3081 | }; | |
ce00a967 | 3082 | |
791badbd | 3083 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3084 | struct cftype *cft) |
8cdea7c0 | 3085 | { |
182446d0 | 3086 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3087 | struct page_counter *counter; |
af36f906 | 3088 | |
3e32cb2e | 3089 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3090 | case _MEM: |
3e32cb2e JW |
3091 | counter = &memcg->memory; |
3092 | break; | |
8c7c6e34 | 3093 | case _MEMSWAP: |
3e32cb2e JW |
3094 | counter = &memcg->memsw; |
3095 | break; | |
510fc4e1 | 3096 | case _KMEM: |
3e32cb2e | 3097 | counter = &memcg->kmem; |
510fc4e1 | 3098 | break; |
d55f90bf | 3099 | case _TCP: |
0db15298 | 3100 | counter = &memcg->tcpmem; |
d55f90bf | 3101 | break; |
8c7c6e34 KH |
3102 | default: |
3103 | BUG(); | |
8c7c6e34 | 3104 | } |
3e32cb2e JW |
3105 | |
3106 | switch (MEMFILE_ATTR(cft->private)) { | |
3107 | case RES_USAGE: | |
3108 | if (counter == &memcg->memory) | |
c12176d3 | 3109 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3110 | if (counter == &memcg->memsw) |
c12176d3 | 3111 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3112 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3113 | case RES_LIMIT: | |
bbec2e15 | 3114 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3115 | case RES_MAX_USAGE: |
3116 | return (u64)counter->watermark * PAGE_SIZE; | |
3117 | case RES_FAILCNT: | |
3118 | return counter->failcnt; | |
3119 | case RES_SOFT_LIMIT: | |
3120 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3121 | default: | |
3122 | BUG(); | |
3123 | } | |
8cdea7c0 | 3124 | } |
510fc4e1 | 3125 | |
84c07d11 | 3126 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3127 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3128 | { |
d6441637 VD |
3129 | int memcg_id; |
3130 | ||
b313aeee VD |
3131 | if (cgroup_memory_nokmem) |
3132 | return 0; | |
3133 | ||
2a4db7eb | 3134 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3135 | BUG_ON(memcg->kmem_state); |
d6441637 | 3136 | |
f3bb3043 | 3137 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3138 | if (memcg_id < 0) |
3139 | return memcg_id; | |
d6441637 | 3140 | |
ef12947c | 3141 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 3142 | /* |
567e9ab2 | 3143 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 3144 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
3145 | * guarantee no one starts accounting before all call sites are |
3146 | * patched. | |
3147 | */ | |
900a38f0 | 3148 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3149 | memcg->kmem_state = KMEM_ONLINE; |
bc2791f8 | 3150 | INIT_LIST_HEAD(&memcg->kmem_caches); |
0b8f73e1 JW |
3151 | |
3152 | return 0; | |
d6441637 VD |
3153 | } |
3154 | ||
8e0a8912 JW |
3155 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3156 | { | |
3157 | struct cgroup_subsys_state *css; | |
3158 | struct mem_cgroup *parent, *child; | |
3159 | int kmemcg_id; | |
3160 | ||
3161 | if (memcg->kmem_state != KMEM_ONLINE) | |
3162 | return; | |
3163 | /* | |
3164 | * Clear the online state before clearing memcg_caches array | |
3165 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
3166 | * guarantees that no cache will be created for this cgroup | |
3167 | * after we are done (see memcg_create_kmem_cache()). | |
3168 | */ | |
3169 | memcg->kmem_state = KMEM_ALLOCATED; | |
3170 | ||
3171 | memcg_deactivate_kmem_caches(memcg); | |
3172 | ||
3173 | kmemcg_id = memcg->kmemcg_id; | |
3174 | BUG_ON(kmemcg_id < 0); | |
3175 | ||
3176 | parent = parent_mem_cgroup(memcg); | |
3177 | if (!parent) | |
3178 | parent = root_mem_cgroup; | |
3179 | ||
3180 | /* | |
3181 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3182 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3183 | * to ones of the parent. After we have finished, all list_lrus | |
3184 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3185 | * ordering is imposed by list_lru_node->lock taken by | |
3186 | * memcg_drain_all_list_lrus(). | |
3187 | */ | |
3a06bb78 | 3188 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3189 | css_for_each_descendant_pre(css, &memcg->css) { |
3190 | child = mem_cgroup_from_css(css); | |
3191 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3192 | child->kmemcg_id = parent->kmemcg_id; | |
3193 | if (!memcg->use_hierarchy) | |
3194 | break; | |
3195 | } | |
3a06bb78 TH |
3196 | rcu_read_unlock(); |
3197 | ||
9bec5c35 | 3198 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3199 | |
3200 | memcg_free_cache_id(kmemcg_id); | |
3201 | } | |
3202 | ||
3203 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3204 | { | |
0b8f73e1 JW |
3205 | /* css_alloc() failed, offlining didn't happen */ |
3206 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3207 | memcg_offline_kmem(memcg); | |
3208 | ||
8e0a8912 JW |
3209 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
3210 | memcg_destroy_kmem_caches(memcg); | |
3211 | static_branch_dec(&memcg_kmem_enabled_key); | |
3212 | WARN_ON(page_counter_read(&memcg->kmem)); | |
3213 | } | |
8e0a8912 | 3214 | } |
d6441637 | 3215 | #else |
0b8f73e1 | 3216 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3217 | { |
3218 | return 0; | |
3219 | } | |
3220 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3221 | { | |
3222 | } | |
3223 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3224 | { | |
3225 | } | |
84c07d11 | 3226 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3227 | |
bbec2e15 RG |
3228 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3229 | unsigned long max) | |
d6441637 | 3230 | { |
b313aeee | 3231 | int ret; |
127424c8 | 3232 | |
bbec2e15 RG |
3233 | mutex_lock(&memcg_max_mutex); |
3234 | ret = page_counter_set_max(&memcg->kmem, max); | |
3235 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3236 | return ret; |
d6441637 | 3237 | } |
510fc4e1 | 3238 | |
bbec2e15 | 3239 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3240 | { |
3241 | int ret; | |
3242 | ||
bbec2e15 | 3243 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3244 | |
bbec2e15 | 3245 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3246 | if (ret) |
3247 | goto out; | |
3248 | ||
0db15298 | 3249 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3250 | /* |
3251 | * The active flag needs to be written after the static_key | |
3252 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3253 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3254 | * for details, and note that we don't mark any socket as | |
3255 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3256 | * |
3257 | * We need to do this, because static_keys will span multiple | |
3258 | * sites, but we can't control their order. If we mark a socket | |
3259 | * as accounted, but the accounting functions are not patched in | |
3260 | * yet, we'll lose accounting. | |
3261 | * | |
2d758073 | 3262 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3263 | * because when this value change, the code to process it is not |
3264 | * patched in yet. | |
3265 | */ | |
3266 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3267 | memcg->tcpmem_active = true; |
d55f90bf VD |
3268 | } |
3269 | out: | |
bbec2e15 | 3270 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3271 | return ret; |
3272 | } | |
d55f90bf | 3273 | |
628f4235 KH |
3274 | /* |
3275 | * The user of this function is... | |
3276 | * RES_LIMIT. | |
3277 | */ | |
451af504 TH |
3278 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3279 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3280 | { |
451af504 | 3281 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3282 | unsigned long nr_pages; |
628f4235 KH |
3283 | int ret; |
3284 | ||
451af504 | 3285 | buf = strstrip(buf); |
650c5e56 | 3286 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3287 | if (ret) |
3288 | return ret; | |
af36f906 | 3289 | |
3e32cb2e | 3290 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3291 | case RES_LIMIT: |
4b3bde4c BS |
3292 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3293 | ret = -EINVAL; | |
3294 | break; | |
3295 | } | |
3e32cb2e JW |
3296 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3297 | case _MEM: | |
bbec2e15 | 3298 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3299 | break; |
3e32cb2e | 3300 | case _MEMSWAP: |
bbec2e15 | 3301 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3302 | break; |
3e32cb2e | 3303 | case _KMEM: |
bbec2e15 | 3304 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3305 | break; |
d55f90bf | 3306 | case _TCP: |
bbec2e15 | 3307 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3308 | break; |
3e32cb2e | 3309 | } |
296c81d8 | 3310 | break; |
3e32cb2e JW |
3311 | case RES_SOFT_LIMIT: |
3312 | memcg->soft_limit = nr_pages; | |
3313 | ret = 0; | |
628f4235 KH |
3314 | break; |
3315 | } | |
451af504 | 3316 | return ret ?: nbytes; |
8cdea7c0 BS |
3317 | } |
3318 | ||
6770c64e TH |
3319 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3320 | size_t nbytes, loff_t off) | |
c84872e1 | 3321 | { |
6770c64e | 3322 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3323 | struct page_counter *counter; |
c84872e1 | 3324 | |
3e32cb2e JW |
3325 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3326 | case _MEM: | |
3327 | counter = &memcg->memory; | |
3328 | break; | |
3329 | case _MEMSWAP: | |
3330 | counter = &memcg->memsw; | |
3331 | break; | |
3332 | case _KMEM: | |
3333 | counter = &memcg->kmem; | |
3334 | break; | |
d55f90bf | 3335 | case _TCP: |
0db15298 | 3336 | counter = &memcg->tcpmem; |
d55f90bf | 3337 | break; |
3e32cb2e JW |
3338 | default: |
3339 | BUG(); | |
3340 | } | |
af36f906 | 3341 | |
3e32cb2e | 3342 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3343 | case RES_MAX_USAGE: |
3e32cb2e | 3344 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3345 | break; |
3346 | case RES_FAILCNT: | |
3e32cb2e | 3347 | counter->failcnt = 0; |
29f2a4da | 3348 | break; |
3e32cb2e JW |
3349 | default: |
3350 | BUG(); | |
29f2a4da | 3351 | } |
f64c3f54 | 3352 | |
6770c64e | 3353 | return nbytes; |
c84872e1 PE |
3354 | } |
3355 | ||
182446d0 | 3356 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3357 | struct cftype *cft) |
3358 | { | |
182446d0 | 3359 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3360 | } |
3361 | ||
02491447 | 3362 | #ifdef CONFIG_MMU |
182446d0 | 3363 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3364 | struct cftype *cft, u64 val) |
3365 | { | |
182446d0 | 3366 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3367 | |
1dfab5ab | 3368 | if (val & ~MOVE_MASK) |
7dc74be0 | 3369 | return -EINVAL; |
ee5e8472 | 3370 | |
7dc74be0 | 3371 | /* |
ee5e8472 GC |
3372 | * No kind of locking is needed in here, because ->can_attach() will |
3373 | * check this value once in the beginning of the process, and then carry | |
3374 | * on with stale data. This means that changes to this value will only | |
3375 | * affect task migrations starting after the change. | |
7dc74be0 | 3376 | */ |
c0ff4b85 | 3377 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3378 | return 0; |
3379 | } | |
02491447 | 3380 | #else |
182446d0 | 3381 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3382 | struct cftype *cft, u64 val) |
3383 | { | |
3384 | return -ENOSYS; | |
3385 | } | |
3386 | #endif | |
7dc74be0 | 3387 | |
406eb0c9 | 3388 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3389 | |
3390 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3391 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3392 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3393 | ||
3394 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
3395 | int nid, unsigned int lru_mask) | |
3396 | { | |
3397 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); | |
3398 | unsigned long nr = 0; | |
3399 | enum lru_list lru; | |
3400 | ||
3401 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3402 | ||
3403 | for_each_lru(lru) { | |
3404 | if (!(BIT(lru) & lru_mask)) | |
3405 | continue; | |
205b20cc | 3406 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); |
113b7dfd JW |
3407 | } |
3408 | return nr; | |
3409 | } | |
3410 | ||
3411 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
3412 | unsigned int lru_mask) | |
3413 | { | |
3414 | unsigned long nr = 0; | |
3415 | enum lru_list lru; | |
3416 | ||
3417 | for_each_lru(lru) { | |
3418 | if (!(BIT(lru) & lru_mask)) | |
3419 | continue; | |
205b20cc | 3420 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); |
113b7dfd JW |
3421 | } |
3422 | return nr; | |
3423 | } | |
3424 | ||
2da8ca82 | 3425 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3426 | { |
25485de6 GT |
3427 | struct numa_stat { |
3428 | const char *name; | |
3429 | unsigned int lru_mask; | |
3430 | }; | |
3431 | ||
3432 | static const struct numa_stat stats[] = { | |
3433 | { "total", LRU_ALL }, | |
3434 | { "file", LRU_ALL_FILE }, | |
3435 | { "anon", LRU_ALL_ANON }, | |
3436 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3437 | }; | |
3438 | const struct numa_stat *stat; | |
406eb0c9 | 3439 | int nid; |
25485de6 | 3440 | unsigned long nr; |
aa9694bb | 3441 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3442 | |
25485de6 GT |
3443 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3444 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3445 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3446 | for_each_node_state(nid, N_MEMORY) { | |
3447 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3448 | stat->lru_mask); | |
3449 | seq_printf(m, " N%d=%lu", nid, nr); | |
3450 | } | |
3451 | seq_putc(m, '\n'); | |
406eb0c9 | 3452 | } |
406eb0c9 | 3453 | |
071aee13 YH |
3454 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3455 | struct mem_cgroup *iter; | |
3456 | ||
3457 | nr = 0; | |
3458 | for_each_mem_cgroup_tree(iter, memcg) | |
3459 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3460 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3461 | for_each_node_state(nid, N_MEMORY) { | |
3462 | nr = 0; | |
3463 | for_each_mem_cgroup_tree(iter, memcg) | |
3464 | nr += mem_cgroup_node_nr_lru_pages( | |
3465 | iter, nid, stat->lru_mask); | |
3466 | seq_printf(m, " N%d=%lu", nid, nr); | |
3467 | } | |
3468 | seq_putc(m, '\n'); | |
406eb0c9 | 3469 | } |
406eb0c9 | 3470 | |
406eb0c9 YH |
3471 | return 0; |
3472 | } | |
3473 | #endif /* CONFIG_NUMA */ | |
3474 | ||
df0e53d0 | 3475 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3476 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
3477 | PGPGIN, |
3478 | PGPGOUT, | |
3479 | PGFAULT, | |
3480 | PGMAJFAULT, | |
3481 | }; | |
3482 | ||
3483 | static const char *const memcg1_event_names[] = { | |
3484 | "pgpgin", | |
3485 | "pgpgout", | |
3486 | "pgfault", | |
3487 | "pgmajfault", | |
3488 | }; | |
3489 | ||
2da8ca82 | 3490 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3491 | { |
aa9694bb | 3492 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 3493 | unsigned long memory, memsw; |
af7c4b0e JW |
3494 | struct mem_cgroup *mi; |
3495 | unsigned int i; | |
406eb0c9 | 3496 | |
71cd3113 | 3497 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c RS |
3498 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3499 | ||
71cd3113 JW |
3500 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
3501 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) | |
1dd3a273 | 3502 | continue; |
71cd3113 | 3503 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], |
205b20cc | 3504 | memcg_page_state_local(memcg, memcg1_stats[i]) * |
71cd3113 | 3505 | PAGE_SIZE); |
1dd3a273 | 3506 | } |
7b854121 | 3507 | |
df0e53d0 JW |
3508 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3509 | seq_printf(m, "%s %lu\n", memcg1_event_names[i], | |
205b20cc | 3510 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
3511 | |
3512 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3513 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
205b20cc | 3514 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 3515 | PAGE_SIZE); |
af7c4b0e | 3516 | |
14067bb3 | 3517 | /* Hierarchical information */ |
3e32cb2e JW |
3518 | memory = memsw = PAGE_COUNTER_MAX; |
3519 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
bbec2e15 RG |
3520 | memory = min(memory, mi->memory.max); |
3521 | memsw = min(memsw, mi->memsw.max); | |
fee7b548 | 3522 | } |
3e32cb2e JW |
3523 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3524 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3525 | if (do_memsw_account()) |
3e32cb2e JW |
3526 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3527 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3528 | |
8de7ecc6 | 3529 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
71cd3113 | 3530 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3531 | continue; |
8de7ecc6 | 3532 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
42a30035 | 3533 | (u64)memcg_page_state(memcg, i) * PAGE_SIZE); |
af7c4b0e JW |
3534 | } |
3535 | ||
8de7ecc6 SB |
3536 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3537 | seq_printf(m, "total_%s %llu\n", memcg1_event_names[i], | |
42a30035 | 3538 | (u64)memcg_events(memcg, i)); |
af7c4b0e | 3539 | |
8de7ecc6 SB |
3540 | for (i = 0; i < NR_LRU_LISTS; i++) |
3541 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], | |
42a30035 JW |
3542 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
3543 | PAGE_SIZE); | |
14067bb3 | 3544 | |
7f016ee8 | 3545 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 3546 | { |
ef8f2327 MG |
3547 | pg_data_t *pgdat; |
3548 | struct mem_cgroup_per_node *mz; | |
89abfab1 | 3549 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3550 | unsigned long recent_rotated[2] = {0, 0}; |
3551 | unsigned long recent_scanned[2] = {0, 0}; | |
