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> | |
a520110e | 28 | #include <linux/pagewalk.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> |
0e4b01df | 60 | #include <linux/psi.h> |
c8713d0b | 61 | #include <linux/seq_buf.h> |
08e552c6 | 62 | #include "internal.h" |
d1a4c0b3 | 63 | #include <net/sock.h> |
4bd2c1ee | 64 | #include <net/ip.h> |
f35c3a8e | 65 | #include "slab.h" |
8cdea7c0 | 66 | |
7c0f6ba6 | 67 | #include <linux/uaccess.h> |
8697d331 | 68 | |
cc8e970c KM |
69 | #include <trace/events/vmscan.h> |
70 | ||
073219e9 TH |
71 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
72 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 73 | |
7d828602 JW |
74 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
75 | ||
a181b0e8 | 76 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 77 | |
f7e1cb6e JW |
78 | /* Socket memory accounting disabled? */ |
79 | static bool cgroup_memory_nosocket; | |
80 | ||
04823c83 VD |
81 | /* Kernel memory accounting disabled? */ |
82 | static bool cgroup_memory_nokmem; | |
83 | ||
21afa38e | 84 | /* Whether the swap controller is active */ |
c255a458 | 85 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 86 | int do_swap_account __read_mostly; |
c077719b | 87 | #else |
a0db00fc | 88 | #define do_swap_account 0 |
c077719b KH |
89 | #endif |
90 | ||
97b27821 TH |
91 | #ifdef CONFIG_CGROUP_WRITEBACK |
92 | static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq); | |
93 | #endif | |
94 | ||
7941d214 JW |
95 | /* Whether legacy memory+swap accounting is active */ |
96 | static bool do_memsw_account(void) | |
97 | { | |
98 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; | |
99 | } | |
100 | ||
a0db00fc KS |
101 | #define THRESHOLDS_EVENTS_TARGET 128 |
102 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
e9f8974f | 103 | |
bb4cc1a8 AM |
104 | /* |
105 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
106 | * their hierarchy representation | |
107 | */ | |
108 | ||
ef8f2327 | 109 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 110 | struct rb_root rb_root; |
fa90b2fd | 111 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
112 | spinlock_t lock; |
113 | }; | |
114 | ||
bb4cc1a8 AM |
115 | struct mem_cgroup_tree { |
116 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
117 | }; | |
118 | ||
119 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
120 | ||
9490ff27 KH |
121 | /* for OOM */ |
122 | struct mem_cgroup_eventfd_list { | |
123 | struct list_head list; | |
124 | struct eventfd_ctx *eventfd; | |
125 | }; | |
2e72b634 | 126 | |
79bd9814 TH |
127 | /* |
128 | * cgroup_event represents events which userspace want to receive. | |
129 | */ | |
3bc942f3 | 130 | struct mem_cgroup_event { |
79bd9814 | 131 | /* |
59b6f873 | 132 | * memcg which the event belongs to. |
79bd9814 | 133 | */ |
59b6f873 | 134 | struct mem_cgroup *memcg; |
79bd9814 TH |
135 | /* |
136 | * eventfd to signal userspace about the event. | |
137 | */ | |
138 | struct eventfd_ctx *eventfd; | |
139 | /* | |
140 | * Each of these stored in a list by the cgroup. | |
141 | */ | |
142 | struct list_head list; | |
fba94807 TH |
143 | /* |
144 | * register_event() callback will be used to add new userspace | |
145 | * waiter for changes related to this event. Use eventfd_signal() | |
146 | * on eventfd to send notification to userspace. | |
147 | */ | |
59b6f873 | 148 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 149 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
150 | /* |
151 | * unregister_event() callback will be called when userspace closes | |
152 | * the eventfd or on cgroup removing. This callback must be set, | |
153 | * if you want provide notification functionality. | |
154 | */ | |
59b6f873 | 155 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 156 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
157 | /* |
158 | * All fields below needed to unregister event when | |
159 | * userspace closes eventfd. | |
160 | */ | |
161 | poll_table pt; | |
162 | wait_queue_head_t *wqh; | |
ac6424b9 | 163 | wait_queue_entry_t wait; |
79bd9814 TH |
164 | struct work_struct remove; |
165 | }; | |
166 | ||
c0ff4b85 R |
167 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
168 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 169 | |
7dc74be0 DN |
170 | /* Stuffs for move charges at task migration. */ |
171 | /* | |
1dfab5ab | 172 | * Types of charges to be moved. |
7dc74be0 | 173 | */ |
1dfab5ab JW |
174 | #define MOVE_ANON 0x1U |
175 | #define MOVE_FILE 0x2U | |
176 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 177 | |
4ffef5fe DN |
178 | /* "mc" and its members are protected by cgroup_mutex */ |
179 | static struct move_charge_struct { | |
b1dd693e | 180 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 181 | struct mm_struct *mm; |
4ffef5fe DN |
182 | struct mem_cgroup *from; |
183 | struct mem_cgroup *to; | |
1dfab5ab | 184 | unsigned long flags; |
4ffef5fe | 185 | unsigned long precharge; |
854ffa8d | 186 | unsigned long moved_charge; |
483c30b5 | 187 | unsigned long moved_swap; |
8033b97c DN |
188 | struct task_struct *moving_task; /* a task moving charges */ |
189 | wait_queue_head_t waitq; /* a waitq for other context */ | |
190 | } mc = { | |
2bd9bb20 | 191 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
192 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
193 | }; | |
4ffef5fe | 194 | |
4e416953 BS |
195 | /* |
196 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
197 | * limit reclaim to prevent infinite loops, if they ever occur. | |
198 | */ | |
a0db00fc | 199 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 200 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 201 | |
217bc319 KH |
202 | enum charge_type { |
203 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 204 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 205 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 206 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
207 | NR_CHARGE_TYPE, |
208 | }; | |
209 | ||
8c7c6e34 | 210 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
211 | enum res_type { |
212 | _MEM, | |
213 | _MEMSWAP, | |
214 | _OOM_TYPE, | |
510fc4e1 | 215 | _KMEM, |
d55f90bf | 216 | _TCP, |
86ae53e1 GC |
217 | }; |
218 | ||
a0db00fc KS |
219 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
220 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 221 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
222 | /* Used for OOM nofiier */ |
223 | #define OOM_CONTROL (0) | |
8c7c6e34 | 224 | |
b05706f1 KT |
225 | /* |
226 | * Iteration constructs for visiting all cgroups (under a tree). If | |
227 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
228 | * be used for reference counting. | |
229 | */ | |
230 | #define for_each_mem_cgroup_tree(iter, root) \ | |
231 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
232 | iter != NULL; \ | |
233 | iter = mem_cgroup_iter(root, iter, NULL)) | |
234 | ||
235 | #define for_each_mem_cgroup(iter) \ | |
236 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
237 | iter != NULL; \ | |
238 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
239 | ||
7775face TH |
240 | static inline bool should_force_charge(void) |
241 | { | |
242 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
243 | (current->flags & PF_EXITING); | |
244 | } | |
245 | ||
70ddf637 AV |
246 | /* Some nice accessors for the vmpressure. */ |
247 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
248 | { | |
249 | if (!memcg) | |
250 | memcg = root_mem_cgroup; | |
251 | return &memcg->vmpressure; | |
252 | } | |
253 | ||
254 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
255 | { | |
256 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
257 | } | |
258 | ||
84c07d11 | 259 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 260 | /* |
f7ce3190 | 261 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
262 | * The main reason for not using cgroup id for this: |
263 | * this works better in sparse environments, where we have a lot of memcgs, | |
264 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
265 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
266 | * 200 entry array for that. | |
55007d84 | 267 | * |
dbcf73e2 VD |
268 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
269 | * will double each time we have to increase it. | |
55007d84 | 270 | */ |
dbcf73e2 VD |
271 | static DEFINE_IDA(memcg_cache_ida); |
272 | int memcg_nr_cache_ids; | |
749c5415 | 273 | |
05257a1a VD |
274 | /* Protects memcg_nr_cache_ids */ |
275 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
276 | ||
277 | void memcg_get_cache_ids(void) | |
278 | { | |
279 | down_read(&memcg_cache_ids_sem); | |
280 | } | |
281 | ||
282 | void memcg_put_cache_ids(void) | |
283 | { | |
284 | up_read(&memcg_cache_ids_sem); | |
285 | } | |
286 | ||
55007d84 GC |
287 | /* |
288 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
289 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
290 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
291 | * tunable, but that is strictly not necessary. | |
292 | * | |
b8627835 | 293 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
294 | * this constant directly from cgroup, but it is understandable that this is |
295 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 296 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
297 | * increase ours as well if it increases. |
298 | */ | |
299 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 300 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 301 | |
d7f25f8a GC |
302 | /* |
303 | * A lot of the calls to the cache allocation functions are expected to be | |
304 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
305 | * conditional to this static branch, we'll have to allow modules that does | |
306 | * kmem_cache_alloc and the such to see this symbol as well | |
307 | */ | |
ef12947c | 308 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 309 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 310 | |
17cc4dfe | 311 | struct workqueue_struct *memcg_kmem_cache_wq; |
0a432dcb | 312 | #endif |
17cc4dfe | 313 | |
0a4465d3 KT |
314 | static int memcg_shrinker_map_size; |
315 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); | |
316 | ||
317 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) | |
318 | { | |
319 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); | |
320 | } | |
321 | ||
322 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, | |
323 | int size, int old_size) | |
324 | { | |
325 | struct memcg_shrinker_map *new, *old; | |
326 | int nid; | |
327 | ||
328 | lockdep_assert_held(&memcg_shrinker_map_mutex); | |
329 | ||
330 | for_each_node(nid) { | |
331 | old = rcu_dereference_protected( | |
332 | mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); | |
333 | /* Not yet online memcg */ | |
334 | if (!old) | |
335 | return 0; | |
336 | ||
86daf94e | 337 | new = kvmalloc_node(sizeof(*new) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
338 | if (!new) |
339 | return -ENOMEM; | |
340 | ||
341 | /* Set all old bits, clear all new bits */ | |
342 | memset(new->map, (int)0xff, old_size); | |
343 | memset((void *)new->map + old_size, 0, size - old_size); | |
344 | ||
345 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); | |
346 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); | |
347 | } | |
348 | ||
349 | return 0; | |
350 | } | |
351 | ||
352 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) | |
353 | { | |
354 | struct mem_cgroup_per_node *pn; | |
355 | struct memcg_shrinker_map *map; | |
356 | int nid; | |
357 | ||
358 | if (mem_cgroup_is_root(memcg)) | |
359 | return; | |
360 | ||
361 | for_each_node(nid) { | |
362 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
363 | map = rcu_dereference_protected(pn->shrinker_map, true); | |
364 | if (map) | |
365 | kvfree(map); | |
366 | rcu_assign_pointer(pn->shrinker_map, NULL); | |
367 | } | |
368 | } | |
369 | ||
370 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
371 | { | |
372 | struct memcg_shrinker_map *map; | |
373 | int nid, size, ret = 0; | |
374 | ||
375 | if (mem_cgroup_is_root(memcg)) | |
376 | return 0; | |
377 | ||
378 | mutex_lock(&memcg_shrinker_map_mutex); | |
379 | size = memcg_shrinker_map_size; | |
380 | for_each_node(nid) { | |
86daf94e | 381 | map = kvzalloc_node(sizeof(*map) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
382 | if (!map) { |
383 | memcg_free_shrinker_maps(memcg); | |
384 | ret = -ENOMEM; | |
385 | break; | |
386 | } | |
387 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); | |
388 | } | |
389 | mutex_unlock(&memcg_shrinker_map_mutex); | |
390 | ||
391 | return ret; | |
392 | } | |
393 | ||
394 | int memcg_expand_shrinker_maps(int new_id) | |
395 | { | |
396 | int size, old_size, ret = 0; | |
397 | struct mem_cgroup *memcg; | |
398 | ||
399 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); | |
400 | old_size = memcg_shrinker_map_size; | |
401 | if (size <= old_size) | |
402 | return 0; | |
403 | ||
404 | mutex_lock(&memcg_shrinker_map_mutex); | |
405 | if (!root_mem_cgroup) | |
406 | goto unlock; | |
407 | ||
408 | for_each_mem_cgroup(memcg) { | |
409 | if (mem_cgroup_is_root(memcg)) | |
410 | continue; | |
411 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); | |
75866af6 VA |
412 | if (ret) { |
413 | mem_cgroup_iter_break(NULL, memcg); | |
0a4465d3 | 414 | goto unlock; |
75866af6 | 415 | } |
0a4465d3 KT |
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 | ||
ad7fa852 TH |
438 | /** |
439 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
440 | * @page: page of interest | |
441 | * | |
442 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
443 | * with @page is returned. The returned css remains associated with @page | |
444 | * until it is released. | |
445 | * | |
446 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
447 | * is returned. | |
ad7fa852 TH |
448 | */ |
449 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
450 | { | |
451 | struct mem_cgroup *memcg; | |
452 | ||
ad7fa852 TH |
453 | memcg = page->mem_cgroup; |
454 | ||
9e10a130 | 455 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
456 | memcg = root_mem_cgroup; |
457 | ||
ad7fa852 TH |
458 | return &memcg->css; |
459 | } | |
460 | ||
2fc04524 VD |
461 | /** |
462 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
463 | * @page: the page | |
464 | * | |
465 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
466 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
467 | * is safe to call this function without holding a reference to @page. | |
468 | * | |
469 | * Note, this function is inherently racy, because there is nothing to prevent | |
470 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
471 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
472 | * do not care (such as procfs interfaces). | |
473 | */ | |
474 | ino_t page_cgroup_ino(struct page *page) | |
475 | { | |
476 | struct mem_cgroup *memcg; | |
477 | unsigned long ino = 0; | |
478 | ||
479 | rcu_read_lock(); | |
221ec5c0 | 480 | if (PageSlab(page) && !PageTail(page)) |
4d96ba35 RG |
481 | memcg = memcg_from_slab_page(page); |
482 | else | |
483 | memcg = READ_ONCE(page->mem_cgroup); | |
2fc04524 VD |
484 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
485 | memcg = parent_mem_cgroup(memcg); | |
486 | if (memcg) | |
487 | ino = cgroup_ino(memcg->css.cgroup); | |
488 | rcu_read_unlock(); | |
489 | return ino; | |
490 | } | |
491 | ||
ef8f2327 MG |
492 | static struct mem_cgroup_per_node * |
493 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 494 | { |
97a6c37b | 495 | int nid = page_to_nid(page); |
f64c3f54 | 496 | |
ef8f2327 | 497 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
498 | } |
499 | ||
ef8f2327 MG |
500 | static struct mem_cgroup_tree_per_node * |
501 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 502 | { |
ef8f2327 | 503 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
504 | } |
505 | ||
ef8f2327 | 506 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
507 | soft_limit_tree_from_page(struct page *page) |
508 | { | |
509 | int nid = page_to_nid(page); | |
bb4cc1a8 | 510 | |
ef8f2327 | 511 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
512 | } |
513 | ||
ef8f2327 MG |
514 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
515 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 516 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
517 | { |
518 | struct rb_node **p = &mctz->rb_root.rb_node; | |
519 | struct rb_node *parent = NULL; | |
ef8f2327 | 520 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 521 | bool rightmost = true; |
bb4cc1a8 AM |
522 | |
523 | if (mz->on_tree) | |
524 | return; | |
525 | ||
526 | mz->usage_in_excess = new_usage_in_excess; | |
527 | if (!mz->usage_in_excess) | |
528 | return; | |
529 | while (*p) { | |
530 | parent = *p; | |
ef8f2327 | 531 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 532 | tree_node); |
fa90b2fd | 533 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 534 | p = &(*p)->rb_left; |
fa90b2fd DB |
535 | rightmost = false; |
536 | } | |
537 | ||
bb4cc1a8 AM |
538 | /* |
539 | * We can't avoid mem cgroups that are over their soft | |
540 | * limit by the same amount | |
541 | */ | |
542 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
543 | p = &(*p)->rb_right; | |
544 | } | |
fa90b2fd DB |
545 | |
546 | if (rightmost) | |
547 | mctz->rb_rightmost = &mz->tree_node; | |
548 | ||
bb4cc1a8 AM |
549 | rb_link_node(&mz->tree_node, parent, p); |
550 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
551 | mz->on_tree = true; | |
552 | } | |
553 | ||
ef8f2327 MG |
554 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
555 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
556 | { |
557 | if (!mz->on_tree) | |
558 | return; | |
fa90b2fd DB |
559 | |
560 | if (&mz->tree_node == mctz->rb_rightmost) | |
561 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
562 | ||
bb4cc1a8 AM |
563 | rb_erase(&mz->tree_node, &mctz->rb_root); |
564 | mz->on_tree = false; | |
565 | } | |
566 | ||
ef8f2327 MG |
567 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
568 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 569 | { |
0a31bc97 JW |
570 | unsigned long flags; |
571 | ||
572 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 573 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 574 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
575 | } |
576 | ||
3e32cb2e JW |
577 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
578 | { | |
579 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 580 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
581 | unsigned long excess = 0; |
582 | ||
583 | if (nr_pages > soft_limit) | |
584 | excess = nr_pages - soft_limit; | |
585 | ||
586 | return excess; | |
587 | } | |
bb4cc1a8 AM |
588 | |
589 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
590 | { | |
3e32cb2e | 591 | unsigned long excess; |
ef8f2327 MG |
592 | struct mem_cgroup_per_node *mz; |
593 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 594 | |
e231875b | 595 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
596 | if (!mctz) |
597 | return; | |
bb4cc1a8 AM |
598 | /* |
599 | * Necessary to update all ancestors when hierarchy is used. | |
600 | * because their event counter is not touched. | |
601 | */ | |
602 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 603 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 604 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
605 | /* |
606 | * We have to update the tree if mz is on RB-tree or | |
607 | * mem is over its softlimit. | |
608 | */ | |
609 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
610 | unsigned long flags; |
611 | ||
612 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
613 | /* if on-tree, remove it */ |
614 | if (mz->on_tree) | |
cf2c8127 | 615 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
616 | /* |
617 | * Insert again. mz->usage_in_excess will be updated. | |
618 | * If excess is 0, no tree ops. | |
619 | */ | |
cf2c8127 | 620 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 621 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
622 | } |
623 | } | |
624 | } | |
625 | ||
626 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
627 | { | |
ef8f2327 MG |
628 | struct mem_cgroup_tree_per_node *mctz; |
629 | struct mem_cgroup_per_node *mz; | |
630 | int nid; | |
bb4cc1a8 | 631 | |
e231875b | 632 | for_each_node(nid) { |
ef8f2327 MG |
633 | mz = mem_cgroup_nodeinfo(memcg, nid); |
634 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
635 | if (mctz) |
636 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
637 | } |
638 | } | |
639 | ||
ef8f2327 MG |
640 | static struct mem_cgroup_per_node * |
641 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 642 | { |
ef8f2327 | 643 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
644 | |
645 | retry: | |
646 | mz = NULL; | |
fa90b2fd | 647 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
648 | goto done; /* Nothing to reclaim from */ |
649 | ||
fa90b2fd DB |
650 | mz = rb_entry(mctz->rb_rightmost, |
651 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
652 | /* |
653 | * Remove the node now but someone else can add it back, | |
654 | * we will to add it back at the end of reclaim to its correct | |
655 | * position in the tree. | |
656 | */ | |
cf2c8127 | 657 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 658 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 659 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
660 | goto retry; |
661 | done: | |
662 | return mz; | |
663 | } | |
664 | ||
ef8f2327 MG |
665 | static struct mem_cgroup_per_node * |
666 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 667 | { |
ef8f2327 | 668 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 669 | |
0a31bc97 | 670 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 671 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 672 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
673 | return mz; |
674 | } | |
675 | ||
db9adbcb JW |
676 | /** |
677 | * __mod_memcg_state - update cgroup memory statistics | |
678 | * @memcg: the memory cgroup | |
679 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
680 | * @val: delta to add to the counter, can be negative | |
681 | */ | |
682 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
683 | { | |
684 | long x; | |
685 | ||
686 | if (mem_cgroup_disabled()) | |
687 | return; | |
688 | ||
689 | x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); | |
690 | if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
691 | struct mem_cgroup *mi; |
692 | ||
766a4c19 YS |
693 | /* |
694 | * Batch local counters to keep them in sync with | |
695 | * the hierarchical ones. | |
696 | */ | |
697 | __this_cpu_add(memcg->vmstats_local->stat[idx], x); | |
42a30035 JW |
698 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
699 | atomic_long_add(x, &mi->vmstats[idx]); | |
db9adbcb JW |
700 | x = 0; |
701 | } | |
702 | __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); | |
703 | } | |
704 | ||
42a30035 JW |
705 | static struct mem_cgroup_per_node * |
706 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
707 | { | |
708 | struct mem_cgroup *parent; | |
709 | ||
710 | parent = parent_mem_cgroup(pn->memcg); | |
711 | if (!parent) | |
712 | return NULL; | |
713 | return mem_cgroup_nodeinfo(parent, nid); | |
714 | } | |
715 | ||
db9adbcb JW |
716 | /** |
717 | * __mod_lruvec_state - update lruvec memory statistics | |
718 | * @lruvec: the lruvec | |
719 | * @idx: the stat item | |
720 | * @val: delta to add to the counter, can be negative | |
721 | * | |
722 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
723 | * function updates the all three counters that are affected by a | |
724 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
725 | */ | |
726 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
727 | int val) | |
728 | { | |
42a30035 | 729 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
db9adbcb | 730 | struct mem_cgroup_per_node *pn; |
42a30035 | 731 | struct mem_cgroup *memcg; |
db9adbcb JW |
732 | long x; |
733 | ||
734 | /* Update node */ | |
42a30035 | 735 | __mod_node_page_state(pgdat, idx, val); |
db9adbcb JW |
736 | |
737 | if (mem_cgroup_disabled()) | |
738 | return; | |
739 | ||
740 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); | |
42a30035 | 741 | memcg = pn->memcg; |
db9adbcb JW |
742 | |
743 | /* Update memcg */ | |
42a30035 | 744 | __mod_memcg_state(memcg, idx, val); |
db9adbcb | 745 | |
b4c46484 RG |
746 | /* Update lruvec */ |
747 | __this_cpu_add(pn->lruvec_stat_local->count[idx], val); | |
748 | ||
db9adbcb JW |
749 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); |
750 | if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
751 | struct mem_cgroup_per_node *pi; |
752 | ||
42a30035 JW |
753 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) |
754 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
755 | x = 0; |
756 | } | |
757 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
758 | } | |
759 | ||
ec9f0238 RG |
760 | void __mod_lruvec_slab_state(void *p, enum node_stat_item idx, int val) |
761 | { | |
4f103c63 | 762 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
763 | struct mem_cgroup *memcg; |
764 | struct lruvec *lruvec; | |
765 | ||
766 | rcu_read_lock(); | |
4f103c63 | 767 | memcg = mem_cgroup_from_obj(p); |
ec9f0238 RG |
768 | |
769 | /* Untracked pages have no memcg, no lruvec. Update only the node */ | |
770 | if (!memcg || memcg == root_mem_cgroup) { | |
771 | __mod_node_page_state(pgdat, idx, val); | |
772 | } else { | |
867e5e1d | 773 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
ec9f0238 RG |
774 | __mod_lruvec_state(lruvec, idx, val); |
775 | } | |
776 | rcu_read_unlock(); | |
777 | } | |
778 | ||
8380ce47 RG |
779 | void mod_memcg_obj_state(void *p, int idx, int val) |
780 | { | |
781 | struct mem_cgroup *memcg; | |
782 | ||
783 | rcu_read_lock(); | |
784 | memcg = mem_cgroup_from_obj(p); | |
785 | if (memcg) | |
786 | mod_memcg_state(memcg, idx, val); | |
787 | rcu_read_unlock(); | |
788 | } | |
789 | ||
db9adbcb JW |
790 | /** |
791 | * __count_memcg_events - account VM events in a cgroup | |
792 | * @memcg: the memory cgroup | |
793 | * @idx: the event item | |
794 | * @count: the number of events that occured | |
795 | */ | |
796 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
797 | unsigned long count) | |
798 | { | |
799 | unsigned long x; | |
800 | ||
801 | if (mem_cgroup_disabled()) | |
802 | return; | |
803 | ||
804 | x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); | |
805 | if (unlikely(x > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
806 | struct mem_cgroup *mi; |
807 | ||
766a4c19 YS |
808 | /* |
809 | * Batch local counters to keep them in sync with | |
810 | * the hierarchical ones. | |
811 | */ | |
812 | __this_cpu_add(memcg->vmstats_local->events[idx], x); | |
42a30035 JW |
813 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
814 | atomic_long_add(x, &mi->vmevents[idx]); | |
db9adbcb JW |
815 | x = 0; |
816 | } | |
817 | __this_cpu_write(memcg->vmstats_percpu->events[idx], x); | |
818 | } | |
819 | ||
42a30035 | 820 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 821 | { |
871789d4 | 822 | return atomic_long_read(&memcg->vmevents[event]); |
e9f8974f JW |
823 | } |
824 | ||
42a30035 JW |
825 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
826 | { | |
815744d7 JW |
827 | long x = 0; |
828 | int cpu; | |
829 | ||
830 | for_each_possible_cpu(cpu) | |
831 | x += per_cpu(memcg->vmstats_local->events[event], cpu); | |
832 | return x; | |
42a30035 JW |
833 | } |
834 | ||
c0ff4b85 | 835 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 836 | struct page *page, |
f627c2f5 | 837 | bool compound, int nr_pages) |
d52aa412 | 838 | { |
b2402857 KH |
839 | /* |
840 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
841 | * counted as CACHE even if it's on ANON LRU. | |
842 | */ | |
0a31bc97 | 843 | if (PageAnon(page)) |
c9019e9b | 844 | __mod_memcg_state(memcg, MEMCG_RSS, nr_pages); |
9a4caf1e | 845 | else { |
c9019e9b | 846 | __mod_memcg_state(memcg, MEMCG_CACHE, nr_pages); |
9a4caf1e | 847 | if (PageSwapBacked(page)) |
c9019e9b | 848 | __mod_memcg_state(memcg, NR_SHMEM, nr_pages); |
9a4caf1e | 849 | } |
55e462b0 | 850 | |
f627c2f5 KS |
851 | if (compound) { |
852 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
c9019e9b | 853 | __mod_memcg_state(memcg, MEMCG_RSS_HUGE, nr_pages); |
f627c2f5 | 854 | } |
b070e65c | 855 | |
e401f176 KH |
856 | /* pagein of a big page is an event. So, ignore page size */ |
857 | if (nr_pages > 0) | |
c9019e9b | 858 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 859 | else { |
c9019e9b | 860 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
861 | nr_pages = -nr_pages; /* for event */ |
862 | } | |
e401f176 | 863 | |
871789d4 | 864 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
865 | } |
866 | ||
f53d7ce3 JW |
867 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
868 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
869 | { |
870 | unsigned long val, next; | |
871 | ||
871789d4 CD |
872 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
873 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 874 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 875 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
876 | switch (target) { |
877 | case MEM_CGROUP_TARGET_THRESH: | |
878 | next = val + THRESHOLDS_EVENTS_TARGET; | |
879 | break; | |
bb4cc1a8 AM |
880 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
881 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
882 | break; | |
f53d7ce3 JW |
883 | default: |
884 | break; | |
885 | } | |
871789d4 | 886 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 887 | return true; |
7a159cc9 | 888 | } |
f53d7ce3 | 889 | return false; |
d2265e6f KH |
890 | } |
891 | ||
892 | /* | |
893 | * Check events in order. | |
894 | * | |
895 | */ | |
c0ff4b85 | 896 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
897 | { |
898 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
899 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
900 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 901 | bool do_softlimit; |
f53d7ce3 | 902 | |
bb4cc1a8 AM |
903 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
904 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 905 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
906 | if (unlikely(do_softlimit)) |
907 | mem_cgroup_update_tree(memcg, page); | |
0a31bc97 | 908 | } |
d2265e6f KH |
909 | } |
910 | ||
cf475ad2 | 911 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 912 | { |
31a78f23 BS |
913 | /* |
914 | * mm_update_next_owner() may clear mm->owner to NULL | |
915 | * if it races with swapoff, page migration, etc. | |
916 | * So this can be called with p == NULL. | |
917 | */ | |
918 | if (unlikely(!p)) | |
919 | return NULL; | |
920 | ||
073219e9 | 921 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 922 | } |
33398cf2 | 923 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 924 | |
d46eb14b SB |
925 | /** |
926 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
927 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
928 | * | |
929 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
930 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
931 | * returned. | |
932 | */ | |
933 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 934 | { |
d46eb14b SB |
935 | struct mem_cgroup *memcg; |
936 | ||
937 | if (mem_cgroup_disabled()) | |
938 | return NULL; | |
0b7f569e | 939 | |
54595fe2 KH |
940 | rcu_read_lock(); |
941 | do { | |
6f6acb00 MH |
942 | /* |
943 | * Page cache insertions can happen withou an | |
944 | * actual mm context, e.g. during disk probing | |
945 | * on boot, loopback IO, acct() writes etc. | |
946 | */ | |
947 | if (unlikely(!mm)) | |
df381975 | 948 | memcg = root_mem_cgroup; |
6f6acb00 MH |
949 | else { |
950 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
951 | if (unlikely(!memcg)) | |
952 | memcg = root_mem_cgroup; | |
953 | } | |
00d484f3 | 954 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 955 | rcu_read_unlock(); |
c0ff4b85 | 956 | return memcg; |
54595fe2 | 957 | } |
d46eb14b SB |
958 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
959 | ||
f745c6f5 SB |
960 | /** |
961 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
962 | * @page: page from which memcg should be extracted. | |
963 | * | |
964 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
965 | * root_mem_cgroup is returned. | |
966 | */ | |
967 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
968 | { | |
969 | struct mem_cgroup *memcg = page->mem_cgroup; | |
970 | ||
971 | if (mem_cgroup_disabled()) | |
972 | return NULL; | |
973 | ||
974 | rcu_read_lock(); | |
8965aa28 SB |
975 | /* Page should not get uncharged and freed memcg under us. */ |
976 | if (!memcg || WARN_ON_ONCE(!css_tryget(&memcg->css))) | |
f745c6f5 SB |
977 | memcg = root_mem_cgroup; |
978 | rcu_read_unlock(); | |
979 | return memcg; | |
980 | } | |
981 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
982 | ||
d46eb14b SB |
983 | /** |
984 | * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg. | |
985 | */ | |
986 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
987 | { | |
988 | if (unlikely(current->active_memcg)) { | |
8965aa28 | 989 | struct mem_cgroup *memcg; |
d46eb14b SB |
990 | |
991 | rcu_read_lock(); | |
8965aa28 SB |
992 | /* current->active_memcg must hold a ref. */ |
993 | if (WARN_ON_ONCE(!css_tryget(¤t->active_memcg->css))) | |
994 | memcg = root_mem_cgroup; | |
995 | else | |
d46eb14b SB |
996 | memcg = current->active_memcg; |
997 | rcu_read_unlock(); | |
998 | return memcg; | |
999 | } | |
1000 | return get_mem_cgroup_from_mm(current->mm); | |