3552 | ||
ef8f2327 MG |
3553 | for_each_online_pgdat(pgdat) { |
3554 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
3555 | rstat = &mz->lruvec.reclaim_stat; | |
7f016ee8 | 3556 | |
ef8f2327 MG |
3557 | recent_rotated[0] += rstat->recent_rotated[0]; |
3558 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3559 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3560 | recent_scanned[1] += rstat->recent_scanned[1]; | |
3561 | } | |
78ccf5b5 JW |
3562 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3563 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3564 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3565 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3566 | } |
3567 | #endif | |
3568 | ||
d2ceb9b7 KH |
3569 | return 0; |
3570 | } | |
3571 | ||
182446d0 TH |
3572 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3573 | struct cftype *cft) | |
a7885eb8 | 3574 | { |
182446d0 | 3575 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3576 | |
1f4c025b | 3577 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3578 | } |
3579 | ||
182446d0 TH |
3580 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3581 | struct cftype *cft, u64 val) | |
a7885eb8 | 3582 | { |
182446d0 | 3583 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3584 | |
3dae7fec | 3585 | if (val > 100) |
a7885eb8 KM |
3586 | return -EINVAL; |
3587 | ||
14208b0e | 3588 | if (css->parent) |
3dae7fec JW |
3589 | memcg->swappiness = val; |
3590 | else | |
3591 | vm_swappiness = val; | |
068b38c1 | 3592 | |
a7885eb8 KM |
3593 | return 0; |
3594 | } | |
3595 | ||
2e72b634 KS |
3596 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3597 | { | |
3598 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3599 | unsigned long usage; |
2e72b634 KS |
3600 | int i; |
3601 | ||
3602 | rcu_read_lock(); | |
3603 | if (!swap) | |
2c488db2 | 3604 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3605 | else |
2c488db2 | 3606 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3607 | |
3608 | if (!t) | |
3609 | goto unlock; | |
3610 | ||
ce00a967 | 3611 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3612 | |
3613 | /* | |
748dad36 | 3614 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3615 | * If it's not true, a threshold was crossed after last |
3616 | * call of __mem_cgroup_threshold(). | |
3617 | */ | |
5407a562 | 3618 | i = t->current_threshold; |
2e72b634 KS |
3619 | |
3620 | /* | |
3621 | * Iterate backward over array of thresholds starting from | |
3622 | * current_threshold and check if a threshold is crossed. | |
3623 | * If none of thresholds below usage is crossed, we read | |
3624 | * only one element of the array here. | |
3625 | */ | |
3626 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3627 | eventfd_signal(t->entries[i].eventfd, 1); | |
3628 | ||
3629 | /* i = current_threshold + 1 */ | |
3630 | i++; | |
3631 | ||
3632 | /* | |
3633 | * Iterate forward over array of thresholds starting from | |
3634 | * current_threshold+1 and check if a threshold is crossed. | |
3635 | * If none of thresholds above usage is crossed, we read | |
3636 | * only one element of the array here. | |
3637 | */ | |
3638 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3639 | eventfd_signal(t->entries[i].eventfd, 1); | |
3640 | ||
3641 | /* Update current_threshold */ | |
5407a562 | 3642 | t->current_threshold = i - 1; |
2e72b634 KS |
3643 | unlock: |
3644 | rcu_read_unlock(); | |
3645 | } | |
3646 | ||
3647 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3648 | { | |
ad4ca5f4 KS |
3649 | while (memcg) { |
3650 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3651 | if (do_memsw_account()) |
ad4ca5f4 KS |
3652 | __mem_cgroup_threshold(memcg, true); |
3653 | ||
3654 | memcg = parent_mem_cgroup(memcg); | |
3655 | } | |
2e72b634 KS |
3656 | } |
3657 | ||
3658 | static int compare_thresholds(const void *a, const void *b) | |
3659 | { | |
3660 | const struct mem_cgroup_threshold *_a = a; | |
3661 | const struct mem_cgroup_threshold *_b = b; | |
3662 | ||
2bff24a3 GT |
3663 | if (_a->threshold > _b->threshold) |
3664 | return 1; | |
3665 | ||
3666 | if (_a->threshold < _b->threshold) | |
3667 | return -1; | |
3668 | ||
3669 | return 0; | |
2e72b634 KS |
3670 | } |
3671 | ||
c0ff4b85 | 3672 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3673 | { |
3674 | struct mem_cgroup_eventfd_list *ev; | |
3675 | ||
2bcf2e92 MH |
3676 | spin_lock(&memcg_oom_lock); |
3677 | ||
c0ff4b85 | 3678 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3679 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3680 | |
3681 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3682 | return 0; |
3683 | } | |
3684 | ||
c0ff4b85 | 3685 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3686 | { |
7d74b06f KH |
3687 | struct mem_cgroup *iter; |
3688 | ||
c0ff4b85 | 3689 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3690 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3691 | } |
3692 | ||
59b6f873 | 3693 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3694 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3695 | { |
2c488db2 KS |
3696 | struct mem_cgroup_thresholds *thresholds; |
3697 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3698 | unsigned long threshold; |
3699 | unsigned long usage; | |
2c488db2 | 3700 | int i, size, ret; |
2e72b634 | 3701 | |
650c5e56 | 3702 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3703 | if (ret) |
3704 | return ret; | |
3705 | ||
3706 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3707 | |
05b84301 | 3708 | if (type == _MEM) { |
2c488db2 | 3709 | thresholds = &memcg->thresholds; |
ce00a967 | 3710 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3711 | } else if (type == _MEMSWAP) { |
2c488db2 | 3712 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3713 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3714 | } else |
2e72b634 KS |
3715 | BUG(); |
3716 | ||
2e72b634 | 3717 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3718 | if (thresholds->primary) |
2e72b634 KS |
3719 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3720 | ||
2c488db2 | 3721 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3722 | |
3723 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 3724 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 3725 | if (!new) { |
2e72b634 KS |
3726 | ret = -ENOMEM; |
3727 | goto unlock; | |
3728 | } | |
2c488db2 | 3729 | new->size = size; |
2e72b634 KS |
3730 | |
3731 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3732 | if (thresholds->primary) { |
3733 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3734 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3735 | } |
3736 | ||
2e72b634 | 3737 | /* Add new threshold */ |
2c488db2 KS |
3738 | new->entries[size - 1].eventfd = eventfd; |
3739 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3740 | |
3741 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3742 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3743 | compare_thresholds, NULL); |
3744 | ||
3745 | /* Find current threshold */ | |
2c488db2 | 3746 | new->current_threshold = -1; |
2e72b634 | 3747 | for (i = 0; i < size; i++) { |
748dad36 | 3748 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3749 | /* |
2c488db2 KS |
3750 | * new->current_threshold will not be used until |
3751 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3752 | * it here. |
3753 | */ | |
2c488db2 | 3754 | ++new->current_threshold; |
748dad36 SZ |
3755 | } else |
3756 | break; | |
2e72b634 KS |
3757 | } |
3758 | ||
2c488db2 KS |
3759 | /* Free old spare buffer and save old primary buffer as spare */ |
3760 | kfree(thresholds->spare); | |
3761 | thresholds->spare = thresholds->primary; | |
3762 | ||
3763 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3764 | |
907860ed | 3765 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3766 | synchronize_rcu(); |
3767 | ||
2e72b634 KS |
3768 | unlock: |
3769 | mutex_unlock(&memcg->thresholds_lock); | |
3770 | ||
3771 | return ret; | |
3772 | } | |
3773 | ||
59b6f873 | 3774 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3775 | struct eventfd_ctx *eventfd, const char *args) |
3776 | { | |
59b6f873 | 3777 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3778 | } |
3779 | ||
59b6f873 | 3780 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3781 | struct eventfd_ctx *eventfd, const char *args) |
3782 | { | |
59b6f873 | 3783 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3784 | } |
3785 | ||
59b6f873 | 3786 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3787 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3788 | { |
2c488db2 KS |
3789 | struct mem_cgroup_thresholds *thresholds; |
3790 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3791 | unsigned long usage; |
2c488db2 | 3792 | int i, j, size; |
2e72b634 KS |
3793 | |
3794 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3795 | |
3796 | if (type == _MEM) { | |
2c488db2 | 3797 | thresholds = &memcg->thresholds; |
ce00a967 | 3798 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3799 | } else if (type == _MEMSWAP) { |
2c488db2 | 3800 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3801 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3802 | } else |
2e72b634 KS |
3803 | BUG(); |
3804 | ||
371528ca AV |
3805 | if (!thresholds->primary) |
3806 | goto unlock; | |
3807 | ||
2e72b634 KS |
3808 | /* Check if a threshold crossed before removing */ |
3809 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3810 | ||
3811 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3812 | size = 0; |
3813 | for (i = 0; i < thresholds->primary->size; i++) { | |
3814 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3815 | size++; |
3816 | } | |
3817 | ||
2c488db2 | 3818 | new = thresholds->spare; |
907860ed | 3819 | |
2e72b634 KS |
3820 | /* Set thresholds array to NULL if we don't have thresholds */ |
3821 | if (!size) { | |
2c488db2 KS |
3822 | kfree(new); |
3823 | new = NULL; | |
907860ed | 3824 | goto swap_buffers; |
2e72b634 KS |
3825 | } |
3826 | ||
2c488db2 | 3827 | new->size = size; |
2e72b634 KS |
3828 | |
3829 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3830 | new->current_threshold = -1; |
3831 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3832 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3833 | continue; |
3834 | ||
2c488db2 | 3835 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3836 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3837 | /* |
2c488db2 | 3838 | * new->current_threshold will not be used |
2e72b634 KS |
3839 | * until rcu_assign_pointer(), so it's safe to increment |
3840 | * it here. | |
3841 | */ | |
2c488db2 | 3842 | ++new->current_threshold; |
2e72b634 KS |
3843 | } |
3844 | j++; | |
3845 | } | |
3846 | ||
907860ed | 3847 | swap_buffers: |
2c488db2 KS |
3848 | /* Swap primary and spare array */ |
3849 | thresholds->spare = thresholds->primary; | |
8c757763 | 3850 | |
2c488db2 | 3851 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3852 | |
907860ed | 3853 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3854 | synchronize_rcu(); |
6611d8d7 MC |
3855 | |
3856 | /* If all events are unregistered, free the spare array */ | |
3857 | if (!new) { | |
3858 | kfree(thresholds->spare); | |
3859 | thresholds->spare = NULL; | |
3860 | } | |
371528ca | 3861 | unlock: |
2e72b634 | 3862 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3863 | } |
c1e862c1 | 3864 | |
59b6f873 | 3865 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3866 | struct eventfd_ctx *eventfd) |
3867 | { | |
59b6f873 | 3868 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3869 | } |
3870 | ||
59b6f873 | 3871 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3872 | struct eventfd_ctx *eventfd) |
3873 | { | |
59b6f873 | 3874 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3875 | } |
3876 | ||
59b6f873 | 3877 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3878 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3879 | { |
9490ff27 | 3880 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3881 | |
9490ff27 KH |
3882 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3883 | if (!event) | |
3884 | return -ENOMEM; | |
3885 | ||
1af8efe9 | 3886 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3887 | |
3888 | event->eventfd = eventfd; | |
3889 | list_add(&event->list, &memcg->oom_notify); | |
3890 | ||
3891 | /* already in OOM ? */ | |
c2b42d3c | 3892 | if (memcg->under_oom) |
9490ff27 | 3893 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3894 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3895 | |
3896 | return 0; | |
3897 | } | |
3898 | ||
59b6f873 | 3899 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3900 | struct eventfd_ctx *eventfd) |
9490ff27 | 3901 | { |
9490ff27 | 3902 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3903 | |
1af8efe9 | 3904 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3905 | |
c0ff4b85 | 3906 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3907 | if (ev->eventfd == eventfd) { |
3908 | list_del(&ev->list); | |
3909 | kfree(ev); | |
3910 | } | |
3911 | } | |
3912 | ||
1af8efe9 | 3913 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3914 | } |
3915 | ||
2da8ca82 | 3916 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3917 | { |
aa9694bb | 3918 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 3919 | |
791badbd | 3920 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3921 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
3922 | seq_printf(sf, "oom_kill %lu\n", |
3923 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
3924 | return 0; |
3925 | } | |
3926 | ||
182446d0 | 3927 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3928 | struct cftype *cft, u64 val) |
3929 | { | |
182446d0 | 3930 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3931 | |
3932 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3933 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3934 | return -EINVAL; |
3935 | ||
c0ff4b85 | 3936 | memcg->oom_kill_disable = val; |
4d845ebf | 3937 | if (!val) |
c0ff4b85 | 3938 | memcg_oom_recover(memcg); |
3dae7fec | 3939 | |
3c11ecf4 KH |
3940 | return 0; |
3941 | } | |
3942 | ||
52ebea74 TH |
3943 | #ifdef CONFIG_CGROUP_WRITEBACK |
3944 | ||
841710aa TH |
3945 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3946 | { | |
3947 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3948 | } | |
3949 | ||
3950 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3951 | { | |
3952 | wb_domain_exit(&memcg->cgwb_domain); | |
3953 | } | |
3954 | ||
2529bb3a TH |
3955 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3956 | { | |
3957 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3958 | } | |
3959 | ||
841710aa TH |
3960 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3961 | { | |
3962 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3963 | ||
3964 | if (!memcg->css.parent) | |
3965 | return NULL; | |
3966 | ||
3967 | return &memcg->cgwb_domain; | |
3968 | } | |
3969 | ||
0b3d6e6f GT |
3970 | /* |
3971 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
3972 | * Keep in sync with memcg_exact_page(). | |
3973 | */ | |
3974 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
3975 | { | |
871789d4 | 3976 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
3977 | int cpu; |
3978 | ||
3979 | for_each_online_cpu(cpu) | |
871789d4 | 3980 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
3981 | if (x < 0) |
3982 | x = 0; | |
3983 | return x; | |
3984 | } | |
3985 | ||
c2aa723a TH |
3986 | /** |
3987 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3988 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3989 | * @pfilepages: out parameter for number of file pages |
3990 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3991 | * @pdirty: out parameter for number of dirty pages |
3992 | * @pwriteback: out parameter for number of pages under writeback | |
3993 | * | |
c5edf9cd TH |
3994 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3995 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3996 | * is a bit more involved. | |
c2aa723a | 3997 | * |
c5edf9cd TH |
3998 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3999 | * headroom is calculated as the lowest headroom of itself and the | |
4000 | * ancestors. Note that this doesn't consider the actual amount of | |
4001 | * available memory in the system. The caller should further cap | |
4002 | * *@pheadroom accordingly. | |
c2aa723a | 4003 | */ |
c5edf9cd TH |
4004 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4005 | unsigned long *pheadroom, unsigned long *pdirty, | |
4006 | unsigned long *pwriteback) | |
c2aa723a TH |
4007 | { |
4008 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4009 | struct mem_cgroup *parent; | |
c2aa723a | 4010 | |
0b3d6e6f | 4011 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a TH |
4012 | |
4013 | /* this should eventually include NR_UNSTABLE_NFS */ | |
0b3d6e6f | 4014 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4015 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4016 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4017 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4018 | |
c2aa723a | 4019 | while ((parent = parent_mem_cgroup(memcg))) { |
bbec2e15 | 4020 | unsigned long ceiling = min(memcg->memory.max, memcg->high); |
c2aa723a TH |
4021 | unsigned long used = page_counter_read(&memcg->memory); |
4022 | ||
c5edf9cd | 4023 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4024 | memcg = parent; |
4025 | } | |
c2aa723a TH |
4026 | } |
4027 | ||
841710aa TH |
4028 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4029 | ||