1001 | } | |
54595fe2 | 1002 | |
5660048c JW |
1003 | /** |
1004 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1005 | * @root: hierarchy root | |
1006 | * @prev: previously returned memcg, NULL on first invocation | |
1007 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1008 | * | |
1009 | * Returns references to children of the hierarchy below @root, or | |
1010 | * @root itself, or %NULL after a full round-trip. | |
1011 | * | |
1012 | * Caller must pass the return value in @prev on subsequent | |
1013 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1014 | * to cancel a hierarchy walk before the round-trip is complete. | |
1015 | * | |
b213b54f | 1016 | * Reclaimers can specify a node and a priority level in @reclaim to |
5660048c | 1017 | * divide up the memcgs in the hierarchy among all concurrent |
b213b54f | 1018 | * reclaimers operating on the same node and priority. |
5660048c | 1019 | */ |
694fbc0f | 1020 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1021 | struct mem_cgroup *prev, |
694fbc0f | 1022 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1023 | { |
33398cf2 | 1024 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 1025 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1026 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1027 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1028 | |
694fbc0f AM |
1029 | if (mem_cgroup_disabled()) |
1030 | return NULL; | |
5660048c | 1031 | |
9f3a0d09 JW |
1032 | if (!root) |
1033 | root = root_mem_cgroup; | |
7d74b06f | 1034 | |
9f3a0d09 | 1035 | if (prev && !reclaim) |
5ac8fb31 | 1036 | pos = prev; |
14067bb3 | 1037 | |
9f3a0d09 JW |
1038 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1039 | if (prev) | |
5ac8fb31 | 1040 | goto out; |
694fbc0f | 1041 | return root; |
9f3a0d09 | 1042 | } |
14067bb3 | 1043 | |
542f85f9 | 1044 | rcu_read_lock(); |
5f578161 | 1045 | |
5ac8fb31 | 1046 | if (reclaim) { |
ef8f2327 | 1047 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1048 | |
ef8f2327 | 1049 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
9da83f3f | 1050 | iter = &mz->iter; |
5ac8fb31 JW |
1051 | |
1052 | if (prev && reclaim->generation != iter->generation) | |
1053 | goto out_unlock; | |
1054 | ||
6df38689 | 1055 | while (1) { |
4db0c3c2 | 1056 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1057 | if (!pos || css_tryget(&pos->css)) |
1058 | break; | |
5ac8fb31 | 1059 | /* |
6df38689 VD |
1060 | * css reference reached zero, so iter->position will |
1061 | * be cleared by ->css_released. However, we should not | |
1062 | * rely on this happening soon, because ->css_released | |
1063 | * is called from a work queue, and by busy-waiting we | |
1064 | * might block it. So we clear iter->position right | |
1065 | * away. | |
5ac8fb31 | 1066 | */ |
6df38689 VD |
1067 | (void)cmpxchg(&iter->position, pos, NULL); |
1068 | } | |
5ac8fb31 JW |
1069 | } |
1070 | ||
1071 | if (pos) | |
1072 | css = &pos->css; | |
1073 | ||
1074 | for (;;) { | |
1075 | css = css_next_descendant_pre(css, &root->css); | |
1076 | if (!css) { | |
1077 | /* | |
1078 | * Reclaimers share the hierarchy walk, and a | |
1079 | * new one might jump in right at the end of | |
1080 | * the hierarchy - make sure they see at least | |
1081 | * one group and restart from the beginning. | |
1082 | */ | |
1083 | if (!prev) | |
1084 | continue; | |
1085 | break; | |
527a5ec9 | 1086 | } |
7d74b06f | 1087 | |
5ac8fb31 JW |
1088 | /* |
1089 | * Verify the css and acquire a reference. The root | |
1090 | * is provided by the caller, so we know it's alive | |
1091 | * and kicking, and don't take an extra reference. | |
1092 | */ | |
1093 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1094 | |
5ac8fb31 JW |
1095 | if (css == &root->css) |
1096 | break; | |
14067bb3 | 1097 | |
0b8f73e1 JW |
1098 | if (css_tryget(css)) |
1099 | break; | |
9f3a0d09 | 1100 | |
5ac8fb31 | 1101 | memcg = NULL; |
9f3a0d09 | 1102 | } |
5ac8fb31 JW |
1103 | |
1104 | if (reclaim) { | |
5ac8fb31 | 1105 | /* |
6df38689 VD |
1106 | * The position could have already been updated by a competing |
1107 | * thread, so check that the value hasn't changed since we read | |
1108 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1109 | */ |
6df38689 VD |
1110 | (void)cmpxchg(&iter->position, pos, memcg); |
1111 | ||
5ac8fb31 JW |
1112 | if (pos) |
1113 | css_put(&pos->css); | |
1114 | ||
1115 | if (!memcg) | |
1116 | iter->generation++; | |
1117 | else if (!prev) | |
1118 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1119 | } |
5ac8fb31 | 1120 | |
542f85f9 MH |
1121 | out_unlock: |
1122 | rcu_read_unlock(); | |
5ac8fb31 | 1123 | out: |
c40046f3 MH |
1124 | if (prev && prev != root) |
1125 | css_put(&prev->css); | |
1126 | ||
9f3a0d09 | 1127 | return memcg; |
14067bb3 | 1128 | } |
7d74b06f | 1129 | |
5660048c JW |
1130 | /** |
1131 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1132 | * @root: hierarchy root | |
1133 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1134 | */ | |
1135 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1136 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1137 | { |
1138 | if (!root) | |
1139 | root = root_mem_cgroup; | |
1140 | if (prev && prev != root) | |
1141 | css_put(&prev->css); | |
1142 | } | |
7d74b06f | 1143 | |
54a83d6b MC |
1144 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1145 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1146 | { |
6df38689 | 1147 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1148 | struct mem_cgroup_per_node *mz; |
1149 | int nid; | |
6df38689 | 1150 | |
54a83d6b MC |
1151 | for_each_node(nid) { |
1152 | mz = mem_cgroup_nodeinfo(from, nid); | |
9da83f3f YS |
1153 | iter = &mz->iter; |
1154 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1155 | } |
1156 | } | |
1157 | ||
54a83d6b MC |
1158 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1159 | { | |
1160 | struct mem_cgroup *memcg = dead_memcg; | |
1161 | struct mem_cgroup *last; | |
1162 | ||
1163 | do { | |
1164 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1165 | last = memcg; | |
1166 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1167 | ||
1168 | /* | |
1169 | * When cgruop1 non-hierarchy mode is used, | |
1170 | * parent_mem_cgroup() does not walk all the way up to the | |
1171 | * cgroup root (root_mem_cgroup). So we have to handle | |
1172 | * dead_memcg from cgroup root separately. | |
1173 | */ | |
1174 | if (last != root_mem_cgroup) | |
1175 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1176 | dead_memcg); | |
1177 | } | |
1178 | ||
7c5f64f8 VD |
1179 | /** |
1180 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1181 | * @memcg: hierarchy root | |
1182 | * @fn: function to call for each task | |
1183 | * @arg: argument passed to @fn | |
1184 | * | |
1185 | * This function iterates over tasks attached to @memcg or to any of its | |
1186 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1187 | * value, the function breaks the iteration loop and returns the value. | |
1188 | * Otherwise, it will iterate over all tasks and return 0. | |
1189 | * | |
1190 | * This function must not be called for the root memory cgroup. | |
1191 | */ | |
1192 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1193 | int (*fn)(struct task_struct *, void *), void *arg) | |
1194 | { | |
1195 | struct mem_cgroup *iter; | |
1196 | int ret = 0; | |
1197 | ||
1198 | BUG_ON(memcg == root_mem_cgroup); | |
1199 | ||
1200 | for_each_mem_cgroup_tree(iter, memcg) { | |
1201 | struct css_task_iter it; | |
1202 | struct task_struct *task; | |
1203 | ||
f168a9a5 | 1204 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1205 | while (!ret && (task = css_task_iter_next(&it))) |
1206 | ret = fn(task, arg); | |
1207 | css_task_iter_end(&it); | |
1208 | if (ret) { | |
1209 | mem_cgroup_iter_break(memcg, iter); | |
1210 | break; | |
1211 | } | |
1212 | } | |
1213 | return ret; | |
1214 | } | |
1215 | ||
925b7673 | 1216 | /** |
dfe0e773 | 1217 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1218 | * @page: the page |
f144c390 | 1219 | * @pgdat: pgdat of the page |
dfe0e773 JW |
1220 | * |
1221 | * This function is only safe when following the LRU page isolation | |
1222 | * and putback protocol: the LRU lock must be held, and the page must | |
1223 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1224 | */ |
599d0c95 | 1225 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1226 | { |
ef8f2327 | 1227 | struct mem_cgroup_per_node *mz; |
925b7673 | 1228 | struct mem_cgroup *memcg; |
bea8c150 | 1229 | struct lruvec *lruvec; |
6d12e2d8 | 1230 | |
bea8c150 | 1231 | if (mem_cgroup_disabled()) { |
867e5e1d | 1232 | lruvec = &pgdat->__lruvec; |
bea8c150 HD |
1233 | goto out; |
1234 | } | |
925b7673 | 1235 | |
1306a85a | 1236 | memcg = page->mem_cgroup; |
7512102c | 1237 | /* |
dfe0e773 | 1238 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1239 | * possibly migrated - before they are charged. |
7512102c | 1240 | */ |
29833315 JW |
1241 | if (!memcg) |
1242 | memcg = root_mem_cgroup; | |
7512102c | 1243 | |
ef8f2327 | 1244 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1245 | lruvec = &mz->lruvec; |
1246 | out: | |
1247 | /* | |
1248 | * Since a node can be onlined after the mem_cgroup was created, | |
1249 | * we have to be prepared to initialize lruvec->zone here; | |
1250 | * and if offlined then reonlined, we need to reinitialize it. | |
1251 | */ | |
599d0c95 MG |
1252 | if (unlikely(lruvec->pgdat != pgdat)) |
1253 | lruvec->pgdat = pgdat; | |
bea8c150 | 1254 | return lruvec; |
08e552c6 | 1255 | } |
b69408e8 | 1256 | |
925b7673 | 1257 | /** |
fa9add64 HD |
1258 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1259 | * @lruvec: mem_cgroup per zone lru vector | |
1260 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1261 | * @zid: zone id of the accounted pages |
fa9add64 | 1262 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1263 | * |
ca707239 HD |
1264 | * This function must be called under lru_lock, just before a page is added |
1265 | * to or just after a page is removed from an lru list (that ordering being | |
1266 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1267 | */ |
fa9add64 | 1268 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1269 | int zid, int nr_pages) |
3f58a829 | 1270 | { |
ef8f2327 | 1271 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1272 | unsigned long *lru_size; |
ca707239 | 1273 | long size; |
3f58a829 MK |
1274 | |
1275 | if (mem_cgroup_disabled()) | |
1276 | return; | |
1277 | ||
ef8f2327 | 1278 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1279 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1280 | |
1281 | if (nr_pages < 0) | |
1282 | *lru_size += nr_pages; | |
1283 | ||
1284 | size = *lru_size; | |
b4536f0c MH |
1285 | if (WARN_ONCE(size < 0, |
1286 | "%s(%p, %d, %d): lru_size %ld\n", | |
1287 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1288 | VM_BUG_ON(1); |
1289 | *lru_size = 0; | |
1290 | } | |
1291 | ||
1292 | if (nr_pages > 0) | |
1293 | *lru_size += nr_pages; | |
08e552c6 | 1294 | } |
544122e5 | 1295 | |
19942822 | 1296 | /** |
9d11ea9f | 1297 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1298 | * @memcg: the memory cgroup |
19942822 | 1299 | * |
9d11ea9f | 1300 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1301 | * pages. |
19942822 | 1302 | */ |
c0ff4b85 | 1303 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1304 | { |
3e32cb2e JW |
1305 | unsigned long margin = 0; |
1306 | unsigned long count; | |
1307 | unsigned long limit; | |
9d11ea9f | 1308 | |
3e32cb2e | 1309 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1310 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1311 | if (count < limit) |
1312 | margin = limit - count; | |
1313 | ||
7941d214 | 1314 | if (do_memsw_account()) { |
3e32cb2e | 1315 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1316 | limit = READ_ONCE(memcg->memsw.max); |
3e32cb2e JW |
1317 | if (count <= limit) |
1318 | margin = min(margin, limit - count); | |
cbedbac3 LR |
1319 | else |
1320 | margin = 0; | |
3e32cb2e JW |
1321 | } |
1322 | ||
1323 | return margin; | |
19942822 JW |
1324 | } |
1325 | ||
32047e2a | 1326 | /* |
bdcbb659 | 1327 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1328 | * |
bdcbb659 QH |
1329 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1330 | * moving cgroups. This is for waiting at high-memory pressure | |
1331 | * caused by "move". | |
32047e2a | 1332 | */ |
c0ff4b85 | 1333 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1334 | { |
2bd9bb20 KH |
1335 | struct mem_cgroup *from; |
1336 | struct mem_cgroup *to; | |
4b534334 | 1337 | bool ret = false; |
2bd9bb20 KH |
1338 | /* |
1339 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1340 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1341 | */ | |
1342 | spin_lock(&mc.lock); | |
1343 | from = mc.from; | |
1344 | to = mc.to; | |
1345 | if (!from) | |
1346 | goto unlock; | |
3e92041d | 1347 | |
2314b42d JW |
1348 | ret = mem_cgroup_is_descendant(from, memcg) || |
1349 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1350 | unlock: |
1351 | spin_unlock(&mc.lock); | |
4b534334 KH |
1352 | return ret; |
1353 | } | |
1354 | ||
c0ff4b85 | 1355 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1356 | { |
1357 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1358 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1359 | DEFINE_WAIT(wait); |
1360 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1361 | /* moving charge context might have finished. */ | |
1362 | if (mc.moving_task) | |
1363 | schedule(); | |
1364 | finish_wait(&mc.waitq, &wait); | |
1365 | return true; | |
1366 | } | |
1367 | } | |
1368 | return false; | |
1369 | } | |
1370 | ||
c8713d0b JW |
1371 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1372 | { | |
1373 | struct seq_buf s; | |
1374 | int i; | |
71cd3113 | 1375 | |
c8713d0b JW |
1376 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1377 | if (!s.buffer) | |
1378 | return NULL; | |
1379 | ||
1380 | /* | |
1381 | * Provide statistics on the state of the memory subsystem as | |
1382 | * well as cumulative event counters that show past behavior. | |
1383 | * | |
1384 | * This list is ordered following a combination of these gradients: | |
1385 | * 1) generic big picture -> specifics and details | |
1386 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1387 | * | |
1388 | * Current memory state: | |
1389 | */ | |
1390 | ||
1391 | seq_buf_printf(&s, "anon %llu\n", | |
1392 | (u64)memcg_page_state(memcg, MEMCG_RSS) * | |
1393 | PAGE_SIZE); | |
1394 | seq_buf_printf(&s, "file %llu\n", | |
1395 | (u64)memcg_page_state(memcg, MEMCG_CACHE) * | |
1396 | PAGE_SIZE); | |
1397 | seq_buf_printf(&s, "kernel_stack %llu\n", | |
1398 | (u64)memcg_page_state(memcg, MEMCG_KERNEL_STACK_KB) * | |
1399 | 1024); | |
1400 | seq_buf_printf(&s, "slab %llu\n", | |
1401 | (u64)(memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) + | |
1402 | memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE)) * | |
1403 | PAGE_SIZE); | |
1404 | seq_buf_printf(&s, "sock %llu\n", | |
1405 | (u64)memcg_page_state(memcg, MEMCG_SOCK) * | |
1406 | PAGE_SIZE); | |
1407 | ||
1408 | seq_buf_printf(&s, "shmem %llu\n", | |
1409 | (u64)memcg_page_state(memcg, NR_SHMEM) * | |
1410 | PAGE_SIZE); | |
1411 | seq_buf_printf(&s, "file_mapped %llu\n", | |
1412 | (u64)memcg_page_state(memcg, NR_FILE_MAPPED) * | |
1413 | PAGE_SIZE); | |
1414 | seq_buf_printf(&s, "file_dirty %llu\n", | |
1415 | (u64)memcg_page_state(memcg, NR_FILE_DIRTY) * | |
1416 | PAGE_SIZE); | |
1417 | seq_buf_printf(&s, "file_writeback %llu\n", | |
1418 | (u64)memcg_page_state(memcg, NR_WRITEBACK) * | |
1419 | PAGE_SIZE); | |
1420 | ||
1421 | /* | |
1422 | * TODO: We should eventually replace our own MEMCG_RSS_HUGE counter | |
1423 | * with the NR_ANON_THP vm counter, but right now it's a pain in the | |
1424 | * arse because it requires migrating the work out of rmap to a place | |
1425 | * where the page->mem_cgroup is set up and stable. | |
1426 | */ | |
1427 | seq_buf_printf(&s, "anon_thp %llu\n", | |
1428 | (u64)memcg_page_state(memcg, MEMCG_RSS_HUGE) * | |
1429 | PAGE_SIZE); | |
1430 | ||
1431 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 1432 | seq_buf_printf(&s, "%s %llu\n", lru_list_name(i), |
c8713d0b JW |
1433 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
1434 | PAGE_SIZE); | |
1435 | ||
1436 | seq_buf_printf(&s, "slab_reclaimable %llu\n", | |
1437 | (u64)memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) * | |
1438 | PAGE_SIZE); | |
1439 | seq_buf_printf(&s, "slab_unreclaimable %llu\n", | |
1440 | (u64)memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE) * | |
1441 | PAGE_SIZE); | |
1442 | ||
1443 | /* Accumulated memory events */ | |
1444 | ||
ebc5d83d KK |
1445 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT), |
1446 | memcg_events(memcg, PGFAULT)); | |
1447 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT), | |
1448 | memcg_events(memcg, PGMAJFAULT)); | |
c8713d0b JW |
1449 | |
1450 | seq_buf_printf(&s, "workingset_refault %lu\n", | |
1451 | memcg_page_state(memcg, WORKINGSET_REFAULT)); | |
1452 | seq_buf_printf(&s, "workingset_activate %lu\n", | |
1453 | memcg_page_state(memcg, WORKINGSET_ACTIVATE)); | |
1454 | seq_buf_printf(&s, "workingset_nodereclaim %lu\n", | |
1455 | memcg_page_state(memcg, WORKINGSET_NODERECLAIM)); | |
1456 | ||
ebc5d83d KK |
1457 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL), |
1458 | memcg_events(memcg, PGREFILL)); | |
c8713d0b JW |
1459 | seq_buf_printf(&s, "pgscan %lu\n", |
1460 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1461 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1462 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1463 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1464 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
ebc5d83d KK |
1465 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE), |
1466 | memcg_events(memcg, PGACTIVATE)); | |
1467 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE), | |
1468 | memcg_events(memcg, PGDEACTIVATE)); | |
1469 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE), | |
1470 | memcg_events(memcg, PGLAZYFREE)); | |
1471 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED), | |
1472 | memcg_events(memcg, PGLAZYFREED)); | |
c8713d0b JW |
1473 | |
1474 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
ebc5d83d | 1475 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC), |
c8713d0b | 1476 | memcg_events(memcg, THP_FAULT_ALLOC)); |
ebc5d83d | 1477 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC), |
c8713d0b JW |
1478 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1479 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1480 | ||
1481 | /* The above should easily fit into one page */ | |
1482 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1483 | ||
1484 | return s.buffer; | |
1485 | } | |
71cd3113 | 1486 | |
58cf188e | 1487 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1488 | /** |
f0c867d9 | 1489 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1490 | * memory controller. | |
e222432b BS |
1491 | * @memcg: The memory cgroup that went over limit |
1492 | * @p: Task that is going to be killed | |
1493 | * | |
1494 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1495 | * enabled | |
1496 | */ | |
f0c867d9 | 1497 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1498 | { |
e222432b BS |
1499 | rcu_read_lock(); |
1500 | ||
f0c867d9 | 1501 | if (memcg) { |
1502 | pr_cont(",oom_memcg="); | |
1503 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1504 | } else | |
1505 | pr_cont(",global_oom"); | |
2415b9f5 | 1506 | if (p) { |
f0c867d9 | 1507 | pr_cont(",task_memcg="); |
2415b9f5 | 1508 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1509 | } |
e222432b | 1510 | rcu_read_unlock(); |
f0c867d9 | 1511 | } |
1512 | ||
1513 | /** | |
1514 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1515 | * memory controller. | |
1516 | * @memcg: The memory cgroup that went over limit | |
1517 | */ | |
1518 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1519 | { | |
c8713d0b | 1520 | char *buf; |
e222432b | 1521 | |
3e32cb2e JW |
1522 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1523 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1524 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1525 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1526 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1527 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1528 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1529 | else { |
1530 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1531 | K((u64)page_counter_read(&memcg->memsw)), | |
1532 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1533 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1534 | K((u64)page_counter_read(&memcg->kmem)), | |
1535 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1536 | } |
c8713d0b JW |
1537 | |
1538 | pr_info("Memory cgroup stats for "); | |
1539 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1540 | pr_cont(":"); | |
1541 | buf = memory_stat_format(memcg); | |
1542 | if (!buf) | |
1543 | return; | |
1544 | pr_info("%s", buf); | |
1545 | kfree(buf); | |
e222432b BS |
1546 | } |
1547 | ||
a63d83f4 DR |
1548 | /* |
1549 | * Return the memory (and swap, if configured) limit for a memcg. | |
1550 | */ | |
bbec2e15 | 1551 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1552 | { |
bbec2e15 | 1553 | unsigned long max; |
f3e8eb70 | 1554 | |
15b42562 | 1555 | max = READ_ONCE(memcg->memory.max); |
9a5a8f19 | 1556 | if (mem_cgroup_swappiness(memcg)) { |
bbec2e15 RG |
1557 | unsigned long memsw_max; |
1558 | unsigned long swap_max; | |
9a5a8f19 | 1559 | |
bbec2e15 | 1560 | memsw_max = memcg->memsw.max; |
32d087cd | 1561 | swap_max = READ_ONCE(memcg->swap.max); |
bbec2e15 RG |
1562 | swap_max = min(swap_max, (unsigned long)total_swap_pages); |
1563 | max = min(max + swap_max, memsw_max); | |
9a5a8f19 | 1564 | } |
bbec2e15 | 1565 | return max; |
a63d83f4 DR |
1566 | } |
1567 | ||
9783aa99 CD |
1568 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1569 | { | |
1570 | return page_counter_read(&memcg->memory); | |
1571 | } | |
1572 | ||
b6e6edcf | 1573 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1574 | int order) |
9cbb78bb | 1575 | { |
6e0fc46d DR |
1576 | struct oom_control oc = { |
1577 | .zonelist = NULL, | |
1578 | .nodemask = NULL, | |
2a966b77 | 1579 | .memcg = memcg, |
6e0fc46d DR |
1580 | .gfp_mask = gfp_mask, |
1581 | .order = order, | |
6e0fc46d | 1582 | }; |
7c5f64f8 | 1583 | bool ret; |
9cbb78bb | 1584 | |
7775face TH |
1585 | if (mutex_lock_killable(&oom_lock)) |
1586 | return true; | |
1587 | /* | |
1588 | * A few threads which were not waiting at mutex_lock_killable() can | |
1589 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1590 | */ | |
1591 | ret = should_force_charge() || out_of_memory(&oc); | |
dc56401f | 1592 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1593 | return ret; |
9cbb78bb DR |
1594 | } |
1595 | ||
0608f43d | 1596 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1597 | pg_data_t *pgdat, |
0608f43d AM |
1598 | gfp_t gfp_mask, |
1599 | unsigned long *total_scanned) | |
1600 | { | |
1601 | struct mem_cgroup *victim = NULL; | |
1602 | int total = 0; | |
1603 | int loop = 0; | |
1604 | unsigned long excess; | |
1605 | unsigned long nr_scanned; | |
1606 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1607 | .pgdat = pgdat, |
0608f43d AM |
1608 | }; |
1609 | ||
3e32cb2e | 1610 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1611 | |
1612 | while (1) { | |
1613 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1614 | if (!victim) { | |
1615 | loop++; | |
1616 | if (loop >= 2) { | |
1617 | /* | |
1618 | * If we have not been able to reclaim | |
1619 | * anything, it might because there are | |
1620 | * no reclaimable pages under this hierarchy | |
1621 | */ | |
1622 | if (!total) | |
1623 | break; | |
1624 | /* | |
1625 | * We want to do more targeted reclaim. | |
1626 | * excess >> 2 is not to excessive so as to | |
1627 | * reclaim too much, nor too less that we keep | |
1628 | * coming back to reclaim from this cgroup | |
1629 | */ | |
1630 | if (total >= (excess >> 2) || | |
1631 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1632 | break; | |
1633 | } | |
1634 | continue; | |
1635 | } | |
a9dd0a83 | 1636 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1637 | pgdat, &nr_scanned); |
0608f43d | 1638 | *total_scanned += nr_scanned; |
3e32cb2e | 1639 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1640 | break; |
6d61ef40 | 1641 | } |
0608f43d AM |
1642 | mem_cgroup_iter_break(root_memcg, victim); |
1643 | return total; | |
6d61ef40 BS |
1644 | } |
1645 | ||
0056f4e6 JW |
1646 | #ifdef CONFIG_LOCKDEP |
1647 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1648 | .name = "memcg_oom_lock", | |
1649 | }; | |
1650 | #endif | |
1651 | ||
fb2a6fc5 JW |
1652 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1653 | ||
867578cb KH |
1654 | /* |
1655 | * Check OOM-Killer is already running under our hierarchy. | |
1656 | * If someone is running, return false. | |
1657 | */ | |
fb2a6fc5 | 1658 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1659 | { |
79dfdacc | 1660 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1661 | |
fb2a6fc5 JW |
1662 | spin_lock(&memcg_oom_lock); |
1663 | ||
9f3a0d09 | 1664 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1665 | if (iter->oom_lock) { |
79dfdacc MH |
1666 | /* |
1667 | * this subtree of our hierarchy is already locked | |
1668 | * so we cannot give a lock. | |
1669 | */ | |
79dfdacc | 1670 | failed = iter; |
9f3a0d09 JW |
1671 | mem_cgroup_iter_break(memcg, iter); |
1672 | break; | |
23751be0 JW |
1673 | } else |
1674 | iter->oom_lock = true; | |
7d74b06f | 1675 | } |
867578cb | 1676 | |
fb2a6fc5 JW |
1677 | if (failed) { |
1678 | /* | |
1679 | * OK, we failed to lock the whole subtree so we have | |
1680 | * to clean up what we set up to the failing subtree | |
1681 | */ | |
1682 | for_each_mem_cgroup_tree(iter, memcg) { | |
1683 | if (iter == failed) { | |
1684 | mem_cgroup_iter_break(memcg, iter); | |
1685 | break; | |
1686 | } | |
1687 | iter->oom_lock = false; | |
79dfdacc | 1688 | } |
0056f4e6 JW |
1689 | } else |
1690 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1691 | |
1692 | spin_unlock(&memcg_oom_lock); | |
1693 | ||
1694 | return !failed; | |
a636b327 | 1695 | } |
0b7f569e | 1696 | |
fb2a6fc5 | 1697 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1698 | { |
7d74b06f KH |
1699 | struct mem_cgroup *iter; |
1700 | ||
fb2a6fc5 | 1701 | spin_lock(&memcg_oom_lock); |
5facae4f | 1702 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1703 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1704 | iter->oom_lock = false; |
fb2a6fc5 | 1705 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1706 | } |
1707 | ||
c0ff4b85 | 1708 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1709 | { |
1710 | struct mem_cgroup *iter; | |
1711 | ||
c2b42d3c | 1712 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1713 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1714 | iter->under_oom++; |
1715 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1716 | } |
1717 | ||
c0ff4b85 | 1718 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1719 | { |
1720 | struct mem_cgroup *iter; | |
1721 | ||
867578cb KH |
1722 | /* |
1723 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1724 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1725 | */ |
c2b42d3c | 1726 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1727 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1728 | if (iter->under_oom > 0) |
1729 | iter->under_oom--; | |
1730 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1731 | } |
1732 | ||
867578cb KH |
1733 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1734 | ||
dc98df5a | 1735 | struct oom_wait_info { |
d79154bb | 1736 | struct mem_cgroup *memcg; |
ac6424b9 | 1737 | wait_queue_entry_t wait; |
dc98df5a KH |
1738 | }; |
1739 | ||
ac6424b9 | 1740 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1741 | unsigned mode, int sync, void *arg) |
1742 | { | |
d79154bb HD |
1743 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1744 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1745 | struct oom_wait_info *oom_wait_info; |
1746 | ||
1747 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1748 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1749 | |
2314b42d JW |
1750 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1751 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1752 | return 0; |
dc98df5a KH |
1753 | return autoremove_wake_function(wait, mode, sync, arg); |
1754 | } | |
1755 | ||
c0ff4b85 | 1756 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1757 | { |
c2b42d3c TH |
1758 | /* |
1759 | * For the following lockless ->under_oom test, the only required | |
1760 | * guarantee is that it must see the state asserted by an OOM when | |
1761 | * this function is called as a result of userland actions | |
1762 | * triggered by the notification of the OOM. This is trivially | |
1763 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1764 | * triggering notification. | |
1765 | */ | |
1766 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1767 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1768 | } |
1769 | ||
29ef680a MH |
1770 | enum oom_status { |
1771 | OOM_SUCCESS, | |
1772 | OOM_FAILED, | |
1773 | OOM_ASYNC, | |
1774 | OOM_SKIPPED | |
1775 | }; | |
1776 | ||
1777 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1778 | { |
7056d3a3 MH |
1779 | enum oom_status ret; |
1780 | bool locked; | |
1781 | ||
29ef680a MH |
1782 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1783 | return OOM_SKIPPED; | |
1784 | ||
7a1adfdd RG |
1785 | memcg_memory_event(memcg, MEMCG_OOM); |
1786 | ||
867578cb | 1787 | /* |
49426420 JW |
1788 | * We are in the middle of the charge context here, so we |
1789 | * don't want to block when potentially sitting on a callstack | |
1790 | * that holds all kinds of filesystem and mm locks. | |
1791 | * | |
29ef680a MH |
1792 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1793 | * handling until the charge can succeed; remember the context and put | |
1794 | * the task to sleep at the end of the page fault when all locks are | |
1795 | * released. | |
49426420 | 1796 | * |
29ef680a MH |
1797 | * On the other hand, in-kernel OOM killer allows for an async victim |
1798 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1799 | * relying on the oom victim to make a forward progress and we can | |
1800 | * invoke the oom killer here. | |
1801 | * | |
1802 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1803 | * victim and then we have to bail out from the charge path. | |
867578cb | 1804 | */ |
29ef680a MH |
1805 | if (memcg->oom_kill_disable) { |
1806 | if (!current->in_user_fault) | |
1807 | return OOM_SKIPPED; | |
1808 | css_get(&memcg->css); | |
1809 | current->memcg_in_oom = memcg; | |
1810 | current->memcg_oom_gfp_mask = mask; | |
1811 | current->memcg_oom_order = order; | |
1812 | ||
1813 | return OOM_ASYNC; | |
1814 | } | |
1815 | ||
7056d3a3 MH |
1816 | mem_cgroup_mark_under_oom(memcg); |
1817 | ||
1818 | locked = mem_cgroup_oom_trylock(memcg); | |
1819 | ||
1820 | if (locked) | |
1821 | mem_cgroup_oom_notify(memcg); | |
1822 | ||
1823 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1824 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1825 | ret = OOM_SUCCESS; |
1826 | else | |
1827 | ret = OOM_FAILED; | |
1828 | ||
1829 | if (locked) | |
1830 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1831 | |
7056d3a3 | 1832 | return ret; |
3812c8c8 JW |
1833 | } |
1834 | ||
1835 | /** | |
1836 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1837 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1838 | * |
49426420 JW |
1839 | * This has to be called at the end of a page fault if the memcg OOM |
1840 | * handler was enabled. | |
3812c8c8 | 1841 | * |
49426420 | 1842 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1843 | * sleep on a waitqueue until the userspace task resolves the |
1844 | * situation. Sleeping directly in the charge context with all kinds | |
1845 | * of locks held is not a good idea, instead we remember an OOM state | |
1846 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1847 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1848 | * |
1849 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1850 | * completed, %false otherwise. |
3812c8c8 | 1851 | */ |
49426420 | 1852 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1853 | { |
626ebc41 | 1854 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1855 | struct oom_wait_info owait; |
49426420 | 1856 | bool locked; |
3812c8c8 JW |
1857 | |
1858 | /* OOM is global, do not handle */ | |
3812c8c8 | 1859 | if (!memcg) |
49426420 | 1860 | return false; |
3812c8c8 | 1861 | |
7c5f64f8 | 1862 | if (!handle) |
49426420 | 1863 | goto cleanup; |
3812c8c8 JW |
1864 | |
1865 | owait.memcg = memcg; | |
1866 | owait.wait.flags = 0; | |
1867 | owait.wait.func = memcg_oom_wake_function; | |
1868 | owait.wait.private = current; | |
2055da97 | 1869 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1870 | |
3812c8c8 | 1871 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1872 | mem_cgroup_mark_under_oom(memcg); |
1873 | ||
1874 | locked = mem_cgroup_oom_trylock(memcg); | |
1875 | ||
1876 | if (locked) | |
1877 | mem_cgroup_oom_notify(memcg); | |
1878 | ||
1879 | if (locked && !memcg->oom_kill_disable) { | |
1880 | mem_cgroup_unmark_under_oom(memcg); | |
1881 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1882 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1883 | current->memcg_oom_order); | |
49426420 | 1884 | } else { |
3812c8c8 | 1885 | schedule(); |
49426420 JW |
1886 | mem_cgroup_unmark_under_oom(memcg); |
1887 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1888 | } | |
1889 | ||
1890 | if (locked) { | |
fb2a6fc5 JW |
1891 | mem_cgroup_oom_unlock(memcg); |
1892 | /* | |
1893 | * There is no guarantee that an OOM-lock contender | |
1894 | * sees the wakeups triggered by the OOM kill | |
1895 | * uncharges. Wake any sleepers explicitely. | |
1896 | */ | |
1897 | memcg_oom_recover(memcg); | |
1898 | } | |
49426420 | 1899 | cleanup: |
626ebc41 | 1900 | current->memcg_in_oom = NULL; |
3812c8c8 | 1901 | css_put(&memcg->css); |
867578cb | 1902 | return true; |
0b7f569e KH |
1903 | } |
1904 | ||
3d8b38eb RG |
1905 | /** |
1906 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
1907 | * @victim: task to be killed by the OOM killer | |
1908 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
1909 | * | |
1910 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
1911 | * by killing all belonging OOM-killable tasks. | |
1912 | * | |
1913 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
1914 | */ | |
1915 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
1916 | struct mem_cgroup *oom_domain) | |
1917 | { | |
1918 | struct mem_cgroup *oom_group = NULL; | |
1919 | struct mem_cgroup *memcg; | |
1920 | ||
1921 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
1922 | return NULL; | |
1923 | ||
1924 | if (!oom_domain) | |
1925 | oom_domain = root_mem_cgroup; | |
1926 | ||
1927 | rcu_read_lock(); | |
1928 | ||
1929 | memcg = mem_cgroup_from_task(victim); | |
1930 | if (memcg == root_mem_cgroup) | |
1931 | goto out; | |
1932 | ||
48fe267c RG |
1933 | /* |
1934 | * If the victim task has been asynchronously moved to a different | |
1935 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
1936 | * In this case it's better to ignore memory.group.oom. | |
1937 | */ | |
1938 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
1939 | goto out; | |
1940 | ||
3d8b38eb RG |
1941 | /* |
1942 | * Traverse the memory cgroup hierarchy from the victim task's | |
1943 | * cgroup up to the OOMing cgroup (or root) to find the | |
1944 | * highest-level memory cgroup with oom.group set. | |
1945 | */ | |
1946 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
1947 | if (memcg->oom_group) | |
1948 | oom_group = memcg; | |
1949 | ||
1950 | if (memcg == oom_domain) | |
1951 | break; | |
1952 | } | |
1953 | ||
1954 | if (oom_group) | |
1955 | css_get(&oom_group->css); | |
1956 | out: | |
1957 | rcu_read_unlock(); | |
1958 | ||
1959 | return oom_group; | |
1960 | } | |
1961 | ||
1962 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
1963 | { | |
1964 | pr_info("Tasks in "); | |
1965 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1966 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
1967 | } | |
1968 | ||
d7365e78 | 1969 | /** |
81f8c3a4 JW |
1970 | * lock_page_memcg - lock a page->mem_cgroup binding |
1971 | * @page: the page | |
32047e2a | 1972 | * |
81f8c3a4 | 1973 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
1974 | * another cgroup. |
1975 | * | |
1976 | * It ensures lifetime of the returned memcg. Caller is responsible | |
1977 | * for the lifetime of the page; __unlock_page_memcg() is available | |
1978 | * when @page might get freed inside the locked section. | |
d69b042f | 1979 | */ |
739f79fc | 1980 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 KH |
1981 | { |
1982 | struct mem_cgroup *memcg; | |
6de22619 | 1983 | unsigned long flags; |
89c06bd5 | 1984 | |
6de22619 JW |
1985 | /* |
1986 | * The RCU lock is held throughout the transaction. The fast | |
1987 | * path can get away without acquiring the memcg->move_lock | |
1988 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
1989 | * |
1990 | * The RCU lock also protects the memcg from being freed when | |
1991 | * the page state that is going to change is the only thing | |
1992 | * preventing the page itself from being freed. E.g. writeback | |
1993 | * doesn't hold a page reference and relies on PG_writeback to | |
1994 | * keep off truncation, migration and so forth. | |
1995 | */ | |
d7365e78 JW |
1996 | rcu_read_lock(); |
1997 | ||
1998 | if (mem_cgroup_disabled()) | |
739f79fc | 1999 | return NULL; |
89c06bd5 | 2000 | again: |
1306a85a | 2001 | memcg = page->mem_cgroup; |
29833315 | 2002 | if (unlikely(!memcg)) |
739f79fc | 2003 | return NULL; |
d7365e78 | 2004 | |
bdcbb659 | 2005 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 2006 | return memcg; |
89c06bd5 | 2007 | |
6de22619 | 2008 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 2009 | if (memcg != page->mem_cgroup) { |
6de22619 | 2010 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2011 | goto again; |
2012 | } | |
6de22619 JW |
2013 | |
2014 | /* | |
2015 | * When charge migration first begins, we can have locked and | |
2016 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 2017 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
2018 | */ |
2019 | memcg->move_lock_task = current; | |
2020 | memcg->move_lock_flags = flags; | |
d7365e78 | 2021 | |
739f79fc | 2022 | return memcg; |
89c06bd5 | 2023 | } |
81f8c3a4 | 2024 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2025 | |
d7365e78 | 2026 | /** |
739f79fc JW |
2027 | * __unlock_page_memcg - unlock and unpin a memcg |
2028 | * @memcg: the memcg | |
2029 | * | |
2030 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2031 | */ |
739f79fc | 2032 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2033 | { |
6de22619 JW |
2034 | if (memcg && memcg->move_lock_task == current) { |
2035 | unsigned long flags = memcg->move_lock_flags; | |
2036 | ||
2037 | memcg->move_lock_task = NULL; | |
2038 | memcg->move_lock_flags = 0; | |
2039 | ||
2040 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2041 | } | |
89c06bd5 | 2042 | |
d7365e78 | 2043 | rcu_read_unlock(); |
89c06bd5 | 2044 | } |
739f79fc JW |
2045 | |
2046 | /** | |
2047 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
2048 | * @page: the page | |
2049 | */ | |
2050 | void unlock_page_memcg(struct page *page) | |
2051 | { | |
2052 | __unlock_page_memcg(page->mem_cgroup); | |
2053 | } | |
81f8c3a4 | 2054 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2055 | |
cdec2e42 KH |
2056 | struct memcg_stock_pcp { |
2057 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2058 | unsigned int nr_pages; |
cdec2e42 | 2059 | struct work_struct work; |
26fe6168 | 2060 | unsigned long flags; |
a0db00fc | 2061 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2062 | }; |
2063 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2064 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2065 | |
a0956d54 SS |
2066 | /** |
2067 | * consume_stock: Try to consume stocked charge on this cpu. | |
2068 | * @memcg: memcg to consume from. | |
2069 | * @nr_pages: how many pages to charge. | |
2070 | * | |
2071 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2072 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2073 | * service an allocation will refill the stock. | |
2074 | * | |
2075 | * returns true if successful, false otherwise. | |
cdec2e42 | 2076 | */ |
a0956d54 | 2077 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2078 | { |
2079 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2080 | unsigned long flags; |
3e32cb2e | 2081 | bool ret = false; |
cdec2e42 | 2082 | |
a983b5eb | 2083 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2084 | return ret; |
a0956d54 | 2085 | |
db2ba40c JW |
2086 | local_irq_save(flags); |
2087 | ||
2088 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2089 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2090 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2091 | ret = true; |
2092 | } | |
db2ba40c JW |
2093 | |
2094 | local_irq_restore(flags); | |
2095 | ||
cdec2e42 KH |
2096 | return ret; |
2097 | } | |
2098 | ||
2099 | /* | |
3e32cb2e | 2100 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2101 | */ |
2102 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2103 | { | |
2104 | struct mem_cgroup *old = stock->cached; | |
2105 | ||
11c9ea4e | 2106 | if (stock->nr_pages) { |
3e32cb2e | 2107 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2108 | if (do_memsw_account()) |
3e32cb2e | 2109 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2110 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2111 | stock->nr_pages = 0; |
cdec2e42 KH |
2112 | } |
2113 | stock->cached = NULL; | |
cdec2e42 KH |
2114 | } |
2115 | ||
cdec2e42 KH |
2116 | static void drain_local_stock(struct work_struct *dummy) |
2117 | { | |
db2ba40c JW |
2118 | struct memcg_stock_pcp *stock; |
2119 | unsigned long flags; | |
2120 | ||
72f0184c MH |
2121 | /* |
2122 | * The only protection from memory hotplug vs. drain_stock races is | |
2123 | * that we always operate on local CPU stock here with IRQ disabled | |
2124 | */ | |
db2ba40c JW |
2125 | local_irq_save(flags); |
2126 | ||
2127 | stock = this_cpu_ptr(&memcg_stock); | |
cdec2e42 | 2128 | drain_stock(stock); |
26fe6168 | 2129 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2130 | |
2131 | local_irq_restore(flags); | |
cdec2e42 KH |
2132 | } |
2133 | ||
2134 | /* | |
3e32cb2e | 2135 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2136 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2137 | */ |
c0ff4b85 | 2138 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2139 | { |
db2ba40c JW |
2140 | struct memcg_stock_pcp *stock; |
2141 | unsigned long flags; | |
2142 | ||
2143 | local_irq_save(flags); | |
cdec2e42 | 2144 | |
db2ba40c | 2145 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2146 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2147 | drain_stock(stock); |
c0ff4b85 | 2148 | stock->cached = memcg; |
cdec2e42 | 2149 | } |
11c9ea4e | 2150 | stock->nr_pages += nr_pages; |
db2ba40c | 2151 | |
a983b5eb | 2152 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2153 | drain_stock(stock); |
2154 | ||
db2ba40c | 2155 | local_irq_restore(flags); |
cdec2e42 KH |
2156 | } |
2157 | ||
2158 | /* | |
c0ff4b85 | 2159 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2160 | * of the hierarchy under it. |
cdec2e42 | 2161 | */ |
6d3d6aa2 | 2162 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2163 | { |
26fe6168 | 2164 | int cpu, curcpu; |
d38144b7 | 2165 | |
6d3d6aa2 JW |
2166 | /* If someone's already draining, avoid adding running more workers. */ |
2167 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2168 | return; | |
72f0184c MH |
2169 | /* |
2170 | * Notify other cpus that system-wide "drain" is running | |
2171 | * We do not care about races with the cpu hotplug because cpu down | |
2172 | * as well as workers from this path always operate on the local | |
2173 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2174 | */ | |
5af12d0e | 2175 | curcpu = get_cpu(); |
cdec2e42 KH |
2176 | for_each_online_cpu(cpu) { |
2177 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2178 | struct mem_cgroup *memcg; |
e1a366be | 2179 | bool flush = false; |
26fe6168 | 2180 | |
e1a366be | 2181 | rcu_read_lock(); |
c0ff4b85 | 2182 | memcg = stock->cached; |
e1a366be RG |
2183 | if (memcg && stock->nr_pages && |
2184 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2185 | flush = true; | |
2186 | rcu_read_unlock(); | |
2187 | ||
2188 | if (flush && | |
2189 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2190 | if (cpu == curcpu) |
2191 | drain_local_stock(&stock->work); | |
2192 | else | |
2193 | schedule_work_on(cpu, &stock->work); | |
2194 | } | |
cdec2e42 | 2195 | } |
5af12d0e | 2196 | put_cpu(); |
9f50fad6 | 2197 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2198 | } |
2199 | ||
308167fc | 2200 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2201 | { |
cdec2e42 | 2202 | struct memcg_stock_pcp *stock; |
42a30035 | 2203 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2204 | |
cdec2e42 KH |
2205 | stock = &per_cpu(memcg_stock, cpu); |
2206 | drain_stock(stock); | |
a983b5eb JW |
2207 | |
2208 | for_each_mem_cgroup(memcg) { | |
2209 | int i; | |
2210 | ||
2211 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2212 | int nid; | |
2213 | long x; | |
2214 | ||
871789d4 | 2215 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
815744d7 | 2216 | if (x) |
42a30035 JW |
2217 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2218 | atomic_long_add(x, &memcg->vmstats[i]); | |
a983b5eb JW |
2219 | |
2220 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2221 | continue; | |
2222 | ||
2223 | for_each_node(nid) { | |
2224 | struct mem_cgroup_per_node *pn; | |
2225 | ||
2226 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2227 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
815744d7 | 2228 | if (x) |
42a30035 JW |
2229 | do { |
2230 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2231 | } while ((pn = parent_nodeinfo(pn, nid))); | |
a983b5eb JW |
2232 | } |
2233 | } | |
2234 | ||
e27be240 | 2235 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2236 | long x; |
2237 | ||
871789d4 | 2238 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
815744d7 | 2239 | if (x) |
42a30035 JW |
2240 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2241 | atomic_long_add(x, &memcg->vmevents[i]); | |
a983b5eb JW |
2242 | } |
2243 | } | |
2244 | ||
308167fc | 2245 | return 0; |
cdec2e42 KH |
2246 | } |
2247 | ||
f7e1cb6e JW |
2248 | static void reclaim_high(struct mem_cgroup *memcg, |
2249 | unsigned int nr_pages, | |
2250 | gfp_t gfp_mask) | |
2251 | { | |
2252 | do { | |
f6f989c5 | 2253 | if (page_counter_read(&memcg->memory) <= READ_ONCE(memcg->high)) |
f7e1cb6e | 2254 | continue; |
e27be240 | 2255 | memcg_memory_event(memcg, MEMCG_HIGH); |
f7e1cb6e JW |
2256 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
2257 | } while ((memcg = parent_mem_cgroup(memcg))); | |
2258 | } | |
2259 | ||
2260 | static void high_work_func(struct work_struct *work) | |
2261 | { | |
2262 | struct mem_cgroup *memcg; | |
2263 | ||
2264 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2265 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2266 | } |
2267 | ||
0e4b01df CD |
2268 | /* |
2269 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2270 | * enough to still cause a significant slowdown in most cases, while still | |
2271 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2272 | */ | |
2273 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2274 | ||
2275 | /* | |
2276 | * When calculating the delay, we use these either side of the exponentiation to | |
2277 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2278 | * below. | |
2279 | * | |
2280 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2281 | * overage ratio to a delay. | |
2282 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down down the | |
2283 | * proposed penalty in order to reduce to a reasonable number of jiffies, and | |
2284 | * to produce a reasonable delay curve. | |
2285 | * | |
2286 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2287 | * reasonable delay curve compared to precision-adjusted overage, not | |
2288 | * penalising heavily at first, but still making sure that growth beyond the | |
2289 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2290 | * example, with a high of 100 megabytes: | |
2291 | * | |
2292 | * +-------+------------------------+ | |
2293 | * | usage | time to allocate in ms | | |
2294 | * +-------+------------------------+ | |
2295 | * | 100M | 0 | | |
2296 | * | 101M | 6 | | |
2297 | * | 102M | 25 | | |
2298 | * | 103M | 57 | | |
2299 | * | 104M | 102 | | |
2300 | * | 105M | 159 | | |
2301 | * | 106M | 230 | | |
2302 | * | 107M | 313 | | |
2303 | * | 108M | 409 | | |
2304 | * | 109M | 518 | | |
2305 | * | 110M | 639 | | |
2306 | * | 111M | 774 | | |
2307 | * | 112M | 921 | | |
2308 | * | 113M | 1081 | | |
2309 | * | 114M | 1254 | | |
2310 | * | 115M | 1439 | | |
2311 | * | 116M | 1638 | | |
2312 | * | 117M | 1849 | | |
2313 | * | 118M | 2000 | | |
2314 | * | 119M | 2000 | | |
2315 | * | 120M | 2000 | | |
2316 | * +-------+------------------------+ | |
2317 | */ | |
2318 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2319 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2320 | ||
b23afb93 | 2321 | /* |
e26733e0 CD |
2322 | * Get the number of jiffies that we should penalise a mischievous cgroup which |
2323 | * is exceeding its memory.high by checking both it and its ancestors. | |
b23afb93 | 2324 | */ |
e26733e0 CD |
2325 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, |
2326 | unsigned int nr_pages) | |
b23afb93 | 2327 | { |
e26733e0 CD |
2328 | unsigned long penalty_jiffies; |
2329 | u64 max_overage = 0; | |
b23afb93 | 2330 | |
e26733e0 CD |
2331 | do { |
2332 | unsigned long usage, high; | |
2333 | u64 overage; | |
b23afb93 | 2334 | |
e26733e0 CD |
2335 | usage = page_counter_read(&memcg->memory); |
2336 | high = READ_ONCE(memcg->high); | |
2337 | ||
2338 | /* | |
2339 | * Prevent division by 0 in overage calculation by acting as if | |
2340 | * it was a threshold of 1 page | |
2341 | */ | |
2342 | high = max(high, 1UL); | |
2343 | ||
2344 | overage = usage - high; | |
2345 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2346 | overage = div64_u64(overage, high); | |
2347 | ||
2348 | if (overage > max_overage) | |
2349 | max_overage = overage; | |
2350 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2351 | !mem_cgroup_is_root(memcg)); | |
2352 | ||
2353 | if (!max_overage) | |
2354 | return 0; | |
0e4b01df CD |
2355 | |
2356 | /* | |
0e4b01df CD |
2357 | * We use overage compared to memory.high to calculate the number of |
2358 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2359 | * fairly lenient on small overages, and increasingly harsh when the | |
2360 | * memcg in question makes it clear that it has no intention of stopping | |
2361 | * its crazy behaviour, so we exponentially increase the delay based on | |
2362 | * overage amount. | |
2363 | */ | |
e26733e0 CD |
2364 | penalty_jiffies = max_overage * max_overage * HZ; |
2365 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2366 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2367 | |
2368 | /* | |
2369 | * Factor in the task's own contribution to the overage, such that four | |
2370 | * N-sized allocations are throttled approximately the same as one | |
2371 | * 4N-sized allocation. | |
2372 | * | |
2373 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2374 | * larger the current charge patch is than that. | |
2375 | */ | |
2376 | penalty_jiffies = penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; | |
2377 | ||
2378 | /* | |
2379 | * Clamp the max delay per usermode return so as to still keep the | |
2380 | * application moving forwards and also permit diagnostics, albeit | |
2381 | * extremely slowly. | |
2382 | */ | |
e26733e0 CD |
2383 | return min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); |
2384 | } | |
2385 | ||
2386 | /* | |
2387 | * Scheduled by try_charge() to be executed from the userland return path | |
2388 | * and reclaims memory over the high limit. | |
2389 | */ | |
2390 | void mem_cgroup_handle_over_high(void) | |
2391 | { | |
2392 | unsigned long penalty_jiffies; | |
2393 | unsigned long pflags; | |
2394 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
2395 | struct mem_cgroup *memcg; | |
2396 | ||
2397 | if (likely(!nr_pages)) | |
2398 | return; | |
2399 | ||
2400 | memcg = get_mem_cgroup_from_mm(current->mm); | |
2401 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
2402 | current->memcg_nr_pages_over_high = 0; | |
2403 | ||
2404 | /* | |
2405 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2406 | * allocators proactively to slow down excessive growth. | |
2407 | */ | |
2408 | penalty_jiffies = calculate_high_delay(memcg, nr_pages); | |
0e4b01df CD |
2409 | |
2410 | /* | |
2411 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2412 | * that it's not even worth doing, in an attempt to be nice to those who | |
2413 | * go only a small amount over their memory.high value and maybe haven't | |
2414 | * been aggressively reclaimed enough yet. | |
2415 | */ | |
2416 | if (penalty_jiffies <= HZ / 100) | |
2417 | goto out; | |
2418 | ||
2419 | /* | |
2420 | * If we exit early, we're guaranteed to die (since | |
2421 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2422 | * need to account for any ill-begotten jiffies to pay them off later. | |
2423 | */ | |
2424 | psi_memstall_enter(&pflags); | |
2425 | schedule_timeout_killable(penalty_jiffies); | |
2426 | psi_memstall_leave(&pflags); | |
2427 | ||
2428 | out: | |
2429 | css_put(&memcg->css); | |
b23afb93 TH |
2430 | } |
2431 | ||
00501b53 JW |
2432 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2433 | unsigned int nr_pages) | |
8a9f3ccd | 2434 | { |
a983b5eb | 2435 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
9b130619 | 2436 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2437 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2438 | struct page_counter *counter; |
6539cc05 | 2439 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2440 | bool may_swap = true; |
2441 | bool drained = false; | |
29ef680a | 2442 | enum oom_status oom_status; |
a636b327 | 2443 | |
ce00a967 | 2444 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2445 | return 0; |
6539cc05 | 2446 | retry: |
b6b6cc72 | 2447 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2448 | return 0; |
8a9f3ccd | 2449 | |
7941d214 | 2450 | if (!do_memsw_account() || |
6071ca52 JW |
2451 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2452 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2453 | goto done_restock; |
7941d214 | 2454 | if (do_memsw_account()) |
3e32cb2e JW |
2455 | page_counter_uncharge(&memcg->memsw, batch); |
2456 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2457 | } else { |
3e32cb2e | 2458 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2459 | may_swap = false; |
3fbe7244 | 2460 | } |
7a81b88c | 2461 | |
6539cc05 JW |
2462 | if (batch > nr_pages) { |
2463 | batch = nr_pages; | |
2464 | goto retry; | |
2465 | } | |
6d61ef40 | 2466 | |
869712fd JW |
2467 | /* |
2468 | * Memcg doesn't have a dedicated reserve for atomic | |
2469 | * allocations. But like the global atomic pool, we need to | |
2470 | * put the burden of reclaim on regular allocation requests | |
2471 | * and let these go through as privileged allocations. | |
2472 | */ | |
2473 | if (gfp_mask & __GFP_ATOMIC) | |
2474 | goto force; | |
2475 | ||
06b078fc JW |
2476 | /* |
2477 | * Unlike in global OOM situations, memcg is not in a physical | |
2478 | * memory shortage. Allow dying and OOM-killed tasks to | |
2479 | * bypass the last charges so that they can exit quickly and | |
2480 | * free their memory. | |
2481 | */ | |
7775face | 2482 | if (unlikely(should_force_charge())) |
10d53c74 | 2483 | goto force; |
06b078fc | 2484 | |
89a28483 JW |
2485 | /* |
2486 | * Prevent unbounded recursion when reclaim operations need to | |
2487 | * allocate memory. This might exceed the limits temporarily, | |
2488 | * but we prefer facilitating memory reclaim and getting back | |
2489 | * under the limit over triggering OOM kills in these cases. | |
2490 | */ | |
2491 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2492 | goto force; | |
2493 | ||
06b078fc JW |
2494 | if (unlikely(task_in_memcg_oom(current))) |
2495 | goto nomem; | |
2496 | ||
d0164adc | 2497 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2498 | goto nomem; |
4b534334 | 2499 | |
e27be240 | 2500 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2501 | |
b70a2a21 JW |
2502 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2503 | gfp_mask, may_swap); | |
6539cc05 | 2504 | |
61e02c74 | 2505 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2506 | goto retry; |
28c34c29 | 2507 | |
b70a2a21 | 2508 | if (!drained) { |
6d3d6aa2 | 2509 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2510 | drained = true; |
2511 | goto retry; | |
2512 | } | |
2513 | ||
28c34c29 JW |
2514 | if (gfp_mask & __GFP_NORETRY) |
2515 | goto nomem; | |
6539cc05 JW |
2516 | /* |
2517 | * Even though the limit is exceeded at this point, reclaim | |
2518 | * may have been able to free some pages. Retry the charge | |
2519 | * before killing the task. | |
2520 | * | |
2521 | * Only for regular pages, though: huge pages are rather | |
2522 | * unlikely to succeed so close to the limit, and we fall back | |
2523 | * to regular pages anyway in case of failure. | |
2524 | */ | |
61e02c74 | 2525 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2526 | goto retry; |
2527 | /* | |
2528 | * At task move, charge accounts can be doubly counted. So, it's | |
2529 | * better to wait until the end of task_move if something is going on. | |
2530 | */ | |
2531 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2532 | goto retry; | |
2533 | ||
9b130619 JW |
2534 | if (nr_retries--) |
2535 | goto retry; | |
2536 | ||
38d38493 | 2537 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2538 | goto nomem; |
2539 | ||
06b078fc | 2540 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2541 | goto force; |
06b078fc | 2542 | |
6539cc05 | 2543 | if (fatal_signal_pending(current)) |
10d53c74 | 2544 | goto force; |
6539cc05 | 2545 | |
29ef680a MH |
2546 | /* |
2547 | * keep retrying as long as the memcg oom killer is able to make | |
2548 | * a forward progress or bypass the charge if the oom killer | |
2549 | * couldn't make any progress. | |
2550 | */ | |
2551 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2552 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2553 | switch (oom_status) { |
2554 | case OOM_SUCCESS: | |
2555 | nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
29ef680a MH |
2556 | goto retry; |
2557 | case OOM_FAILED: | |
2558 | goto force; | |
2559 | default: | |
2560 | goto nomem; | |
2561 | } | |
7a81b88c | 2562 | nomem: |
6d1fdc48 | 2563 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2564 | return -ENOMEM; |
10d53c74 TH |
2565 | force: |
2566 | /* | |
2567 | * The allocation either can't fail or will lead to more memory | |
2568 | * being freed very soon. Allow memory usage go over the limit | |
2569 | * temporarily by force charging it. | |
2570 | */ | |
2571 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2572 | if (do_memsw_account()) |
10d53c74 TH |
2573 | page_counter_charge(&memcg->memsw, nr_pages); |
2574 | css_get_many(&memcg->css, nr_pages); | |
2575 | ||
2576 | return 0; | |
6539cc05 JW |
2577 | |
2578 | done_restock: | |
e8ea14cc | 2579 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2580 | if (batch > nr_pages) |
2581 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2582 | |
241994ed | 2583 | /* |
b23afb93 TH |
2584 | * If the hierarchy is above the normal consumption range, schedule |
2585 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2586 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2587 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2588 | * not recorded as it most likely matches current's and won't | |
2589 | * change in the meantime. As high limit is checked again before | |
2590 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2591 | */ |
2592 | do { | |
f6f989c5 | 2593 | if (page_counter_read(&memcg->memory) > READ_ONCE(memcg->high)) { |
f7e1cb6e JW |
2594 | /* Don't bother a random interrupted task */ |
2595 | if (in_interrupt()) { | |
2596 | schedule_work(&memcg->high_work); | |
2597 | break; | |
2598 | } | |
9516a18a | 2599 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2600 | set_notify_resume(current); |
2601 | break; | |
2602 | } | |
241994ed | 2603 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2604 | |
2605 | return 0; | |
7a81b88c | 2606 | } |
8a9f3ccd | 2607 | |
00501b53 | 2608 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2609 | { |
ce00a967 JW |
2610 | if (mem_cgroup_is_root(memcg)) |
2611 | return; | |
2612 | ||
3e32cb2e | 2613 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2614 | if (do_memsw_account()) |
3e32cb2e | 2615 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2616 | |
e8ea14cc | 2617 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2618 | } |
2619 | ||
0a31bc97 JW |
2620 | static void lock_page_lru(struct page *page, int *isolated) |
2621 | { | |
f4b7e272 | 2622 | pg_data_t *pgdat = page_pgdat(page); |
0a31bc97 | 2623 | |
f4b7e272 | 2624 | spin_lock_irq(&pgdat->lru_lock); |
0a31bc97 JW |
2625 | if (PageLRU(page)) { |
2626 | struct lruvec *lruvec; | |
2627 | ||
f4b7e272 | 2628 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
0a31bc97 JW |
2629 | ClearPageLRU(page); |
2630 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2631 | *isolated = 1; | |
2632 | } else | |
2633 | *isolated = 0; | |
2634 | } | |
2635 | ||
2636 | static void unlock_page_lru(struct page *page, int isolated) | |
2637 | { | |
f4b7e272 | 2638 | pg_data_t *pgdat = page_pgdat(page); |
0a31bc97 JW |
2639 | |
2640 | if (isolated) { | |
2641 | struct lruvec *lruvec; | |
2642 | ||
f4b7e272 | 2643 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
0a31bc97 JW |
2644 | VM_BUG_ON_PAGE(PageLRU(page), page); |
2645 | SetPageLRU(page); | |
2646 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2647 | } | |
f4b7e272 | 2648 | spin_unlock_irq(&pgdat->lru_lock); |
0a31bc97 JW |
2649 | } |
2650 | ||
00501b53 | 2651 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2652 | bool lrucare) |
7a81b88c | 2653 | { |
0a31bc97 | 2654 | int isolated; |
9ce70c02 | 2655 | |
1306a85a | 2656 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2657 | |
2658 | /* | |
2659 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2660 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2661 | */ | |
0a31bc97 JW |
2662 | if (lrucare) |
2663 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2664 | |
0a31bc97 JW |
2665 | /* |
2666 | * Nobody should be changing or seriously looking at | |
1306a85a | 2667 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2668 | * |
2669 | * - the page is uncharged | |
2670 | * | |
2671 | * - the page is off-LRU | |
2672 | * | |
2673 | * - an anonymous fault has exclusive page access, except for | |
2674 | * a locked page table | |
2675 | * | |
2676 | * - a page cache insertion, a swapin fault, or a migration | |
2677 | * have the page locked | |
2678 | */ | |
1306a85a | 2679 | page->mem_cgroup = memcg; |
9ce70c02 | 2680 | |
0a31bc97 JW |
2681 | if (lrucare) |
2682 | unlock_page_lru(page, isolated); | |
7a81b88c | 2683 | } |
66e1707b | 2684 | |
84c07d11 | 2685 | #ifdef CONFIG_MEMCG_KMEM |
8380ce47 RG |
2686 | /* |
2687 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2688 | * | |
2689 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
2690 | * cgroup_mutex, etc. | |
2691 | */ | |
2692 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2693 | { | |
2694 | struct page *page; | |
2695 | ||
2696 | if (mem_cgroup_disabled()) | |
2697 | return NULL; | |
2698 | ||
2699 | page = virt_to_head_page(p); | |
2700 | ||
2701 | /* | |
2702 | * Slab pages don't have page->mem_cgroup set because corresponding | |
2703 | * kmem caches can be reparented during the lifetime. That's why | |
2704 | * memcg_from_slab_page() should be used instead. | |
2705 | */ | |
2706 | if (PageSlab(page)) | |
2707 | return memcg_from_slab_page(page); | |
2708 | ||
2709 | /* All other pages use page->mem_cgroup */ | |
2710 | return page->mem_cgroup; | |
2711 | } | |
2712 | ||
f3bb3043 | 2713 | static int memcg_alloc_cache_id(void) |
55007d84 | 2714 | { |
f3bb3043 VD |
2715 | int id, size; |
2716 | int err; | |
2717 | ||
dbcf73e2 | 2718 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2719 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2720 | if (id < 0) | |
2721 | return id; | |
55007d84 | 2722 | |
dbcf73e2 | 2723 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2724 | return id; |
2725 | ||
2726 | /* | |
2727 | * There's no space for the new id in memcg_caches arrays, | |
2728 | * so we have to grow them. | |
2729 | */ | |
05257a1a | 2730 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2731 | |
2732 | size = 2 * (id + 1); | |
55007d84 GC |
2733 | if (size < MEMCG_CACHES_MIN_SIZE) |
2734 | size = MEMCG_CACHES_MIN_SIZE; | |
2735 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2736 | size = MEMCG_CACHES_MAX_SIZE; | |
2737 | ||
f3bb3043 | 2738 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2739 | if (!err) |
2740 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2741 | if (!err) |
2742 | memcg_nr_cache_ids = size; | |
2743 | ||
2744 | up_write(&memcg_cache_ids_sem); | |
2745 | ||
f3bb3043 | 2746 | if (err) { |
dbcf73e2 | 2747 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2748 | return err; |
2749 | } | |
2750 | return id; | |
2751 | } | |
2752 | ||
2753 | static void memcg_free_cache_id(int id) | |
2754 | { | |
dbcf73e2 | 2755 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2756 | } |
2757 | ||
d5b3cf71 | 2758 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2759 | struct mem_cgroup *memcg; |
2760 | struct kmem_cache *cachep; | |
2761 | struct work_struct work; | |
2762 | }; | |
2763 | ||
d5b3cf71 | 2764 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2765 | { |
d5b3cf71 VD |
2766 | struct memcg_kmem_cache_create_work *cw = |
2767 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2768 | struct mem_cgroup *memcg = cw->memcg; |
2769 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2770 | |
d5b3cf71 | 2771 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2772 | |
5722d094 | 2773 | css_put(&memcg->css); |
d7f25f8a GC |
2774 | kfree(cw); |
2775 | } | |
2776 | ||
2777 | /* | |
2778 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2779 | */ |
85cfb245 | 2780 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
d5b3cf71 | 2781 | struct kmem_cache *cachep) |
d7f25f8a | 2782 | { |
d5b3cf71 | 2783 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2784 | |