4030 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4031 | { | |
4032 | return 0; | |
4033 | } | |
4034 | ||
4035 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4036 | { | |
4037 | } | |
4038 | ||
2529bb3a TH |
4039 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4040 | { | |
4041 | } | |
4042 | ||
52ebea74 TH |
4043 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4044 | ||
3bc942f3 TH |
4045 | /* |
4046 | * DO NOT USE IN NEW FILES. | |
4047 | * | |
4048 | * "cgroup.event_control" implementation. | |
4049 | * | |
4050 | * This is way over-engineered. It tries to support fully configurable | |
4051 | * events for each user. Such level of flexibility is completely | |
4052 | * unnecessary especially in the light of the planned unified hierarchy. | |
4053 | * | |
4054 | * Please deprecate this and replace with something simpler if at all | |
4055 | * possible. | |
4056 | */ | |
4057 | ||
79bd9814 TH |
4058 | /* |
4059 | * Unregister event and free resources. | |
4060 | * | |
4061 | * Gets called from workqueue. | |
4062 | */ | |
3bc942f3 | 4063 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4064 | { |
3bc942f3 TH |
4065 | struct mem_cgroup_event *event = |
4066 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4067 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4068 | |
4069 | remove_wait_queue(event->wqh, &event->wait); | |
4070 | ||
59b6f873 | 4071 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4072 | |
4073 | /* Notify userspace the event is going away. */ | |
4074 | eventfd_signal(event->eventfd, 1); | |
4075 | ||
4076 | eventfd_ctx_put(event->eventfd); | |
4077 | kfree(event); | |
59b6f873 | 4078 | css_put(&memcg->css); |
79bd9814 TH |
4079 | } |
4080 | ||
4081 | /* | |
a9a08845 | 4082 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4083 | * |
4084 | * Called with wqh->lock held and interrupts disabled. | |
4085 | */ | |
ac6424b9 | 4086 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4087 | int sync, void *key) |
79bd9814 | 4088 | { |
3bc942f3 TH |
4089 | struct mem_cgroup_event *event = |
4090 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4091 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4092 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4093 | |
a9a08845 | 4094 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4095 | /* |
4096 | * If the event has been detached at cgroup removal, we | |
4097 | * can simply return knowing the other side will cleanup | |
4098 | * for us. | |
4099 | * | |
4100 | * We can't race against event freeing since the other | |
4101 | * side will require wqh->lock via remove_wait_queue(), | |
4102 | * which we hold. | |
4103 | */ | |
fba94807 | 4104 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4105 | if (!list_empty(&event->list)) { |
4106 | list_del_init(&event->list); | |
4107 | /* | |
4108 | * We are in atomic context, but cgroup_event_remove() | |
4109 | * may sleep, so we have to call it in workqueue. | |
4110 | */ | |
4111 | schedule_work(&event->remove); | |
4112 | } | |
fba94807 | 4113 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4114 | } |
4115 | ||
4116 | return 0; | |
4117 | } | |
4118 | ||
3bc942f3 | 4119 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4120 | wait_queue_head_t *wqh, poll_table *pt) |
4121 | { | |
3bc942f3 TH |
4122 | struct mem_cgroup_event *event = |
4123 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4124 | |
4125 | event->wqh = wqh; | |
4126 | add_wait_queue(wqh, &event->wait); | |
4127 | } | |
4128 | ||
4129 | /* | |
3bc942f3 TH |
4130 | * DO NOT USE IN NEW FILES. |
4131 | * | |
79bd9814 TH |
4132 | * Parse input and register new cgroup event handler. |
4133 | * | |
4134 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4135 | * Interpretation of args is defined by control file implementation. | |
4136 | */ | |
451af504 TH |
4137 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4138 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4139 | { |
451af504 | 4140 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4141 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4142 | struct mem_cgroup_event *event; |
79bd9814 TH |
4143 | struct cgroup_subsys_state *cfile_css; |
4144 | unsigned int efd, cfd; | |
4145 | struct fd efile; | |
4146 | struct fd cfile; | |
fba94807 | 4147 | const char *name; |
79bd9814 TH |
4148 | char *endp; |
4149 | int ret; | |
4150 | ||
451af504 TH |
4151 | buf = strstrip(buf); |
4152 | ||
4153 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4154 | if (*endp != ' ') |
4155 | return -EINVAL; | |
451af504 | 4156 | buf = endp + 1; |
79bd9814 | 4157 | |
451af504 | 4158 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4159 | if ((*endp != ' ') && (*endp != '\0')) |
4160 | return -EINVAL; | |
451af504 | 4161 | buf = endp + 1; |
79bd9814 TH |
4162 | |
4163 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4164 | if (!event) | |
4165 | return -ENOMEM; | |
4166 | ||
59b6f873 | 4167 | event->memcg = memcg; |
79bd9814 | 4168 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4169 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4170 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4171 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4172 | |
4173 | efile = fdget(efd); | |
4174 | if (!efile.file) { | |
4175 | ret = -EBADF; | |
4176 | goto out_kfree; | |
4177 | } | |
4178 | ||
4179 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4180 | if (IS_ERR(event->eventfd)) { | |
4181 | ret = PTR_ERR(event->eventfd); | |
4182 | goto out_put_efile; | |
4183 | } | |
4184 | ||
4185 | cfile = fdget(cfd); | |
4186 | if (!cfile.file) { | |
4187 | ret = -EBADF; | |
4188 | goto out_put_eventfd; | |
4189 | } | |
4190 | ||
4191 | /* the process need read permission on control file */ | |
4192 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4193 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4194 | if (ret < 0) | |
4195 | goto out_put_cfile; | |
4196 | ||
fba94807 TH |
4197 | /* |
4198 | * Determine the event callbacks and set them in @event. This used | |
4199 | * to be done via struct cftype but cgroup core no longer knows | |
4200 | * about these events. The following is crude but the whole thing | |
4201 | * is for compatibility anyway. | |
3bc942f3 TH |
4202 | * |
4203 | * DO NOT ADD NEW FILES. | |
fba94807 | 4204 | */ |
b583043e | 4205 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4206 | |
4207 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4208 | event->register_event = mem_cgroup_usage_register_event; | |
4209 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4210 | } else if (!strcmp(name, "memory.oom_control")) { | |
4211 | event->register_event = mem_cgroup_oom_register_event; | |
4212 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4213 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4214 | event->register_event = vmpressure_register_event; | |
4215 | event->unregister_event = vmpressure_unregister_event; | |
4216 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4217 | event->register_event = memsw_cgroup_usage_register_event; |
4218 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4219 | } else { |
4220 | ret = -EINVAL; | |
4221 | goto out_put_cfile; | |
4222 | } | |
4223 | ||
79bd9814 | 4224 | /* |
b5557c4c TH |
4225 | * Verify @cfile should belong to @css. Also, remaining events are |
4226 | * automatically removed on cgroup destruction but the removal is | |
4227 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4228 | */ |
b583043e | 4229 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4230 | &memory_cgrp_subsys); |
79bd9814 | 4231 | ret = -EINVAL; |
5a17f543 | 4232 | if (IS_ERR(cfile_css)) |
79bd9814 | 4233 | goto out_put_cfile; |
5a17f543 TH |
4234 | if (cfile_css != css) { |
4235 | css_put(cfile_css); | |
79bd9814 | 4236 | goto out_put_cfile; |
5a17f543 | 4237 | } |
79bd9814 | 4238 | |
451af504 | 4239 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4240 | if (ret) |
4241 | goto out_put_css; | |
4242 | ||
9965ed17 | 4243 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4244 | |
fba94807 TH |
4245 | spin_lock(&memcg->event_list_lock); |
4246 | list_add(&event->list, &memcg->event_list); | |
4247 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4248 | |
4249 | fdput(cfile); | |
4250 | fdput(efile); | |
4251 | ||
451af504 | 4252 | return nbytes; |
79bd9814 TH |
4253 | |
4254 | out_put_css: | |
b5557c4c | 4255 | css_put(css); |
79bd9814 TH |
4256 | out_put_cfile: |
4257 | fdput(cfile); | |
4258 | out_put_eventfd: | |
4259 | eventfd_ctx_put(event->eventfd); | |
4260 | out_put_efile: | |
4261 | fdput(efile); | |
4262 | out_kfree: | |
4263 | kfree(event); | |
4264 | ||
4265 | return ret; | |
4266 | } | |
4267 | ||
241994ed | 4268 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4269 | { |
0eea1030 | 4270 | .name = "usage_in_bytes", |
8c7c6e34 | 4271 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4272 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4273 | }, |
c84872e1 PE |
4274 | { |
4275 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4276 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4277 | .write = mem_cgroup_reset, |
791badbd | 4278 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4279 | }, |
8cdea7c0 | 4280 | { |
0eea1030 | 4281 | .name = "limit_in_bytes", |
8c7c6e34 | 4282 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4283 | .write = mem_cgroup_write, |
791badbd | 4284 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4285 | }, |
296c81d8 BS |
4286 | { |
4287 | .name = "soft_limit_in_bytes", | |
4288 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4289 | .write = mem_cgroup_write, |
791badbd | 4290 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4291 | }, |
8cdea7c0 BS |
4292 | { |
4293 | .name = "failcnt", | |
8c7c6e34 | 4294 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4295 | .write = mem_cgroup_reset, |
791badbd | 4296 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4297 | }, |
d2ceb9b7 KH |
4298 | { |
4299 | .name = "stat", | |
2da8ca82 | 4300 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4301 | }, |
c1e862c1 KH |
4302 | { |
4303 | .name = "force_empty", | |
6770c64e | 4304 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4305 | }, |
18f59ea7 BS |
4306 | { |
4307 | .name = "use_hierarchy", | |
4308 | .write_u64 = mem_cgroup_hierarchy_write, | |
4309 | .read_u64 = mem_cgroup_hierarchy_read, | |
4310 | }, | |
79bd9814 | 4311 | { |
3bc942f3 | 4312 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4313 | .write = memcg_write_event_control, |
7dbdb199 | 4314 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4315 | }, |
a7885eb8 KM |
4316 | { |
4317 | .name = "swappiness", | |
4318 | .read_u64 = mem_cgroup_swappiness_read, | |
4319 | .write_u64 = mem_cgroup_swappiness_write, | |
4320 | }, | |
7dc74be0 DN |
4321 | { |
4322 | .name = "move_charge_at_immigrate", | |
4323 | .read_u64 = mem_cgroup_move_charge_read, | |
4324 | .write_u64 = mem_cgroup_move_charge_write, | |
4325 | }, | |
9490ff27 KH |
4326 | { |
4327 | .name = "oom_control", | |
2da8ca82 | 4328 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4329 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4330 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4331 | }, | |
70ddf637 AV |
4332 | { |
4333 | .name = "pressure_level", | |
70ddf637 | 4334 | }, |
406eb0c9 YH |
4335 | #ifdef CONFIG_NUMA |
4336 | { | |
4337 | .name = "numa_stat", | |
2da8ca82 | 4338 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4339 | }, |
4340 | #endif | |
510fc4e1 GC |
4341 | { |
4342 | .name = "kmem.limit_in_bytes", | |
4343 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4344 | .write = mem_cgroup_write, |
791badbd | 4345 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4346 | }, |
4347 | { | |
4348 | .name = "kmem.usage_in_bytes", | |
4349 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4350 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4351 | }, |
4352 | { | |
4353 | .name = "kmem.failcnt", | |
4354 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4355 | .write = mem_cgroup_reset, |
791badbd | 4356 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4357 | }, |
4358 | { | |
4359 | .name = "kmem.max_usage_in_bytes", | |
4360 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4361 | .write = mem_cgroup_reset, |
791badbd | 4362 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4363 | }, |
5b365771 | 4364 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
749c5415 GC |
4365 | { |
4366 | .name = "kmem.slabinfo", | |
bc2791f8 TH |
4367 | .seq_start = memcg_slab_start, |
4368 | .seq_next = memcg_slab_next, | |
4369 | .seq_stop = memcg_slab_stop, | |
b047501c | 4370 | .seq_show = memcg_slab_show, |
749c5415 GC |
4371 | }, |
4372 | #endif | |
d55f90bf VD |
4373 | { |
4374 | .name = "kmem.tcp.limit_in_bytes", | |
4375 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4376 | .write = mem_cgroup_write, | |
4377 | .read_u64 = mem_cgroup_read_u64, | |
4378 | }, | |
4379 | { | |
4380 | .name = "kmem.tcp.usage_in_bytes", | |
4381 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4382 | .read_u64 = mem_cgroup_read_u64, | |
4383 | }, | |
4384 | { | |
4385 | .name = "kmem.tcp.failcnt", | |
4386 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4387 | .write = mem_cgroup_reset, | |
4388 | .read_u64 = mem_cgroup_read_u64, | |
4389 | }, | |
4390 | { | |
4391 | .name = "kmem.tcp.max_usage_in_bytes", | |
4392 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4393 | .write = mem_cgroup_reset, | |
4394 | .read_u64 = mem_cgroup_read_u64, | |
4395 | }, | |
6bc10349 | 4396 | { }, /* terminate */ |
af36f906 | 4397 | }; |
8c7c6e34 | 4398 | |
73f576c0 JW |
4399 | /* |
4400 | * Private memory cgroup IDR | |
4401 | * | |
4402 | * Swap-out records and page cache shadow entries need to store memcg | |
4403 | * references in constrained space, so we maintain an ID space that is | |
4404 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4405 | * memory-controlled cgroups to 64k. | |
4406 | * | |
4407 | * However, there usually are many references to the oflline CSS after | |
4408 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4409 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4410 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4411 | * relatively small ID space and prevent the creation of new cgroups | |
4412 | * even when there are much fewer than 64k cgroups - possibly none. | |
4413 | * | |
4414 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4415 | * be freed and recycled when it's no longer needed, which is usually | |
4416 | * when the CSS is offlined. | |
4417 | * | |
4418 | * The only exception to that are records of swapped out tmpfs/shmem | |
4419 | * pages that need to be attributed to live ancestors on swapin. But | |
4420 | * those references are manageable from userspace. | |
4421 | */ | |
4422 | ||
4423 | static DEFINE_IDR(mem_cgroup_idr); | |
4424 | ||
7e97de0b KT |
4425 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
4426 | { | |
4427 | if (memcg->id.id > 0) { | |
4428 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
4429 | memcg->id.id = 0; | |
4430 | } | |
4431 | } | |
4432 | ||
615d66c3 | 4433 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4434 | { |
1c2d479a | 4435 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
4436 | } |
4437 | ||
615d66c3 | 4438 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4439 | { |
1c2d479a | 4440 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 4441 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
4442 | |
4443 | /* Memcg ID pins CSS */ | |
4444 | css_put(&memcg->css); | |
4445 | } | |
4446 | } | |
4447 | ||
615d66c3 VD |
4448 | static inline void mem_cgroup_id_get(struct mem_cgroup *memcg) |
4449 | { | |
4450 | mem_cgroup_id_get_many(memcg, 1); | |
4451 | } | |
4452 | ||
4453 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) | |
4454 | { | |
4455 | mem_cgroup_id_put_many(memcg, 1); | |
4456 | } | |
4457 | ||
73f576c0 JW |
4458 | /** |
4459 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4460 | * @id: the memcg id to look up | |
4461 | * | |
4462 | * Caller must hold rcu_read_lock(). | |
4463 | */ | |
4464 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4465 | { | |
4466 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4467 | return idr_find(&mem_cgroup_idr, id); | |
4468 | } | |
4469 | ||
ef8f2327 | 4470 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4471 | { |
4472 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 4473 | int tmp = node; |
1ecaab2b KH |
4474 | /* |
4475 | * This routine is called against possible nodes. | |
4476 | * But it's BUG to call kmalloc() against offline node. | |
4477 | * | |
4478 | * TODO: this routine can waste much memory for nodes which will | |
4479 | * never be onlined. It's better to use memory hotplug callback | |
4480 | * function. | |
4481 | */ | |
41e3355d KH |
4482 | if (!node_state(node, N_NORMAL_MEMORY)) |
4483 | tmp = -1; | |
17295c88 | 4484 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4485 | if (!pn) |
4486 | return 1; | |
1ecaab2b | 4487 | |
815744d7 JW |
4488 | pn->lruvec_stat_local = alloc_percpu(struct lruvec_stat); |
4489 | if (!pn->lruvec_stat_local) { | |
4490 | kfree(pn); | |
4491 | return 1; | |
4492 | } | |
4493 | ||
a983b5eb JW |
4494 | pn->lruvec_stat_cpu = alloc_percpu(struct lruvec_stat); |
4495 | if (!pn->lruvec_stat_cpu) { | |
815744d7 | 4496 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c JW |
4497 | kfree(pn); |
4498 | return 1; | |
4499 | } | |
4500 | ||
ef8f2327 MG |
4501 | lruvec_init(&pn->lruvec); |
4502 | pn->usage_in_excess = 0; | |