f0a3a24b RG |
2785 | if (!css_tryget_online(&memcg->css)) |
2786 | return; | |
2787 | ||
c892fd82 | 2788 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT | __GFP_NOWARN); |
8135be5a | 2789 | if (!cw) |
d7f25f8a | 2790 | return; |
8135be5a | 2791 | |
d7f25f8a GC |
2792 | cw->memcg = memcg; |
2793 | cw->cachep = cachep; | |
d5b3cf71 | 2794 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2795 | |
17cc4dfe | 2796 | queue_work(memcg_kmem_cache_wq, &cw->work); |
d7f25f8a GC |
2797 | } |
2798 | ||
45264778 VD |
2799 | static inline bool memcg_kmem_bypass(void) |
2800 | { | |
2801 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
2802 | return true; | |
2803 | return false; | |
2804 | } | |
2805 | ||
2806 | /** | |
2807 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation | |
2808 | * @cachep: the original global kmem cache | |
2809 | * | |
d7f25f8a GC |
2810 | * Return the kmem_cache we're supposed to use for a slab allocation. |
2811 | * We try to use the current memcg's version of the cache. | |
2812 | * | |
45264778 VD |
2813 | * If the cache does not exist yet, if we are the first user of it, we |
2814 | * create it asynchronously in a workqueue and let the current allocation | |
2815 | * go through with the original cache. | |
d7f25f8a | 2816 | * |
45264778 VD |
2817 | * This function takes a reference to the cache it returns to assure it |
2818 | * won't get destroyed while we are working with it. Once the caller is | |
2819 | * done with it, memcg_kmem_put_cache() must be called to release the | |
2820 | * reference. | |
d7f25f8a | 2821 | */ |
45264778 | 2822 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2823 | { |
2824 | struct mem_cgroup *memcg; | |
959c8963 | 2825 | struct kmem_cache *memcg_cachep; |
f0a3a24b | 2826 | struct memcg_cache_array *arr; |
2a4db7eb | 2827 | int kmemcg_id; |
d7f25f8a | 2828 | |
f7ce3190 | 2829 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2830 | |
45264778 | 2831 | if (memcg_kmem_bypass()) |
230e9fc2 VD |
2832 | return cachep; |
2833 | ||
f0a3a24b RG |
2834 | rcu_read_lock(); |
2835 | ||
2836 | if (unlikely(current->active_memcg)) | |
2837 | memcg = current->active_memcg; | |
2838 | else | |
2839 | memcg = mem_cgroup_from_task(current); | |
2840 | ||
2841 | if (!memcg || memcg == root_mem_cgroup) | |
2842 | goto out_unlock; | |
2843 | ||
4db0c3c2 | 2844 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2845 | if (kmemcg_id < 0) |
f0a3a24b | 2846 | goto out_unlock; |
d7f25f8a | 2847 | |
f0a3a24b RG |
2848 | arr = rcu_dereference(cachep->memcg_params.memcg_caches); |
2849 | ||
2850 | /* | |
2851 | * Make sure we will access the up-to-date value. The code updating | |
2852 | * memcg_caches issues a write barrier to match the data dependency | |
2853 | * barrier inside READ_ONCE() (see memcg_create_kmem_cache()). | |
2854 | */ | |
2855 | memcg_cachep = READ_ONCE(arr->entries[kmemcg_id]); | |
ca0dde97 LZ |
2856 | |
2857 | /* | |
2858 | * If we are in a safe context (can wait, and not in interrupt | |
2859 | * context), we could be be predictable and return right away. | |
2860 | * This would guarantee that the allocation being performed | |
2861 | * already belongs in the new cache. | |
2862 | * | |
2863 | * However, there are some clashes that can arrive from locking. | |
2864 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2865 | * memcg_create_kmem_cache, this means no further allocation |
2866 | * could happen with the slab_mutex held. So it's better to | |
2867 | * defer everything. | |
f0a3a24b RG |
2868 | * |
2869 | * If the memcg is dying or memcg_cache is about to be released, | |
2870 | * don't bother creating new kmem_caches. Because memcg_cachep | |
2871 | * is ZEROed as the fist step of kmem offlining, we don't need | |
2872 | * percpu_ref_tryget_live() here. css_tryget_online() check in | |
2873 | * memcg_schedule_kmem_cache_create() will prevent us from | |
2874 | * creation of a new kmem_cache. | |
ca0dde97 | 2875 | */ |
f0a3a24b RG |
2876 | if (unlikely(!memcg_cachep)) |
2877 | memcg_schedule_kmem_cache_create(memcg, cachep); | |
2878 | else if (percpu_ref_tryget(&memcg_cachep->memcg_params.refcnt)) | |
2879 | cachep = memcg_cachep; | |
2880 | out_unlock: | |
2881 | rcu_read_unlock(); | |
ca0dde97 | 2882 | return cachep; |
d7f25f8a | 2883 | } |
d7f25f8a | 2884 | |
45264778 VD |
2885 | /** |
2886 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache | |
2887 | * @cachep: the cache returned by memcg_kmem_get_cache | |
2888 | */ | |
2889 | void memcg_kmem_put_cache(struct kmem_cache *cachep) | |
8135be5a VD |
2890 | { |
2891 | if (!is_root_cache(cachep)) | |
f0a3a24b | 2892 | percpu_ref_put(&cachep->memcg_params.refcnt); |
8135be5a VD |
2893 | } |
2894 | ||
45264778 | 2895 | /** |
4b13f64d | 2896 | * __memcg_kmem_charge: charge a number of kernel pages to a memcg |
10eaec2f | 2897 | * @memcg: memory cgroup to charge |
45264778 | 2898 | * @gfp: reclaim mode |
92d0510c | 2899 | * @nr_pages: number of pages to charge |
45264778 VD |
2900 | * |
2901 | * Returns 0 on success, an error code on failure. | |
2902 | */ | |
4b13f64d RG |
2903 | int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp, |
2904 | unsigned int nr_pages) | |
7ae1e1d0 | 2905 | { |
f3ccb2c4 | 2906 | struct page_counter *counter; |
7ae1e1d0 GC |
2907 | int ret; |
2908 | ||
f3ccb2c4 | 2909 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2910 | if (ret) |
f3ccb2c4 | 2911 | return ret; |
52c29b04 JW |
2912 | |
2913 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2914 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
e55d9d9b MH |
2915 | |
2916 | /* | |
2917 | * Enforce __GFP_NOFAIL allocation because callers are not | |
2918 | * prepared to see failures and likely do not have any failure | |
2919 | * handling code. | |
2920 | */ | |
2921 | if (gfp & __GFP_NOFAIL) { | |
2922 | page_counter_charge(&memcg->kmem, nr_pages); | |
2923 | return 0; | |
2924 | } | |
52c29b04 JW |
2925 | cancel_charge(memcg, nr_pages); |
2926 | return -ENOMEM; | |
7ae1e1d0 | 2927 | } |
f3ccb2c4 | 2928 | return 0; |
7ae1e1d0 GC |
2929 | } |
2930 | ||
4b13f64d RG |
2931 | /** |
2932 | * __memcg_kmem_uncharge: uncharge a number of kernel pages from a memcg | |
2933 | * @memcg: memcg to uncharge | |
2934 | * @nr_pages: number of pages to uncharge | |
2935 | */ | |
2936 | void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages) | |
2937 | { | |
2938 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2939 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2940 | ||
2941 | page_counter_uncharge(&memcg->memory, nr_pages); | |
2942 | if (do_memsw_account()) | |
2943 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
2944 | } | |
2945 | ||
45264778 | 2946 | /** |
f4b00eab | 2947 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
2948 | * @page: page to charge |
2949 | * @gfp: reclaim mode | |
2950 | * @order: allocation order | |
2951 | * | |
2952 | * Returns 0 on success, an error code on failure. | |
2953 | */ | |
f4b00eab | 2954 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2955 | { |
f3ccb2c4 | 2956 | struct mem_cgroup *memcg; |
fcff7d7e | 2957 | int ret = 0; |
7ae1e1d0 | 2958 | |
60cd4bcd | 2959 | if (memcg_kmem_bypass()) |
45264778 VD |
2960 | return 0; |
2961 | ||
d46eb14b | 2962 | memcg = get_mem_cgroup_from_current(); |
c4159a75 | 2963 | if (!mem_cgroup_is_root(memcg)) { |
4b13f64d | 2964 | ret = __memcg_kmem_charge(memcg, gfp, 1 << order); |
4d96ba35 RG |
2965 | if (!ret) { |
2966 | page->mem_cgroup = memcg; | |
c4159a75 | 2967 | __SetPageKmemcg(page); |
4d96ba35 | 2968 | } |
c4159a75 | 2969 | } |
7ae1e1d0 | 2970 | css_put(&memcg->css); |
d05e83a6 | 2971 | return ret; |
7ae1e1d0 | 2972 | } |
49a18eae | 2973 | |
45264778 | 2974 | /** |
f4b00eab | 2975 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
2976 | * @page: page to uncharge |
2977 | * @order: allocation order | |
2978 | */ | |
f4b00eab | 2979 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 2980 | { |
1306a85a | 2981 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2982 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2983 | |
7ae1e1d0 GC |
2984 | if (!memcg) |
2985 | return; | |
2986 | ||
309381fe | 2987 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
4b13f64d | 2988 | __memcg_kmem_uncharge(memcg, nr_pages); |
1306a85a | 2989 | page->mem_cgroup = NULL; |
c4159a75 VD |
2990 | |
2991 | /* slab pages do not have PageKmemcg flag set */ | |
2992 | if (PageKmemcg(page)) | |
2993 | __ClearPageKmemcg(page); | |
2994 | ||
f3ccb2c4 | 2995 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2996 | } |
84c07d11 | 2997 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 2998 | |
ca3e0214 KH |
2999 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3000 | ||
ca3e0214 KH |
3001 | /* |
3002 | * Because tail pages are not marked as "used", set it. We're under | |
f4b7e272 | 3003 | * pgdat->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 3004 | */ |
e94c8a9c | 3005 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 3006 | { |
e94c8a9c | 3007 | int i; |
ca3e0214 | 3008 | |
3d37c4a9 KH |
3009 | if (mem_cgroup_disabled()) |
3010 | return; | |
b070e65c | 3011 | |
29833315 | 3012 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 3013 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 3014 | |
c9019e9b | 3015 | __mod_memcg_state(head->mem_cgroup, MEMCG_RSS_HUGE, -HPAGE_PMD_NR); |
ca3e0214 | 3016 | } |
12d27107 | 3017 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3018 | |
c255a458 | 3019 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
3020 | /** |
3021 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3022 | * @entry: swap entry to be moved | |
3023 | * @from: mem_cgroup which the entry is moved from | |
3024 | * @to: mem_cgroup which the entry is moved to | |
3025 | * | |
3026 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3027 | * as the mem_cgroup's id of @from. | |
3028 | * | |
3029 | * Returns 0 on success, -EINVAL on failure. | |
3030 | * | |
3e32cb2e | 3031 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3032 | * both res and memsw, and called css_get(). |
3033 | */ | |
3034 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3035 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3036 | { |
3037 | unsigned short old_id, new_id; | |
3038 | ||
34c00c31 LZ |
3039 | old_id = mem_cgroup_id(from); |
3040 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3041 | |
3042 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3043 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3044 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3045 | return 0; |
3046 | } | |
3047 | return -EINVAL; | |
3048 | } | |
3049 | #else | |
3050 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3051 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3052 | { |
3053 | return -EINVAL; | |
3054 | } | |
8c7c6e34 | 3055 | #endif |
d13d1443 | 3056 | |
bbec2e15 | 3057 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3058 | |
bbec2e15 RG |
3059 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3060 | unsigned long max, bool memsw) | |
628f4235 | 3061 | { |
3e32cb2e | 3062 | bool enlarge = false; |
bb4a7ea2 | 3063 | bool drained = false; |
3e32cb2e | 3064 | int ret; |
c054a78c YZ |
3065 | bool limits_invariant; |
3066 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3067 | |
3e32cb2e | 3068 | do { |
628f4235 KH |
3069 | if (signal_pending(current)) { |
3070 | ret = -EINTR; | |
3071 | break; | |
3072 | } | |
3e32cb2e | 3073 | |
bbec2e15 | 3074 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3075 | /* |
3076 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3077 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3078 | */ |
15b42562 | 3079 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3080 | max <= memcg->memsw.max; |
c054a78c | 3081 | if (!limits_invariant) { |
bbec2e15 | 3082 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3083 | ret = -EINVAL; |
8c7c6e34 KH |
3084 | break; |
3085 | } | |
bbec2e15 | 3086 | if (max > counter->max) |
3e32cb2e | 3087 | enlarge = true; |
bbec2e15 RG |
3088 | ret = page_counter_set_max(counter, max); |
3089 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3090 | |
3091 | if (!ret) | |
3092 | break; | |
3093 | ||
bb4a7ea2 SB |
3094 | if (!drained) { |
3095 | drain_all_stock(memcg); | |
3096 | drained = true; | |
3097 | continue; | |
3098 | } | |
3099 | ||
1ab5c056 AR |
3100 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
3101 | GFP_KERNEL, !memsw)) { | |
3102 | ret = -EBUSY; | |
3103 | break; | |
3104 | } | |
3105 | } while (true); | |
3e32cb2e | 3106 | |
3c11ecf4 KH |
3107 | if (!ret && enlarge) |
3108 | memcg_oom_recover(memcg); | |
3e32cb2e | 3109 | |
628f4235 KH |
3110 | return ret; |
3111 | } | |
3112 | ||
ef8f2327 | 3113 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3114 | gfp_t gfp_mask, |
3115 | unsigned long *total_scanned) | |
3116 | { | |
3117 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3118 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3119 | unsigned long reclaimed; |
3120 | int loop = 0; | |
ef8f2327 | 3121 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3122 | unsigned long excess; |
0608f43d AM |
3123 | unsigned long nr_scanned; |
3124 | ||
3125 | if (order > 0) | |
3126 | return 0; | |
3127 | ||
ef8f2327 | 3128 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
3129 | |
3130 | /* | |
3131 | * Do not even bother to check the largest node if the root | |
3132 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3133 | * are acceptable as soft limit is best effort anyway. | |
3134 | */ | |
bfc7228b | 3135 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3136 | return 0; |
3137 | ||
0608f43d AM |
3138 | /* |
3139 | * This loop can run a while, specially if mem_cgroup's continuously | |
3140 | * keep exceeding their soft limit and putting the system under | |
3141 | * pressure | |
3142 | */ | |
3143 | do { | |
3144 | if (next_mz) | |
3145 | mz = next_mz; | |
3146 | else | |
3147 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3148 | if (!mz) | |
3149 | break; | |
3150 | ||
3151 | nr_scanned = 0; | |
ef8f2327 | 3152 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
3153 | gfp_mask, &nr_scanned); |
3154 | nr_reclaimed += reclaimed; | |
3155 | *total_scanned += nr_scanned; | |
0a31bc97 | 3156 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3157 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3158 | |
3159 | /* | |
3160 | * If we failed to reclaim anything from this memory cgroup | |
3161 | * it is time to move on to the next cgroup | |
3162 | */ | |
3163 | next_mz = NULL; | |
bc2f2e7f VD |
3164 | if (!reclaimed) |
3165 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3166 | ||
3e32cb2e | 3167 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3168 | /* |
3169 | * One school of thought says that we should not add | |
3170 | * back the node to the tree if reclaim returns 0. | |
3171 | * But our reclaim could return 0, simply because due | |
3172 | * to priority we are exposing a smaller subset of | |
3173 | * memory to reclaim from. Consider this as a longer | |
3174 | * term TODO. | |
3175 | */ | |
3176 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3177 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3178 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3179 | css_put(&mz->memcg->css); |
3180 | loop++; | |
3181 | /* | |
3182 | * Could not reclaim anything and there are no more | |
3183 | * mem cgroups to try or we seem to be looping without | |
3184 | * reclaiming anything. | |
3185 | */ | |
3186 | if (!nr_reclaimed && | |
3187 | (next_mz == NULL || | |
3188 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3189 | break; | |
3190 | } while (!nr_reclaimed); | |
3191 | if (next_mz) | |
3192 | css_put(&next_mz->memcg->css); | |
3193 | return nr_reclaimed; | |
3194 | } | |
3195 | ||
ea280e7b TH |
3196 | /* |
3197 | * Test whether @memcg has children, dead or alive. Note that this | |
3198 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3199 | * returns %true if there are child csses according to the cgroup | |
3200 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3201 | */ | |
b5f99b53 GC |
3202 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3203 | { | |
ea280e7b TH |
3204 | bool ret; |
3205 | ||
ea280e7b TH |
3206 | rcu_read_lock(); |
3207 | ret = css_next_child(NULL, &memcg->css); | |
3208 | rcu_read_unlock(); | |
3209 | return ret; | |
b5f99b53 GC |
3210 | } |
3211 | ||
c26251f9 | 3212 | /* |
51038171 | 3213 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3214 | * |
3215 | * Caller is responsible for holding css reference for memcg. | |
3216 | */ | |
3217 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3218 | { | |
3219 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3220 | |
c1e862c1 KH |
3221 | /* we call try-to-free pages for make this cgroup empty */ |
3222 | lru_add_drain_all(); | |
d12c60f6 JS |
3223 | |
3224 | drain_all_stock(memcg); | |
3225 | ||
f817ed48 | 3226 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3227 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3228 | int progress; |
c1e862c1 | 3229 | |
c26251f9 MH |
3230 | if (signal_pending(current)) |
3231 | return -EINTR; | |
3232 | ||
b70a2a21 JW |
3233 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3234 | GFP_KERNEL, true); | |
c1e862c1 | 3235 | if (!progress) { |
f817ed48 | 3236 | nr_retries--; |
c1e862c1 | 3237 | /* maybe some writeback is necessary */ |
8aa7e847 | 3238 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3239 | } |
f817ed48 KH |
3240 | |
3241 | } | |
ab5196c2 MH |
3242 | |
3243 | return 0; | |
cc847582 KH |
3244 | } |
3245 | ||
6770c64e TH |
3246 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3247 | char *buf, size_t nbytes, | |
3248 | loff_t off) | |
c1e862c1 | 3249 | { |
6770c64e | 3250 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3251 | |
d8423011 MH |
3252 | if (mem_cgroup_is_root(memcg)) |
3253 | return -EINVAL; | |
6770c64e | 3254 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3255 | } |
3256 | ||
182446d0 TH |
3257 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3258 | struct cftype *cft) | |
18f59ea7 | 3259 | { |
182446d0 | 3260 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3261 | } |
3262 | ||
182446d0 TH |
3263 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3264 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3265 | { |
3266 | int retval = 0; | |
182446d0 | 3267 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3268 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3269 | |
567fb435 | 3270 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 3271 | return 0; |
567fb435 | 3272 | |
18f59ea7 | 3273 | /* |
af901ca1 | 3274 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3275 | * in the child subtrees. If it is unset, then the change can |
3276 | * occur, provided the current cgroup has no children. | |
3277 | * | |
3278 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3279 | * set if there are no children. | |
3280 | */ | |
c0ff4b85 | 3281 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3282 | (val == 1 || val == 0)) { |
ea280e7b | 3283 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3284 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3285 | else |
3286 | retval = -EBUSY; | |
3287 | } else | |
3288 | retval = -EINVAL; | |
567fb435 | 3289 | |
18f59ea7 BS |
3290 | return retval; |
3291 | } | |
3292 | ||
6f646156 | 3293 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3294 | { |
42a30035 | 3295 | unsigned long val; |
ce00a967 | 3296 | |
3e32cb2e | 3297 | if (mem_cgroup_is_root(memcg)) { |
42a30035 JW |
3298 | val = memcg_page_state(memcg, MEMCG_CACHE) + |
3299 | memcg_page_state(memcg, MEMCG_RSS); | |
3300 | if (swap) | |
3301 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3302 | } else { |
ce00a967 | 3303 | if (!swap) |
3e32cb2e | 3304 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3305 | else |
3e32cb2e | 3306 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3307 | } |
c12176d3 | 3308 | return val; |
ce00a967 JW |
3309 | } |
3310 | ||
3e32cb2e JW |
3311 | enum { |
3312 | RES_USAGE, | |
3313 | RES_LIMIT, | |
3314 | RES_MAX_USAGE, | |
3315 | RES_FAILCNT, | |
3316 | RES_SOFT_LIMIT, | |
3317 | }; | |
ce00a967 | 3318 | |
791badbd | 3319 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3320 | struct cftype *cft) |
8cdea7c0 | 3321 | { |
182446d0 | 3322 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3323 | struct page_counter *counter; |
af36f906 | 3324 | |
3e32cb2e | 3325 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3326 | case _MEM: |
3e32cb2e JW |
3327 | counter = &memcg->memory; |
3328 | break; | |
8c7c6e34 | 3329 | case _MEMSWAP: |
3e32cb2e JW |
3330 | counter = &memcg->memsw; |
3331 | break; | |
510fc4e1 | 3332 | case _KMEM: |
3e32cb2e | 3333 | counter = &memcg->kmem; |
510fc4e1 | 3334 | break; |
d55f90bf | 3335 | case _TCP: |
0db15298 | 3336 | counter = &memcg->tcpmem; |
d55f90bf | 3337 | break; |
8c7c6e34 KH |
3338 | default: |
3339 | BUG(); | |
8c7c6e34 | 3340 | } |
3e32cb2e JW |
3341 | |
3342 | switch (MEMFILE_ATTR(cft->private)) { | |
3343 | case RES_USAGE: | |
3344 | if (counter == &memcg->memory) | |
c12176d3 | 3345 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3346 | if (counter == &memcg->memsw) |
c12176d3 | 3347 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3348 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3349 | case RES_LIMIT: | |
bbec2e15 | 3350 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3351 | case RES_MAX_USAGE: |
3352 | return (u64)counter->watermark * PAGE_SIZE; | |
3353 | case RES_FAILCNT: | |
3354 | return counter->failcnt; | |
3355 | case RES_SOFT_LIMIT: | |
3356 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3357 | default: | |
3358 | BUG(); | |
3359 | } | |
8cdea7c0 | 3360 | } |
510fc4e1 | 3361 | |
4a87e2a2 | 3362 | static void memcg_flush_percpu_vmstats(struct mem_cgroup *memcg) |
c350a99e | 3363 | { |
4a87e2a2 | 3364 | unsigned long stat[MEMCG_NR_STAT] = {0}; |
c350a99e RG |
3365 | struct mem_cgroup *mi; |
3366 | int node, cpu, i; | |
c350a99e RG |
3367 | |
3368 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3369 | for (i = 0; i < MEMCG_NR_STAT; i++) |
6c1c2808 | 3370 | stat[i] += per_cpu(memcg->vmstats_percpu->stat[i], cpu); |
c350a99e RG |
3371 | |
3372 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
4a87e2a2 | 3373 | for (i = 0; i < MEMCG_NR_STAT; i++) |
c350a99e RG |
3374 | atomic_long_add(stat[i], &mi->vmstats[i]); |
3375 | ||
3376 | for_each_node(node) { | |
3377 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; | |
3378 | struct mem_cgroup_per_node *pi; | |
3379 | ||
4a87e2a2 | 3380 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3381 | stat[i] = 0; |
3382 | ||
3383 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3384 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
6c1c2808 SB |
3385 | stat[i] += per_cpu( |
3386 | pn->lruvec_stat_cpu->count[i], cpu); | |
c350a99e RG |
3387 | |
3388 | for (pi = pn; pi; pi = parent_nodeinfo(pi, node)) | |
4a87e2a2 | 3389 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3390 | atomic_long_add(stat[i], &pi->lruvec_stat[i]); |
3391 | } | |
3392 | } | |
3393 | ||
bb65f89b RG |
3394 | static void memcg_flush_percpu_vmevents(struct mem_cgroup *memcg) |
3395 | { | |
3396 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
3397 | struct mem_cgroup *mi; | |
3398 | int cpu, i; | |
3399 | ||
3400 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3401 | events[i] = 0; | |
3402 | ||
3403 | for_each_online_cpu(cpu) | |
3404 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
6c1c2808 SB |
3405 | events[i] += per_cpu(memcg->vmstats_percpu->events[i], |
3406 | cpu); | |
bb65f89b RG |
3407 | |
3408 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
3409 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3410 | atomic_long_add(events[i], &mi->vmevents[i]); | |
3411 | } | |
3412 | ||
84c07d11 | 3413 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3414 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3415 | { |
d6441637 VD |
3416 | int memcg_id; |
3417 | ||
b313aeee VD |
3418 | if (cgroup_memory_nokmem) |
3419 | return 0; | |
3420 | ||
2a4db7eb | 3421 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3422 | BUG_ON(memcg->kmem_state); |
d6441637 | 3423 | |
f3bb3043 | 3424 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3425 | if (memcg_id < 0) |
3426 | return memcg_id; | |
d6441637 | 3427 | |
ef12947c | 3428 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 3429 | /* |
567e9ab2 | 3430 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 3431 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
3432 | * guarantee no one starts accounting before all call sites are |
3433 | * patched. | |
3434 | */ | |
900a38f0 | 3435 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3436 | memcg->kmem_state = KMEM_ONLINE; |
bc2791f8 | 3437 | INIT_LIST_HEAD(&memcg->kmem_caches); |
0b8f73e1 JW |
3438 | |
3439 | return 0; | |
d6441637 VD |
3440 | } |
3441 | ||
8e0a8912 JW |
3442 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3443 | { | |
3444 | struct cgroup_subsys_state *css; | |
3445 | struct mem_cgroup *parent, *child; | |
3446 | int kmemcg_id; | |
3447 | ||
3448 | if (memcg->kmem_state != KMEM_ONLINE) | |
3449 | return; | |
3450 | /* | |
3451 | * Clear the online state before clearing memcg_caches array | |
3452 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
3453 | * guarantees that no cache will be created for this cgroup | |
3454 | * after we are done (see memcg_create_kmem_cache()). | |
3455 | */ | |
3456 | memcg->kmem_state = KMEM_ALLOCATED; | |
3457 | ||
8e0a8912 JW |
3458 | parent = parent_mem_cgroup(memcg); |
3459 | if (!parent) | |
3460 | parent = root_mem_cgroup; | |
3461 | ||
bee07b33 | 3462 | /* |
4a87e2a2 | 3463 | * Deactivate and reparent kmem_caches. |
bee07b33 | 3464 | */ |
fb2f2b0a RG |
3465 | memcg_deactivate_kmem_caches(memcg, parent); |
3466 | ||
3467 | kmemcg_id = memcg->kmemcg_id; | |
3468 | BUG_ON(kmemcg_id < 0); | |
3469 | ||
8e0a8912 JW |
3470 | /* |
3471 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3472 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3473 | * to ones of the parent. After we have finished, all list_lrus | |
3474 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3475 | * ordering is imposed by list_lru_node->lock taken by | |
3476 | * memcg_drain_all_list_lrus(). | |
3477 | */ | |
3a06bb78 | 3478 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3479 | css_for_each_descendant_pre(css, &memcg->css) { |
3480 | child = mem_cgroup_from_css(css); | |
3481 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3482 | child->kmemcg_id = parent->kmemcg_id; | |
3483 | if (!memcg->use_hierarchy) | |
3484 | break; | |
3485 | } | |
3a06bb78 TH |
3486 | rcu_read_unlock(); |
3487 | ||
9bec5c35 | 3488 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3489 | |
3490 | memcg_free_cache_id(kmemcg_id); | |
3491 | } | |
3492 | ||
3493 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3494 | { | |
0b8f73e1 JW |
3495 | /* css_alloc() failed, offlining didn't happen */ |
3496 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3497 | memcg_offline_kmem(memcg); | |
3498 | ||
8e0a8912 | 3499 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
f0a3a24b | 3500 | WARN_ON(!list_empty(&memcg->kmem_caches)); |
8e0a8912 | 3501 | static_branch_dec(&memcg_kmem_enabled_key); |
8e0a8912 | 3502 | } |
8e0a8912 | 3503 | } |
d6441637 | 3504 | #else |
0b8f73e1 | 3505 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3506 | { |
3507 | return 0; | |
3508 | } | |
3509 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3510 | { | |
3511 | } | |
3512 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3513 | { | |
3514 | } | |
84c07d11 | 3515 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3516 | |
bbec2e15 RG |
3517 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3518 | unsigned long max) | |
d6441637 | 3519 | { |
b313aeee | 3520 | int ret; |
127424c8 | 3521 | |
bbec2e15 RG |
3522 | mutex_lock(&memcg_max_mutex); |
3523 | ret = page_counter_set_max(&memcg->kmem, max); | |
3524 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3525 | return ret; |
d6441637 | 3526 | } |
510fc4e1 | 3527 | |
bbec2e15 | 3528 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3529 | { |
3530 | int ret; | |
3531 | ||
bbec2e15 | 3532 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3533 | |
bbec2e15 | 3534 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3535 | if (ret) |
3536 | goto out; | |
3537 | ||
0db15298 | 3538 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3539 | /* |
3540 | * The active flag needs to be written after the static_key | |
3541 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3542 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3543 | * for details, and note that we don't mark any socket as | |
3544 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3545 | * |
3546 | * We need to do this, because static_keys will span multiple | |
3547 | * sites, but we can't control their order. If we mark a socket | |
3548 | * as accounted, but the accounting functions are not patched in | |
3549 | * yet, we'll lose accounting. | |
3550 | * | |
2d758073 | 3551 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3552 | * because when this value change, the code to process it is not |
3553 | * patched in yet. | |
3554 | */ | |
3555 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3556 | memcg->tcpmem_active = true; |
d55f90bf VD |
3557 | } |
3558 | out: | |
bbec2e15 | 3559 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3560 | return ret; |
3561 | } | |
d55f90bf | 3562 | |
628f4235 KH |
3563 | /* |
3564 | * The user of this function is... | |
3565 | * RES_LIMIT. | |
3566 | */ | |
451af504 TH |
3567 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3568 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3569 | { |
451af504 | 3570 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3571 | unsigned long nr_pages; |
628f4235 KH |
3572 | int ret; |
3573 | ||
451af504 | 3574 | buf = strstrip(buf); |
650c5e56 | 3575 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3576 | if (ret) |
3577 | return ret; | |
af36f906 | 3578 | |
3e32cb2e | 3579 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3580 | case RES_LIMIT: |
4b3bde4c BS |
3581 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3582 | ret = -EINVAL; | |
3583 | break; | |
3584 | } | |
3e32cb2e JW |
3585 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3586 | case _MEM: | |
bbec2e15 | 3587 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3588 | break; |
3e32cb2e | 3589 | case _MEMSWAP: |
bbec2e15 | 3590 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3591 | break; |
3e32cb2e | 3592 | case _KMEM: |
0158115f MH |
3593 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " |
3594 | "Please report your usecase to linux-mm@kvack.org if you " | |
3595 | "depend on this functionality.\n"); | |
bbec2e15 | 3596 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3597 | break; |
d55f90bf | 3598 | case _TCP: |
bbec2e15 | 3599 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3600 | break; |
3e32cb2e | 3601 | } |
296c81d8 | 3602 | break; |
3e32cb2e JW |
3603 | case RES_SOFT_LIMIT: |
3604 | memcg->soft_limit = nr_pages; | |
3605 | ret = 0; | |
628f4235 KH |
3606 | break; |
3607 | } | |
451af504 | 3608 | return ret ?: nbytes; |
8cdea7c0 BS |
3609 | } |
3610 | ||
6770c64e TH |
3611 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3612 | size_t nbytes, loff_t off) | |
c84872e1 | 3613 | { |
6770c64e | 3614 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3615 | struct page_counter *counter; |
c84872e1 | 3616 | |
3e32cb2e JW |
3617 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3618 | case _MEM: | |
3619 | counter = &memcg->memory; | |
3620 | break; | |
3621 | case _MEMSWAP: | |
3622 | counter = &memcg->memsw; | |
3623 | break; | |
3624 | case _KMEM: | |
3625 | counter = &memcg->kmem; | |
3626 | break; | |
d55f90bf | 3627 | case _TCP: |
0db15298 | 3628 | counter = &memcg->tcpmem; |
d55f90bf | 3629 | break; |
3e32cb2e JW |
3630 | default: |
3631 | BUG(); | |
3632 | } | |
af36f906 | 3633 | |
3e32cb2e | 3634 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3635 | case RES_MAX_USAGE: |
3e32cb2e | 3636 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3637 | break; |
3638 | case RES_FAILCNT: | |
3e32cb2e | 3639 | counter->failcnt = 0; |
29f2a4da | 3640 | break; |
3e32cb2e JW |
3641 | default: |
3642 | BUG(); | |
29f2a4da | 3643 | } |
f64c3f54 | 3644 | |
6770c64e | 3645 | return nbytes; |
c84872e1 PE |
3646 | } |
3647 | ||
182446d0 | 3648 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3649 | struct cftype *cft) |
3650 | { | |
182446d0 | 3651 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3652 | } |
3653 | ||
02491447 | 3654 | #ifdef CONFIG_MMU |
182446d0 | 3655 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3656 | struct cftype *cft, u64 val) |
3657 | { | |
182446d0 | 3658 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3659 | |
1dfab5ab | 3660 | if (val & ~MOVE_MASK) |
7dc74be0 | 3661 | return -EINVAL; |
ee5e8472 | 3662 | |
7dc74be0 | 3663 | /* |
ee5e8472 GC |
3664 | * No kind of locking is needed in here, because ->can_attach() will |
3665 | * check this value once in the beginning of the process, and then carry | |
3666 | * on with stale data. This means that changes to this value will only | |
3667 | * affect task migrations starting after the change. | |
7dc74be0 | 3668 | */ |
c0ff4b85 | 3669 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3670 | return 0; |
3671 | } | |
02491447 | 3672 | #else |
182446d0 | 3673 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3674 | struct cftype *cft, u64 val) |