4503 | pn->on_tree = false; | |
4504 | pn->memcg = memcg; | |
4505 | ||
54f72fe0 | 4506 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4507 | return 0; |
4508 | } | |
4509 | ||
ef8f2327 | 4510 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4511 | { |
00f3ca2c JW |
4512 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
4513 | ||
4eaf431f MH |
4514 | if (!pn) |
4515 | return; | |
4516 | ||
a983b5eb | 4517 | free_percpu(pn->lruvec_stat_cpu); |
815744d7 | 4518 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c | 4519 | kfree(pn); |
1ecaab2b KH |
4520 | } |
4521 | ||
40e952f9 | 4522 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4523 | { |
c8b2a36f | 4524 | int node; |
59927fb9 | 4525 | |
c8b2a36f | 4526 | for_each_node(node) |
ef8f2327 | 4527 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 4528 | free_percpu(memcg->vmstats_percpu); |
815744d7 | 4529 | free_percpu(memcg->vmstats_local); |
8ff69e2c | 4530 | kfree(memcg); |
59927fb9 | 4531 | } |
3afe36b1 | 4532 | |
40e952f9 TE |
4533 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
4534 | { | |
4535 | memcg_wb_domain_exit(memcg); | |
4536 | __mem_cgroup_free(memcg); | |
4537 | } | |
4538 | ||
0b8f73e1 | 4539 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4540 | { |
d142e3e6 | 4541 | struct mem_cgroup *memcg; |
b9726c26 | 4542 | unsigned int size; |
6d12e2d8 | 4543 | int node; |
8cdea7c0 | 4544 | |
0b8f73e1 JW |
4545 | size = sizeof(struct mem_cgroup); |
4546 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4547 | ||
4548 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4549 | if (!memcg) |
0b8f73e1 JW |
4550 | return NULL; |
4551 | ||
73f576c0 JW |
4552 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
4553 | 1, MEM_CGROUP_ID_MAX, | |
4554 | GFP_KERNEL); | |
4555 | if (memcg->id.id < 0) | |
4556 | goto fail; | |
4557 | ||
815744d7 JW |
4558 | memcg->vmstats_local = alloc_percpu(struct memcg_vmstats_percpu); |
4559 | if (!memcg->vmstats_local) | |
4560 | goto fail; | |
4561 | ||
871789d4 CD |
4562 | memcg->vmstats_percpu = alloc_percpu(struct memcg_vmstats_percpu); |
4563 | if (!memcg->vmstats_percpu) | |
0b8f73e1 | 4564 | goto fail; |
78fb7466 | 4565 | |
3ed28fa1 | 4566 | for_each_node(node) |
ef8f2327 | 4567 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 4568 | goto fail; |
f64c3f54 | 4569 | |
0b8f73e1 JW |
4570 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4571 | goto fail; | |
28dbc4b6 | 4572 | |
f7e1cb6e | 4573 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4574 | memcg->last_scanned_node = MAX_NUMNODES; |
4575 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4576 | mutex_init(&memcg->thresholds_lock); |
4577 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4578 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4579 | INIT_LIST_HEAD(&memcg->event_list); |
4580 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4581 | memcg->socket_pressure = jiffies; |
84c07d11 | 4582 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 4583 | memcg->kmemcg_id = -1; |
900a38f0 | 4584 | #endif |
52ebea74 TH |
4585 | #ifdef CONFIG_CGROUP_WRITEBACK |
4586 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4587 | #endif | |
73f576c0 | 4588 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
4589 | return memcg; |
4590 | fail: | |
7e97de0b | 4591 | mem_cgroup_id_remove(memcg); |
40e952f9 | 4592 | __mem_cgroup_free(memcg); |
0b8f73e1 | 4593 | return NULL; |
d142e3e6 GC |
4594 | } |
4595 | ||
0b8f73e1 JW |
4596 | static struct cgroup_subsys_state * __ref |
4597 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4598 | { |
0b8f73e1 JW |
4599 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4600 | struct mem_cgroup *memcg; | |
4601 | long error = -ENOMEM; | |
d142e3e6 | 4602 | |
0b8f73e1 JW |
4603 | memcg = mem_cgroup_alloc(); |
4604 | if (!memcg) | |
4605 | return ERR_PTR(error); | |
d142e3e6 | 4606 | |
0b8f73e1 JW |
4607 | memcg->high = PAGE_COUNTER_MAX; |
4608 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4609 | if (parent) { | |
4610 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4611 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4612 | } | |
4613 | if (parent && parent->use_hierarchy) { | |
4614 | memcg->use_hierarchy = true; | |
3e32cb2e | 4615 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4616 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4617 | page_counter_init(&memcg->memsw, &parent->memsw); |
4618 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4619 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4620 | } else { |
3e32cb2e | 4621 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4622 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4623 | page_counter_init(&memcg->memsw, NULL); |
4624 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4625 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4626 | /* |
4627 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4628 | * much sense so let cgroup subsystem know about this | |
4629 | * unfortunate state in our controller. | |
4630 | */ | |
d142e3e6 | 4631 | if (parent != root_mem_cgroup) |
073219e9 | 4632 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4633 | } |
d6441637 | 4634 | |
0b8f73e1 JW |
4635 | /* The following stuff does not apply to the root */ |
4636 | if (!parent) { | |
4637 | root_mem_cgroup = memcg; | |
4638 | return &memcg->css; | |
4639 | } | |
4640 | ||
b313aeee | 4641 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
4642 | if (error) |
4643 | goto fail; | |
127424c8 | 4644 | |
f7e1cb6e | 4645 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4646 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4647 | |
0b8f73e1 JW |
4648 | return &memcg->css; |
4649 | fail: | |
7e97de0b | 4650 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 4651 | mem_cgroup_free(memcg); |
ea3a9645 | 4652 | return ERR_PTR(-ENOMEM); |
0b8f73e1 JW |
4653 | } |
4654 | ||
73f576c0 | 4655 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 4656 | { |
58fa2a55 VD |
4657 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
4658 | ||
0a4465d3 KT |
4659 | /* |
4660 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
4661 | * by the time the maps are allocated. So, we allocate maps | |
4662 | * here, when for_each_mem_cgroup() can't skip it. | |
4663 | */ | |
4664 | if (memcg_alloc_shrinker_maps(memcg)) { | |
4665 | mem_cgroup_id_remove(memcg); | |
4666 | return -ENOMEM; | |
4667 | } | |
4668 | ||
73f576c0 | 4669 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 4670 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 4671 | css_get(css); |
2f7dd7a4 | 4672 | return 0; |
8cdea7c0 BS |
4673 | } |
4674 | ||
eb95419b | 4675 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4676 | { |
eb95419b | 4677 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4678 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4679 | |
4680 | /* | |
4681 | * Unregister events and notify userspace. | |
4682 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4683 | * directory to avoid race between userspace and kernelspace. | |
4684 | */ | |
fba94807 TH |
4685 | spin_lock(&memcg->event_list_lock); |
4686 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4687 | list_del_init(&event->list); |
4688 | schedule_work(&event->remove); | |
4689 | } | |
fba94807 | 4690 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4691 | |
bf8d5d52 | 4692 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4693 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 4694 | |
567e9ab2 | 4695 | memcg_offline_kmem(memcg); |
52ebea74 | 4696 | wb_memcg_offline(memcg); |
73f576c0 | 4697 | |
591edfb1 RG |
4698 | drain_all_stock(memcg); |
4699 | ||
73f576c0 | 4700 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
4701 | } |
4702 | ||
6df38689 VD |
4703 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4704 | { | |
4705 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4706 | ||
4707 | invalidate_reclaim_iterators(memcg); | |
4708 | } | |
4709 | ||
eb95419b | 4710 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4711 | { |
eb95419b | 4712 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4713 | |
f7e1cb6e | 4714 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4715 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4716 | |
0db15298 | 4717 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4718 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4719 | |
0b8f73e1 JW |
4720 | vmpressure_cleanup(&memcg->vmpressure); |
4721 | cancel_work_sync(&memcg->high_work); | |
4722 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 4723 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 4724 | memcg_free_kmem(memcg); |
0b8f73e1 | 4725 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4726 | } |
4727 | ||
1ced953b TH |
4728 | /** |
4729 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4730 | * @css: the target css | |
4731 | * | |
4732 | * Reset the states of the mem_cgroup associated with @css. This is | |
4733 | * invoked when the userland requests disabling on the default hierarchy | |
4734 | * but the memcg is pinned through dependency. The memcg should stop | |
4735 | * applying policies and should revert to the vanilla state as it may be | |
4736 | * made visible again. | |
4737 | * | |
4738 | * The current implementation only resets the essential configurations. | |
4739 | * This needs to be expanded to cover all the visible parts. | |
4740 | */ | |
4741 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4742 | { | |
4743 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4744 | ||
bbec2e15 RG |
4745 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
4746 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
4747 | page_counter_set_max(&memcg->memsw, PAGE_COUNTER_MAX); | |
4748 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); | |
4749 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 4750 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4751 | page_counter_set_low(&memcg->memory, 0); |
241994ed | 4752 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4753 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4754 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4755 | } |
4756 | ||
02491447 | 4757 | #ifdef CONFIG_MMU |
7dc74be0 | 4758 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4759 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4760 | { |
05b84301 | 4761 | int ret; |
9476db97 | 4762 | |
d0164adc MG |
4763 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4764 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4765 | if (!ret) { |
854ffa8d | 4766 | mc.precharge += count; |
854ffa8d DN |
4767 | return ret; |
4768 | } | |
9476db97 | 4769 | |
3674534b | 4770 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 4771 | while (count--) { |
3674534b | 4772 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 4773 | if (ret) |
38c5d72f | 4774 | return ret; |
854ffa8d | 4775 | mc.precharge++; |
9476db97 | 4776 | cond_resched(); |
854ffa8d | 4777 | } |
9476db97 | 4778 | return 0; |
4ffef5fe DN |
4779 | } |
4780 | ||
4ffef5fe DN |
4781 | union mc_target { |
4782 | struct page *page; | |
02491447 | 4783 | swp_entry_t ent; |
4ffef5fe DN |
4784 | }; |
4785 | ||
4ffef5fe | 4786 | enum mc_target_type { |
8d32ff84 | 4787 | MC_TARGET_NONE = 0, |
4ffef5fe | 4788 | MC_TARGET_PAGE, |
02491447 | 4789 | MC_TARGET_SWAP, |
c733a828 | 4790 | MC_TARGET_DEVICE, |
4ffef5fe DN |
4791 | }; |
4792 | ||
90254a65 DN |
4793 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4794 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4795 | { |
c733a828 | 4796 | struct page *page = _vm_normal_page(vma, addr, ptent, true); |
4ffef5fe | 4797 | |
90254a65 DN |
4798 | if (!page || !page_mapped(page)) |
4799 | return NULL; | |
4800 | if (PageAnon(page)) { | |
1dfab5ab | 4801 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4802 | return NULL; |
1dfab5ab JW |
4803 | } else { |
4804 | if (!(mc.flags & MOVE_FILE)) | |
4805 | return NULL; | |
4806 | } | |
90254a65 DN |
4807 | if (!get_page_unless_zero(page)) |
4808 | return NULL; | |
4809 | ||
4810 | return page; | |
4811 | } | |
4812 | ||
c733a828 | 4813 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 4814 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 4815 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 4816 | { |
90254a65 DN |
4817 | struct page *page = NULL; |
4818 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4819 | ||
1dfab5ab | 4820 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4821 | return NULL; |
c733a828 JG |
4822 | |
4823 | /* | |
4824 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
4825 | * a device and because they are not accessible by CPU they are store | |
4826 | * as special swap entry in the CPU page table. | |
4827 | */ | |
4828 | if (is_device_private_entry(ent)) { | |
4829 | page = device_private_entry_to_page(ent); | |
4830 | /* | |
4831 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
4832 | * a refcount of 1 when free (unlike normal page) | |
4833 | */ | |
4834 | if (!page_ref_add_unless(page, 1, 1)) | |
4835 | return NULL; | |
4836 | return page; | |
4837 | } | |
4838 | ||
4b91355e KH |
4839 | /* |
4840 | * Because lookup_swap_cache() updates some statistics counter, | |
4841 | * we call find_get_page() with swapper_space directly. | |
4842 | */ | |
f6ab1f7f | 4843 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
7941d214 | 4844 | if (do_memsw_account()) |
90254a65 DN |
4845 | entry->val = ent.val; |
4846 | ||
4847 | return page; | |
4848 | } | |
4b91355e KH |
4849 | #else |
4850 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 4851 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
4852 | { |
4853 | return NULL; | |
4854 | } | |
4855 | #endif | |
90254a65 | 4856 | |
87946a72 DN |
4857 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4858 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4859 | { | |
4860 | struct page *page = NULL; | |
87946a72 DN |
4861 | struct address_space *mapping; |
4862 | pgoff_t pgoff; | |
4863 | ||
4864 | if (!vma->vm_file) /* anonymous vma */ | |
4865 | return NULL; | |
1dfab5ab | 4866 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4867 | return NULL; |
4868 | ||
87946a72 | 4869 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4870 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4871 | |
4872 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4873 | #ifdef CONFIG_SWAP |
4874 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4875 | if (shmem_mapping(mapping)) { |
4876 | page = find_get_entry(mapping, pgoff); | |
3159f943 | 4877 | if (xa_is_value(page)) { |
139b6a6f | 4878 | swp_entry_t swp = radix_to_swp_entry(page); |
7941d214 | 4879 | if (do_memsw_account()) |
139b6a6f | 4880 | *entry = swp; |
f6ab1f7f HY |
4881 | page = find_get_page(swap_address_space(swp), |
4882 | swp_offset(swp)); | |
139b6a6f JW |
4883 | } |
4884 | } else | |
4885 | page = find_get_page(mapping, pgoff); | |
4886 | #else | |
4887 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4888 | #endif |
87946a72 DN |
4889 | return page; |
4890 | } | |
4891 | ||
b1b0deab CG |
4892 | /** |
4893 | * mem_cgroup_move_account - move account of the page | |
4894 | * @page: the page | |
25843c2b | 4895 | * @compound: charge the page as compound or small page |
b1b0deab CG |
4896 | * @from: mem_cgroup which the page is moved from. |
4897 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4898 | * | |
3ac808fd | 4899 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
4900 | * |
4901 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4902 | * from old cgroup. | |
4903 | */ | |
4904 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 4905 | bool compound, |
b1b0deab CG |
4906 | struct mem_cgroup *from, |
4907 | struct mem_cgroup *to) | |
4908 | { | |
4909 | unsigned long flags; | |
f627c2f5 | 4910 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 4911 | int ret; |
c4843a75 | 4912 | bool anon; |
b1b0deab CG |
4913 | |
4914 | VM_BUG_ON(from == to); | |
4915 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 4916 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
4917 | |
4918 | /* | |
6a93ca8f | 4919 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 4920 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 4921 | */ |
f627c2f5 | 4922 | ret = -EBUSY; |
b1b0deab CG |
4923 | if (!trylock_page(page)) |
4924 | goto out; | |
4925 | ||
4926 | ret = -EINVAL; | |
4927 | if (page->mem_cgroup != from) | |
4928 | goto out_unlock; | |
4929 | ||
c4843a75 GT |
4930 | anon = PageAnon(page); |
4931 | ||
b1b0deab CG |
4932 | spin_lock_irqsave(&from->move_lock, flags); |
4933 | ||
c4843a75 | 4934 | if (!anon && page_mapped(page)) { |
c9019e9b JW |
4935 | __mod_memcg_state(from, NR_FILE_MAPPED, -nr_pages); |
4936 | __mod_memcg_state(to, NR_FILE_MAPPED, nr_pages); | |
b1b0deab CG |
4937 | } |
4938 | ||
c4843a75 GT |
4939 | /* |
4940 | * move_lock grabbed above and caller set from->moving_account, so | |
ccda7f43 | 4941 | * mod_memcg_page_state will serialize updates to PageDirty. |
c4843a75 GT |
4942 | * So mapping should be stable for dirty pages. |
4943 | */ | |
4944 | if (!anon && PageDirty(page)) { | |
4945 | struct address_space *mapping = page_mapping(page); | |
4946 | ||
4947 | if (mapping_cap_account_dirty(mapping)) { | |
c9019e9b JW |
4948 | __mod_memcg_state(from, NR_FILE_DIRTY, -nr_pages); |
4949 | __mod_memcg_state(to, NR_FILE_DIRTY, nr_pages); | |
c4843a75 GT |
4950 | } |
4951 | } | |
4952 | ||
b1b0deab | 4953 | if (PageWriteback(page)) { |
c9019e9b JW |
4954 | __mod_memcg_state(from, NR_WRITEBACK, -nr_pages); |
4955 | __mod_memcg_state(to, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
4956 | } |
4957 | ||
4958 | /* | |
4959 | * It is safe to change page->mem_cgroup here because the page | |
4960 | * is referenced, charged, and isolated - we can't race with | |
4961 | * uncharging, charging, migration, or LRU putback. | |
4962 | */ | |
4963 | ||
4964 | /* caller should have done css_get */ | |
4965 | page->mem_cgroup = to; | |
4966 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4967 | ||
4968 | ret = 0; | |
4969 | ||
4970 | local_irq_disable(); | |
f627c2f5 | 4971 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 4972 | memcg_check_events(to, page); |
f627c2f5 | 4973 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
4974 | memcg_check_events(from, page); |
4975 | local_irq_enable(); | |
4976 | out_unlock: | |
4977 | unlock_page(page); | |
4978 | out: | |
4979 | return ret; | |
4980 | } | |
4981 | ||
7cf7806c LR |
4982 | /** |
4983 | * get_mctgt_type - get target type of moving charge | |
4984 | * @vma: the vma the pte to be checked belongs | |
4985 | * @addr: the address corresponding to the pte to be checked | |
4986 | * @ptent: the pte to be checked | |
4987 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
4988 | * | |
4989 | * Returns | |
4990 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4991 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4992 | * move charge. if @target is not NULL, the page is stored in target->page | |
4993 | * with extra refcnt got(Callers should handle it). | |
4994 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
4995 | * target for charge migration. if @target is not NULL, the entry is stored | |
4996 | * in target->ent. | |
df6ad698 JG |
4997 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PUBLIC |
4998 | * or MEMORY_DEVICE_PRIVATE (so ZONE_DEVICE page and thus not on the lru). | |
4999 | * For now we such page is charge like a regular page would be as for all | |
5000 | * intent and purposes it is just special memory taking the place of a | |
5001 | * regular page. | |
c733a828 JG |
5002 | * |
5003 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5004 | * |
5005 | * Called with pte lock held. | |
5006 | */ | |
5007 | ||
8d32ff84 | 5008 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5009 | unsigned long addr, pte_t ptent, union mc_target *target) |
5010 | { | |
5011 | struct page *page = NULL; | |
8d32ff84 | 5012 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5013 | swp_entry_t ent = { .val = 0 }; |
5014 | ||
5015 | if (pte_present(ptent)) | |
5016 | page = mc_handle_present_pte(vma, addr, ptent); | |
5017 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5018 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5019 | else if (pte_none(ptent)) |
87946a72 | 5020 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5021 | |
5022 | if (!page && !ent.val) | |
8d32ff84 | 5023 | return ret; |
02491447 | 5024 | if (page) { |
02491447 | 5025 | /* |
0a31bc97 | 5026 | * Do only loose check w/o serialization. |
1306a85a | 5027 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5028 | * not under LRU exclusion. |
02491447 | 5029 | */ |
1306a85a | 5030 | if (page->mem_cgroup == mc.from) { |
02491447 | 5031 | ret = MC_TARGET_PAGE; |
df6ad698 JG |
5032 | if (is_device_private_page(page) || |
5033 | is_device_public_page(page)) | |
c733a828 | 5034 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5035 | if (target) |
5036 | target->page = page; | |
5037 | } | |
5038 | if (!ret || !target) | |
5039 | put_page(page); | |
5040 | } | |
3e14a57b HY |
5041 | /* |
5042 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5043 | * But we cannot move a tail-page in a THP. | |
5044 | */ | |
5045 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5046 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5047 | ret = MC_TARGET_SWAP; |
5048 | if (target) | |
5049 | target->ent = ent; | |
4ffef5fe | 5050 | } |
4ffef5fe DN |
5051 | return ret; |
5052 | } | |
5053 | ||
12724850 NH |
5054 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5055 | /* | |
d6810d73 HY |
5056 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5057 | * not support them for now. | |
12724850 NH |
5058 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5059 | */ | |
5060 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5061 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5062 | { | |
5063 | struct page *page = NULL; | |
12724850 NH |
5064 | enum mc_target_type ret = MC_TARGET_NONE; |
5065 | ||
84c3fc4e ZY |
5066 | if (unlikely(is_swap_pmd(pmd))) { |
5067 | VM_BUG_ON(thp_migration_supported() && | |
5068 | !is_pmd_migration_entry(pmd)); | |
5069 | return ret; | |
5070 | } | |
12724850 | 5071 | page = pmd_page(pmd); |
309381fe | 5072 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5073 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5074 | return ret; |
1306a85a | 5075 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
5076 | ret = MC_TARGET_PAGE; |
5077 | if (target) { | |
5078 | get_page(page); | |
5079 | target->page = page; | |
5080 | } | |
5081 | } | |
5082 | return ret; | |
5083 | } | |
5084 | #else | |
5085 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5086 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5087 | { | |
5088 | return MC_TARGET_NONE; | |
5089 | } | |
5090 | #endif | |
5091 | ||
4ffef5fe DN |
5092 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5093 | unsigned long addr, unsigned long end, | |
5094 | struct mm_walk *walk) | |
5095 | { | |
26bcd64a | 5096 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5097 | pte_t *pte; |
5098 | spinlock_t *ptl; | |
5099 | ||
b6ec57f4 KS |
5100 | ptl = pmd_trans_huge_lock(pmd, vma); |
5101 | if (ptl) { | |
c733a828 JG |
5102 | /* |
5103 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
5104 | * support transparent huge page with MEMORY_DEVICE_PUBLIC or | |
5105 | * MEMORY_DEVICE_PRIVATE but this might change. | |
5106 | */ | |
12724850 NH |
5107 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5108 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5109 | spin_unlock(ptl); |
1a5a9906 | 5110 | return 0; |
12724850 | 5111 | } |
03319327 | 5112 | |
45f83cef AA |
5113 | if (pmd_trans_unstable(pmd)) |
5114 | return 0; | |
4ffef5fe DN |
5115 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5116 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5117 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5118 | mc.precharge++; /* increment precharge temporarily */ |
5119 | pte_unmap_unlock(pte - 1, ptl); | |
5120 | cond_resched(); | |
5121 | ||
7dc74be0 DN |
5122 | return 0; |
5123 | } | |
5124 | ||
4ffef5fe DN |
5125 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5126 | { | |
5127 | unsigned long precharge; | |
4ffef5fe | 5128 | |
26bcd64a NH |
5129 | struct mm_walk mem_cgroup_count_precharge_walk = { |
5130 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5131 | .mm = mm, | |
5132 | }; | |
dfe076b0 | 5133 | down_read(&mm->mmap_sem); |
0247f3f4 JM |
5134 | walk_page_range(0, mm->highest_vm_end, |
5135 | &mem_cgroup_count_precharge_walk); | |
dfe076b0 | 5136 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5137 | |
5138 | precharge = mc.precharge; | |
5139 | mc.precharge = 0; | |
5140 | ||
5141 | return precharge; | |
5142 | } | |
5143 | ||
4ffef5fe DN |
5144 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5145 | { | |
dfe076b0 DN |
5146 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5147 | ||
5148 | VM_BUG_ON(mc.moving_task); | |
5149 | mc.moving_task = current; | |
5150 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5151 | } |
5152 | ||
dfe076b0 DN |
5153 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5154 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5155 | { |
2bd9bb20 KH |
5156 | struct mem_cgroup *from = mc.from; |
5157 | struct mem_cgroup *to = mc.to; | |
5158 | ||
4ffef5fe | 5159 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5160 | if (mc.precharge) { |
00501b53 | 5161 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5162 | mc.precharge = 0; |
5163 | } | |
5164 | /* | |
5165 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5166 | * we must uncharge here. | |
5167 | */ | |
5168 | if (mc.moved_charge) { | |
00501b53 | 5169 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5170 | mc.moved_charge = 0; |
4ffef5fe | 5171 | } |
483c30b5 DN |
5172 | /* we must fixup refcnts and charges */ |
5173 | if (mc.moved_swap) { | |
483c30b5 | 5174 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5175 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5176 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5177 | |
615d66c3 VD |
5178 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5179 | ||
05b84301 | 5180 | /* |
3e32cb2e JW |
5181 | * we charged both to->memory and to->memsw, so we |
5182 | * should uncharge to->memory. | |
05b84301 | 5183 | */ |
ce00a967 | 5184 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5185 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5186 | ||
615d66c3 VD |
5187 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
5188 | css_put_many(&mc.to->css, mc.moved_swap); | |
3e32cb2e | 5189 | |
483c30b5 DN |
5190 | mc.moved_swap = 0; |
5191 | } | |
dfe076b0 DN |
5192 | memcg_oom_recover(from); |
5193 | memcg_oom_recover(to); | |
5194 | wake_up_all(&mc.waitq); | |
5195 | } | |
5196 | ||
5197 | static void mem_cgroup_clear_mc(void) | |
5198 | { | |
264a0ae1 TH |
5199 | struct mm_struct *mm = mc.mm; |
5200 | ||
dfe076b0 DN |
5201 | /* |
5202 | * we must clear moving_task before waking up waiters at the end of | |
5203 | * task migration. | |
5204 | */ | |
5205 | mc.moving_task = NULL; | |
5206 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5207 | spin_lock(&mc.lock); |
4ffef5fe DN |
5208 | mc.from = NULL; |
5209 | mc.to = NULL; | |
264a0ae1 | 5210 | mc.mm = NULL; |
2bd9bb20 | 5211 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5212 | |
5213 | mmput(mm); | |
4ffef5fe DN |
5214 | } |
5215 | ||
1f7dd3e5 | 5216 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5217 | { |
1f7dd3e5 | 5218 | struct cgroup_subsys_state *css; |
eed67d75 | 5219 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5220 | struct mem_cgroup *from; |
4530eddb | 5221 | struct task_struct *leader, *p; |
9f2115f9 | 5222 | struct mm_struct *mm; |
1dfab5ab | 5223 | unsigned long move_flags; |
9f2115f9 | 5224 | int ret = 0; |
7dc74be0 | 5225 | |
1f7dd3e5 TH |
5226 | /* charge immigration isn't supported on the default hierarchy */ |
5227 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5228 | return 0; |
5229 | ||
4530eddb TH |
5230 | /* |
5231 | * Multi-process migrations only happen on the default hierarchy | |
5232 | * where charge immigration is not used. Perform charge | |
5233 | * immigration if @tset contains a leader and whine if there are | |
5234 | * multiple. | |
5235 | */ | |
5236 | p = NULL; | |
1f7dd3e5 | 5237 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5238 | WARN_ON_ONCE(p); |
5239 | p = leader; | |
1f7dd3e5 | 5240 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5241 | } |
5242 | if (!p) | |
5243 | return 0; | |
5244 | ||
1f7dd3e5 TH |
5245 | /* |
5246 | * We are now commited to this value whatever it is. Changes in this | |
5247 | * tunable will only affect upcoming migrations, not the current one. | |
5248 | * So we need to save it, and keep it going. | |
5249 | */ | |
5250 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5251 | if (!move_flags) | |
5252 | return 0; | |
5253 | ||
9f2115f9 TH |
5254 | from = mem_cgroup_from_task(p); |
5255 | ||
5256 | VM_BUG_ON(from == memcg); | |
5257 | ||
5258 | mm = get_task_mm(p); | |
5259 | if (!mm) | |
5260 | return 0; | |
5261 | /* We move charges only when we move a owner of the mm */ | |
5262 | if (mm->owner == p) { | |
5263 | VM_BUG_ON(mc.from); | |
5264 | VM_BUG_ON(mc.to); | |
5265 | VM_BUG_ON(mc.precharge); | |
5266 | VM_BUG_ON(mc.moved_charge); | |
5267 | VM_BUG_ON(mc.moved_swap); | |
5268 | ||
5269 | spin_lock(&mc.lock); | |
264a0ae1 | 5270 | mc.mm = mm; |
9f2115f9 TH |
5271 | mc.from = from; |
5272 | mc.to = memcg; | |
5273 | mc.flags = move_flags; | |
5274 | spin_unlock(&mc.lock); | |
5275 | /* We set mc.moving_task later */ | |
5276 | ||
5277 | ret = mem_cgroup_precharge_mc(mm); | |
5278 | if (ret) | |
5279 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5280 | } else { |
5281 | mmput(mm); | |
7dc74be0 DN |
5282 | } |
5283 | return ret; | |
5284 | } | |
5285 | ||
1f7dd3e5 | 5286 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5287 | { |
4e2f245d JW |
5288 | if (mc.to) |
5289 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5290 | } |
5291 | ||
4ffef5fe DN |
5292 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5293 | unsigned long addr, unsigned long end, | |
5294 | struct mm_walk *walk) | |
7dc74be0 | 5295 | { |
4ffef5fe | 5296 | int ret = 0; |
26bcd64a | 5297 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5298 | pte_t *pte; |
5299 | spinlock_t *ptl; | |
12724850 NH |
5300 | enum mc_target_type target_type; |
5301 | union mc_target target; | |
5302 | struct page *page; | |
4ffef5fe | 5303 | |
b6ec57f4 KS |
5304 | ptl = pmd_trans_huge_lock(pmd, vma); |
5305 | if (ptl) { | |
62ade86a | 5306 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5307 | spin_unlock(ptl); |
12724850 NH |
5308 | return 0; |
5309 | } | |
5310 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5311 | if (target_type == MC_TARGET_PAGE) { | |
5312 | page = target.page; | |
5313 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5314 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5315 | mc.from, mc.to)) { |
12724850 NH |
5316 | mc.precharge -= HPAGE_PMD_NR; |
5317 | mc.moved_charge += HPAGE_PMD_NR; | |
5318 | } | |
5319 | putback_lru_page(page); | |
5320 | } | |
5321 | put_page(page); | |
c733a828 JG |
5322 | } else if (target_type == MC_TARGET_DEVICE) { |
5323 | page = target.page; | |
5324 | if (!mem_cgroup_move_account(page, true, | |
5325 | mc.from, mc.to)) { | |
5326 | mc.precharge -= HPAGE_PMD_NR; | |
5327 | mc.moved_charge += HPAGE_PMD_NR; | |
5328 | } | |
5329 | put_page(page); | |
12724850 | 5330 | } |
bf929152 | 5331 | spin_unlock(ptl); |
1a5a9906 | 5332 | return 0; |
12724850 NH |
5333 | } |
5334 | ||
45f83cef AA |
5335 | if (pmd_trans_unstable(pmd)) |
5336 | return 0; | |
4ffef5fe DN |
5337 | retry: |
5338 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5339 | for (; addr != end; addr += PAGE_SIZE) { | |
5340 | pte_t ptent = *(pte++); | |
c733a828 | 5341 | bool device = false; |
02491447 | 5342 | swp_entry_t ent; |
4ffef5fe DN |
5343 | |
5344 | if (!mc.precharge) | |
5345 | break; | |
5346 | ||
8d32ff84 | 5347 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
5348 | case MC_TARGET_DEVICE: |
5349 | device = true; | |
5350 | /* fall through */ | |
4ffef5fe DN |
5351 | case MC_TARGET_PAGE: |
5352 | page = target.page; | |
53f9263b KS |
5353 | /* |
5354 | * We can have a part of the split pmd here. Moving it | |
5355 | * can be done but it would be too convoluted so simply | |
5356 | * ignore such a partial THP and keep it in original | |
5357 | * memcg. There should be somebody mapping the head. | |
5358 | */ | |
5359 | if (PageTransCompound(page)) | |
5360 | goto put; | |
c733a828 | 5361 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 5362 | goto put; |
f627c2f5 KS |
5363 | if (!mem_cgroup_move_account(page, false, |
5364 | mc.from, mc.to)) { | |
4ffef5fe | 5365 | mc.precharge--; |
854ffa8d DN |
5366 | /* we uncharge from mc.from later. */ |
5367 | mc.moved_charge++; | |
4ffef5fe | 5368 | } |
c733a828 JG |
5369 | if (!device) |
5370 | putback_lru_page(page); | |
8d32ff84 | 5371 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5372 | put_page(page); |
5373 | break; | |
02491447 DN |
5374 | case MC_TARGET_SWAP: |
5375 | ent = target.ent; | |
e91cbb42 | 5376 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5377 | mc.precharge--; |
483c30b5 DN |
5378 | /* we fixup refcnts and charges later. */ |
5379 | mc.moved_swap++; | |
5380 | } | |
02491447 | 5381 | break; |
4ffef5fe DN |
5382 | default: |
5383 | break; | |
5384 | } | |
5385 | } | |
5386 | pte_unmap_unlock(pte - 1, ptl); | |
5387 | cond_resched(); | |
5388 | ||
5389 | if (addr != end) { | |
5390 | /* | |
5391 | * We have consumed all precharges we got in can_attach(). | |
5392 | * We try charge one by one, but don't do any additional | |
5393 | * charges to mc.to if we have failed in charge once in attach() | |
5394 | * phase. | |
5395 | */ | |
854ffa8d | 5396 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5397 | if (!ret) |
5398 | goto retry; | |
5399 | } | |
5400 | ||
5401 | return ret; | |
5402 | } | |
5403 | ||
264a0ae1 | 5404 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 5405 | { |
26bcd64a NH |
5406 | struct mm_walk mem_cgroup_move_charge_walk = { |
5407 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
264a0ae1 | 5408 | .mm = mc.mm, |
26bcd64a | 5409 | }; |
4ffef5fe DN |
5410 | |
5411 | lru_add_drain_all(); | |
312722cb | 5412 | /* |
81f8c3a4 JW |
5413 | * Signal lock_page_memcg() to take the memcg's move_lock |
5414 | * while we're moving its pages to another memcg. Then wait | |
5415 | * for already started RCU-only updates to finish. | |
312722cb JW |
5416 | */ |
5417 | atomic_inc(&mc.from->moving_account); | |
5418 | synchronize_rcu(); | |
dfe076b0 | 5419 | retry: |
264a0ae1 | 5420 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
5421 | /* |
5422 | * Someone who are holding the mmap_sem might be waiting in | |
5423 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5424 | * and retry. Because we cancel precharges, we might not be able | |
5425 | * to move enough charges, but moving charge is a best-effort | |