3675 | { | |
3676 | return -ENOSYS; | |
3677 | } | |
3678 | #endif | |
7dc74be0 | 3679 | |
406eb0c9 | 3680 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3681 | |
3682 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3683 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3684 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3685 | ||
3686 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
3687 | int nid, unsigned int lru_mask) | |
3688 | { | |
867e5e1d | 3689 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
3690 | unsigned long nr = 0; |
3691 | enum lru_list lru; | |
3692 | ||
3693 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3694 | ||
3695 | for_each_lru(lru) { | |
3696 | if (!(BIT(lru) & lru_mask)) | |
3697 | continue; | |
205b20cc | 3698 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); |
113b7dfd JW |
3699 | } |
3700 | return nr; | |
3701 | } | |
3702 | ||
3703 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
3704 | unsigned int lru_mask) | |
3705 | { | |
3706 | unsigned long nr = 0; | |
3707 | enum lru_list lru; | |
3708 | ||
3709 | for_each_lru(lru) { | |
3710 | if (!(BIT(lru) & lru_mask)) | |
3711 | continue; | |
205b20cc | 3712 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); |
113b7dfd JW |
3713 | } |
3714 | return nr; | |
3715 | } | |
3716 | ||
2da8ca82 | 3717 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3718 | { |
25485de6 GT |
3719 | struct numa_stat { |
3720 | const char *name; | |
3721 | unsigned int lru_mask; | |
3722 | }; | |
3723 | ||
3724 | static const struct numa_stat stats[] = { | |
3725 | { "total", LRU_ALL }, | |
3726 | { "file", LRU_ALL_FILE }, | |
3727 | { "anon", LRU_ALL_ANON }, | |
3728 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3729 | }; | |
3730 | const struct numa_stat *stat; | |
406eb0c9 | 3731 | int nid; |
25485de6 | 3732 | unsigned long nr; |
aa9694bb | 3733 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3734 | |
25485de6 GT |
3735 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3736 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3737 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3738 | for_each_node_state(nid, N_MEMORY) { | |
3739 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3740 | stat->lru_mask); | |
3741 | seq_printf(m, " N%d=%lu", nid, nr); | |
3742 | } | |
3743 | seq_putc(m, '\n'); | |
406eb0c9 | 3744 | } |
406eb0c9 | 3745 | |
071aee13 YH |
3746 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3747 | struct mem_cgroup *iter; | |
3748 | ||
3749 | nr = 0; | |
3750 | for_each_mem_cgroup_tree(iter, memcg) | |
3751 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3752 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3753 | for_each_node_state(nid, N_MEMORY) { | |
3754 | nr = 0; | |
3755 | for_each_mem_cgroup_tree(iter, memcg) | |
3756 | nr += mem_cgroup_node_nr_lru_pages( | |
3757 | iter, nid, stat->lru_mask); | |
3758 | seq_printf(m, " N%d=%lu", nid, nr); | |
3759 | } | |
3760 | seq_putc(m, '\n'); | |
406eb0c9 | 3761 | } |
406eb0c9 | 3762 | |
406eb0c9 YH |
3763 | return 0; |
3764 | } | |
3765 | #endif /* CONFIG_NUMA */ | |
3766 | ||
c8713d0b JW |
3767 | static const unsigned int memcg1_stats[] = { |
3768 | MEMCG_CACHE, | |
3769 | MEMCG_RSS, | |
3770 | MEMCG_RSS_HUGE, | |
3771 | NR_SHMEM, | |
3772 | NR_FILE_MAPPED, | |
3773 | NR_FILE_DIRTY, | |
3774 | NR_WRITEBACK, | |
3775 | MEMCG_SWAP, | |
3776 | }; | |
3777 | ||
3778 | static const char *const memcg1_stat_names[] = { | |
3779 | "cache", | |
3780 | "rss", | |
3781 | "rss_huge", | |
3782 | "shmem", | |
3783 | "mapped_file", | |
3784 | "dirty", | |
3785 | "writeback", | |
3786 | "swap", | |
3787 | }; | |
3788 | ||
df0e53d0 | 3789 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3790 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
3791 | PGPGIN, |
3792 | PGPGOUT, | |
3793 | PGFAULT, | |
3794 | PGMAJFAULT, | |
3795 | }; | |
3796 | ||
2da8ca82 | 3797 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3798 | { |
aa9694bb | 3799 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 3800 | unsigned long memory, memsw; |
af7c4b0e JW |
3801 | struct mem_cgroup *mi; |
3802 | unsigned int i; | |
406eb0c9 | 3803 | |
71cd3113 | 3804 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 3805 | |
71cd3113 JW |
3806 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
3807 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) | |
1dd3a273 | 3808 | continue; |
71cd3113 | 3809 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], |
205b20cc | 3810 | memcg_page_state_local(memcg, memcg1_stats[i]) * |
71cd3113 | 3811 | PAGE_SIZE); |
1dd3a273 | 3812 | } |
7b854121 | 3813 | |
df0e53d0 | 3814 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d | 3815 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
205b20cc | 3816 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
3817 | |
3818 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 3819 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
205b20cc | 3820 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 3821 | PAGE_SIZE); |
af7c4b0e | 3822 | |
14067bb3 | 3823 | /* Hierarchical information */ |
3e32cb2e JW |
3824 | memory = memsw = PAGE_COUNTER_MAX; |
3825 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
3826 | memory = min(memory, READ_ONCE(mi->memory.max)); |
3827 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 3828 | } |
3e32cb2e JW |
3829 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3830 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3831 | if (do_memsw_account()) |
3e32cb2e JW |
3832 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3833 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3834 | |
8de7ecc6 | 3835 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
71cd3113 | 3836 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3837 | continue; |
8de7ecc6 | 3838 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
dd923990 YS |
3839 | (u64)memcg_page_state(memcg, memcg1_stats[i]) * |
3840 | PAGE_SIZE); | |
af7c4b0e JW |
3841 | } |
3842 | ||
8de7ecc6 | 3843 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d KK |
3844 | seq_printf(m, "total_%s %llu\n", |
3845 | vm_event_name(memcg1_events[i]), | |
dd923990 | 3846 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 3847 | |
8de7ecc6 | 3848 | for (i = 0; i < NR_LRU_LISTS; i++) |
ebc5d83d | 3849 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
42a30035 JW |
3850 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
3851 | PAGE_SIZE); | |
14067bb3 | 3852 | |
7f016ee8 | 3853 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 3854 | { |
ef8f2327 MG |
3855 | pg_data_t *pgdat; |
3856 | struct mem_cgroup_per_node *mz; | |
89abfab1 | 3857 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3858 | unsigned long recent_rotated[2] = {0, 0}; |
3859 | unsigned long recent_scanned[2] = {0, 0}; | |
3860 | ||
ef8f2327 MG |
3861 | for_each_online_pgdat(pgdat) { |
3862 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
3863 | rstat = &mz->lruvec.reclaim_stat; | |
7f016ee8 | 3864 | |
ef8f2327 MG |
3865 | recent_rotated[0] += rstat->recent_rotated[0]; |
3866 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3867 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3868 | recent_scanned[1] += rstat->recent_scanned[1]; | |
3869 | } | |
78ccf5b5 JW |
3870 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3871 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3872 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3873 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3874 | } |
3875 | #endif | |
3876 | ||
d2ceb9b7 KH |
3877 | return 0; |
3878 | } | |
3879 | ||
182446d0 TH |
3880 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3881 | struct cftype *cft) | |
a7885eb8 | 3882 | { |
182446d0 | 3883 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3884 | |
1f4c025b | 3885 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3886 | } |
3887 | ||
182446d0 TH |
3888 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3889 | struct cftype *cft, u64 val) | |
a7885eb8 | 3890 | { |
182446d0 | 3891 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3892 | |
3dae7fec | 3893 | if (val > 100) |
a7885eb8 KM |
3894 | return -EINVAL; |
3895 | ||
14208b0e | 3896 | if (css->parent) |
3dae7fec JW |
3897 | memcg->swappiness = val; |
3898 | else | |
3899 | vm_swappiness = val; | |
068b38c1 | 3900 | |
a7885eb8 KM |
3901 | return 0; |
3902 | } | |
3903 | ||
2e72b634 KS |
3904 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3905 | { | |
3906 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3907 | unsigned long usage; |
2e72b634 KS |
3908 | int i; |
3909 | ||
3910 | rcu_read_lock(); | |
3911 | if (!swap) | |
2c488db2 | 3912 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3913 | else |
2c488db2 | 3914 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3915 | |
3916 | if (!t) | |
3917 | goto unlock; | |
3918 | ||
ce00a967 | 3919 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3920 | |
3921 | /* | |
748dad36 | 3922 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3923 | * If it's not true, a threshold was crossed after last |
3924 | * call of __mem_cgroup_threshold(). | |
3925 | */ | |
5407a562 | 3926 | i = t->current_threshold; |
2e72b634 KS |
3927 | |
3928 | /* | |
3929 | * Iterate backward over array of thresholds starting from | |
3930 | * current_threshold and check if a threshold is crossed. | |
3931 | * If none of thresholds below usage is crossed, we read | |
3932 | * only one element of the array here. | |
3933 | */ | |
3934 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3935 | eventfd_signal(t->entries[i].eventfd, 1); | |
3936 | ||
3937 | /* i = current_threshold + 1 */ | |
3938 | i++; | |
3939 | ||
3940 | /* | |
3941 | * Iterate forward over array of thresholds starting from | |
3942 | * current_threshold+1 and check if a threshold is crossed. | |
3943 | * If none of thresholds above usage is crossed, we read | |
3944 | * only one element of the array here. | |
3945 | */ | |
3946 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3947 | eventfd_signal(t->entries[i].eventfd, 1); | |
3948 | ||
3949 | /* Update current_threshold */ | |
5407a562 | 3950 | t->current_threshold = i - 1; |
2e72b634 KS |
3951 | unlock: |
3952 | rcu_read_unlock(); | |
3953 | } | |
3954 | ||
3955 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3956 | { | |
ad4ca5f4 KS |
3957 | while (memcg) { |
3958 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3959 | if (do_memsw_account()) |
ad4ca5f4 KS |
3960 | __mem_cgroup_threshold(memcg, true); |
3961 | ||
3962 | memcg = parent_mem_cgroup(memcg); | |
3963 | } | |
2e72b634 KS |
3964 | } |
3965 | ||
3966 | static int compare_thresholds(const void *a, const void *b) | |
3967 | { | |
3968 | const struct mem_cgroup_threshold *_a = a; | |
3969 | const struct mem_cgroup_threshold *_b = b; | |
3970 | ||
2bff24a3 GT |
3971 | if (_a->threshold > _b->threshold) |
3972 | return 1; | |
3973 | ||
3974 | if (_a->threshold < _b->threshold) | |
3975 | return -1; | |
3976 | ||
3977 | return 0; | |
2e72b634 KS |
3978 | } |
3979 | ||
c0ff4b85 | 3980 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3981 | { |
3982 | struct mem_cgroup_eventfd_list *ev; | |
3983 | ||
2bcf2e92 MH |
3984 | spin_lock(&memcg_oom_lock); |
3985 | ||
c0ff4b85 | 3986 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3987 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3988 | |
3989 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3990 | return 0; |
3991 | } | |
3992 | ||
c0ff4b85 | 3993 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3994 | { |
7d74b06f KH |
3995 | struct mem_cgroup *iter; |
3996 | ||
c0ff4b85 | 3997 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3998 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3999 | } |
4000 | ||
59b6f873 | 4001 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4002 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4003 | { |
2c488db2 KS |
4004 | struct mem_cgroup_thresholds *thresholds; |
4005 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4006 | unsigned long threshold; |
4007 | unsigned long usage; | |
2c488db2 | 4008 | int i, size, ret; |
2e72b634 | 4009 | |
650c5e56 | 4010 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4011 | if (ret) |
4012 | return ret; | |
4013 | ||
4014 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4015 | |
05b84301 | 4016 | if (type == _MEM) { |
2c488db2 | 4017 | thresholds = &memcg->thresholds; |
ce00a967 | 4018 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4019 | } else if (type == _MEMSWAP) { |
2c488db2 | 4020 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4021 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4022 | } else |
2e72b634 KS |
4023 | BUG(); |
4024 | ||
2e72b634 | 4025 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4026 | if (thresholds->primary) |
2e72b634 KS |
4027 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4028 | ||
2c488db2 | 4029 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4030 | |
4031 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4032 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4033 | if (!new) { |
2e72b634 KS |
4034 | ret = -ENOMEM; |
4035 | goto unlock; | |
4036 | } | |
2c488db2 | 4037 | new->size = size; |
2e72b634 KS |
4038 | |
4039 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4040 | if (thresholds->primary) { |
4041 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4042 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4043 | } |
4044 | ||
2e72b634 | 4045 | /* Add new threshold */ |
2c488db2 KS |
4046 | new->entries[size - 1].eventfd = eventfd; |
4047 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4048 | |
4049 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4050 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4051 | compare_thresholds, NULL); |
4052 | ||
4053 | /* Find current threshold */ | |
2c488db2 | 4054 | new->current_threshold = -1; |
2e72b634 | 4055 | for (i = 0; i < size; i++) { |
748dad36 | 4056 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4057 | /* |
2c488db2 KS |
4058 | * new->current_threshold will not be used until |
4059 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4060 | * it here. |
4061 | */ | |
2c488db2 | 4062 | ++new->current_threshold; |
748dad36 SZ |
4063 | } else |
4064 | break; | |
2e72b634 KS |
4065 | } |
4066 | ||
2c488db2 KS |
4067 | /* Free old spare buffer and save old primary buffer as spare */ |
4068 | kfree(thresholds->spare); | |
4069 | thresholds->spare = thresholds->primary; | |
4070 | ||
4071 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4072 | |
907860ed | 4073 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4074 | synchronize_rcu(); |
4075 | ||
2e72b634 KS |
4076 | unlock: |
4077 | mutex_unlock(&memcg->thresholds_lock); | |
4078 | ||
4079 | return ret; | |
4080 | } | |
4081 | ||
59b6f873 | 4082 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4083 | struct eventfd_ctx *eventfd, const char *args) |
4084 | { | |
59b6f873 | 4085 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4086 | } |
4087 | ||
59b6f873 | 4088 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4089 | struct eventfd_ctx *eventfd, const char *args) |
4090 | { | |
59b6f873 | 4091 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4092 | } |
4093 | ||
59b6f873 | 4094 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4095 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4096 | { |
2c488db2 KS |
4097 | struct mem_cgroup_thresholds *thresholds; |
4098 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4099 | unsigned long usage; |
7d36665a | 4100 | int i, j, size, entries; |
2e72b634 KS |
4101 | |
4102 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4103 | |
4104 | if (type == _MEM) { | |
2c488db2 | 4105 | thresholds = &memcg->thresholds; |
ce00a967 | 4106 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4107 | } else if (type == _MEMSWAP) { |
2c488db2 | 4108 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4109 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4110 | } else |
2e72b634 KS |
4111 | BUG(); |
4112 | ||
371528ca AV |
4113 | if (!thresholds->primary) |
4114 | goto unlock; | |
4115 | ||
2e72b634 KS |
4116 | /* Check if a threshold crossed before removing */ |
4117 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4118 | ||
4119 | /* Calculate new number of threshold */ | |
7d36665a | 4120 | size = entries = 0; |
2c488db2 KS |
4121 | for (i = 0; i < thresholds->primary->size; i++) { |
4122 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4123 | size++; |
7d36665a CX |
4124 | else |
4125 | entries++; | |
2e72b634 KS |
4126 | } |
4127 | ||
2c488db2 | 4128 | new = thresholds->spare; |
907860ed | 4129 | |
7d36665a CX |
4130 | /* If no items related to eventfd have been cleared, nothing to do */ |
4131 | if (!entries) | |
4132 | goto unlock; | |
4133 | ||
2e72b634 KS |
4134 | /* Set thresholds array to NULL if we don't have thresholds */ |
4135 | if (!size) { | |
2c488db2 KS |
4136 | kfree(new); |
4137 | new = NULL; | |
907860ed | 4138 | goto swap_buffers; |
2e72b634 KS |
4139 | } |
4140 | ||
2c488db2 | 4141 | new->size = size; |
2e72b634 KS |
4142 | |
4143 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4144 | new->current_threshold = -1; |
4145 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4146 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4147 | continue; |
4148 | ||
2c488db2 | 4149 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4150 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4151 | /* |
2c488db2 | 4152 | * new->current_threshold will not be used |
2e72b634 KS |
4153 | * until rcu_assign_pointer(), so it's safe to increment |
4154 | * it here. | |
4155 | */ | |
2c488db2 | 4156 | ++new->current_threshold; |
2e72b634 KS |
4157 | } |
4158 | j++; | |
4159 | } | |
4160 | ||
907860ed | 4161 | swap_buffers: |
2c488db2 KS |
4162 | /* Swap primary and spare array */ |
4163 | thresholds->spare = thresholds->primary; | |
8c757763 | 4164 | |
2c488db2 | 4165 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4166 | |
907860ed | 4167 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4168 | synchronize_rcu(); |
6611d8d7 MC |
4169 | |
4170 | /* If all events are unregistered, free the spare array */ | |
4171 | if (!new) { | |
4172 | kfree(thresholds->spare); | |
4173 | thresholds->spare = NULL; | |
4174 | } | |
371528ca | 4175 | unlock: |
2e72b634 | 4176 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4177 | } |
c1e862c1 | 4178 | |
59b6f873 | 4179 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4180 | struct eventfd_ctx *eventfd) |
4181 | { | |
59b6f873 | 4182 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4183 | } |
4184 | ||
59b6f873 | 4185 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4186 | struct eventfd_ctx *eventfd) |
4187 | { | |
59b6f873 | 4188 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4189 | } |
4190 | ||
59b6f873 | 4191 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4192 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4193 | { |
9490ff27 | 4194 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4195 | |
9490ff27 KH |
4196 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4197 | if (!event) | |
4198 | return -ENOMEM; | |
4199 | ||
1af8efe9 | 4200 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4201 | |
4202 | event->eventfd = eventfd; | |
4203 | list_add(&event->list, &memcg->oom_notify); | |
4204 | ||
4205 | /* already in OOM ? */ | |
c2b42d3c | 4206 | if (memcg->under_oom) |
9490ff27 | 4207 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4208 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4209 | |
4210 | return 0; | |
4211 | } | |
4212 | ||
59b6f873 | 4213 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4214 | struct eventfd_ctx *eventfd) |
9490ff27 | 4215 | { |
9490ff27 | 4216 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4217 | |
1af8efe9 | 4218 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4219 | |
c0ff4b85 | 4220 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4221 | if (ev->eventfd == eventfd) { |
4222 | list_del(&ev->list); | |
4223 | kfree(ev); | |
4224 | } | |
4225 | } | |
4226 | ||
1af8efe9 | 4227 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4228 | } |
4229 | ||
2da8ca82 | 4230 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4231 | { |
aa9694bb | 4232 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4233 | |
791badbd | 4234 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4235 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4236 | seq_printf(sf, "oom_kill %lu\n", |
4237 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4238 | return 0; |
4239 | } | |
4240 | ||
182446d0 | 4241 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4242 | struct cftype *cft, u64 val) |
4243 | { | |
182446d0 | 4244 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4245 | |
4246 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4247 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4248 | return -EINVAL; |
4249 | ||
c0ff4b85 | 4250 | memcg->oom_kill_disable = val; |
4d845ebf | 4251 | if (!val) |
c0ff4b85 | 4252 | memcg_oom_recover(memcg); |
3dae7fec | 4253 | |
3c11ecf4 KH |
4254 | return 0; |
4255 | } | |
4256 | ||
52ebea74 TH |
4257 | #ifdef CONFIG_CGROUP_WRITEBACK |
4258 | ||
3a8e9ac8 TH |
4259 | #include <trace/events/writeback.h> |
4260 | ||
841710aa TH |
4261 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4262 | { | |
4263 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4264 | } | |
4265 | ||
4266 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4267 | { | |
4268 | wb_domain_exit(&memcg->cgwb_domain); | |
4269 | } | |
4270 | ||
2529bb3a TH |
4271 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4272 | { | |
4273 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4274 | } | |
4275 | ||
841710aa TH |
4276 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4277 | { | |
4278 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4279 | ||
4280 | if (!memcg->css.parent) | |
4281 | return NULL; | |
4282 | ||
4283 | return &memcg->cgwb_domain; | |
4284 | } | |
4285 | ||
0b3d6e6f GT |
4286 | /* |
4287 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
4288 | * Keep in sync with memcg_exact_page(). | |
4289 | */ | |
4290 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
4291 | { | |
871789d4 | 4292 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
4293 | int cpu; |
4294 | ||
4295 | for_each_online_cpu(cpu) | |
871789d4 | 4296 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
4297 | if (x < 0) |
4298 | x = 0; | |
4299 | return x; | |
4300 | } | |
4301 | ||
c2aa723a TH |
4302 | /** |
4303 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4304 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4305 | * @pfilepages: out parameter for number of file pages |
4306 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4307 | * @pdirty: out parameter for number of dirty pages |
4308 | * @pwriteback: out parameter for number of pages under writeback | |
4309 | * | |
c5edf9cd TH |
4310 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4311 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4312 | * is a bit more involved. | |
c2aa723a | 4313 | * |
c5edf9cd TH |
4314 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4315 | * headroom is calculated as the lowest headroom of itself and the | |
4316 | * ancestors. Note that this doesn't consider the actual amount of | |
4317 | * available memory in the system. The caller should further cap | |
4318 | * *@pheadroom accordingly. | |
c2aa723a | 4319 | */ |
c5edf9cd TH |
4320 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4321 | unsigned long *pheadroom, unsigned long *pdirty, | |
4322 | unsigned long *pwriteback) | |
c2aa723a TH |
4323 | { |
4324 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4325 | struct mem_cgroup *parent; | |
c2aa723a | 4326 | |
0b3d6e6f | 4327 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a TH |
4328 | |
4329 | /* this should eventually include NR_UNSTABLE_NFS */ | |
0b3d6e6f | 4330 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4331 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4332 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4333 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4334 | |
c2aa723a | 4335 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 4336 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
f6f989c5 | 4337 | READ_ONCE(memcg->high)); |
c2aa723a TH |
4338 | unsigned long used = page_counter_read(&memcg->memory); |
4339 | ||
c5edf9cd | 4340 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4341 | memcg = parent; |
4342 | } | |
c2aa723a TH |
4343 | } |
4344 | ||
97b27821 TH |
4345 | /* |
4346 | * Foreign dirty flushing | |
4347 | * | |
4348 | * There's an inherent mismatch between memcg and writeback. The former | |
4349 | * trackes ownership per-page while the latter per-inode. This was a | |
4350 | * deliberate design decision because honoring per-page ownership in the | |
4351 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4352 | * and deemed unnecessary given that write-sharing an inode across | |
4353 | * different cgroups isn't a common use-case. | |
4354 | * | |
4355 | * Combined with inode majority-writer ownership switching, this works well | |
4356 | * enough in most cases but there are some pathological cases. For | |
4357 | * example, let's say there are two cgroups A and B which keep writing to | |
4358 | * different but confined parts of the same inode. B owns the inode and | |
4359 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4360 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4361 | * triggering background writeback. A will be slowed down without a way to | |
4362 | * make writeback of the dirty pages happen. | |
4363 | * | |
4364 | * Conditions like the above can lead to a cgroup getting repatedly and | |
4365 | * severely throttled after making some progress after each | |
4366 | * dirty_expire_interval while the underyling IO device is almost | |
4367 | * completely idle. | |
4368 | * | |
4369 | * Solving this problem completely requires matching the ownership tracking | |
4370 | * granularities between memcg and writeback in either direction. However, | |
4371 | * the more egregious behaviors can be avoided by simply remembering the | |
4372 | * most recent foreign dirtying events and initiating remote flushes on | |
4373 | * them when local writeback isn't enough to keep the memory clean enough. | |
4374 | * | |
4375 | * The following two functions implement such mechanism. When a foreign | |
4376 | * page - a page whose memcg and writeback ownerships don't match - is | |
4377 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4378 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4379 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4380 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4381 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4382 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4383 | * limited to MEMCG_CGWB_FRN_CNT. | |
4384 | * | |
4385 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4386 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4387 | * records are lockless and racy. | |
4388 | */ | |
4389 | void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, | |
4390 | struct bdi_writeback *wb) | |
4391 | { | |
4392 | struct mem_cgroup *memcg = page->mem_cgroup; | |
4393 | struct memcg_cgwb_frn *frn; | |
4394 | u64 now = get_jiffies_64(); | |
4395 | u64 oldest_at = now; | |
4396 | int oldest = -1; | |
4397 | int i; | |
4398 | ||
3a8e9ac8 TH |
4399 | trace_track_foreign_dirty(page, wb); |
4400 | ||
97b27821 TH |
4401 | /* |
4402 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4403 | * using it. If not replace the oldest one which isn't being | |
4404 | * written out. | |
4405 | */ | |
4406 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4407 | frn = &memcg->cgwb_frn[i]; | |
4408 | if (frn->bdi_id == wb->bdi->id && | |
4409 | frn->memcg_id == wb->memcg_css->id) | |
4410 | break; | |
4411 | if (time_before64(frn->at, oldest_at) && | |
4412 | atomic_read(&frn->done.cnt) == 1) { | |
4413 | oldest = i; | |
4414 | oldest_at = frn->at; | |
4415 | } | |
4416 | } | |
4417 | ||
4418 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4419 | /* | |
4420 | * Re-using an existing one. Update timestamp lazily to | |
4421 | * avoid making the cacheline hot. We want them to be | |
4422 | * reasonably up-to-date and significantly shorter than | |
4423 | * dirty_expire_interval as that's what expires the record. | |
4424 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4425 | */ | |
4426 | unsigned long update_intv = | |
4427 | min_t(unsigned long, HZ, | |
4428 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4429 | ||
4430 | if (time_before64(frn->at, now - update_intv)) | |
4431 | frn->at = now; | |
4432 | } else if (oldest >= 0) { | |
4433 | /* replace the oldest free one */ | |
4434 | frn = &memcg->cgwb_frn[oldest]; | |
4435 | frn->bdi_id = wb->bdi->id; | |
4436 | frn->memcg_id = wb->memcg_css->id; | |
4437 | frn->at = now; | |
4438 | } | |
4439 | } | |
4440 | ||
4441 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4442 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4443 | { | |
4444 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4445 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4446 | u64 now = jiffies_64; | |
4447 | int i; | |
4448 | ||
4449 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4450 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4451 | ||
4452 | /* | |
4453 | * If the record is older than dirty_expire_interval, | |
4454 | * writeback on it has already started. No need to kick it | |
4455 | * off again. Also, don't start a new one if there's | |
4456 | * already one in flight. | |
4457 | */ | |
4458 | if (time_after64(frn->at, now - intv) && | |
4459 | atomic_read(&frn->done.cnt) == 1) { | |
4460 | frn->at = 0; | |
3a8e9ac8 | 4461 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
97b27821 TH |
4462 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0, |
4463 | WB_REASON_FOREIGN_FLUSH, | |
4464 | &frn->done); | |
4465 | } | |
4466 | } | |
4467 | } | |
4468 | ||
841710aa TH |
4469 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4470 | ||
4471 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4472 | { | |
4473 | return 0; | |
4474 | } | |
4475 | ||
4476 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4477 | { | |
4478 | } | |
4479 | ||
2529bb3a TH |
4480 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4481 | { | |
4482 | } | |
4483 | ||
52ebea74 TH |
4484 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4485 | ||
3bc942f3 TH |
4486 | /* |
4487 | * DO NOT USE IN NEW FILES. | |
4488 | * | |
4489 | * "cgroup.event_control" implementation. | |
4490 | * | |
4491 | * This is way over-engineered. It tries to support fully configurable | |
4492 | * events for each user. Such level of flexibility is completely | |
4493 | * unnecessary especially in the light of the planned unified hierarchy. | |
4494 | * | |
4495 | * Please deprecate this and replace with something simpler if at all | |
4496 | * possible. | |
4497 | */ | |
4498 | ||
79bd9814 TH |
4499 | /* |
4500 | * Unregister event and free resources. | |
4501 | * | |
4502 | * Gets called from workqueue. | |
4503 | */ | |
3bc942f3 | 4504 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4505 | { |
3bc942f3 TH |
4506 | struct mem_cgroup_event *event = |
4507 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4508 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4509 | |
4510 | remove_wait_queue(event->wqh, &event->wait); | |
4511 | ||
59b6f873 | 4512 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4513 | |
4514 | /* Notify userspace the event is going away. */ | |
4515 | eventfd_signal(event->eventfd, 1); | |
4516 | ||
4517 | eventfd_ctx_put(event->eventfd); | |
4518 | kfree(event); | |
59b6f873 | 4519 | css_put(&memcg->css); |
79bd9814 TH |
4520 | } |
4521 | ||
4522 | /* | |
a9a08845 | 4523 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4524 | * |
4525 | * Called with wqh->lock held and interrupts disabled. | |
4526 | */ | |
ac6424b9 | 4527 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4528 | int sync, void *key) |
79bd9814 | 4529 | { |
3bc942f3 TH |
4530 | struct mem_cgroup_event *event = |
4531 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4532 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4533 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4534 | |
a9a08845 | 4535 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4536 | /* |
4537 | * If the event has been detached at cgroup removal, we | |
4538 | * can simply return knowing the other side will cleanup | |
4539 | * for us. | |
4540 | * | |
4541 | * We can't race against event freeing since the other | |
4542 | * side will require wqh->lock via remove_wait_queue(), | |
4543 | * which we hold. | |
4544 | */ | |
fba94807 | 4545 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4546 | if (!list_empty(&event->list)) { |
4547 | list_del_init(&event->list); | |
4548 | /* | |
4549 | * We are in atomic context, but cgroup_event_remove() | |
4550 | * may sleep, so we have to call it in workqueue. | |
4551 | */ | |
4552 | schedule_work(&event->remove); | |
4553 | } | |
fba94807 | 4554 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4555 | } |
4556 | ||
4557 | return 0; | |
4558 | } | |
4559 | ||
3bc942f3 | 4560 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4561 | wait_queue_head_t *wqh, poll_table *pt) |
4562 | { | |
3bc942f3 TH |
4563 | struct mem_cgroup_event *event = |
4564 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4565 | |
4566 | event->wqh = wqh; | |
4567 | add_wait_queue(wqh, &event->wait); | |
4568 | } | |
4569 | ||
4570 | /* | |
3bc942f3 TH |
4571 | * DO NOT USE IN NEW FILES. |
4572 | * | |
79bd9814 TH |
4573 | * Parse input and register new cgroup event handler. |
4574 | * | |
4575 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4576 | * Interpretation of args is defined by control file implementation. | |
4577 | */ | |
451af504 TH |
4578 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4579 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4580 | { |
451af504 | 4581 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4582 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4583 | struct mem_cgroup_event *event; |
79bd9814 TH |
4584 | struct cgroup_subsys_state *cfile_css; |
4585 | unsigned int efd, cfd; | |
4586 | struct fd efile; | |
4587 | struct fd cfile; | |
fba94807 | 4588 | const char *name; |
79bd9814 TH |
4589 | char *endp; |
4590 | int ret; | |
4591 | ||
451af504 TH |
4592 | buf = strstrip(buf); |
4593 | ||
4594 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4595 | if (*endp != ' ') |
4596 | return -EINVAL; | |
451af504 | 4597 | buf = endp + 1; |
79bd9814 | 4598 | |
451af504 | 4599 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4600 | if ((*endp != ' ') && (*endp != '\0')) |
4601 | return -EINVAL; | |
451af504 | 4602 | buf = endp + 1; |
79bd9814 TH |
4603 | |
4604 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4605 | if (!event) | |
4606 | return -ENOMEM; | |
4607 | ||
59b6f873 | 4608 | event->memcg = memcg; |
79bd9814 | 4609 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4610 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4611 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4612 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4613 | |
4614 | efile = fdget(efd); | |
4615 | if (!efile.file) { | |
4616 | ret = -EBADF; | |
4617 | goto out_kfree; | |
4618 | } | |
4619 | ||
4620 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4621 | if (IS_ERR(event->eventfd)) { | |
4622 | ret = PTR_ERR(event->eventfd); | |
4623 | goto out_put_efile; | |
4624 | } | |
4625 | ||
4626 | cfile = fdget(cfd); | |
4627 | if (!cfile.file) { | |
4628 | ret = -EBADF; | |
4629 | goto out_put_eventfd; | |
4630 | } | |
4631 | ||
4632 | /* the process need read permission on control file */ | |
4633 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4634 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4635 | if (ret < 0) | |
4636 | goto out_put_cfile; | |
4637 | ||
fba94807 TH |
4638 | /* |
4639 | * Determine the event callbacks and set them in @event. This used | |
4640 | * to be done via struct cftype but cgroup core no longer knows | |
4641 | * about these events. The following is crude but the whole thing | |
4642 | * is for compatibility anyway. | |
3bc942f3 TH |
4643 | * |
4644 | * DO NOT ADD NEW FILES. | |
fba94807 | 4645 | */ |
b583043e | 4646 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4647 | |
4648 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4649 | event->register_event = mem_cgroup_usage_register_event; | |
4650 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4651 | } else if (!strcmp(name, "memory.oom_control")) { | |
4652 | event->register_event = mem_cgroup_oom_register_event; | |
4653 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4654 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4655 | event->register_event = vmpressure_register_event; | |
4656 | event->unregister_event = vmpressure_unregister_event; | |
4657 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4658 | event->register_event = memsw_cgroup_usage_register_event; |
4659 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4660 | } else { |
4661 | ret = -EINVAL; | |
4662 | goto out_put_cfile; | |
4663 | } | |
4664 | ||
79bd9814 | 4665 | /* |
b5557c4c TH |
4666 | * Verify @cfile should belong to @css. Also, remaining events are |
4667 | * automatically removed on cgroup destruction but the removal is | |
4668 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4669 | */ |
b583043e | 4670 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4671 | &memory_cgrp_subsys); |
79bd9814 | 4672 | ret = -EINVAL; |
5a17f543 | 4673 | if (IS_ERR(cfile_css)) |
79bd9814 | 4674 | goto out_put_cfile; |
5a17f543 TH |
4675 | if (cfile_css != css) { |
4676 | css_put(cfile_css); | |
79bd9814 | 4677 | goto out_put_cfile; |
5a17f543 | 4678 | } |
79bd9814 | 4679 | |
451af504 | 4680 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4681 | if (ret) |
4682 | goto out_put_css; | |
4683 | ||
9965ed17 | 4684 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4685 | |
fba94807 TH |
4686 | spin_lock(&memcg->event_list_lock); |
4687 | list_add(&event->list, &memcg->event_list); | |
4688 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4689 | |
4690 | fdput(cfile); | |
4691 | fdput(efile); | |
4692 | ||
451af504 | 4693 | return nbytes; |
79bd9814 TH |
4694 | |
4695 | out_put_css: | |
b5557c4c | 4696 | css_put(css); |
79bd9814 TH |
4697 | out_put_cfile: |
4698 | fdput(cfile); | |
4699 | out_put_eventfd: | |
4700 | eventfd_ctx_put(event->eventfd); | |
4701 | out_put_efile: | |
4702 | fdput(efile); | |
4703 | out_kfree: | |
4704 | kfree(event); | |
4705 | ||
4706 | return ret; | |
4707 | } | |
4708 | ||
241994ed | 4709 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4710 | { |
0eea1030 | 4711 | .name = "usage_in_bytes", |
8c7c6e34 | 4712 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4713 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4714 | }, |
c84872e1 PE |
4715 | { |
4716 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4717 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4718 | .write = mem_cgroup_reset, |
791badbd | 4719 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4720 | }, |
8cdea7c0 | 4721 | { |
0eea1030 | 4722 | .name = "limit_in_bytes", |
8c7c6e34 | 4723 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4724 | .write = mem_cgroup_write, |
791badbd | 4725 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4726 | }, |
296c81d8 BS |
4727 | { |
4728 | .name = "soft_limit_in_bytes", | |
4729 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4730 | .write = mem_cgroup_write, |
791badbd | 4731 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4732 | }, |
8cdea7c0 BS |
4733 | { |
4734 | .name = "failcnt", | |
8c7c6e34 | 4735 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4736 | .write = mem_cgroup_reset, |
791badbd | 4737 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4738 | }, |
d2ceb9b7 KH |
4739 | { |
4740 | .name = "stat", | |
2da8ca82 | 4741 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4742 | }, |
c1e862c1 KH |
4743 | { |
4744 | .name = "force_empty", | |
6770c64e | 4745 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4746 | }, |
18f59ea7 BS |
4747 | { |
4748 | .name = "use_hierarchy", | |
4749 | .write_u64 = mem_cgroup_hierarchy_write, | |
4750 | .read_u64 = mem_cgroup_hierarchy_read, | |
4751 | }, | |
79bd9814 | 4752 | { |
3bc942f3 | 4753 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4754 | .write = memcg_write_event_control, |
7dbdb199 | 4755 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4756 | }, |
a7885eb8 KM |
4757 | { |
4758 | .name = "swappiness", | |
4759 | .read_u64 = mem_cgroup_swappiness_read, | |
4760 | .write_u64 = mem_cgroup_swappiness_write, | |
4761 | }, | |
7dc74be0 DN |
4762 | { |
4763 | .name = "move_charge_at_immigrate", | |
4764 | .read_u64 = mem_cgroup_move_charge_read, | |
4765 | .write_u64 = mem_cgroup_move_charge_write, | |
4766 | }, | |
9490ff27 KH |
4767 | { |
4768 | .name = "oom_control", | |
2da8ca82 | 4769 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4770 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4771 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4772 | }, | |
70ddf637 AV |
4773 | { |
4774 | .name = "pressure_level", | |
70ddf637 | 4775 | }, |
406eb0c9 YH |
4776 | #ifdef CONFIG_NUMA |
4777 | { | |
4778 | .name = "numa_stat", | |
2da8ca82 | 4779 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4780 | }, |
4781 | #endif | |
510fc4e1 GC |
4782 | { |
4783 | .name = "kmem.limit_in_bytes", | |
4784 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4785 | .write = mem_cgroup_write, |
791badbd | 4786 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4787 | }, |
4788 | { | |
4789 | .name = "kmem.usage_in_bytes", | |
4790 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4791 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4792 | }, |
4793 | { | |
4794 | .name = "kmem.failcnt", | |
4795 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4796 | .write = mem_cgroup_reset, |
791badbd | 4797 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4798 | }, |
4799 | { | |
4800 | .name = "kmem.max_usage_in_bytes", | |
4801 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4802 | .write = mem_cgroup_reset, |
791badbd | 4803 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4804 | }, |
a87425a3 YS |
4805 | #if defined(CONFIG_MEMCG_KMEM) && \ |
4806 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) | |
749c5415 GC |
4807 | { |
4808 | .name = "kmem.slabinfo", | |
bc2791f8 TH |
4809 | .seq_start = memcg_slab_start, |
4810 | .seq_next = memcg_slab_next, | |
4811 | .seq_stop = memcg_slab_stop, | |
b047501c | 4812 | .seq_show = memcg_slab_show, |
749c5415 GC |
4813 | }, |
4814 | #endif | |
d55f90bf VD |
4815 | { |
4816 | .name = "kmem.tcp.limit_in_bytes", | |
4817 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4818 | .write = mem_cgroup_write, | |
4819 | .read_u64 = mem_cgroup_read_u64, | |
4820 | }, | |
4821 | { | |
4822 | .name = "kmem.tcp.usage_in_bytes", | |
4823 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4824 | .read_u64 = mem_cgroup_read_u64, | |
4825 | }, | |
4826 | { | |
4827 | .name = "kmem.tcp.failcnt", | |
4828 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4829 | .write = mem_cgroup_reset, | |
4830 | .read_u64 = mem_cgroup_read_u64, | |
4831 | }, | |
4832 | { | |
4833 | .name = "kmem.tcp.max_usage_in_bytes", | |
4834 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4835 | .write = mem_cgroup_reset, | |
4836 | .read_u64 = mem_cgroup_read_u64, | |
4837 | }, | |
6bc10349 | 4838 | { }, /* terminate */ |
af36f906 | 4839 | }; |
8c7c6e34 | 4840 | |
73f576c0 JW |
4841 | /* |
4842 | * Private memory cgroup IDR | |
4843 | * | |
4844 | * Swap-out records and page cache shadow entries need to store memcg | |
4845 | * references in constrained space, so we maintain an ID space that is | |
4846 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4847 | * memory-controlled cgroups to 64k. | |
4848 | * | |
4849 | * However, there usually are many references to the oflline CSS after | |
4850 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4851 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4852 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4853 | * relatively small ID space and prevent the creation of new cgroups | |
4854 | * even when there are much fewer than 64k cgroups - possibly none. | |
4855 | * | |
4856 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4857 | * be freed and recycled when it's no longer needed, which is usually | |
4858 | * when the CSS is offlined. | |
4859 | * | |
4860 | * The only exception to that are records of swapped out tmpfs/shmem | |
4861 | * pages that need to be attributed to live ancestors on swapin. But | |
4862 | * those references are manageable from userspace. | |
4863 | */ | |
4864 | ||
4865 | static DEFINE_IDR(mem_cgroup_idr); | |
4866 | ||
7e97de0b KT |
4867 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
4868 | { | |
4869 | if (memcg->id.id > 0) { | |
4870 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
4871 | memcg->id.id = 0; | |
4872 | } | |
4873 | } | |
4874 | ||
c1514c0a VF |
4875 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
4876 | unsigned int n) | |
73f576c0 | 4877 | { |
1c2d479a | 4878 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
4879 | } |
4880 | ||
615d66c3 | 4881 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4882 | { |
1c2d479a | 4883 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 4884 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
4885 | |
4886 | /* Memcg ID pins CSS */ | |
4887 | css_put(&memcg->css); | |
4888 | } | |
4889 | } | |
4890 | ||
615d66c3 VD |
4891 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
4892 | { | |
4893 | mem_cgroup_id_put_many(memcg, 1); | |
4894 | } | |
4895 | ||
73f576c0 JW |
4896 | /** |
4897 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4898 | * @id: the memcg id to look up | |
4899 | * | |
4900 | * Caller must hold rcu_read_lock(). | |
4901 | */ | |
4902 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4903 | { | |
4904 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4905 | return idr_find(&mem_cgroup_idr, id); | |
4906 | } | |
4907 | ||
ef8f2327 | 4908 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4909 | { |
4910 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 4911 | int tmp = node; |
1ecaab2b KH |
4912 | /* |
4913 | * This routine is called against possible nodes. | |
4914 | * But it's BUG to call kmalloc() against offline node. | |
4915 | * | |
4916 | * TODO: this routine can waste much memory for nodes which will | |
4917 | * never be onlined. It's better to use memory hotplug callback | |
4918 | * function. | |
4919 | */ | |
41e3355d KH |
4920 | if (!node_state(node, N_NORMAL_MEMORY)) |
4921 | tmp = -1; | |
17295c88 | 4922 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4923 | if (!pn) |
4924 | return 1; | |
1ecaab2b | 4925 | |
815744d7 JW |
4926 | pn->lruvec_stat_local = alloc_percpu(struct lruvec_stat); |
4927 | if (!pn->lruvec_stat_local) { | |
4928 | kfree(pn); | |
4929 | return 1; | |
4930 | } | |
4931 | ||
a983b5eb JW |
4932 | pn->lruvec_stat_cpu = alloc_percpu(struct lruvec_stat); |
4933 | if (!pn->lruvec_stat_cpu) { | |
815744d7 | 4934 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c JW |
4935 | kfree(pn); |
4936 | return 1; | |
4937 | } | |
4938 | ||
ef8f2327 MG |
4939 | lruvec_init(&pn->lruvec); |
4940 | pn->usage_in_excess = 0; | |
4941 | pn->on_tree = false; | |
4942 | pn->memcg = memcg; | |
4943 | ||
54f72fe0 | 4944 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4945 | return 0; |
4946 | } | |
4947 | ||
ef8f2327 | 4948 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4949 | { |
00f3ca2c JW |
4950 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
4951 | ||
4eaf431f MH |
4952 | if (!pn) |
4953 | return; | |
4954 | ||
a983b5eb | 4955 | free_percpu(pn->lruvec_stat_cpu); |
815744d7 | 4956 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c | 4957 | kfree(pn); |
1ecaab2b KH |
4958 | } |
4959 | ||
40e952f9 | 4960 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4961 | { |
c8b2a36f | 4962 | int node; |
59927fb9 | 4963 | |
c8b2a36f | 4964 | for_each_node(node) |
ef8f2327 | 4965 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 4966 | free_percpu(memcg->vmstats_percpu); |
815744d7 | 4967 | free_percpu(memcg->vmstats_local); |
8ff69e2c | 4968 | kfree(memcg); |
59927fb9 | 4969 | } |
3afe36b1 | 4970 | |
40e952f9 TE |
4971 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
4972 | { | |
4973 | memcg_wb_domain_exit(memcg); | |
7961eee3 SB |
4974 | /* |
4975 | * Flush percpu vmstats and vmevents to guarantee the value correctness | |
4976 | * on parent's and all ancestor levels. | |
4977 | */ | |
4a87e2a2 | 4978 | memcg_flush_percpu_vmstats(memcg); |
7961eee3 | 4979 | memcg_flush_percpu_vmevents(memcg); |
40e952f9 TE |
4980 | __mem_cgroup_free(memcg); |
4981 | } | |
4982 | ||
0b8f73e1 | 4983 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4984 | { |
d142e3e6 | 4985 | struct mem_cgroup *memcg; |
b9726c26 | 4986 | unsigned int size; |
6d12e2d8 | 4987 | int node; |
97b27821 | 4988 | int __maybe_unused i; |
8cdea7c0 | 4989 | |
0b8f73e1 JW |
4990 | size = sizeof(struct mem_cgroup); |
4991 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4992 | ||
4993 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4994 | if (!memcg) |
0b8f73e1 JW |
4995 | return NULL; |
4996 | ||
73f576c0 JW |
4997 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
4998 | 1, MEM_CGROUP_ID_MAX, | |
4999 | GFP_KERNEL); | |
5000 | if (memcg->id.id < 0) | |
5001 | goto fail; | |
5002 | ||
815744d7 JW |
5003 | memcg->vmstats_local = alloc_percpu(struct memcg_vmstats_percpu); |
5004 | if (!memcg->vmstats_local) | |
5005 | goto fail; | |
5006 | ||
871789d4 CD |
5007 | memcg->vmstats_percpu = alloc_percpu(struct memcg_vmstats_percpu); |
5008 | if (!memcg->vmstats_percpu) | |
0b8f73e1 | 5009 | goto fail; |
78fb7466 | 5010 | |
3ed28fa1 | 5011 | for_each_node(node) |
ef8f2327 | 5012 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5013 | goto fail; |
f64c3f54 | 5014 | |
0b8f73e1 JW |
5015 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5016 | goto fail; | |
28dbc4b6 | 5017 | |
f7e1cb6e | 5018 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5019 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5020 | mutex_init(&memcg->thresholds_lock); |
5021 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5022 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5023 | INIT_LIST_HEAD(&memcg->event_list); |
5024 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5025 | memcg->socket_pressure = jiffies; |
84c07d11 | 5026 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5027 | memcg->kmemcg_id = -1; |
900a38f0 | 5028 | #endif |
52ebea74 TH |
5029 | #ifdef CONFIG_CGROUP_WRITEBACK |
5030 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5031 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5032 | memcg->cgwb_frn[i].done = | |
5033 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5034 | #endif |
5035 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5036 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5037 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5038 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5039 | #endif |
73f576c0 | 5040 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
5041 | return memcg; |
5042 | fail: | |
7e97de0b | 5043 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5044 | __mem_cgroup_free(memcg); |
0b8f73e1 | 5045 | return NULL; |
d142e3e6 GC |
5046 | } |
5047 | ||
0b8f73e1 JW |
5048 | static struct cgroup_subsys_state * __ref |
5049 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5050 | { |
0b8f73e1 JW |
5051 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
5052 | struct mem_cgroup *memcg; | |
5053 | long error = -ENOMEM; | |
d142e3e6 | 5054 | |
0b8f73e1 JW |
5055 | memcg = mem_cgroup_alloc(); |
5056 | if (!memcg) | |
5057 | return ERR_PTR(error); | |
d142e3e6 | 5058 | |
f6f989c5 | 5059 | WRITE_ONCE(memcg->high, PAGE_COUNTER_MAX); |
0b8f73e1 JW |
5060 | memcg->soft_limit = PAGE_COUNTER_MAX; |
5061 | if (parent) { | |
5062 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5063 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
5064 | } | |
5065 | if (parent && parent->use_hierarchy) { | |
5066 | memcg->use_hierarchy = true; | |
3e32cb2e | 5067 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5068 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
5069 | page_counter_init(&memcg->memsw, &parent->memsw); |
5070 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 5071 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5072 | } else { |
3e32cb2e | 5073 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 5074 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
5075 | page_counter_init(&memcg->memsw, NULL); |
5076 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 5077 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
5078 | /* |
5079 | * Deeper hierachy with use_hierarchy == false doesn't make | |
5080 | * much sense so let cgroup subsystem know about this | |
5081 | * unfortunate state in our controller. | |
5082 | */ | |
d142e3e6 | 5083 | if (parent != root_mem_cgroup) |
073219e9 | 5084 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 5085 | } |
d6441637 | 5086 | |
0b8f73e1 JW |
5087 | /* The following stuff does not apply to the root */ |
5088 | if (!parent) { | |
fb2f2b0a RG |
5089 | #ifdef CONFIG_MEMCG_KMEM |
5090 | INIT_LIST_HEAD(&memcg->kmem_caches); | |
5091 | #endif | |
0b8f73e1 JW |
5092 | root_mem_cgroup = memcg; |
5093 | return &memcg->css; | |
5094 | } | |
5095 | ||
b313aeee | 5096 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
5097 | if (error) |
5098 | goto fail; | |
127424c8 | 5099 | |
f7e1cb6e | 5100 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5101 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5102 | |
0b8f73e1 JW |
5103 | return &memcg->css; |
5104 | fail: | |
7e97de0b | 5105 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 5106 | mem_cgroup_free(memcg); |
ea3a9645 | 5107 | return ERR_PTR(-ENOMEM); |
0b8f73e1 JW |
5108 | } |
5109 | ||
73f576c0 | 5110 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5111 | { |
58fa2a55 VD |
5112 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5113 | ||
0a4465d3 KT |
5114 | /* |
5115 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
5116 | * by the time the maps are allocated. So, we allocate maps | |
5117 | * here, when for_each_mem_cgroup() can't skip it. | |
5118 | */ | |
5119 | if (memcg_alloc_shrinker_maps(memcg)) { | |
5120 | mem_cgroup_id_remove(memcg); | |
5121 | return -ENOMEM; | |
5122 | } | |
5123 | ||
73f576c0 | 5124 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5125 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5126 | css_get(css); |
2f7dd7a4 | 5127 | return 0; |
8cdea7c0 BS |
5128 | } |
5129 | ||
eb95419b | 5130 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5131 | { |
eb95419b | 5132 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5133 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5134 | |
5135 | /* | |
5136 | * Unregister events and notify userspace. | |
5137 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5138 | * directory to avoid race between userspace and kernelspace. | |
5139 | */ | |
fba94807 TH |
5140 | spin_lock(&memcg->event_list_lock); |
5141 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5142 | list_del_init(&event->list); |
5143 | schedule_work(&event->remove); | |
5144 | } | |
fba94807 | 5145 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5146 | |
bf8d5d52 | 5147 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5148 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5149 | |
567e9ab2 | 5150 | memcg_offline_kmem(memcg); |
52ebea74 | 5151 | wb_memcg_offline(memcg); |
73f576c0 | 5152 | |
591edfb1 RG |
5153 | drain_all_stock(memcg); |
5154 | ||
73f576c0 | 5155 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5156 | } |
5157 | ||
6df38689 VD |
5158 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5159 | { | |
5160 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5161 | ||
5162 | invalidate_reclaim_iterators(memcg); | |
5163 | } | |
5164 | ||
eb95419b | 5165 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5166 | { |
eb95419b | 5167 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5168 | int __maybe_unused i; |
c268e994 | 5169 | |
97b27821 TH |
5170 | #ifdef CONFIG_CGROUP_WRITEBACK |
5171 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5172 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5173 | #endif | |
f7e1cb6e | 5174 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5175 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5176 | |
0db15298 | 5177 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5178 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5179 | |
0b8f73e1 JW |
5180 | vmpressure_cleanup(&memcg->vmpressure); |
5181 | cancel_work_sync(&memcg->high_work); | |
5182 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 5183 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 5184 | memcg_free_kmem(memcg); |
0b8f73e1 | 5185 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5186 | } |
5187 | ||
1ced953b TH |
5188 | /** |
5189 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5190 | * @css: the target css | |
5191 | * | |
5192 | * Reset the states of the mem_cgroup associated with @css. This is | |
5193 | * invoked when the userland requests disabling on the default hierarchy | |
5194 | * but the memcg is pinned through dependency. The memcg should stop | |
5195 | * applying policies and should revert to the vanilla state as it may be | |
5196 | * made visible again. | |
5197 | * | |
5198 | * The current implementation only resets the essential configurations. | |
5199 | * This needs to be expanded to cover all the visible parts. | |
5200 | */ | |
5201 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5202 | { | |
5203 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5204 | ||
bbec2e15 RG |
5205 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5206 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
5207 | page_counter_set_max(&memcg->memsw, PAGE_COUNTER_MAX); | |
5208 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); | |
5209 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5210 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5211 | page_counter_set_low(&memcg->memory, 0); |
f6f989c5 | 5212 | WRITE_ONCE(memcg->high, PAGE_COUNTER_MAX); |
24d404dc | 5213 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 5214 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5215 | } |
5216 | ||
02491447 | 5217 | #ifdef CONFIG_MMU |
7dc74be0 | 5218 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5219 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5220 | { |
05b84301 | 5221 | int ret; |
9476db97 | 5222 | |
d0164adc MG |
5223 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5224 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5225 | if (!ret) { |
854ffa8d | 5226 | mc.precharge += count; |
854ffa8d DN |
5227 | return ret; |
5228 | } | |
9476db97 | 5229 | |
3674534b | 5230 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5231 | while (count--) { |
3674534b | 5232 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5233 | if (ret) |
38c5d72f | 5234 | return ret; |
854ffa8d | 5235 | mc.precharge++; |
9476db97 | 5236 | cond_resched(); |
854ffa8d | 5237 | } |
9476db97 | 5238 | return 0; |
4ffef5fe DN |
5239 | } |
5240 | ||
4ffef5fe DN |
5241 | union mc_target { |
5242 | struct page *page; | |
02491447 | 5243 | swp_entry_t ent; |
4ffef5fe DN |
5244 | }; |
5245 | ||
4ffef5fe | 5246 | enum mc_target_type { |
8d32ff84 | 5247 | MC_TARGET_NONE = 0, |
4ffef5fe | 5248 | MC_TARGET_PAGE, |
02491447 | 5249 | MC_TARGET_SWAP, |
c733a828 | 5250 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5251 | }; |
5252 | ||
90254a65 DN |
5253 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5254 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5255 | { |
25b2995a | 5256 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5257 | |
90254a65 DN |
5258 | if (!page || !page_mapped(page)) |
5259 | return NULL; | |
5260 | if (PageAnon(page)) { | |
1dfab5ab | 5261 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5262 | return NULL; |
1dfab5ab JW |
5263 | } else { |
5264 | if (!(mc.flags & MOVE_FILE)) | |
5265 | return NULL; | |
5266 | } | |
90254a65 DN |
5267 | if (!get_page_unless_zero(page)) |
5268 | return NULL; | |
5269 | ||
5270 | return page; | |
5271 | } | |
5272 | ||
c733a828 | 5273 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5274 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5275 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5276 | { |
90254a65 DN |
5277 | struct page *page = NULL; |
5278 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5279 | ||
1dfab5ab | 5280 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 5281 | return NULL; |
c733a828 JG |
5282 | |
5283 | /* | |
5284 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
5285 | * a device and because they are not accessible by CPU they are store | |
5286 | * as special swap entry in the CPU page table. | |
5287 | */ | |
5288 | if (is_device_private_entry(ent)) { | |
5289 | page = device_private_entry_to_page(ent); | |
5290 | /* | |
5291 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
5292 | * a refcount of 1 when free (unlike normal page) | |
5293 | */ | |
5294 | if (!page_ref_add_unless(page, 1, 1)) | |
5295 | return NULL; | |
5296 | return page; | |
5297 | } | |
5298 | ||
4b91355e KH |
5299 | /* |
5300 | * Because lookup_swap_cache() updates some statistics counter, | |
5301 | * we call find_get_page() with swapper_space directly. | |
5302 | */ | |
f6ab1f7f | 5303 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
7941d214 | 5304 | if (do_memsw_account()) |
90254a65 DN |
5305 | entry->val = ent.val; |
5306 | ||
5307 | return page; | |
5308 | } | |
4b91355e KH |
5309 | #else |
5310 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5311 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5312 | { |
5313 | return NULL; | |
5314 | } | |
5315 | #endif | |
90254a65 | 5316 | |
87946a72 DN |
5317 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5318 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5319 | { | |
5320 | struct page *page = NULL; | |
87946a72 DN |
5321 | struct address_space *mapping; |
5322 | pgoff_t pgoff; | |
5323 | ||
5324 | if (!vma->vm_file) /* anonymous vma */ | |
5325 | return NULL; | |
1dfab5ab | 5326 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5327 | return NULL; |
5328 | ||
87946a72 | 5329 | mapping = vma->vm_file->f_mapping; |
0661a336 | 5330 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
5331 | |
5332 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5333 | #ifdef CONFIG_SWAP |
5334 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
5335 | if (shmem_mapping(mapping)) { |
5336 | page = find_get_entry(mapping, pgoff); | |
3159f943 | 5337 | if (xa_is_value(page)) { |
139b6a6f | 5338 | swp_entry_t swp = radix_to_swp_entry(page); |
7941d214 | 5339 | if (do_memsw_account()) |
139b6a6f | 5340 | *entry = swp; |
f6ab1f7f HY |
5341 | page = find_get_page(swap_address_space(swp), |
5342 | swp_offset(swp)); | |
139b6a6f JW |
5343 | } |
5344 | } else | |
5345 | page = find_get_page(mapping, pgoff); | |
5346 | #else | |
5347 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 5348 | #endif |
87946a72 DN |
5349 | return page; |
5350 | } | |
5351 | ||
b1b0deab CG |
5352 | /** |
5353 | * mem_cgroup_move_account - move account of the page | |
5354 | * @page: the page | |
25843c2b | 5355 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5356 | * @from: mem_cgroup which the page is moved from. |
5357 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5358 | * | |
3ac808fd | 5359 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5360 | * |
5361 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5362 | * from old cgroup. | |
5363 | */ | |
5364 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5365 | bool compound, |
b1b0deab CG |
5366 | struct mem_cgroup *from, |
5367 | struct mem_cgroup *to) | |
5368 | { | |
ae8af438 KK |
5369 | struct lruvec *from_vec, *to_vec; |
5370 | struct pglist_data *pgdat; | |
b1b0deab | 5371 | unsigned long flags; |
f627c2f5 | 5372 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 5373 | int ret; |
c4843a75 | 5374 | bool anon; |
b1b0deab CG |
5375 | |
5376 | VM_BUG_ON(from == to); | |
5377 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5378 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5379 | |
5380 | /* | |
6a93ca8f | 5381 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 5382 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 5383 | */ |
f627c2f5 | 5384 | ret = -EBUSY; |
b1b0deab CG |
5385 | if (!trylock_page(page)) |
5386 | goto out; | |
5387 | ||
5388 | ret = -EINVAL; | |
5389 | if (page->mem_cgroup != from) | |
5390 | goto out_unlock; | |
5391 | ||
c4843a75 GT |
5392 | anon = PageAnon(page); |
5393 | ||
ae8af438 | 5394 | pgdat = page_pgdat(page); |
867e5e1d JW |
5395 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5396 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5397 | |
b1b0deab CG |
5398 | spin_lock_irqsave(&from->move_lock, flags); |
5399 | ||
c4843a75 | 5400 | if (!anon && page_mapped(page)) { |
ae8af438 KK |
5401 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); |
5402 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
b1b0deab CG |
5403 | } |
5404 | ||
c4843a75 GT |
5405 | /* |
5406 | * move_lock grabbed above and caller set from->moving_account, so | |
ccda7f43 | 5407 | * mod_memcg_page_state will serialize updates to PageDirty. |
c4843a75 GT |
5408 | * So mapping should be stable for dirty pages. |
5409 | */ | |
5410 | if (!anon && PageDirty(page)) { | |
5411 | struct address_space *mapping = page_mapping(page); | |
5412 | ||
5413 | if (mapping_cap_account_dirty(mapping)) { | |
ae8af438 KK |
5414 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, -nr_pages); |
5415 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, nr_pages); | |
c4843a75 GT |
5416 | } |
5417 | } | |
5418 | ||
b1b0deab | 5419 | if (PageWriteback(page)) { |
ae8af438 KK |
5420 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5421 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5422 | } |
5423 | ||
5424 | /* | |
5425 | * It is safe to change page->mem_cgroup here because the page | |
5426 | * is referenced, charged, and isolated - we can't race with | |
5427 | * uncharging, charging, migration, or LRU putback. | |
5428 | */ | |
5429 | ||
5430 | /* caller should have done css_get */ | |
5431 | page->mem_cgroup = to; | |
87eaceb3 | 5432 | |
b1b0deab CG |
5433 | spin_unlock_irqrestore(&from->move_lock, flags); |
5434 | ||
5435 | ret = 0; | |
5436 | ||
5437 | local_irq_disable(); | |
f627c2f5 | 5438 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 5439 | memcg_check_events(to, page); |
f627c2f5 | 5440 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