5426 | * feature anyway, so it wouldn't be a big problem. | |
5427 | */ | |
5428 | __mem_cgroup_clear_mc(); | |
5429 | cond_resched(); | |
5430 | goto retry; | |
5431 | } | |
26bcd64a NH |
5432 | /* |
5433 | * When we have consumed all precharges and failed in doing | |
5434 | * additional charge, the page walk just aborts. | |
5435 | */ | |
0247f3f4 JM |
5436 | walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk); |
5437 | ||
264a0ae1 | 5438 | up_read(&mc.mm->mmap_sem); |
312722cb | 5439 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5440 | } |
5441 | ||
264a0ae1 | 5442 | static void mem_cgroup_move_task(void) |
67e465a7 | 5443 | { |
264a0ae1 TH |
5444 | if (mc.to) { |
5445 | mem_cgroup_move_charge(); | |
a433658c | 5446 | mem_cgroup_clear_mc(); |
264a0ae1 | 5447 | } |
67e465a7 | 5448 | } |
5cfb80a7 | 5449 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5450 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5451 | { |
5452 | return 0; | |
5453 | } | |
1f7dd3e5 | 5454 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5455 | { |
5456 | } | |
264a0ae1 | 5457 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
5458 | { |
5459 | } | |
5460 | #endif | |
67e465a7 | 5461 | |
f00baae7 TH |
5462 | /* |
5463 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5464 | * to verify whether we're attached to the default hierarchy on each mount |
5465 | * attempt. | |
f00baae7 | 5466 | */ |
eb95419b | 5467 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5468 | { |
5469 | /* | |
aa6ec29b | 5470 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5471 | * guarantees that @root doesn't have any children, so turning it |
5472 | * on for the root memcg is enough. | |
5473 | */ | |
9e10a130 | 5474 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5475 | root_mem_cgroup->use_hierarchy = true; |
5476 | else | |
5477 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5478 | } |
5479 | ||
677dc973 CD |
5480 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
5481 | { | |
5482 | if (value == PAGE_COUNTER_MAX) | |
5483 | seq_puts(m, "max\n"); | |
5484 | else | |
5485 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
5486 | ||
5487 | return 0; | |
5488 | } | |
5489 | ||
241994ed JW |
5490 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5491 | struct cftype *cft) | |
5492 | { | |
f5fc3c5d JW |
5493 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5494 | ||
5495 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5496 | } |
5497 | ||
bf8d5d52 RG |
5498 | static int memory_min_show(struct seq_file *m, void *v) |
5499 | { | |
677dc973 CD |
5500 | return seq_puts_memcg_tunable(m, |
5501 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
5502 | } |
5503 | ||
5504 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
5505 | char *buf, size_t nbytes, loff_t off) | |
5506 | { | |
5507 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5508 | unsigned long min; | |
5509 | int err; | |
5510 | ||
5511 | buf = strstrip(buf); | |
5512 | err = page_counter_memparse(buf, "max", &min); | |
5513 | if (err) | |
5514 | return err; | |
5515 | ||
5516 | page_counter_set_min(&memcg->memory, min); | |
5517 | ||
5518 | return nbytes; | |
5519 | } | |
5520 | ||
241994ed JW |
5521 | static int memory_low_show(struct seq_file *m, void *v) |
5522 | { | |
677dc973 CD |
5523 | return seq_puts_memcg_tunable(m, |
5524 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
5525 | } |
5526 | ||
5527 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5528 | char *buf, size_t nbytes, loff_t off) | |
5529 | { | |
5530 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5531 | unsigned long low; | |
5532 | int err; | |
5533 | ||
5534 | buf = strstrip(buf); | |
d2973697 | 5535 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5536 | if (err) |
5537 | return err; | |
5538 | ||
23067153 | 5539 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
5540 | |
5541 | return nbytes; | |
5542 | } | |
5543 | ||
5544 | static int memory_high_show(struct seq_file *m, void *v) | |
5545 | { | |
677dc973 | 5546 | return seq_puts_memcg_tunable(m, READ_ONCE(mem_cgroup_from_seq(m)->high)); |
241994ed JW |
5547 | } |
5548 | ||
5549 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5550 | char *buf, size_t nbytes, loff_t off) | |
5551 | { | |
5552 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 5553 | unsigned long nr_pages; |
241994ed JW |
5554 | unsigned long high; |
5555 | int err; | |
5556 | ||
5557 | buf = strstrip(buf); | |
d2973697 | 5558 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5559 | if (err) |
5560 | return err; | |
5561 | ||
5562 | memcg->high = high; | |
5563 | ||
588083bb JW |
5564 | nr_pages = page_counter_read(&memcg->memory); |
5565 | if (nr_pages > high) | |
5566 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5567 | GFP_KERNEL, true); | |
5568 | ||
2529bb3a | 5569 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5570 | return nbytes; |
5571 | } | |
5572 | ||
5573 | static int memory_max_show(struct seq_file *m, void *v) | |
5574 | { | |
677dc973 CD |
5575 | return seq_puts_memcg_tunable(m, |
5576 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
5577 | } |
5578 | ||
5579 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5580 | char *buf, size_t nbytes, loff_t off) | |
5581 | { | |
5582 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
5583 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5584 | bool drained = false; | |
241994ed JW |
5585 | unsigned long max; |
5586 | int err; | |
5587 | ||
5588 | buf = strstrip(buf); | |
d2973697 | 5589 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5590 | if (err) |
5591 | return err; | |
5592 | ||
bbec2e15 | 5593 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
5594 | |
5595 | for (;;) { | |
5596 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5597 | ||
5598 | if (nr_pages <= max) | |
5599 | break; | |
5600 | ||
5601 | if (signal_pending(current)) { | |
5602 | err = -EINTR; | |
5603 | break; | |
5604 | } | |
5605 | ||
5606 | if (!drained) { | |
5607 | drain_all_stock(memcg); | |
5608 | drained = true; | |
5609 | continue; | |
5610 | } | |
5611 | ||
5612 | if (nr_reclaims) { | |
5613 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5614 | GFP_KERNEL, true)) | |
5615 | nr_reclaims--; | |
5616 | continue; | |
5617 | } | |
5618 | ||
e27be240 | 5619 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
5620 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
5621 | break; | |
5622 | } | |
241994ed | 5623 | |
2529bb3a | 5624 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5625 | return nbytes; |
5626 | } | |
5627 | ||
5628 | static int memory_events_show(struct seq_file *m, void *v) | |
5629 | { | |
aa9694bb | 5630 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 5631 | |
e27be240 JW |
5632 | seq_printf(m, "low %lu\n", |
5633 | atomic_long_read(&memcg->memory_events[MEMCG_LOW])); | |
5634 | seq_printf(m, "high %lu\n", | |
5635 | atomic_long_read(&memcg->memory_events[MEMCG_HIGH])); | |
5636 | seq_printf(m, "max %lu\n", | |
5637 | atomic_long_read(&memcg->memory_events[MEMCG_MAX])); | |
5638 | seq_printf(m, "oom %lu\n", | |
5639 | atomic_long_read(&memcg->memory_events[MEMCG_OOM])); | |
fe6bdfc8 RG |
5640 | seq_printf(m, "oom_kill %lu\n", |
5641 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
241994ed JW |
5642 | |
5643 | return 0; | |
5644 | } | |
5645 | ||
587d9f72 JW |
5646 | static int memory_stat_show(struct seq_file *m, void *v) |
5647 | { | |
aa9694bb | 5648 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
587d9f72 JW |
5649 | int i; |
5650 | ||
5651 | /* | |
5652 | * Provide statistics on the state of the memory subsystem as | |
5653 | * well as cumulative event counters that show past behavior. | |
5654 | * | |
5655 | * This list is ordered following a combination of these gradients: | |
5656 | * 1) generic big picture -> specifics and details | |
5657 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
5658 | * | |
5659 | * Current memory state: | |
5660 | */ | |
5661 | ||
5662 | seq_printf(m, "anon %llu\n", | |
42a30035 | 5663 | (u64)memcg_page_state(memcg, MEMCG_RSS) * PAGE_SIZE); |
587d9f72 | 5664 | seq_printf(m, "file %llu\n", |
42a30035 | 5665 | (u64)memcg_page_state(memcg, MEMCG_CACHE) * PAGE_SIZE); |
12580e4b | 5666 | seq_printf(m, "kernel_stack %llu\n", |
42a30035 | 5667 | (u64)memcg_page_state(memcg, MEMCG_KERNEL_STACK_KB) * 1024); |
27ee57c9 | 5668 | seq_printf(m, "slab %llu\n", |
42a30035 JW |
5669 | (u64)(memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) + |
5670 | memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE)) * | |
5671 | PAGE_SIZE); | |
b2807f07 | 5672 | seq_printf(m, "sock %llu\n", |
42a30035 | 5673 | (u64)memcg_page_state(memcg, MEMCG_SOCK) * PAGE_SIZE); |
587d9f72 | 5674 | |
9a4caf1e | 5675 | seq_printf(m, "shmem %llu\n", |
42a30035 | 5676 | (u64)memcg_page_state(memcg, NR_SHMEM) * PAGE_SIZE); |
587d9f72 | 5677 | seq_printf(m, "file_mapped %llu\n", |
42a30035 | 5678 | (u64)memcg_page_state(memcg, NR_FILE_MAPPED) * PAGE_SIZE); |
587d9f72 | 5679 | seq_printf(m, "file_dirty %llu\n", |
42a30035 | 5680 | (u64)memcg_page_state(memcg, NR_FILE_DIRTY) * PAGE_SIZE); |
587d9f72 | 5681 | seq_printf(m, "file_writeback %llu\n", |
42a30035 | 5682 | (u64)memcg_page_state(memcg, NR_WRITEBACK) * PAGE_SIZE); |
587d9f72 | 5683 | |
1ff9e6e1 CD |
5684 | /* |
5685 | * TODO: We should eventually replace our own MEMCG_RSS_HUGE counter | |
5686 | * with the NR_ANON_THP vm counter, but right now it's a pain in the | |
5687 | * arse because it requires migrating the work out of rmap to a place | |
5688 | * where the page->mem_cgroup is set up and stable. | |
5689 | */ | |
5690 | seq_printf(m, "anon_thp %llu\n", | |
42a30035 | 5691 | (u64)memcg_page_state(memcg, MEMCG_RSS_HUGE) * PAGE_SIZE); |
1ff9e6e1 | 5692 | |
8de7ecc6 SB |
5693 | for (i = 0; i < NR_LRU_LISTS; i++) |
5694 | seq_printf(m, "%s %llu\n", mem_cgroup_lru_names[i], | |
42a30035 JW |
5695 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
5696 | PAGE_SIZE); | |
587d9f72 | 5697 | |
27ee57c9 | 5698 | seq_printf(m, "slab_reclaimable %llu\n", |
42a30035 JW |
5699 | (u64)memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) * |
5700 | PAGE_SIZE); | |
27ee57c9 | 5701 | seq_printf(m, "slab_unreclaimable %llu\n", |
42a30035 JW |
5702 | (u64)memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE) * |
5703 | PAGE_SIZE); | |
27ee57c9 | 5704 | |
587d9f72 JW |
5705 | /* Accumulated memory events */ |
5706 | ||
42a30035 JW |
5707 | seq_printf(m, "pgfault %lu\n", memcg_events(memcg, PGFAULT)); |
5708 | seq_printf(m, "pgmajfault %lu\n", memcg_events(memcg, PGMAJFAULT)); | |
587d9f72 | 5709 | |
e9b257ed | 5710 | seq_printf(m, "workingset_refault %lu\n", |
42a30035 | 5711 | memcg_page_state(memcg, WORKINGSET_REFAULT)); |
e9b257ed | 5712 | seq_printf(m, "workingset_activate %lu\n", |
42a30035 | 5713 | memcg_page_state(memcg, WORKINGSET_ACTIVATE)); |
e9b257ed | 5714 | seq_printf(m, "workingset_nodereclaim %lu\n", |
42a30035 JW |
5715 | memcg_page_state(memcg, WORKINGSET_NODERECLAIM)); |
5716 | ||
5717 | seq_printf(m, "pgrefill %lu\n", memcg_events(memcg, PGREFILL)); | |
5718 | seq_printf(m, "pgscan %lu\n", memcg_events(memcg, PGSCAN_KSWAPD) + | |
5719 | memcg_events(memcg, PGSCAN_DIRECT)); | |
5720 | seq_printf(m, "pgsteal %lu\n", memcg_events(memcg, PGSTEAL_KSWAPD) + | |
5721 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
5722 | seq_printf(m, "pgactivate %lu\n", memcg_events(memcg, PGACTIVATE)); | |
5723 | seq_printf(m, "pgdeactivate %lu\n", memcg_events(memcg, PGDEACTIVATE)); | |
5724 | seq_printf(m, "pglazyfree %lu\n", memcg_events(memcg, PGLAZYFREE)); | |
5725 | seq_printf(m, "pglazyfreed %lu\n", memcg_events(memcg, PGLAZYFREED)); | |
2262185c | 5726 | |
1ff9e6e1 | 5727 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
42a30035 JW |
5728 | seq_printf(m, "thp_fault_alloc %lu\n", |
5729 | memcg_events(memcg, THP_FAULT_ALLOC)); | |
1ff9e6e1 | 5730 | seq_printf(m, "thp_collapse_alloc %lu\n", |
42a30035 | 5731 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1ff9e6e1 CD |
5732 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
5733 | ||
587d9f72 JW |
5734 | return 0; |
5735 | } | |
5736 | ||
3d8b38eb RG |
5737 | static int memory_oom_group_show(struct seq_file *m, void *v) |
5738 | { | |
aa9694bb | 5739 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
5740 | |
5741 | seq_printf(m, "%d\n", memcg->oom_group); | |
5742 | ||
5743 | return 0; | |
5744 | } | |
5745 | ||
5746 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
5747 | char *buf, size_t nbytes, loff_t off) | |
5748 | { | |
5749 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5750 | int ret, oom_group; | |
5751 | ||
5752 | buf = strstrip(buf); | |
5753 | if (!buf) | |
5754 | return -EINVAL; | |
5755 | ||
5756 | ret = kstrtoint(buf, 0, &oom_group); | |
5757 | if (ret) | |
5758 | return ret; | |
5759 | ||
5760 | if (oom_group != 0 && oom_group != 1) | |
5761 | return -EINVAL; | |
5762 | ||
5763 | memcg->oom_group = oom_group; | |
5764 | ||
5765 | return nbytes; | |
5766 | } | |
5767 | ||
241994ed JW |
5768 | static struct cftype memory_files[] = { |
5769 | { | |
5770 | .name = "current", | |
f5fc3c5d | 5771 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5772 | .read_u64 = memory_current_read, |
5773 | }, | |
bf8d5d52 RG |
5774 | { |
5775 | .name = "min", | |
5776 | .flags = CFTYPE_NOT_ON_ROOT, | |
5777 | .seq_show = memory_min_show, | |
5778 | .write = memory_min_write, | |
5779 | }, | |
241994ed JW |
5780 | { |
5781 | .name = "low", | |
5782 | .flags = CFTYPE_NOT_ON_ROOT, | |
5783 | .seq_show = memory_low_show, | |
5784 | .write = memory_low_write, | |
5785 | }, | |
5786 | { | |
5787 | .name = "high", | |
5788 | .flags = CFTYPE_NOT_ON_ROOT, | |
5789 | .seq_show = memory_high_show, | |
5790 | .write = memory_high_write, | |
5791 | }, | |
5792 | { | |
5793 | .name = "max", | |
5794 | .flags = CFTYPE_NOT_ON_ROOT, | |
5795 | .seq_show = memory_max_show, | |
5796 | .write = memory_max_write, | |
5797 | }, | |
5798 | { | |
5799 | .name = "events", | |
5800 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5801 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5802 | .seq_show = memory_events_show, |
5803 | }, | |
587d9f72 JW |
5804 | { |
5805 | .name = "stat", | |
5806 | .flags = CFTYPE_NOT_ON_ROOT, | |
5807 | .seq_show = memory_stat_show, | |
5808 | }, | |
3d8b38eb RG |
5809 | { |
5810 | .name = "oom.group", | |
5811 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
5812 | .seq_show = memory_oom_group_show, | |
5813 | .write = memory_oom_group_write, | |
5814 | }, | |
241994ed JW |
5815 | { } /* terminate */ |
5816 | }; | |
5817 | ||
073219e9 | 5818 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5819 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5820 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5821 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5822 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5823 | .css_free = mem_cgroup_css_free, |
1ced953b | 5824 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5825 | .can_attach = mem_cgroup_can_attach, |
5826 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 5827 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5828 | .bind = mem_cgroup_bind, |
241994ed JW |
5829 | .dfl_cftypes = memory_files, |
5830 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5831 | .early_init = 0, |
8cdea7c0 | 5832 | }; |
c077719b | 5833 | |
241994ed | 5834 | /** |
bf8d5d52 | 5835 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 5836 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
5837 | * @memcg: the memory cgroup to check |
5838 | * | |
23067153 RG |
5839 | * WARNING: This function is not stateless! It can only be used as part |
5840 | * of a top-down tree iteration, not for isolated queries. | |
34c81057 | 5841 | * |
bf8d5d52 RG |
5842 | * Returns one of the following: |
5843 | * MEMCG_PROT_NONE: cgroup memory is not protected | |
5844 | * MEMCG_PROT_LOW: cgroup memory is protected as long there is | |
5845 | * an unprotected supply of reclaimable memory from other cgroups. | |
5846 | * MEMCG_PROT_MIN: cgroup memory is protected | |
34c81057 | 5847 | * |
bf8d5d52 | 5848 | * @root is exclusive; it is never protected when looked at directly |
34c81057 | 5849 | * |
bf8d5d52 RG |
5850 | * To provide a proper hierarchical behavior, effective memory.min/low values |
5851 | * are used. Below is the description of how effective memory.low is calculated. | |
5852 | * Effective memory.min values is calculated in the same way. | |
34c81057 | 5853 | * |
23067153 RG |
5854 | * Effective memory.low is always equal or less than the original memory.low. |
5855 | * If there is no memory.low overcommittment (which is always true for | |
5856 | * top-level memory cgroups), these two values are equal. | |
5857 | * Otherwise, it's a part of parent's effective memory.low, | |
5858 | * calculated as a cgroup's memory.low usage divided by sum of sibling's | |
5859 | * memory.low usages, where memory.low usage is the size of actually | |
5860 | * protected memory. | |
34c81057 | 5861 | * |
23067153 RG |
5862 | * low_usage |
5863 | * elow = min( memory.low, parent->elow * ------------------ ), | |
5864 | * siblings_low_usage | |
34c81057 | 5865 | * |
23067153 RG |
5866 | * | memory.current, if memory.current < memory.low |
5867 | * low_usage = | | |
82ede7ee | 5868 | * | 0, otherwise. |
34c81057 | 5869 | * |
23067153 RG |
5870 | * |
5871 | * Such definition of the effective memory.low provides the expected | |
5872 | * hierarchical behavior: parent's memory.low value is limiting | |