5441 | memcg_check_events(from, page); |
5442 | local_irq_enable(); | |
5443 | out_unlock: | |
5444 | unlock_page(page); | |
5445 | out: | |
5446 | return ret; | |
5447 | } | |
5448 | ||
7cf7806c LR |
5449 | /** |
5450 | * get_mctgt_type - get target type of moving charge | |
5451 | * @vma: the vma the pte to be checked belongs | |
5452 | * @addr: the address corresponding to the pte to be checked | |
5453 | * @ptent: the pte to be checked | |
5454 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5455 | * | |
5456 | * Returns | |
5457 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5458 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5459 | * move charge. if @target is not NULL, the page is stored in target->page | |
5460 | * with extra refcnt got(Callers should handle it). | |
5461 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5462 | * target for charge migration. if @target is not NULL, the entry is stored | |
5463 | * in target->ent. | |
25b2995a CH |
5464 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
5465 | * (so ZONE_DEVICE page and thus not on the lru). | |
df6ad698 JG |
5466 | * For now we such page is charge like a regular page would be as for all |
5467 | * intent and purposes it is just special memory taking the place of a | |
5468 | * regular page. | |
c733a828 JG |
5469 | * |
5470 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5471 | * |
5472 | * Called with pte lock held. | |
5473 | */ | |
5474 | ||
8d32ff84 | 5475 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5476 | unsigned long addr, pte_t ptent, union mc_target *target) |
5477 | { | |
5478 | struct page *page = NULL; | |
8d32ff84 | 5479 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5480 | swp_entry_t ent = { .val = 0 }; |
5481 | ||
5482 | if (pte_present(ptent)) | |
5483 | page = mc_handle_present_pte(vma, addr, ptent); | |
5484 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5485 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5486 | else if (pte_none(ptent)) |
87946a72 | 5487 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5488 | |
5489 | if (!page && !ent.val) | |
8d32ff84 | 5490 | return ret; |
02491447 | 5491 | if (page) { |
02491447 | 5492 | /* |
0a31bc97 | 5493 | * Do only loose check w/o serialization. |
1306a85a | 5494 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5495 | * not under LRU exclusion. |
02491447 | 5496 | */ |
1306a85a | 5497 | if (page->mem_cgroup == mc.from) { |
02491447 | 5498 | ret = MC_TARGET_PAGE; |
25b2995a | 5499 | if (is_device_private_page(page)) |
c733a828 | 5500 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5501 | if (target) |
5502 | target->page = page; | |
5503 | } | |
5504 | if (!ret || !target) | |
5505 | put_page(page); | |
5506 | } | |
3e14a57b HY |
5507 | /* |
5508 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5509 | * But we cannot move a tail-page in a THP. | |
5510 | */ | |
5511 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5512 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5513 | ret = MC_TARGET_SWAP; |
5514 | if (target) | |
5515 | target->ent = ent; | |
4ffef5fe | 5516 | } |
4ffef5fe DN |
5517 | return ret; |
5518 | } | |
5519 | ||
12724850 NH |
5520 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5521 | /* | |
d6810d73 HY |
5522 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5523 | * not support them for now. | |
12724850 NH |
5524 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5525 | */ | |
5526 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5527 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5528 | { | |
5529 | struct page *page = NULL; | |
12724850 NH |
5530 | enum mc_target_type ret = MC_TARGET_NONE; |
5531 | ||
84c3fc4e ZY |
5532 | if (unlikely(is_swap_pmd(pmd))) { |
5533 | VM_BUG_ON(thp_migration_supported() && | |
5534 | !is_pmd_migration_entry(pmd)); | |
5535 | return ret; | |
5536 | } | |
12724850 | 5537 | page = pmd_page(pmd); |
309381fe | 5538 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5539 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5540 | return ret; |
1306a85a | 5541 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
5542 | ret = MC_TARGET_PAGE; |
5543 | if (target) { | |
5544 | get_page(page); | |
5545 | target->page = page; | |
5546 | } | |
5547 | } | |
5548 | return ret; | |
5549 | } | |
5550 | #else | |
5551 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5552 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5553 | { | |
5554 | return MC_TARGET_NONE; | |
5555 | } | |
5556 | #endif | |
5557 | ||
4ffef5fe DN |
5558 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5559 | unsigned long addr, unsigned long end, | |
5560 | struct mm_walk *walk) | |
5561 | { | |
26bcd64a | 5562 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5563 | pte_t *pte; |
5564 | spinlock_t *ptl; | |
5565 | ||
b6ec57f4 KS |
5566 | ptl = pmd_trans_huge_lock(pmd, vma); |
5567 | if (ptl) { | |
c733a828 JG |
5568 | /* |
5569 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5570 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5571 | * this might change. | |
c733a828 | 5572 | */ |
12724850 NH |
5573 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5574 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5575 | spin_unlock(ptl); |
1a5a9906 | 5576 | return 0; |
12724850 | 5577 | } |
03319327 | 5578 | |
45f83cef AA |
5579 | if (pmd_trans_unstable(pmd)) |
5580 | return 0; | |
4ffef5fe DN |
5581 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5582 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5583 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5584 | mc.precharge++; /* increment precharge temporarily */ |
5585 | pte_unmap_unlock(pte - 1, ptl); | |
5586 | cond_resched(); | |
5587 | ||
7dc74be0 DN |
5588 | return 0; |
5589 | } | |
5590 | ||
7b86ac33 CH |
5591 | static const struct mm_walk_ops precharge_walk_ops = { |
5592 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5593 | }; | |
5594 | ||
4ffef5fe DN |
5595 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5596 | { | |
5597 | unsigned long precharge; | |
4ffef5fe | 5598 | |
dfe076b0 | 5599 | down_read(&mm->mmap_sem); |
7b86ac33 | 5600 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
dfe076b0 | 5601 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5602 | |
5603 | precharge = mc.precharge; | |
5604 | mc.precharge = 0; | |
5605 | ||
5606 | return precharge; | |
5607 | } | |
5608 | ||
4ffef5fe DN |
5609 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5610 | { | |
dfe076b0 DN |
5611 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5612 | ||
5613 | VM_BUG_ON(mc.moving_task); | |
5614 | mc.moving_task = current; | |
5615 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5616 | } |
5617 | ||
dfe076b0 DN |
5618 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5619 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5620 | { |
2bd9bb20 KH |
5621 | struct mem_cgroup *from = mc.from; |
5622 | struct mem_cgroup *to = mc.to; | |
5623 | ||
4ffef5fe | 5624 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5625 | if (mc.precharge) { |
00501b53 | 5626 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5627 | mc.precharge = 0; |
5628 | } | |
5629 | /* | |
5630 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5631 | * we must uncharge here. | |
5632 | */ | |
5633 | if (mc.moved_charge) { | |
00501b53 | 5634 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5635 | mc.moved_charge = 0; |
4ffef5fe | 5636 | } |
483c30b5 DN |
5637 | /* we must fixup refcnts and charges */ |
5638 | if (mc.moved_swap) { | |
483c30b5 | 5639 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5640 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5641 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5642 | |
615d66c3 VD |
5643 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5644 | ||
05b84301 | 5645 | /* |
3e32cb2e JW |
5646 | * we charged both to->memory and to->memsw, so we |
5647 | * should uncharge to->memory. | |
05b84301 | 5648 | */ |
ce00a967 | 5649 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5650 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5651 | ||
615d66c3 VD |
5652 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
5653 | css_put_many(&mc.to->css, mc.moved_swap); | |
3e32cb2e | 5654 | |
483c30b5 DN |
5655 | mc.moved_swap = 0; |
5656 | } | |
dfe076b0 DN |
5657 | memcg_oom_recover(from); |
5658 | memcg_oom_recover(to); | |
5659 | wake_up_all(&mc.waitq); | |
5660 | } | |
5661 | ||
5662 | static void mem_cgroup_clear_mc(void) | |
5663 | { | |
264a0ae1 TH |
5664 | struct mm_struct *mm = mc.mm; |
5665 | ||
dfe076b0 DN |
5666 | /* |
5667 | * we must clear moving_task before waking up waiters at the end of | |
5668 | * task migration. | |
5669 | */ | |
5670 | mc.moving_task = NULL; | |
5671 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5672 | spin_lock(&mc.lock); |
4ffef5fe DN |
5673 | mc.from = NULL; |
5674 | mc.to = NULL; | |
264a0ae1 | 5675 | mc.mm = NULL; |
2bd9bb20 | 5676 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5677 | |
5678 | mmput(mm); | |
4ffef5fe DN |
5679 | } |
5680 | ||
1f7dd3e5 | 5681 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5682 | { |
1f7dd3e5 | 5683 | struct cgroup_subsys_state *css; |
eed67d75 | 5684 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5685 | struct mem_cgroup *from; |
4530eddb | 5686 | struct task_struct *leader, *p; |
9f2115f9 | 5687 | struct mm_struct *mm; |
1dfab5ab | 5688 | unsigned long move_flags; |
9f2115f9 | 5689 | int ret = 0; |
7dc74be0 | 5690 | |
1f7dd3e5 TH |
5691 | /* charge immigration isn't supported on the default hierarchy */ |
5692 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5693 | return 0; |
5694 | ||
4530eddb TH |
5695 | /* |
5696 | * Multi-process migrations only happen on the default hierarchy | |
5697 | * where charge immigration is not used. Perform charge | |
5698 | * immigration if @tset contains a leader and whine if there are | |
5699 | * multiple. | |
5700 | */ | |
5701 | p = NULL; | |
1f7dd3e5 | 5702 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5703 | WARN_ON_ONCE(p); |
5704 | p = leader; | |
1f7dd3e5 | 5705 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5706 | } |
5707 | if (!p) | |
5708 | return 0; | |
5709 | ||
1f7dd3e5 TH |
5710 | /* |
5711 | * We are now commited to this value whatever it is. Changes in this | |
5712 | * tunable will only affect upcoming migrations, not the current one. | |
5713 | * So we need to save it, and keep it going. | |
5714 | */ | |
5715 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5716 | if (!move_flags) | |
5717 | return 0; | |
5718 | ||
9f2115f9 TH |
5719 | from = mem_cgroup_from_task(p); |
5720 | ||
5721 | VM_BUG_ON(from == memcg); | |
5722 | ||
5723 | mm = get_task_mm(p); | |
5724 | if (!mm) | |
5725 | return 0; | |
5726 | /* We move charges only when we move a owner of the mm */ | |
5727 | if (mm->owner == p) { | |
5728 | VM_BUG_ON(mc.from); | |
5729 | VM_BUG_ON(mc.to); | |
5730 | VM_BUG_ON(mc.precharge); | |
5731 | VM_BUG_ON(mc.moved_charge); | |
5732 | VM_BUG_ON(mc.moved_swap); | |
5733 | ||
5734 | spin_lock(&mc.lock); | |
264a0ae1 | 5735 | mc.mm = mm; |
9f2115f9 TH |
5736 | mc.from = from; |
5737 | mc.to = memcg; | |
5738 | mc.flags = move_flags; | |
5739 | spin_unlock(&mc.lock); | |
5740 | /* We set mc.moving_task later */ | |
5741 | ||
5742 | ret = mem_cgroup_precharge_mc(mm); | |
5743 | if (ret) | |
5744 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5745 | } else { |
5746 | mmput(mm); | |
7dc74be0 DN |
5747 | } |
5748 | return ret; | |
5749 | } | |
5750 | ||
1f7dd3e5 | 5751 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5752 | { |
4e2f245d JW |
5753 | if (mc.to) |
5754 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5755 | } |
5756 | ||
4ffef5fe DN |
5757 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5758 | unsigned long addr, unsigned long end, | |
5759 | struct mm_walk *walk) | |
7dc74be0 | 5760 | { |
4ffef5fe | 5761 | int ret = 0; |
26bcd64a | 5762 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5763 | pte_t *pte; |
5764 | spinlock_t *ptl; | |
12724850 NH |
5765 | enum mc_target_type target_type; |
5766 | union mc_target target; | |
5767 | struct page *page; | |
4ffef5fe | 5768 | |
b6ec57f4 KS |
5769 | ptl = pmd_trans_huge_lock(pmd, vma); |
5770 | if (ptl) { | |
62ade86a | 5771 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5772 | spin_unlock(ptl); |
12724850 NH |
5773 | return 0; |
5774 | } | |
5775 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5776 | if (target_type == MC_TARGET_PAGE) { | |
5777 | page = target.page; | |
5778 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5779 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5780 | mc.from, mc.to)) { |
12724850 NH |
5781 | mc.precharge -= HPAGE_PMD_NR; |
5782 | mc.moved_charge += HPAGE_PMD_NR; | |
5783 | } | |
5784 | putback_lru_page(page); | |
5785 | } | |
5786 | put_page(page); | |
c733a828 JG |
5787 | } else if (target_type == MC_TARGET_DEVICE) { |
5788 | page = target.page; | |
5789 | if (!mem_cgroup_move_account(page, true, | |
5790 | mc.from, mc.to)) { | |
5791 | mc.precharge -= HPAGE_PMD_NR; | |
5792 | mc.moved_charge += HPAGE_PMD_NR; | |
5793 | } | |
5794 | put_page(page); | |
12724850 | 5795 | } |
bf929152 | 5796 | spin_unlock(ptl); |
1a5a9906 | 5797 | return 0; |
12724850 NH |
5798 | } |
5799 | ||
45f83cef AA |
5800 | if (pmd_trans_unstable(pmd)) |
5801 | return 0; | |
4ffef5fe DN |
5802 | retry: |
5803 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5804 | for (; addr != end; addr += PAGE_SIZE) { | |
5805 | pte_t ptent = *(pte++); | |
c733a828 | 5806 | bool device = false; |
02491447 | 5807 | swp_entry_t ent; |
4ffef5fe DN |
5808 | |
5809 | if (!mc.precharge) | |
5810 | break; | |
5811 | ||
8d32ff84 | 5812 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
5813 | case MC_TARGET_DEVICE: |
5814 | device = true; | |
5815 | /* fall through */ | |
4ffef5fe DN |
5816 | case MC_TARGET_PAGE: |
5817 | page = target.page; | |
53f9263b KS |
5818 | /* |
5819 | * We can have a part of the split pmd here. Moving it | |
5820 | * can be done but it would be too convoluted so simply | |
5821 | * ignore such a partial THP and keep it in original | |
5822 | * memcg. There should be somebody mapping the head. | |
5823 | */ | |
5824 | if (PageTransCompound(page)) | |
5825 | goto put; | |
c733a828 | 5826 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 5827 | goto put; |
f627c2f5 KS |
5828 | if (!mem_cgroup_move_account(page, false, |
5829 | mc.from, mc.to)) { | |
4ffef5fe | 5830 | mc.precharge--; |
854ffa8d DN |
5831 | /* we uncharge from mc.from later. */ |
5832 | mc.moved_charge++; | |
4ffef5fe | 5833 | } |
c733a828 JG |
5834 | if (!device) |
5835 | putback_lru_page(page); | |
8d32ff84 | 5836 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5837 | put_page(page); |
5838 | break; | |
02491447 DN |
5839 | case MC_TARGET_SWAP: |
5840 | ent = target.ent; | |
e91cbb42 | 5841 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5842 | mc.precharge--; |
483c30b5 DN |
5843 | /* we fixup refcnts and charges later. */ |
5844 | mc.moved_swap++; | |
5845 | } | |
02491447 | 5846 | break; |
4ffef5fe DN |
5847 | default: |
5848 | break; | |
5849 | } | |
5850 | } | |
5851 | pte_unmap_unlock(pte - 1, ptl); | |
5852 | cond_resched(); | |
5853 | ||
5854 | if (addr != end) { | |
5855 | /* | |
5856 | * We have consumed all precharges we got in can_attach(). | |
5857 | * We try charge one by one, but don't do any additional | |
5858 | * charges to mc.to if we have failed in charge once in attach() | |
5859 | * phase. | |
5860 | */ | |
854ffa8d | 5861 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5862 | if (!ret) |
5863 | goto retry; | |
5864 | } | |
5865 | ||
5866 | return ret; | |
5867 | } | |
5868 | ||
7b86ac33 CH |
5869 | static const struct mm_walk_ops charge_walk_ops = { |
5870 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5871 | }; | |
5872 | ||
264a0ae1 | 5873 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 5874 | { |
4ffef5fe | 5875 | lru_add_drain_all(); |
312722cb | 5876 | /* |
81f8c3a4 JW |
5877 | * Signal lock_page_memcg() to take the memcg's move_lock |
5878 | * while we're moving its pages to another memcg. Then wait | |
5879 | * for already started RCU-only updates to finish. | |
312722cb JW |
5880 | */ |
5881 | atomic_inc(&mc.from->moving_account); | |
5882 | synchronize_rcu(); | |
dfe076b0 | 5883 | retry: |
264a0ae1 | 5884 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
5885 | /* |
5886 | * Someone who are holding the mmap_sem might be waiting in | |
5887 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5888 | * and retry. Because we cancel precharges, we might not be able | |
5889 | * to move enough charges, but moving charge is a best-effort | |
5890 | * feature anyway, so it wouldn't be a big problem. | |
5891 | */ | |
5892 | __mem_cgroup_clear_mc(); | |
5893 | cond_resched(); | |
5894 | goto retry; | |
5895 | } | |
26bcd64a NH |
5896 | /* |
5897 | * When we have consumed all precharges and failed in doing | |
5898 | * additional charge, the page walk just aborts. | |
5899 | */ | |
7b86ac33 CH |
5900 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
5901 | NULL); | |
0247f3f4 | 5902 | |
264a0ae1 | 5903 | up_read(&mc.mm->mmap_sem); |
312722cb | 5904 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5905 | } |
5906 | ||
264a0ae1 | 5907 | static void mem_cgroup_move_task(void) |
67e465a7 | 5908 | { |
264a0ae1 TH |
5909 | if (mc.to) { |
5910 | mem_cgroup_move_charge(); | |
a433658c | 5911 | mem_cgroup_clear_mc(); |
264a0ae1 | 5912 | } |
67e465a7 | 5913 | } |
5cfb80a7 | 5914 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5915 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5916 | { |
5917 | return 0; | |
5918 | } | |
1f7dd3e5 | 5919 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5920 | { |
5921 | } | |
264a0ae1 | 5922 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
5923 | { |
5924 | } | |
5925 | #endif | |
67e465a7 | 5926 | |
f00baae7 TH |
5927 | /* |
5928 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5929 | * to verify whether we're attached to the default hierarchy on each mount |
5930 | * attempt. | |
f00baae7 | 5931 | */ |
eb95419b | 5932 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5933 | { |
5934 | /* | |
aa6ec29b | 5935 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5936 | * guarantees that @root doesn't have any children, so turning it |
5937 | * on for the root memcg is enough. | |
5938 | */ | |
9e10a130 | 5939 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5940 | root_mem_cgroup->use_hierarchy = true; |
5941 | else | |
5942 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5943 | } |
5944 | ||
677dc973 CD |
5945 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
5946 | { | |
5947 | if (value == PAGE_COUNTER_MAX) | |
5948 | seq_puts(m, "max\n"); | |
5949 | else | |
5950 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
5951 | ||
5952 | return 0; | |
5953 | } | |
5954 | ||
241994ed JW |
5955 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5956 | struct cftype *cft) | |
5957 | { | |
f5fc3c5d JW |
5958 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5959 | ||
5960 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5961 | } |
5962 | ||
bf8d5d52 RG |
5963 | static int memory_min_show(struct seq_file *m, void *v) |
5964 | { | |
677dc973 CD |
5965 | return seq_puts_memcg_tunable(m, |
5966 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
5967 | } |
5968 | ||
5969 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
5970 | char *buf, size_t nbytes, loff_t off) | |
5971 | { | |
5972 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5973 | unsigned long min; | |
5974 | int err; | |
5975 | ||
5976 | buf = strstrip(buf); | |
5977 | err = page_counter_memparse(buf, "max", &min); | |
5978 | if (err) | |
5979 | return err; | |
5980 | ||
5981 | page_counter_set_min(&memcg->memory, min); | |
5982 | ||
5983 | return nbytes; | |
5984 | } | |
5985 | ||
241994ed JW |
5986 | static int memory_low_show(struct seq_file *m, void *v) |
5987 | { | |
677dc973 CD |
5988 | return seq_puts_memcg_tunable(m, |
5989 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
5990 | } |
5991 | ||
5992 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5993 | char *buf, size_t nbytes, loff_t off) | |
5994 | { | |
5995 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5996 | unsigned long low; | |
5997 | int err; | |
5998 | ||
5999 | buf = strstrip(buf); | |
d2973697 | 6000 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6001 | if (err) |
6002 | return err; | |
6003 | ||
23067153 | 6004 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6005 | |
6006 | return nbytes; | |
6007 | } | |
6008 | ||
6009 | static int memory_high_show(struct seq_file *m, void *v) | |
6010 | { | |
677dc973 | 6011 | return seq_puts_memcg_tunable(m, READ_ONCE(mem_cgroup_from_seq(m)->high)); |
241994ed JW |
6012 | } |
6013 | ||
6014 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6015 | char *buf, size_t nbytes, loff_t off) | |
6016 | { | |
6017 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8c8c383c JW |
6018 | unsigned int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6019 | bool drained = false; | |
241994ed JW |
6020 | unsigned long high; |
6021 | int err; | |
6022 | ||
6023 | buf = strstrip(buf); | |
d2973697 | 6024 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6025 | if (err) |
6026 | return err; | |
6027 | ||
f6f989c5 | 6028 | WRITE_ONCE(memcg->high, high); |
241994ed | 6029 | |
8c8c383c JW |
6030 | for (;;) { |
6031 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6032 | unsigned long reclaimed; | |
6033 | ||
6034 | if (nr_pages <= high) | |
6035 | break; | |
6036 | ||
6037 | if (signal_pending(current)) | |
6038 | break; | |
6039 | ||
6040 | if (!drained) { | |
6041 | drain_all_stock(memcg); | |
6042 | drained = true; | |
6043 | continue; | |
6044 | } | |
6045 | ||
6046 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
6047 | GFP_KERNEL, true); | |
6048 | ||
6049 | if (!reclaimed && !nr_retries--) | |
6050 | break; | |
6051 | } | |
588083bb | 6052 | |
241994ed JW |
6053 | return nbytes; |
6054 | } | |
6055 | ||
6056 | static int memory_max_show(struct seq_file *m, void *v) | |
6057 | { | |
677dc973 CD |
6058 | return seq_puts_memcg_tunable(m, |
6059 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6060 | } |
6061 | ||
6062 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6063 | char *buf, size_t nbytes, loff_t off) | |
6064 | { | |
6065 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
6066 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
6067 | bool drained = false; | |
241994ed JW |
6068 | unsigned long max; |
6069 | int err; | |
6070 | ||
6071 | buf = strstrip(buf); | |
d2973697 | 6072 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6073 | if (err) |
6074 | return err; | |
6075 | ||
bbec2e15 | 6076 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6077 | |
6078 | for (;;) { | |
6079 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6080 | ||
6081 | if (nr_pages <= max) | |
6082 | break; | |
6083 | ||
7249c9f0 | 6084 | if (signal_pending(current)) |
b6e6edcf | 6085 | break; |
b6e6edcf JW |
6086 | |
6087 | if (!drained) { | |
6088 | drain_all_stock(memcg); | |
6089 | drained = true; | |
6090 | continue; | |
6091 | } | |
6092 | ||
6093 | if (nr_reclaims) { | |
6094 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
6095 | GFP_KERNEL, true)) | |
6096 | nr_reclaims--; | |
6097 | continue; | |
6098 | } | |
6099 | ||
e27be240 | 6100 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6101 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6102 | break; | |
6103 | } | |
241994ed | 6104 | |
2529bb3a | 6105 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6106 | return nbytes; |
6107 | } | |
6108 | ||
1e577f97 SB |
6109 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6110 | { | |
6111 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6112 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6113 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6114 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6115 | seq_printf(m, "oom_kill %lu\n", | |
6116 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
6117 | } | |
6118 | ||
241994ed JW |
6119 | static int memory_events_show(struct seq_file *m, void *v) |
6120 | { | |
aa9694bb | 6121 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6122 | |
1e577f97 SB |
6123 | __memory_events_show(m, memcg->memory_events); |
6124 | return 0; | |
6125 | } | |
6126 | ||
6127 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6128 | { | |
6129 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6130 | |
1e577f97 | 6131 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6132 | return 0; |
6133 | } | |
6134 | ||
587d9f72 JW |
6135 | static int memory_stat_show(struct seq_file *m, void *v) |
6136 | { | |
aa9694bb | 6137 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 6138 | char *buf; |
1ff9e6e1 | 6139 | |
c8713d0b JW |
6140 | buf = memory_stat_format(memcg); |
6141 | if (!buf) | |
6142 | return -ENOMEM; | |
6143 | seq_puts(m, buf); | |
6144 | kfree(buf); | |
587d9f72 JW |
6145 | return 0; |
6146 | } | |
6147 | ||
3d8b38eb RG |
6148 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6149 | { | |
aa9694bb | 6150 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6151 | |
6152 | seq_printf(m, "%d\n", memcg->oom_group); | |
6153 | ||
6154 | return 0; | |
6155 | } | |
6156 | ||
6157 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6158 | char *buf, size_t nbytes, loff_t off) | |
6159 | { | |
6160 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6161 | int ret, oom_group; | |
6162 | ||
6163 | buf = strstrip(buf); | |
6164 | if (!buf) | |
6165 | return -EINVAL; | |
6166 | ||
6167 | ret = kstrtoint(buf, 0, &oom_group); | |
6168 | if (ret) | |
6169 | return ret; | |
6170 | ||
6171 | if (oom_group != 0 && oom_group != 1) | |
6172 | return -EINVAL; | |
6173 | ||
6174 | memcg->oom_group = oom_group; | |
6175 | ||
6176 | return nbytes; | |
6177 | } | |
6178 | ||
241994ed JW |
6179 | static struct cftype memory_files[] = { |
6180 | { | |
6181 | .name = "current", | |
f5fc3c5d | 6182 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6183 | .read_u64 = memory_current_read, |
6184 | }, | |
bf8d5d52 RG |
6185 | { |
6186 | .name = "min", | |
6187 | .flags = CFTYPE_NOT_ON_ROOT, | |
6188 | .seq_show = memory_min_show, | |
6189 | .write = memory_min_write, | |
6190 | }, | |
241994ed JW |
6191 | { |
6192 | .name = "low", | |
6193 | .flags = CFTYPE_NOT_ON_ROOT, | |
6194 | .seq_show = memory_low_show, | |
6195 | .write = memory_low_write, | |
6196 | }, | |
6197 | { | |
6198 | .name = "high", | |
6199 | .flags = CFTYPE_NOT_ON_ROOT, | |
6200 | .seq_show = memory_high_show, | |
6201 | .write = memory_high_write, | |
6202 | }, | |
6203 | { | |
6204 | .name = "max", | |
6205 | .flags = CFTYPE_NOT_ON_ROOT, | |
6206 | .seq_show = memory_max_show, | |
6207 | .write = memory_max_write, | |
6208 | }, | |
6209 | { | |
6210 | .name = "events", | |
6211 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6212 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6213 | .seq_show = memory_events_show, |
6214 | }, | |
1e577f97 SB |
6215 | { |
6216 | .name = "events.local", | |
6217 | .flags = CFTYPE_NOT_ON_ROOT, | |
6218 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6219 | .seq_show = memory_events_local_show, | |
6220 | }, | |
587d9f72 JW |
6221 | { |
6222 | .name = "stat", | |
6223 | .flags = CFTYPE_NOT_ON_ROOT, | |
6224 | .seq_show = memory_stat_show, | |
6225 | }, | |
3d8b38eb RG |
6226 | { |
6227 | .name = "oom.group", | |
6228 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6229 | .seq_show = memory_oom_group_show, | |
6230 | .write = memory_oom_group_write, | |
6231 | }, | |
241994ed JW |
6232 | { } /* terminate */ |
6233 | }; | |
6234 | ||
073219e9 | 6235 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6236 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6237 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6238 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6239 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6240 | .css_free = mem_cgroup_css_free, |
1ced953b | 6241 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6242 | .can_attach = mem_cgroup_can_attach, |
6243 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 6244 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 6245 | .bind = mem_cgroup_bind, |
241994ed JW |
6246 | .dfl_cftypes = memory_files, |
6247 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6248 | .early_init = 0, |
8cdea7c0 | 6249 | }; |
c077719b | 6250 | |
bc50bcc6 JW |
6251 | /* |
6252 | * This function calculates an individual cgroup's effective | |
6253 | * protection which is derived from its own memory.min/low, its | |
6254 | * parent's and siblings' settings, as well as the actual memory | |
6255 | * distribution in the tree. | |
6256 | * | |
6257 | * The following rules apply to the effective protection values: | |
6258 | * | |
6259 | * 1. At the first level of reclaim, effective protection is equal to | |
6260 | * the declared protection in memory.min and memory.low. | |
6261 | * | |
6262 | * 2. To enable safe delegation of the protection configuration, at | |
6263 | * subsequent levels the effective protection is capped to the | |
6264 | * parent's effective protection. | |
6265 | * | |
6266 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6267 | * user is allowed to overcommit the declared protection at a given | |
6268 | * level. If that is the case, the parent's effective protection is | |
6269 | * distributed to the children in proportion to how much protection | |
6270 | * they have declared and how much of it they are utilizing. | |
6271 | * | |
6272 | * This makes distribution proportional, but also work-conserving: | |
6273 | * if one cgroup claims much more protection than it uses memory, | |
6274 | * the unused remainder is available to its siblings. | |
6275 | * | |
6276 | * 4. Conversely, when the declared protection is undercommitted at a | |
6277 | * given level, the distribution of the larger parental protection | |
6278 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6279 | * is capped to its own memory.min/low setting. | |
6280 | * | |
8a931f80 JW |
6281 | * 5. However, to allow protecting recursive subtrees from each other |
6282 | * without having to declare each individual cgroup's fixed share | |
6283 | * of the ancestor's claim to protection, any unutilized - | |
6284 | * "floating" - protection from up the tree is distributed in | |
6285 | * proportion to each cgroup's *usage*. This makes the protection | |
6286 | * neutral wrt sibling cgroups and lets them compete freely over | |
6287 | * the shared parental protection budget, but it protects the | |
6288 | * subtree as a whole from neighboring subtrees. | |
6289 | * | |
6290 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6291 | * against immediate siblings whereas 5. is about protecting against | |
6292 | * neighboring subtrees. | |
bc50bcc6 JW |
6293 | */ |
6294 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6295 | unsigned long parent_usage, |
bc50bcc6 JW |
6296 | unsigned long setting, |
6297 | unsigned long parent_effective, | |
6298 | unsigned long siblings_protected) | |
6299 | { | |
6300 | unsigned long protected; | |
8a931f80 | 6301 | unsigned long ep; |
bc50bcc6 JW |
6302 | |
6303 | protected = min(usage, setting); | |
6304 | /* | |
6305 | * If all cgroups at this level combined claim and use more | |
6306 | * protection then what the parent affords them, distribute | |
6307 | * shares in proportion to utilization. | |
6308 | * | |
6309 | * We are using actual utilization rather than the statically | |
6310 | * claimed protection in order to be work-conserving: claimed | |
6311 | * but unused protection is available to siblings that would | |
6312 | * otherwise get a smaller chunk than what they claimed. | |
6313 | */ | |
6314 | if (siblings_protected > parent_effective) | |
6315 | return protected * parent_effective / siblings_protected; | |
6316 | ||
6317 | /* | |
6318 | * Ok, utilized protection of all children is within what the | |
6319 | * parent affords them, so we know whatever this child claims | |
6320 | * and utilizes is effectively protected. | |
6321 | * | |
6322 | * If there is unprotected usage beyond this value, reclaim | |
6323 | * will apply pressure in proportion to that amount. | |
6324 | * | |
6325 | * If there is unutilized protection, the cgroup will be fully | |
6326 | * shielded from reclaim, but we do return a smaller value for | |
6327 | * protection than what the group could enjoy in theory. This | |
6328 | * is okay. With the overcommit distribution above, effective | |
6329 | * protection is always dependent on how memory is actually | |
6330 | * consumed among the siblings anyway. | |
6331 | */ | |
8a931f80 JW |
6332 | ep = protected; |
6333 | ||
6334 | /* | |
6335 | * If the children aren't claiming (all of) the protection | |
6336 | * afforded to them by the parent, distribute the remainder in | |
6337 | * proportion to the (unprotected) memory of each cgroup. That | |
6338 | * way, cgroups that aren't explicitly prioritized wrt each | |
6339 | * other compete freely over the allowance, but they are | |
6340 | * collectively protected from neighboring trees. | |
6341 | * | |
6342 | * We're using unprotected memory for the weight so that if | |
6343 | * some cgroups DO claim explicit protection, we don't protect | |
6344 | * the same bytes twice. | |
6345 | */ | |
6346 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6347 | return ep; | |
6348 | ||
6349 | if (parent_effective > siblings_protected && usage > protected) { | |