5873 | * children, unprotected memory is reclaimed first and cgroups, | |
5874 | * which are not using their guarantee do not affect actual memory | |
5875 | * distribution. | |
5876 | * | |
5877 | * For example, if there are memcgs A, A/B, A/C, A/D and A/E: | |
5878 | * | |
5879 | * A A/memory.low = 2G, A/memory.current = 6G | |
5880 | * //\\ | |
5881 | * BC DE B/memory.low = 3G B/memory.current = 2G | |
5882 | * C/memory.low = 1G C/memory.current = 2G | |
5883 | * D/memory.low = 0 D/memory.current = 2G | |
5884 | * E/memory.low = 10G E/memory.current = 0 | |
5885 | * | |
5886 | * and the memory pressure is applied, the following memory distribution | |
5887 | * is expected (approximately): | |
5888 | * | |
5889 | * A/memory.current = 2G | |
5890 | * | |
5891 | * B/memory.current = 1.3G | |
5892 | * C/memory.current = 0.6G | |
5893 | * D/memory.current = 0 | |
5894 | * E/memory.current = 0 | |
5895 | * | |
5896 | * These calculations require constant tracking of the actual low usages | |
bf8d5d52 RG |
5897 | * (see propagate_protected_usage()), as well as recursive calculation of |
5898 | * effective memory.low values. But as we do call mem_cgroup_protected() | |
23067153 RG |
5899 | * path for each memory cgroup top-down from the reclaim, |
5900 | * it's possible to optimize this part, and save calculated elow | |
5901 | * for next usage. This part is intentionally racy, but it's ok, | |
5902 | * as memory.low is a best-effort mechanism. | |
241994ed | 5903 | */ |
bf8d5d52 RG |
5904 | enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root, |
5905 | struct mem_cgroup *memcg) | |
241994ed | 5906 | { |
23067153 | 5907 | struct mem_cgroup *parent; |
bf8d5d52 RG |
5908 | unsigned long emin, parent_emin; |
5909 | unsigned long elow, parent_elow; | |
5910 | unsigned long usage; | |
23067153 | 5911 | |
241994ed | 5912 | if (mem_cgroup_disabled()) |
bf8d5d52 | 5913 | return MEMCG_PROT_NONE; |
241994ed | 5914 | |
34c81057 SC |
5915 | if (!root) |
5916 | root = root_mem_cgroup; | |
5917 | if (memcg == root) | |
bf8d5d52 | 5918 | return MEMCG_PROT_NONE; |
241994ed | 5919 | |
23067153 | 5920 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 RG |
5921 | if (!usage) |
5922 | return MEMCG_PROT_NONE; | |
5923 | ||
5924 | emin = memcg->memory.min; | |
5925 | elow = memcg->memory.low; | |
34c81057 | 5926 | |
bf8d5d52 | 5927 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
5928 | /* No parent means a non-hierarchical mode on v1 memcg */ |
5929 | if (!parent) | |
5930 | return MEMCG_PROT_NONE; | |
5931 | ||
23067153 RG |
5932 | if (parent == root) |
5933 | goto exit; | |
5934 | ||
bf8d5d52 RG |
5935 | parent_emin = READ_ONCE(parent->memory.emin); |
5936 | emin = min(emin, parent_emin); | |
5937 | if (emin && parent_emin) { | |
5938 | unsigned long min_usage, siblings_min_usage; | |
5939 | ||
5940 | min_usage = min(usage, memcg->memory.min); | |
5941 | siblings_min_usage = atomic_long_read( | |
5942 | &parent->memory.children_min_usage); | |
5943 | ||
5944 | if (min_usage && siblings_min_usage) | |
5945 | emin = min(emin, parent_emin * min_usage / | |
5946 | siblings_min_usage); | |
5947 | } | |
5948 | ||
23067153 RG |
5949 | parent_elow = READ_ONCE(parent->memory.elow); |
5950 | elow = min(elow, parent_elow); | |
bf8d5d52 RG |
5951 | if (elow && parent_elow) { |
5952 | unsigned long low_usage, siblings_low_usage; | |
23067153 | 5953 | |
bf8d5d52 RG |
5954 | low_usage = min(usage, memcg->memory.low); |
5955 | siblings_low_usage = atomic_long_read( | |
5956 | &parent->memory.children_low_usage); | |
23067153 | 5957 | |
bf8d5d52 RG |
5958 | if (low_usage && siblings_low_usage) |
5959 | elow = min(elow, parent_elow * low_usage / | |
5960 | siblings_low_usage); | |
5961 | } | |
23067153 | 5962 | |
23067153 | 5963 | exit: |
bf8d5d52 | 5964 | memcg->memory.emin = emin; |
23067153 | 5965 | memcg->memory.elow = elow; |
bf8d5d52 RG |
5966 | |
5967 | if (usage <= emin) | |
5968 | return MEMCG_PROT_MIN; | |
5969 | else if (usage <= elow) | |
5970 | return MEMCG_PROT_LOW; | |
5971 | else | |
5972 | return MEMCG_PROT_NONE; | |
241994ed JW |
5973 | } |
5974 | ||
00501b53 JW |
5975 | /** |
5976 | * mem_cgroup_try_charge - try charging a page | |
5977 | * @page: page to charge | |
5978 | * @mm: mm context of the victim | |
5979 | * @gfp_mask: reclaim mode | |
5980 | * @memcgp: charged memcg return | |
25843c2b | 5981 | * @compound: charge the page as compound or small page |
00501b53 JW |
5982 | * |
5983 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5984 | * pages according to @gfp_mask if necessary. | |
5985 | * | |
5986 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5987 | * Otherwise, an error code is returned. | |
5988 | * | |
5989 | * After page->mapping has been set up, the caller must finalize the | |
5990 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5991 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5992 | */ | |
5993 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
5994 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
5995 | bool compound) | |
00501b53 JW |
5996 | { |
5997 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 5998 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5999 | int ret = 0; |
6000 | ||
6001 | if (mem_cgroup_disabled()) | |
6002 | goto out; | |
6003 | ||
6004 | if (PageSwapCache(page)) { | |
00501b53 JW |
6005 | /* |
6006 | * Every swap fault against a single page tries to charge the | |
6007 | * page, bail as early as possible. shmem_unuse() encounters | |
6008 | * already charged pages, too. The USED bit is protected by | |
6009 | * the page lock, which serializes swap cache removal, which | |
6010 | * in turn serializes uncharging. | |
6011 | */ | |
e993d905 | 6012 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 6013 | if (compound_head(page)->mem_cgroup) |
00501b53 | 6014 | goto out; |
e993d905 | 6015 | |
37e84351 | 6016 | if (do_swap_account) { |
e993d905 VD |
6017 | swp_entry_t ent = { .val = page_private(page), }; |
6018 | unsigned short id = lookup_swap_cgroup_id(ent); | |
6019 | ||
6020 | rcu_read_lock(); | |
6021 | memcg = mem_cgroup_from_id(id); | |
6022 | if (memcg && !css_tryget_online(&memcg->css)) | |
6023 | memcg = NULL; | |
6024 | rcu_read_unlock(); | |
6025 | } | |
00501b53 JW |
6026 | } |
6027 | ||
00501b53 JW |
6028 | if (!memcg) |
6029 | memcg = get_mem_cgroup_from_mm(mm); | |
6030 | ||
6031 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
6032 | ||
6033 | css_put(&memcg->css); | |
00501b53 JW |
6034 | out: |
6035 | *memcgp = memcg; | |
6036 | return ret; | |
6037 | } | |
6038 | ||
2cf85583 TH |
6039 | int mem_cgroup_try_charge_delay(struct page *page, struct mm_struct *mm, |
6040 | gfp_t gfp_mask, struct mem_cgroup **memcgp, | |
6041 | bool compound) | |
6042 | { | |
6043 | struct mem_cgroup *memcg; | |
6044 | int ret; | |
6045 | ||
6046 | ret = mem_cgroup_try_charge(page, mm, gfp_mask, memcgp, compound); | |
6047 | memcg = *memcgp; | |
6048 | mem_cgroup_throttle_swaprate(memcg, page_to_nid(page), gfp_mask); | |
6049 | return ret; | |
6050 | } | |
6051 | ||
00501b53 JW |
6052 | /** |
6053 | * mem_cgroup_commit_charge - commit a page charge | |
6054 | * @page: page to charge | |
6055 | * @memcg: memcg to charge the page to | |
6056 | * @lrucare: page might be on LRU already | |
25843c2b | 6057 | * @compound: charge the page as compound or small page |
00501b53 JW |
6058 | * |
6059 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
6060 | * after page->mapping has been set up. This must happen atomically | |
6061 | * as part of the page instantiation, i.e. under the page table lock | |
6062 | * for anonymous pages, under the page lock for page and swap cache. | |
6063 | * | |
6064 | * In addition, the page must not be on the LRU during the commit, to | |
6065 | * prevent racing with task migration. If it might be, use @lrucare. | |
6066 | * | |
6067 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
6068 | */ | |
6069 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 6070 | bool lrucare, bool compound) |
00501b53 | 6071 | { |
f627c2f5 | 6072 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6073 | |
6074 | VM_BUG_ON_PAGE(!page->mapping, page); | |
6075 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
6076 | ||
6077 | if (mem_cgroup_disabled()) | |
6078 | return; | |
6079 | /* | |
6080 | * Swap faults will attempt to charge the same page multiple | |
6081 | * times. But reuse_swap_page() might have removed the page | |
6082 | * from swapcache already, so we can't check PageSwapCache(). | |
6083 | */ | |
6084 | if (!memcg) | |
6085 | return; | |
6086 | ||
6abb5a86 JW |
6087 | commit_charge(page, memcg, lrucare); |
6088 | ||
6abb5a86 | 6089 | local_irq_disable(); |
f627c2f5 | 6090 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
6091 | memcg_check_events(memcg, page); |
6092 | local_irq_enable(); | |
00501b53 | 6093 | |
7941d214 | 6094 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
6095 | swp_entry_t entry = { .val = page_private(page) }; |
6096 | /* | |
6097 | * The swap entry might not get freed for a long time, | |
6098 | * let's not wait for it. The page already received a | |
6099 | * memory+swap charge, drop the swap entry duplicate. | |
6100 | */ | |
38d8b4e6 | 6101 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 JW |
6102 | } |
6103 | } | |
6104 | ||
6105 | /** | |
6106 | * mem_cgroup_cancel_charge - cancel a page charge | |
6107 | * @page: page to charge | |
6108 | * @memcg: memcg to charge the page to | |
25843c2b | 6109 | * @compound: charge the page as compound or small page |
00501b53 JW |
6110 | * |
6111 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
6112 | */ | |
f627c2f5 KS |
6113 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
6114 | bool compound) | |
00501b53 | 6115 | { |
f627c2f5 | 6116 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6117 | |
6118 | if (mem_cgroup_disabled()) | |
6119 | return; | |
6120 | /* | |
6121 | * Swap faults will attempt to charge the same page multiple | |
6122 | * times. But reuse_swap_page() might have removed the page | |
6123 | * from swapcache already, so we can't check PageSwapCache(). | |
6124 | */ | |
6125 | if (!memcg) | |
6126 | return; | |
6127 | ||
00501b53 JW |
6128 | cancel_charge(memcg, nr_pages); |
6129 | } | |
6130 | ||
a9d5adee JG |
6131 | struct uncharge_gather { |
6132 | struct mem_cgroup *memcg; | |
6133 | unsigned long pgpgout; | |
6134 | unsigned long nr_anon; | |
6135 | unsigned long nr_file; | |
6136 | unsigned long nr_kmem; | |
6137 | unsigned long nr_huge; | |
6138 | unsigned long nr_shmem; | |
6139 | struct page *dummy_page; | |
6140 | }; | |
6141 | ||
6142 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6143 | { |
a9d5adee JG |
6144 | memset(ug, 0, sizeof(*ug)); |
6145 | } | |
6146 | ||
6147 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6148 | { | |
6149 | unsigned long nr_pages = ug->nr_anon + ug->nr_file + ug->nr_kmem; | |
747db954 JW |
6150 | unsigned long flags; |
6151 | ||
a9d5adee JG |
6152 | if (!mem_cgroup_is_root(ug->memcg)) { |
6153 | page_counter_uncharge(&ug->memcg->memory, nr_pages); | |
7941d214 | 6154 | if (do_memsw_account()) |
a9d5adee JG |
6155 | page_counter_uncharge(&ug->memcg->memsw, nr_pages); |
6156 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) | |
6157 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6158 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6159 | } |
747db954 JW |
6160 | |
6161 | local_irq_save(flags); | |
c9019e9b JW |
6162 | __mod_memcg_state(ug->memcg, MEMCG_RSS, -ug->nr_anon); |
6163 | __mod_memcg_state(ug->memcg, MEMCG_CACHE, -ug->nr_file); | |
6164 | __mod_memcg_state(ug->memcg, MEMCG_RSS_HUGE, -ug->nr_huge); | |
6165 | __mod_memcg_state(ug->memcg, NR_SHMEM, -ug->nr_shmem); | |
6166 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); | |
871789d4 | 6167 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, nr_pages); |
a9d5adee | 6168 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6169 | local_irq_restore(flags); |
e8ea14cc | 6170 | |
a9d5adee JG |
6171 | if (!mem_cgroup_is_root(ug->memcg)) |
6172 | css_put_many(&ug->memcg->css, nr_pages); | |
6173 | } | |
6174 | ||
6175 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6176 | { | |
6177 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
3f2eb028 JG |
6178 | VM_BUG_ON_PAGE(page_count(page) && !is_zone_device_page(page) && |
6179 | !PageHWPoison(page) , page); | |
a9d5adee JG |
6180 | |
6181 | if (!page->mem_cgroup) | |
6182 | return; | |
6183 | ||
6184 | /* | |
6185 | * Nobody should be changing or seriously looking at | |
6186 | * page->mem_cgroup at this point, we have fully | |
6187 | * exclusive access to the page. | |
6188 | */ | |
6189 | ||
6190 | if (ug->memcg != page->mem_cgroup) { | |
6191 | if (ug->memcg) { | |
6192 | uncharge_batch(ug); | |
6193 | uncharge_gather_clear(ug); | |
6194 | } | |
6195 | ug->memcg = page->mem_cgroup; | |
6196 | } | |
6197 | ||
6198 | if (!PageKmemcg(page)) { | |
6199 | unsigned int nr_pages = 1; | |
6200 | ||
6201 | if (PageTransHuge(page)) { | |
6202 | nr_pages <<= compound_order(page); | |
6203 | ug->nr_huge += nr_pages; | |
6204 | } | |
6205 | if (PageAnon(page)) | |
6206 | ug->nr_anon += nr_pages; | |
6207 | else { | |
6208 | ug->nr_file += nr_pages; | |
6209 | if (PageSwapBacked(page)) | |
6210 | ug->nr_shmem += nr_pages; | |
6211 | } | |
6212 | ug->pgpgout++; | |
6213 | } else { | |
6214 | ug->nr_kmem += 1 << compound_order(page); | |
6215 | __ClearPageKmemcg(page); | |
6216 | } | |
6217 | ||
6218 | ug->dummy_page = page; | |
6219 | page->mem_cgroup = NULL; | |
747db954 JW |
6220 | } |
6221 | ||
6222 | static void uncharge_list(struct list_head *page_list) | |
6223 | { | |
a9d5adee | 6224 | struct uncharge_gather ug; |
747db954 | 6225 | struct list_head *next; |
a9d5adee JG |
6226 | |
6227 | uncharge_gather_clear(&ug); | |
747db954 | 6228 | |
8b592656 JW |
6229 | /* |
6230 | * Note that the list can be a single page->lru; hence the | |
6231 | * do-while loop instead of a simple list_for_each_entry(). | |
6232 | */ | |
747db954 JW |
6233 | next = page_list->next; |
6234 | do { | |
a9d5adee JG |
6235 | struct page *page; |
6236 | ||
747db954 JW |
6237 | page = list_entry(next, struct page, lru); |
6238 | next = page->lru.next; | |
6239 | ||
a9d5adee | 6240 | uncharge_page(page, &ug); |
747db954 JW |
6241 | } while (next != page_list); |
6242 | ||
a9d5adee JG |
6243 | if (ug.memcg) |
6244 | uncharge_batch(&ug); | |
747db954 JW |
6245 | } |
6246 | ||
0a31bc97 JW |
6247 | /** |
6248 | * mem_cgroup_uncharge - uncharge a page | |
6249 | * @page: page to uncharge | |
6250 | * | |
6251 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
6252 | * mem_cgroup_commit_charge(). | |
6253 | */ | |
6254 | void mem_cgroup_uncharge(struct page *page) | |
6255 | { | |
a9d5adee JG |
6256 | struct uncharge_gather ug; |
6257 | ||
0a31bc97 JW |
6258 | if (mem_cgroup_disabled()) |
6259 | return; | |
6260 | ||
747db954 | 6261 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6262 | if (!page->mem_cgroup) |
0a31bc97 JW |
6263 | return; |
6264 | ||
a9d5adee JG |
6265 | uncharge_gather_clear(&ug); |
6266 | uncharge_page(page, &ug); | |
6267 | uncharge_batch(&ug); | |
747db954 | 6268 | } |
0a31bc97 | 6269 | |
747db954 JW |
6270 | /** |
6271 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6272 | * @page_list: list of pages to uncharge | |
6273 | * | |
6274 | * Uncharge a list of pages previously charged with | |
6275 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
6276 | */ | |
6277 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6278 | { | |
6279 | if (mem_cgroup_disabled()) | |
6280 | return; | |
0a31bc97 | 6281 | |
747db954 JW |
6282 | if (!list_empty(page_list)) |
6283 | uncharge_list(page_list); | |
0a31bc97 JW |
6284 | } |
6285 | ||
6286 | /** | |
6a93ca8f JW |
6287 | * mem_cgroup_migrate - charge a page's replacement |
6288 | * @oldpage: currently circulating page | |
6289 | * @newpage: replacement page | |
0a31bc97 | 6290 | * |
6a93ca8f JW |
6291 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6292 | * be uncharged upon free. | |
0a31bc97 JW |
6293 | * |
6294 | * Both pages must be locked, @newpage->mapping must be set up. | |
6295 | */ | |
6a93ca8f | 6296 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6297 | { |
29833315 | 6298 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
6299 | unsigned int nr_pages; |
6300 | bool compound; | |
d93c4130 | 6301 | unsigned long flags; |
0a31bc97 JW |
6302 | |
6303 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6304 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6305 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6306 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6307 | newpage); | |
0a31bc97 JW |
6308 | |
6309 | if (mem_cgroup_disabled()) | |
6310 | return; | |
6311 | ||
6312 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6313 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6314 | return; |
6315 | ||
45637bab | 6316 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6317 | memcg = oldpage->mem_cgroup; |
29833315 | 6318 | if (!memcg) |
0a31bc97 JW |
6319 | return; |
6320 | ||
44b7a8d3 JW |
6321 | /* Force-charge the new page. The old one will be freed soon */ |
6322 | compound = PageTransHuge(newpage); | |
6323 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
6324 | ||
6325 | page_counter_charge(&memcg->memory, nr_pages); | |
6326 | if (do_memsw_account()) | |
6327 | page_counter_charge(&memcg->memsw, nr_pages); | |
6328 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 6329 | |
9cf7666a | 6330 | commit_charge(newpage, memcg, false); |
44b7a8d3 | 6331 | |
d93c4130 | 6332 | local_irq_save(flags); |
44b7a8d3 JW |
6333 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
6334 | memcg_check_events(memcg, newpage); | |
d93c4130 | 6335 | local_irq_restore(flags); |
0a31bc97 JW |
6336 | } |
6337 | ||
ef12947c | 6338 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6339 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6340 | ||
2d758073 | 6341 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6342 | { |
6343 | struct mem_cgroup *memcg; | |
6344 | ||
2d758073 JW |
6345 | if (!mem_cgroup_sockets_enabled) |
6346 | return; | |
6347 | ||
edbe69ef RG |
6348 | /* |
6349 | * Socket cloning can throw us here with sk_memcg already | |
6350 | * filled. It won't however, necessarily happen from | |
6351 | * process context. So the test for root memcg given | |
6352 | * the current task's memcg won't help us in this case. | |
6353 | * | |
6354 | * Respecting the original socket's memcg is a better | |
6355 | * decision in this case. | |
6356 | */ | |
6357 | if (sk->sk_memcg) { | |
6358 | css_get(&sk->sk_memcg->css); | |
6359 | return; | |
6360 | } | |
6361 | ||
11092087 JW |
6362 | rcu_read_lock(); |
6363 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6364 | if (memcg == root_mem_cgroup) |
6365 | goto out; | |
0db15298 | 6366 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6367 | goto out; |
f7e1cb6e | 6368 | if (css_tryget_online(&memcg->css)) |
11092087 | 6369 | sk->sk_memcg = memcg; |
f7e1cb6e | 6370 | out: |
11092087 JW |
6371 | rcu_read_unlock(); |
6372 | } | |
11092087 | 6373 | |
2d758073 | 6374 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6375 | { |
2d758073 JW |
6376 | if (sk->sk_memcg) |
6377 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6378 | } |
6379 | ||
6380 | /** | |
6381 | * mem_cgroup_charge_skmem - charge socket memory | |
6382 | * @memcg: memcg to charge | |
6383 | * @nr_pages: number of pages to charge | |
6384 | * | |
6385 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6386 | * @memcg's configured limit, %false if the charge had to be forced. | |
6387 | */ | |
6388 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6389 | { | |
f7e1cb6e | 6390 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6391 | |
f7e1cb6e | 6392 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6393 | struct page_counter *fail; |
f7e1cb6e | 6394 | |
0db15298 JW |
6395 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6396 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6397 | return true; |
6398 | } | |
0db15298 JW |
6399 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6400 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6401 | return false; |
11092087 | 6402 | } |
d886f4e4 | 6403 | |
f7e1cb6e JW |
6404 | /* Don't block in the packet receive path */ |
6405 | if (in_softirq()) | |
6406 | gfp_mask = GFP_NOWAIT; | |
6407 | ||
c9019e9b | 6408 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6409 | |
f7e1cb6e JW |
6410 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6411 | return true; | |
6412 | ||
6413 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6414 | return false; |
6415 | } | |
6416 | ||
6417 | /** | |
6418 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
6419 | * @memcg: memcg to uncharge |
6420 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
6421 | */ |
6422 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6423 | { | |
f7e1cb6e | 6424 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6425 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
6426 | return; |
6427 | } | |
d886f4e4 | 6428 | |
c9019e9b | 6429 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 6430 | |
475d0487 | 6431 | refill_stock(memcg, nr_pages); |
11092087 JW |
6432 | } |
6433 | ||
f7e1cb6e JW |
6434 | static int __init cgroup_memory(char *s) |
6435 | { | |
6436 | char *token; | |
6437 | ||
6438 | while ((token = strsep(&s, ",")) != NULL) { | |
6439 | if (!*token) | |
6440 | continue; | |
6441 | if (!strcmp(token, "nosocket")) | |
6442 | cgroup_memory_nosocket = true; | |
04823c83 VD |
6443 | if (!strcmp(token, "nokmem")) |
6444 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
6445 | } |
6446 | return 0; | |
6447 | } | |
6448 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 6449 | |
2d11085e | 6450 | /* |
1081312f MH |
6451 | * subsys_initcall() for memory controller. |
6452 | * | |
308167fc SAS |
6453 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
6454 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
6455 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
6456 | * should be initialized from here. | |
2d11085e MH |
6457 | */ |
6458 | static int __init mem_cgroup_init(void) | |
6459 | { | |
95a045f6 JW |
6460 | int cpu, node; |
6461 | ||
84c07d11 | 6462 | #ifdef CONFIG_MEMCG_KMEM |
13583c3d VD |
6463 | /* |
6464 | * Kmem cache creation is mostly done with the slab_mutex held, | |
17cc4dfe TH |
6465 | * so use a workqueue with limited concurrency to avoid stalling |
6466 | * all worker threads in case lots of cgroups are created and | |
6467 | * destroyed simultaneously. | |
13583c3d | 6468 | */ |
17cc4dfe TH |
6469 | memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1); |
6470 | BUG_ON(!memcg_kmem_cache_wq); | |
13583c3d VD |
6471 | #endif |
6472 | ||
308167fc SAS |
6473 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
6474 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
6475 | |
6476 | for_each_possible_cpu(cpu) | |
6477 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
6478 | drain_local_stock); | |
6479 | ||
6480 | for_each_node(node) { | |
6481 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
6482 | |
6483 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
6484 | node_online(node) ? node : NUMA_NO_NODE); | |
6485 | ||
ef8f2327 | 6486 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 6487 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 6488 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
6489 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
6490 | } | |
6491 | ||
2d11085e MH |
6492 | return 0; |
6493 | } | |
6494 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
6495 | |
6496 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
6497 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
6498 | { | |
1c2d479a | 6499 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
6500 | /* |
6501 | * The root cgroup cannot be destroyed, so it's refcount must | |
6502 | * always be >= 1. | |
6503 | */ | |
6504 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
6505 | VM_BUG_ON(1); | |
6506 | break; | |
6507 | } | |
6508 | memcg = parent_mem_cgroup(memcg); | |
6509 | if (!memcg) | |
6510 | memcg = root_mem_cgroup; | |
6511 | } | |
6512 | return memcg; | |
6513 | } | |
6514 | ||
21afa38e JW |
6515 | /** |
6516 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
6517 | * @page: page whose memsw charge to transfer | |
6518 | * @entry: swap entry to move the charge to | |
6519 | * | |
6520 | * Transfer the memsw charge of @page to @entry. | |
6521 | */ | |
6522 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
6523 | { | |
1f47b61f | 6524 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 6525 | unsigned int nr_entries; |
21afa38e JW |
6526 | unsigned short oldid; |
6527 | ||
6528 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6529 | VM_BUG_ON_PAGE(page_count(page), page); | |
6530 | ||
7941d214 | 6531 | if (!do_memsw_account()) |
21afa38e JW |
6532 | return; |
6533 | ||
6534 | memcg = page->mem_cgroup; | |
6535 | ||
6536 | /* Readahead page, never charged */ | |
6537 | if (!memcg) | |
6538 | return; | |
6539 | ||
1f47b61f VD |
6540 | /* |
6541 | * In case the memcg owning these pages has been offlined and doesn't | |
6542 | * have an ID allocated to it anymore, charge the closest online | |
6543 | * ancestor for the swap instead and transfer the memory+swap charge. | |
6544 | */ | |
6545 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
d6810d73 HY |
6546 | nr_entries = hpage_nr_pages(page); |
6547 | /* Get references for the tail pages, too */ | |
6548 | if (nr_entries > 1) | |
6549 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
6550 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
6551 | nr_entries); | |
21afa38e | 6552 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6553 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
6554 | |
6555 | page->mem_cgroup = NULL; | |
6556 | ||
6557 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 6558 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 6559 | |
1f47b61f VD |
6560 | if (memcg != swap_memcg) { |
6561 | if (!mem_cgroup_is_root(swap_memcg)) | |
d6810d73 HY |
6562 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
6563 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
6564 | } |
6565 | ||
ce9ce665 SAS |
6566 | /* |
6567 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 6568 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 6569 | * important here to have the interrupts disabled because it is the |
b93b0163 | 6570 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
6571 | */ |
6572 | VM_BUG_ON(!irqs_disabled()); | |
d6810d73 HY |
6573 | mem_cgroup_charge_statistics(memcg, page, PageTransHuge(page), |
6574 | -nr_entries); | |
21afa38e | 6575 | memcg_check_events(memcg, page); |
73f576c0 JW |
6576 | |
6577 | if (!mem_cgroup_is_root(memcg)) | |
d08afa14 | 6578 | css_put_many(&memcg->css, nr_entries); |
21afa38e JW |
6579 | } |
6580 | ||
38d8b4e6 HY |
6581 | /** |
6582 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
6583 | * @page: page being added to swap |
6584 | * @entry: swap entry to charge | |
6585 | * | |
38d8b4e6 | 6586 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
6587 | * |
6588 | * Returns 0 on success, -ENOMEM on failure. | |
6589 | */ | |
6590 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
6591 | { | |
38d8b4e6 | 6592 | unsigned int nr_pages = hpage_nr_pages(page); |
37e84351 | 6593 | struct page_counter *counter; |
38d8b4e6 | 6594 | struct mem_cgroup *memcg; |
37e84351 VD |
6595 | unsigned short oldid; |
6596 | ||
6597 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
6598 | return 0; | |
6599 | ||
6600 | memcg = page->mem_cgroup; | |
6601 | ||
6602 | /* Readahead page, never charged */ | |
6603 | if (!memcg) | |
6604 | return 0; | |
6605 | ||
f3a53a3a TH |
6606 | if (!entry.val) { |
6607 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 6608 | return 0; |
f3a53a3a | 6609 | } |
bb98f2c5 | 6610 | |
1f47b61f VD |
6611 | memcg = mem_cgroup_id_get_online(memcg); |
6612 | ||
37e84351 | 6613 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 6614 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
6615 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
6616 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 6617 | mem_cgroup_id_put(memcg); |
37e84351 | 6618 | return -ENOMEM; |
1f47b61f | 6619 | } |
37e84351 | 6620 | |
38d8b4e6 HY |
6621 | /* Get references for the tail pages, too */ |
6622 | if (nr_pages > 1) | |
6623 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
6624 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 6625 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6626 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 6627 | |
37e84351 VD |
6628 | return 0; |
6629 | } | |
6630 | ||
21afa38e | 6631 | /** |
38d8b4e6 | 6632 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 6633 | * @entry: swap entry to uncharge |
38d8b4e6 | 6634 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 6635 | */ |
38d8b4e6 | 6636 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
6637 | { |
6638 | struct mem_cgroup *memcg; | |
6639 | unsigned short id; | |
6640 | ||
37e84351 | 6641 | if (!do_swap_account) |
21afa38e JW |
6642 | return; |
6643 | ||
38d8b4e6 | 6644 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 6645 | rcu_read_lock(); |
adbe427b | 6646 | memcg = mem_cgroup_from_id(id); |
21afa38e | 6647 | if (memcg) { |
37e84351 VD |
6648 | if (!mem_cgroup_is_root(memcg)) { |
6649 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
38d8b4e6 | 6650 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 6651 | else |
38d8b4e6 | 6652 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 6653 | } |
c9019e9b | 6654 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 6655 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
6656 | } |
6657 | rcu_read_unlock(); | |
6658 | } | |
6659 | ||
d8b38438 VD |
6660 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
6661 | { | |
6662 | long nr_swap_pages = get_nr_swap_pages(); | |
6663 | ||
6664 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6665 | return nr_swap_pages; | |
6666 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
6667 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 6668 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
6669 | page_counter_read(&memcg->swap)); |
6670 | return nr_swap_pages; | |
6671 | } | |
6672 | ||
5ccc5aba VD |
6673 | bool mem_cgroup_swap_full(struct page *page) |
6674 | { | |
6675 | struct mem_cgroup *memcg; | |
6676 | ||
6677 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
6678 | ||
6679 | if (vm_swap_full()) | |
6680 | return true; | |
6681 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6682 | return false; | |
6683 | ||
6684 | memcg = page->mem_cgroup; | |
6685 | if (!memcg) | |
6686 | return false; | |
6687 | ||
6688 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
bbec2e15 | 6689 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.max) |
5ccc5aba VD |
6690 | return true; |
6691 | ||
6692 | return false; | |
6693 | } | |
6694 | ||
21afa38e JW |
6695 | /* for remember boot option*/ |
6696 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
6697 | static int really_do_swap_account __initdata = 1; | |
6698 | #else | |
6699 | static int really_do_swap_account __initdata; | |
6700 | #endif | |
6701 | ||
6702 | static int __init enable_swap_account(char *s) | |
6703 | { | |
6704 | if (!strcmp(s, "1")) | |
6705 | really_do_swap_account = 1; | |
6706 | else if (!strcmp(s, "0")) | |
6707 | really_do_swap_account = 0; | |
6708 | return 1; | |
6709 | } | |
6710 | __setup("swapaccount=", enable_swap_account); | |
6711 | ||
37e84351 VD |
6712 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
6713 | struct cftype *cft) | |
6714 | { | |
6715 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6716 | ||
6717 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
6718 | } | |
6719 | ||
6720 | static int swap_max_show(struct seq_file *m, void *v) | |
6721 | { | |
677dc973 CD |
6722 | return seq_puts_memcg_tunable(m, |
6723 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
6724 | } |
6725 | ||
6726 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
6727 | char *buf, size_t nbytes, loff_t off) | |
6728 | { | |
6729 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6730 | unsigned long max; | |
6731 | int err; | |
6732 | ||
6733 | buf = strstrip(buf); | |
6734 | err = page_counter_memparse(buf, "max", &max); | |
6735 | if (err) | |
6736 | return err; | |
6737 | ||
be09102b | 6738 | xchg(&memcg->swap.max, max); |
37e84351 VD |
6739 | |
6740 | return nbytes; | |
6741 | } | |
6742 | ||
f3a53a3a TH |
6743 | static int swap_events_show(struct seq_file *m, void *v) |
6744 | { | |
aa9694bb | 6745 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a TH |
6746 | |
6747 | seq_printf(m, "max %lu\n", | |
6748 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
6749 | seq_printf(m, "fail %lu\n", | |
6750 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
6751 | ||
6752 | return 0; | |
6753 | } | |
6754 | ||
37e84351 VD |
6755 | static struct cftype swap_files[] = { |
6756 | { | |
6757 | .name = "swap.current", | |
6758 | .flags = CFTYPE_NOT_ON_ROOT, | |
6759 | .read_u64 = swap_current_read, | |
6760 | }, | |
6761 | { | |
6762 | .name = "swap.max", | |
6763 | .flags = CFTYPE_NOT_ON_ROOT, | |
6764 | .seq_show = swap_max_show, | |
6765 | .write = swap_max_write, | |
6766 | }, | |
f3a53a3a TH |
6767 | { |
6768 | .name = "swap.events", | |
6769 | .flags = CFTYPE_NOT_ON_ROOT, | |
6770 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
6771 | .seq_show = swap_events_show, | |
6772 | }, | |
37e84351 VD |
6773 | { } /* terminate */ |
6774 | }; | |
6775 | ||
21afa38e JW |
6776 | static struct cftype memsw_cgroup_files[] = { |
6777 | { | |
6778 | .name = "memsw.usage_in_bytes", | |
6779 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6780 | .read_u64 = mem_cgroup_read_u64, | |
6781 | }, | |
6782 | { | |
6783 | .name = "memsw.max_usage_in_bytes", | |
6784 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6785 | .write = mem_cgroup_reset, | |
6786 | .read_u64 = mem_cgroup_read_u64, | |
6787 | }, | |
6788 | { | |
6789 | .name = "memsw.limit_in_bytes", | |
6790 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6791 | .write = mem_cgroup_write, | |
6792 | .read_u64 = mem_cgroup_read_u64, | |
6793 | }, | |
6794 | { | |
6795 | .name = "memsw.failcnt", | |
6796 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6797 | .write = mem_cgroup_reset, | |
6798 | .read_u64 = mem_cgroup_read_u64, | |
6799 | }, | |
6800 | { }, /* terminate */ | |
6801 | }; | |
6802 | ||
6803 | static int __init mem_cgroup_swap_init(void) | |
6804 | { | |
6805 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6806 | do_swap_account = 1; | |
37e84351 VD |
6807 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
6808 | swap_files)); | |
21afa38e JW |
6809 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6810 | memsw_cgroup_files)); | |
6811 | } | |
6812 | return 0; | |
6813 | } | |
6814 | subsys_initcall(mem_cgroup_swap_init); | |
6815 | ||
6816 | #endif /* CONFIG_MEMCG_SWAP */ |