6350 | unsigned long unclaimed; | |
6351 | ||
6352 | unclaimed = parent_effective - siblings_protected; | |
6353 | unclaimed *= usage - protected; | |
6354 | unclaimed /= parent_usage - siblings_protected; | |
6355 | ||
6356 | ep += unclaimed; | |
6357 | } | |
6358 | ||
6359 | return ep; | |
bc50bcc6 JW |
6360 | } |
6361 | ||
241994ed | 6362 | /** |
bf8d5d52 | 6363 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 6364 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6365 | * @memcg: the memory cgroup to check |
6366 | * | |
23067153 RG |
6367 | * WARNING: This function is not stateless! It can only be used as part |
6368 | * of a top-down tree iteration, not for isolated queries. | |
34c81057 | 6369 | * |
bf8d5d52 RG |
6370 | * Returns one of the following: |
6371 | * MEMCG_PROT_NONE: cgroup memory is not protected | |
6372 | * MEMCG_PROT_LOW: cgroup memory is protected as long there is | |
6373 | * an unprotected supply of reclaimable memory from other cgroups. | |
6374 | * MEMCG_PROT_MIN: cgroup memory is protected | |
241994ed | 6375 | */ |
bf8d5d52 RG |
6376 | enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root, |
6377 | struct mem_cgroup *memcg) | |
241994ed | 6378 | { |
8a931f80 | 6379 | unsigned long usage, parent_usage; |
23067153 RG |
6380 | struct mem_cgroup *parent; |
6381 | ||
241994ed | 6382 | if (mem_cgroup_disabled()) |
bf8d5d52 | 6383 | return MEMCG_PROT_NONE; |
241994ed | 6384 | |
34c81057 SC |
6385 | if (!root) |
6386 | root = root_mem_cgroup; | |
6387 | if (memcg == root) | |
bf8d5d52 | 6388 | return MEMCG_PROT_NONE; |
241994ed | 6389 | |
23067153 | 6390 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 RG |
6391 | if (!usage) |
6392 | return MEMCG_PROT_NONE; | |
6393 | ||
bf8d5d52 | 6394 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
6395 | /* No parent means a non-hierarchical mode on v1 memcg */ |
6396 | if (!parent) | |
6397 | return MEMCG_PROT_NONE; | |
6398 | ||
bc50bcc6 | 6399 | if (parent == root) { |
c3d53200 | 6400 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
bc50bcc6 JW |
6401 | memcg->memory.elow = memcg->memory.low; |
6402 | goto out; | |
bf8d5d52 RG |
6403 | } |
6404 | ||
8a931f80 JW |
6405 | parent_usage = page_counter_read(&parent->memory); |
6406 | ||
b3a7822e | 6407 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6408 | READ_ONCE(memcg->memory.min), |
6409 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6410 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6411 | |
b3a7822e | 6412 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
bc50bcc6 | 6413 | memcg->memory.low, READ_ONCE(parent->memory.elow), |
b3a7822e | 6414 | atomic_long_read(&parent->memory.children_low_usage))); |
23067153 | 6415 | |
bc50bcc6 JW |
6416 | out: |
6417 | if (usage <= memcg->memory.emin) | |
bf8d5d52 | 6418 | return MEMCG_PROT_MIN; |
bc50bcc6 | 6419 | else if (usage <= memcg->memory.elow) |
bf8d5d52 RG |
6420 | return MEMCG_PROT_LOW; |
6421 | else | |
6422 | return MEMCG_PROT_NONE; | |
241994ed JW |
6423 | } |
6424 | ||
00501b53 JW |
6425 | /** |
6426 | * mem_cgroup_try_charge - try charging a page | |
6427 | * @page: page to charge | |
6428 | * @mm: mm context of the victim | |
6429 | * @gfp_mask: reclaim mode | |
6430 | * @memcgp: charged memcg return | |
25843c2b | 6431 | * @compound: charge the page as compound or small page |
00501b53 JW |
6432 | * |
6433 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6434 | * pages according to @gfp_mask if necessary. | |
6435 | * | |
6436 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
6437 | * Otherwise, an error code is returned. | |
6438 | * | |
6439 | * After page->mapping has been set up, the caller must finalize the | |
6440 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
6441 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
6442 | */ | |
6443 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
6444 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
6445 | bool compound) | |
00501b53 JW |
6446 | { |
6447 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 6448 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6449 | int ret = 0; |
6450 | ||
6451 | if (mem_cgroup_disabled()) | |
6452 | goto out; | |
6453 | ||
6454 | if (PageSwapCache(page)) { | |
00501b53 JW |
6455 | /* |
6456 | * Every swap fault against a single page tries to charge the | |
6457 | * page, bail as early as possible. shmem_unuse() encounters | |
6458 | * already charged pages, too. The USED bit is protected by | |
6459 | * the page lock, which serializes swap cache removal, which | |
6460 | * in turn serializes uncharging. | |
6461 | */ | |
e993d905 | 6462 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 6463 | if (compound_head(page)->mem_cgroup) |
00501b53 | 6464 | goto out; |
e993d905 | 6465 | |
37e84351 | 6466 | if (do_swap_account) { |
e993d905 VD |
6467 | swp_entry_t ent = { .val = page_private(page), }; |
6468 | unsigned short id = lookup_swap_cgroup_id(ent); | |
6469 | ||
6470 | rcu_read_lock(); | |
6471 | memcg = mem_cgroup_from_id(id); | |
6472 | if (memcg && !css_tryget_online(&memcg->css)) | |
6473 | memcg = NULL; | |
6474 | rcu_read_unlock(); | |
6475 | } | |
00501b53 JW |
6476 | } |
6477 | ||
00501b53 JW |
6478 | if (!memcg) |
6479 | memcg = get_mem_cgroup_from_mm(mm); | |
6480 | ||
6481 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
6482 | ||
6483 | css_put(&memcg->css); | |
00501b53 JW |
6484 | out: |
6485 | *memcgp = memcg; | |
6486 | return ret; | |
6487 | } | |
6488 | ||
2cf85583 TH |
6489 | int mem_cgroup_try_charge_delay(struct page *page, struct mm_struct *mm, |
6490 | gfp_t gfp_mask, struct mem_cgroup **memcgp, | |
6491 | bool compound) | |
6492 | { | |
6493 | struct mem_cgroup *memcg; | |
6494 | int ret; | |
6495 | ||
6496 | ret = mem_cgroup_try_charge(page, mm, gfp_mask, memcgp, compound); | |
6497 | memcg = *memcgp; | |
6498 | mem_cgroup_throttle_swaprate(memcg, page_to_nid(page), gfp_mask); | |
6499 | return ret; | |
6500 | } | |
6501 | ||
00501b53 JW |
6502 | /** |
6503 | * mem_cgroup_commit_charge - commit a page charge | |
6504 | * @page: page to charge | |
6505 | * @memcg: memcg to charge the page to | |
6506 | * @lrucare: page might be on LRU already | |
25843c2b | 6507 | * @compound: charge the page as compound or small page |
00501b53 JW |
6508 | * |
6509 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
6510 | * after page->mapping has been set up. This must happen atomically | |
6511 | * as part of the page instantiation, i.e. under the page table lock | |
6512 | * for anonymous pages, under the page lock for page and swap cache. | |
6513 | * | |
6514 | * In addition, the page must not be on the LRU during the commit, to | |
6515 | * prevent racing with task migration. If it might be, use @lrucare. | |
6516 | * | |
6517 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
6518 | */ | |
6519 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 6520 | bool lrucare, bool compound) |
00501b53 | 6521 | { |
f627c2f5 | 6522 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6523 | |
6524 | VM_BUG_ON_PAGE(!page->mapping, page); | |
6525 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
6526 | ||
6527 | if (mem_cgroup_disabled()) | |
6528 | return; | |
6529 | /* | |
6530 | * Swap faults will attempt to charge the same page multiple | |
6531 | * times. But reuse_swap_page() might have removed the page | |
6532 | * from swapcache already, so we can't check PageSwapCache(). | |
6533 | */ | |
6534 | if (!memcg) | |
6535 | return; | |
6536 | ||
6abb5a86 JW |
6537 | commit_charge(page, memcg, lrucare); |
6538 | ||
6abb5a86 | 6539 | local_irq_disable(); |
f627c2f5 | 6540 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
6541 | memcg_check_events(memcg, page); |
6542 | local_irq_enable(); | |
00501b53 | 6543 | |
7941d214 | 6544 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
6545 | swp_entry_t entry = { .val = page_private(page) }; |
6546 | /* | |
6547 | * The swap entry might not get freed for a long time, | |
6548 | * let's not wait for it. The page already received a | |
6549 | * memory+swap charge, drop the swap entry duplicate. | |
6550 | */ | |
38d8b4e6 | 6551 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 JW |
6552 | } |
6553 | } | |
6554 | ||
6555 | /** | |
6556 | * mem_cgroup_cancel_charge - cancel a page charge | |
6557 | * @page: page to charge | |
6558 | * @memcg: memcg to charge the page to | |
25843c2b | 6559 | * @compound: charge the page as compound or small page |
00501b53 JW |
6560 | * |
6561 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
6562 | */ | |
f627c2f5 KS |
6563 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
6564 | bool compound) | |
00501b53 | 6565 | { |
f627c2f5 | 6566 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6567 | |
6568 | if (mem_cgroup_disabled()) | |
6569 | return; | |
6570 | /* | |
6571 | * Swap faults will attempt to charge the same page multiple | |
6572 | * times. But reuse_swap_page() might have removed the page | |
6573 | * from swapcache already, so we can't check PageSwapCache(). | |
6574 | */ | |
6575 | if (!memcg) | |
6576 | return; | |
6577 | ||
00501b53 JW |
6578 | cancel_charge(memcg, nr_pages); |
6579 | } | |
6580 | ||
a9d5adee JG |
6581 | struct uncharge_gather { |
6582 | struct mem_cgroup *memcg; | |
6583 | unsigned long pgpgout; | |
6584 | unsigned long nr_anon; | |
6585 | unsigned long nr_file; | |
6586 | unsigned long nr_kmem; | |
6587 | unsigned long nr_huge; | |
6588 | unsigned long nr_shmem; | |
6589 | struct page *dummy_page; | |
6590 | }; | |
6591 | ||
6592 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6593 | { |
a9d5adee JG |
6594 | memset(ug, 0, sizeof(*ug)); |
6595 | } | |
6596 | ||
6597 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6598 | { | |
6599 | unsigned long nr_pages = ug->nr_anon + ug->nr_file + ug->nr_kmem; | |
747db954 JW |
6600 | unsigned long flags; |
6601 | ||
a9d5adee JG |
6602 | if (!mem_cgroup_is_root(ug->memcg)) { |
6603 | page_counter_uncharge(&ug->memcg->memory, nr_pages); | |
7941d214 | 6604 | if (do_memsw_account()) |
a9d5adee JG |
6605 | page_counter_uncharge(&ug->memcg->memsw, nr_pages); |
6606 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) | |
6607 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6608 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6609 | } |
747db954 JW |
6610 | |
6611 | local_irq_save(flags); | |
c9019e9b JW |
6612 | __mod_memcg_state(ug->memcg, MEMCG_RSS, -ug->nr_anon); |
6613 | __mod_memcg_state(ug->memcg, MEMCG_CACHE, -ug->nr_file); | |
6614 | __mod_memcg_state(ug->memcg, MEMCG_RSS_HUGE, -ug->nr_huge); | |
6615 | __mod_memcg_state(ug->memcg, NR_SHMEM, -ug->nr_shmem); | |
6616 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); | |
871789d4 | 6617 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, nr_pages); |
a9d5adee | 6618 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6619 | local_irq_restore(flags); |
e8ea14cc | 6620 | |
a9d5adee JG |
6621 | if (!mem_cgroup_is_root(ug->memcg)) |
6622 | css_put_many(&ug->memcg->css, nr_pages); | |
6623 | } | |
6624 | ||
6625 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6626 | { | |
6627 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
3f2eb028 JG |
6628 | VM_BUG_ON_PAGE(page_count(page) && !is_zone_device_page(page) && |
6629 | !PageHWPoison(page) , page); | |
a9d5adee JG |
6630 | |
6631 | if (!page->mem_cgroup) | |
6632 | return; | |
6633 | ||
6634 | /* | |
6635 | * Nobody should be changing or seriously looking at | |
6636 | * page->mem_cgroup at this point, we have fully | |
6637 | * exclusive access to the page. | |
6638 | */ | |
6639 | ||
6640 | if (ug->memcg != page->mem_cgroup) { | |
6641 | if (ug->memcg) { | |
6642 | uncharge_batch(ug); | |
6643 | uncharge_gather_clear(ug); | |
6644 | } | |
6645 | ug->memcg = page->mem_cgroup; | |
6646 | } | |
6647 | ||
6648 | if (!PageKmemcg(page)) { | |
6649 | unsigned int nr_pages = 1; | |
6650 | ||
6651 | if (PageTransHuge(page)) { | |
d8c6546b | 6652 | nr_pages = compound_nr(page); |
a9d5adee JG |
6653 | ug->nr_huge += nr_pages; |
6654 | } | |
6655 | if (PageAnon(page)) | |
6656 | ug->nr_anon += nr_pages; | |
6657 | else { | |
6658 | ug->nr_file += nr_pages; | |
6659 | if (PageSwapBacked(page)) | |
6660 | ug->nr_shmem += nr_pages; | |
6661 | } | |
6662 | ug->pgpgout++; | |
6663 | } else { | |
d8c6546b | 6664 | ug->nr_kmem += compound_nr(page); |
a9d5adee JG |
6665 | __ClearPageKmemcg(page); |
6666 | } | |
6667 | ||
6668 | ug->dummy_page = page; | |
6669 | page->mem_cgroup = NULL; | |
747db954 JW |
6670 | } |
6671 | ||
6672 | static void uncharge_list(struct list_head *page_list) | |
6673 | { | |
a9d5adee | 6674 | struct uncharge_gather ug; |
747db954 | 6675 | struct list_head *next; |
a9d5adee JG |
6676 | |
6677 | uncharge_gather_clear(&ug); | |
747db954 | 6678 | |
8b592656 JW |
6679 | /* |
6680 | * Note that the list can be a single page->lru; hence the | |
6681 | * do-while loop instead of a simple list_for_each_entry(). | |
6682 | */ | |
747db954 JW |
6683 | next = page_list->next; |
6684 | do { | |
a9d5adee JG |
6685 | struct page *page; |
6686 | ||
747db954 JW |
6687 | page = list_entry(next, struct page, lru); |
6688 | next = page->lru.next; | |
6689 | ||
a9d5adee | 6690 | uncharge_page(page, &ug); |
747db954 JW |
6691 | } while (next != page_list); |
6692 | ||
a9d5adee JG |
6693 | if (ug.memcg) |
6694 | uncharge_batch(&ug); | |
747db954 JW |
6695 | } |
6696 | ||
0a31bc97 JW |
6697 | /** |
6698 | * mem_cgroup_uncharge - uncharge a page | |
6699 | * @page: page to uncharge | |
6700 | * | |
6701 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
6702 | * mem_cgroup_commit_charge(). | |
6703 | */ | |
6704 | void mem_cgroup_uncharge(struct page *page) | |
6705 | { | |
a9d5adee JG |
6706 | struct uncharge_gather ug; |
6707 | ||
0a31bc97 JW |
6708 | if (mem_cgroup_disabled()) |
6709 | return; | |
6710 | ||
747db954 | 6711 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6712 | if (!page->mem_cgroup) |
0a31bc97 JW |
6713 | return; |
6714 | ||
a9d5adee JG |
6715 | uncharge_gather_clear(&ug); |
6716 | uncharge_page(page, &ug); | |
6717 | uncharge_batch(&ug); | |
747db954 | 6718 | } |
0a31bc97 | 6719 | |
747db954 JW |
6720 | /** |
6721 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6722 | * @page_list: list of pages to uncharge | |
6723 | * | |
6724 | * Uncharge a list of pages previously charged with | |
6725 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
6726 | */ | |
6727 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6728 | { | |
6729 | if (mem_cgroup_disabled()) | |
6730 | return; | |
0a31bc97 | 6731 | |
747db954 JW |
6732 | if (!list_empty(page_list)) |
6733 | uncharge_list(page_list); | |
0a31bc97 JW |
6734 | } |
6735 | ||
6736 | /** | |
6a93ca8f JW |
6737 | * mem_cgroup_migrate - charge a page's replacement |
6738 | * @oldpage: currently circulating page | |
6739 | * @newpage: replacement page | |
0a31bc97 | 6740 | * |
6a93ca8f JW |
6741 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6742 | * be uncharged upon free. | |
0a31bc97 JW |
6743 | * |
6744 | * Both pages must be locked, @newpage->mapping must be set up. | |
6745 | */ | |
6a93ca8f | 6746 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6747 | { |
29833315 | 6748 | struct mem_cgroup *memcg; |
44b7a8d3 | 6749 | unsigned int nr_pages; |
d93c4130 | 6750 | unsigned long flags; |
0a31bc97 JW |
6751 | |
6752 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6753 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6754 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6755 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6756 | newpage); | |
0a31bc97 JW |
6757 | |
6758 | if (mem_cgroup_disabled()) | |
6759 | return; | |
6760 | ||
6761 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6762 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6763 | return; |
6764 | ||
45637bab | 6765 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6766 | memcg = oldpage->mem_cgroup; |
29833315 | 6767 | if (!memcg) |
0a31bc97 JW |
6768 | return; |
6769 | ||
44b7a8d3 | 6770 | /* Force-charge the new page. The old one will be freed soon */ |
92855270 | 6771 | nr_pages = hpage_nr_pages(newpage); |
44b7a8d3 JW |
6772 | |
6773 | page_counter_charge(&memcg->memory, nr_pages); | |
6774 | if (do_memsw_account()) | |
6775 | page_counter_charge(&memcg->memsw, nr_pages); | |
6776 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 6777 | |
9cf7666a | 6778 | commit_charge(newpage, memcg, false); |
44b7a8d3 | 6779 | |
d93c4130 | 6780 | local_irq_save(flags); |
92855270 KC |
6781 | mem_cgroup_charge_statistics(memcg, newpage, PageTransHuge(newpage), |
6782 | nr_pages); | |
44b7a8d3 | 6783 | memcg_check_events(memcg, newpage); |
d93c4130 | 6784 | local_irq_restore(flags); |
0a31bc97 JW |
6785 | } |
6786 | ||
ef12947c | 6787 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6788 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6789 | ||
2d758073 | 6790 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6791 | { |
6792 | struct mem_cgroup *memcg; | |
6793 | ||
2d758073 JW |
6794 | if (!mem_cgroup_sockets_enabled) |
6795 | return; | |
6796 | ||
e876ecc6 SB |
6797 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
6798 | if (in_interrupt()) | |
6799 | return; | |
6800 | ||
11092087 JW |
6801 | rcu_read_lock(); |
6802 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6803 | if (memcg == root_mem_cgroup) |
6804 | goto out; | |
0db15298 | 6805 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6806 | goto out; |
8965aa28 | 6807 | if (css_tryget(&memcg->css)) |
11092087 | 6808 | sk->sk_memcg = memcg; |
f7e1cb6e | 6809 | out: |
11092087 JW |
6810 | rcu_read_unlock(); |
6811 | } | |
11092087 | 6812 | |
2d758073 | 6813 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6814 | { |
2d758073 JW |
6815 | if (sk->sk_memcg) |
6816 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6817 | } |
6818 | ||
6819 | /** | |
6820 | * mem_cgroup_charge_skmem - charge socket memory | |
6821 | * @memcg: memcg to charge | |
6822 | * @nr_pages: number of pages to charge | |
6823 | * | |
6824 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6825 | * @memcg's configured limit, %false if the charge had to be forced. | |
6826 | */ | |
6827 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6828 | { | |
f7e1cb6e | 6829 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6830 | |
f7e1cb6e | 6831 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6832 | struct page_counter *fail; |
f7e1cb6e | 6833 | |
0db15298 JW |
6834 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6835 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6836 | return true; |
6837 | } | |
0db15298 JW |
6838 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6839 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6840 | return false; |
11092087 | 6841 | } |
d886f4e4 | 6842 | |
f7e1cb6e JW |
6843 | /* Don't block in the packet receive path */ |
6844 | if (in_softirq()) | |
6845 | gfp_mask = GFP_NOWAIT; | |
6846 | ||
c9019e9b | 6847 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6848 | |
f7e1cb6e JW |
6849 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6850 | return true; | |
6851 | ||
6852 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6853 | return false; |
6854 | } | |
6855 | ||
6856 | /** | |
6857 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
6858 | * @memcg: memcg to uncharge |
6859 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
6860 | */ |
6861 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6862 | { | |
f7e1cb6e | 6863 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6864 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
6865 | return; |
6866 | } | |
d886f4e4 | 6867 | |
c9019e9b | 6868 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 6869 | |
475d0487 | 6870 | refill_stock(memcg, nr_pages); |
11092087 JW |
6871 | } |
6872 | ||
f7e1cb6e JW |
6873 | static int __init cgroup_memory(char *s) |
6874 | { | |
6875 | char *token; | |
6876 | ||
6877 | while ((token = strsep(&s, ",")) != NULL) { | |
6878 | if (!*token) | |
6879 | continue; | |
6880 | if (!strcmp(token, "nosocket")) | |
6881 | cgroup_memory_nosocket = true; | |
04823c83 VD |
6882 | if (!strcmp(token, "nokmem")) |
6883 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
6884 | } |
6885 | return 0; | |
6886 | } | |
6887 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 6888 | |
2d11085e | 6889 | /* |
1081312f MH |
6890 | * subsys_initcall() for memory controller. |
6891 | * | |
308167fc SAS |
6892 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
6893 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
6894 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
6895 | * should be initialized from here. | |
2d11085e MH |
6896 | */ |
6897 | static int __init mem_cgroup_init(void) | |
6898 | { | |
95a045f6 JW |
6899 | int cpu, node; |
6900 | ||
84c07d11 | 6901 | #ifdef CONFIG_MEMCG_KMEM |
13583c3d VD |
6902 | /* |
6903 | * Kmem cache creation is mostly done with the slab_mutex held, | |
17cc4dfe TH |
6904 | * so use a workqueue with limited concurrency to avoid stalling |
6905 | * all worker threads in case lots of cgroups are created and | |
6906 | * destroyed simultaneously. | |
13583c3d | 6907 | */ |
17cc4dfe TH |
6908 | memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1); |
6909 | BUG_ON(!memcg_kmem_cache_wq); | |
13583c3d VD |
6910 | #endif |
6911 | ||
308167fc SAS |
6912 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
6913 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
6914 | |
6915 | for_each_possible_cpu(cpu) | |
6916 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
6917 | drain_local_stock); | |
6918 | ||
6919 | for_each_node(node) { | |
6920 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
6921 | |
6922 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
6923 | node_online(node) ? node : NUMA_NO_NODE); | |
6924 | ||
ef8f2327 | 6925 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 6926 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 6927 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
6928 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
6929 | } | |
6930 | ||
2d11085e MH |
6931 | return 0; |
6932 | } | |
6933 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
6934 | |
6935 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
6936 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
6937 | { | |
1c2d479a | 6938 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
6939 | /* |
6940 | * The root cgroup cannot be destroyed, so it's refcount must | |
6941 | * always be >= 1. | |
6942 | */ | |
6943 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
6944 | VM_BUG_ON(1); | |
6945 | break; | |
6946 | } | |
6947 | memcg = parent_mem_cgroup(memcg); | |
6948 | if (!memcg) | |
6949 | memcg = root_mem_cgroup; | |
6950 | } | |
6951 | return memcg; | |
6952 | } | |
6953 | ||
21afa38e JW |
6954 | /** |
6955 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
6956 | * @page: page whose memsw charge to transfer | |
6957 | * @entry: swap entry to move the charge to | |
6958 | * | |
6959 | * Transfer the memsw charge of @page to @entry. | |
6960 | */ | |
6961 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
6962 | { | |
1f47b61f | 6963 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 6964 | unsigned int nr_entries; |
21afa38e JW |
6965 | unsigned short oldid; |
6966 | ||
6967 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6968 | VM_BUG_ON_PAGE(page_count(page), page); | |
6969 | ||
7941d214 | 6970 | if (!do_memsw_account()) |
21afa38e JW |
6971 | return; |
6972 | ||
6973 | memcg = page->mem_cgroup; | |
6974 | ||
6975 | /* Readahead page, never charged */ | |
6976 | if (!memcg) | |
6977 | return; | |
6978 | ||
1f47b61f VD |
6979 | /* |
6980 | * In case the memcg owning these pages has been offlined and doesn't | |
6981 | * have an ID allocated to it anymore, charge the closest online | |
6982 | * ancestor for the swap instead and transfer the memory+swap charge. | |
6983 | */ | |
6984 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
d6810d73 HY |
6985 | nr_entries = hpage_nr_pages(page); |
6986 | /* Get references for the tail pages, too */ | |
6987 | if (nr_entries > 1) | |
6988 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
6989 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
6990 | nr_entries); | |
21afa38e | 6991 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6992 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
6993 | |
6994 | page->mem_cgroup = NULL; | |
6995 | ||
6996 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 6997 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 6998 | |
1f47b61f VD |
6999 | if (memcg != swap_memcg) { |
7000 | if (!mem_cgroup_is_root(swap_memcg)) | |
d6810d73 HY |
7001 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7002 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7003 | } |
7004 | ||
ce9ce665 SAS |
7005 | /* |
7006 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7007 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7008 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7009 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
7010 | */ |
7011 | VM_BUG_ON(!irqs_disabled()); | |
d6810d73 HY |
7012 | mem_cgroup_charge_statistics(memcg, page, PageTransHuge(page), |
7013 | -nr_entries); | |
21afa38e | 7014 | memcg_check_events(memcg, page); |
73f576c0 JW |
7015 | |
7016 | if (!mem_cgroup_is_root(memcg)) | |
d08afa14 | 7017 | css_put_many(&memcg->css, nr_entries); |
21afa38e JW |
7018 | } |
7019 | ||
38d8b4e6 HY |
7020 | /** |
7021 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
7022 | * @page: page being added to swap |
7023 | * @entry: swap entry to charge | |
7024 | * | |
38d8b4e6 | 7025 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
7026 | * |
7027 | * Returns 0 on success, -ENOMEM on failure. | |
7028 | */ | |
7029 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
7030 | { | |
38d8b4e6 | 7031 | unsigned int nr_pages = hpage_nr_pages(page); |
37e84351 | 7032 | struct page_counter *counter; |
38d8b4e6 | 7033 | struct mem_cgroup *memcg; |
37e84351 VD |
7034 | unsigned short oldid; |
7035 | ||
7036 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
7037 | return 0; | |
7038 | ||
7039 | memcg = page->mem_cgroup; | |
7040 | ||
7041 | /* Readahead page, never charged */ | |
7042 | if (!memcg) | |
7043 | return 0; | |
7044 | ||
f3a53a3a TH |
7045 | if (!entry.val) { |
7046 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7047 | return 0; |
f3a53a3a | 7048 | } |
bb98f2c5 | 7049 | |
1f47b61f VD |
7050 | memcg = mem_cgroup_id_get_online(memcg); |
7051 | ||
37e84351 | 7052 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7053 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7054 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7055 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7056 | mem_cgroup_id_put(memcg); |
37e84351 | 7057 | return -ENOMEM; |
1f47b61f | 7058 | } |
37e84351 | 7059 | |
38d8b4e6 HY |
7060 | /* Get references for the tail pages, too */ |
7061 | if (nr_pages > 1) | |
7062 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7063 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 7064 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7065 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7066 | |
37e84351 VD |
7067 | return 0; |
7068 | } | |
7069 | ||
21afa38e | 7070 | /** |
38d8b4e6 | 7071 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7072 | * @entry: swap entry to uncharge |
38d8b4e6 | 7073 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7074 | */ |
38d8b4e6 | 7075 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7076 | { |
7077 | struct mem_cgroup *memcg; | |
7078 | unsigned short id; | |
7079 | ||
37e84351 | 7080 | if (!do_swap_account) |
21afa38e JW |
7081 | return; |
7082 | ||
38d8b4e6 | 7083 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7084 | rcu_read_lock(); |
adbe427b | 7085 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7086 | if (memcg) { |
37e84351 VD |
7087 | if (!mem_cgroup_is_root(memcg)) { |
7088 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
38d8b4e6 | 7089 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7090 | else |
38d8b4e6 | 7091 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7092 | } |
c9019e9b | 7093 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7094 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7095 | } |
7096 | rcu_read_unlock(); | |
7097 | } | |
7098 | ||
d8b38438 VD |
7099 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7100 | { | |
7101 | long nr_swap_pages = get_nr_swap_pages(); | |
7102 | ||
7103 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
7104 | return nr_swap_pages; | |
7105 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7106 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7107 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7108 | page_counter_read(&memcg->swap)); |
7109 | return nr_swap_pages; | |
7110 | } | |
7111 | ||
5ccc5aba VD |
7112 | bool mem_cgroup_swap_full(struct page *page) |
7113 | { | |
7114 | struct mem_cgroup *memcg; | |
7115 | ||
7116 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
7117 | ||
7118 | if (vm_swap_full()) | |
7119 | return true; | |
7120 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
7121 | return false; | |
7122 | ||
7123 | memcg = page->mem_cgroup; | |
7124 | if (!memcg) | |
7125 | return false; | |
7126 | ||
7127 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
32d087cd CD |
7128 | if (page_counter_read(&memcg->swap) * 2 >= |
7129 | READ_ONCE(memcg->swap.max)) | |
5ccc5aba VD |
7130 | return true; |
7131 | ||
7132 | return false; | |
7133 | } | |
7134 | ||
21afa38e JW |
7135 | /* for remember boot option*/ |
7136 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
7137 | static int really_do_swap_account __initdata = 1; | |
7138 | #else | |
7139 | static int really_do_swap_account __initdata; | |
7140 | #endif | |
7141 | ||
7142 | static int __init enable_swap_account(char *s) | |
7143 | { | |
7144 | if (!strcmp(s, "1")) | |
7145 | really_do_swap_account = 1; | |
7146 | else if (!strcmp(s, "0")) | |
7147 | really_do_swap_account = 0; | |
7148 | return 1; | |
7149 | } | |
7150 | __setup("swapaccount=", enable_swap_account); | |
7151 | ||
37e84351 VD |
7152 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7153 | struct cftype *cft) | |
7154 | { | |
7155 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7156 | ||
7157 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7158 | } | |
7159 | ||
7160 | static int swap_max_show(struct seq_file *m, void *v) | |
7161 | { | |
677dc973 CD |
7162 | return seq_puts_memcg_tunable(m, |
7163 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7164 | } |
7165 | ||
7166 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7167 | char *buf, size_t nbytes, loff_t off) | |
7168 | { | |
7169 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7170 | unsigned long max; | |
7171 | int err; | |
7172 | ||
7173 | buf = strstrip(buf); | |
7174 | err = page_counter_memparse(buf, "max", &max); | |
7175 | if (err) | |
7176 | return err; | |
7177 | ||
be09102b | 7178 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7179 | |
7180 | return nbytes; | |
7181 | } | |
7182 | ||
f3a53a3a TH |
7183 | static int swap_events_show(struct seq_file *m, void *v) |
7184 | { | |
aa9694bb | 7185 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a TH |
7186 | |
7187 | seq_printf(m, "max %lu\n", | |
7188 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7189 | seq_printf(m, "fail %lu\n", | |
7190 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7191 | ||
7192 | return 0; | |
7193 | } | |
7194 | ||
37e84351 VD |
7195 | static struct cftype swap_files[] = { |
7196 | { | |
7197 | .name = "swap.current", | |
7198 | .flags = CFTYPE_NOT_ON_ROOT, | |
7199 | .read_u64 = swap_current_read, | |
7200 | }, | |
7201 | { | |
7202 | .name = "swap.max", | |
7203 | .flags = CFTYPE_NOT_ON_ROOT, | |
7204 | .seq_show = swap_max_show, | |
7205 | .write = swap_max_write, | |
7206 | }, | |
f3a53a3a TH |
7207 | { |
7208 | .name = "swap.events", | |
7209 | .flags = CFTYPE_NOT_ON_ROOT, | |
7210 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7211 | .seq_show = swap_events_show, | |
7212 | }, | |
37e84351 VD |
7213 | { } /* terminate */ |
7214 | }; | |
7215 | ||
21afa38e JW |
7216 | static struct cftype memsw_cgroup_files[] = { |
7217 | { | |
7218 | .name = "memsw.usage_in_bytes", | |
7219 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7220 | .read_u64 = mem_cgroup_read_u64, | |
7221 | }, | |
7222 | { | |
7223 | .name = "memsw.max_usage_in_bytes", | |
7224 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7225 | .write = mem_cgroup_reset, | |
7226 | .read_u64 = mem_cgroup_read_u64, | |
7227 | }, | |
7228 | { | |
7229 | .name = "memsw.limit_in_bytes", | |
7230 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7231 | .write = mem_cgroup_write, | |
7232 | .read_u64 = mem_cgroup_read_u64, | |
7233 | }, | |
7234 | { | |
7235 | .name = "memsw.failcnt", | |
7236 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7237 | .write = mem_cgroup_reset, | |
7238 | .read_u64 = mem_cgroup_read_u64, | |
7239 | }, | |
7240 | { }, /* terminate */ | |
7241 | }; | |
7242 | ||
7243 | static int __init mem_cgroup_swap_init(void) | |
7244 | { | |
7245 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
7246 | do_swap_account = 1; | |
37e84351 VD |
7247 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
7248 | swap_files)); | |
21afa38e JW |
7249 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
7250 | memsw_cgroup_files)); | |
7251 | } | |
7252 | return 0; | |
7253 | } | |
7254 | subsys_initcall(mem_cgroup_swap_init); | |
7255 | ||
7256 | #endif /* CONFIG_MEMCG_SWAP */ |