]>
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 |
eccb52e7 | 86 | bool cgroup_memory_noswap __read_mostly; |
c077719b | 87 | #else |
eccb52e7 | 88 | #define cgroup_memory_noswap 1 |
2d1c4980 | 89 | #endif |
c077719b | 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 | { | |
eccb52e7 | 98 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_noswap; |
7941d214 JW |
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 |
bf4f0599 RG |
260 | extern spinlock_t css_set_lock; |
261 | ||
262 | static void obj_cgroup_release(struct percpu_ref *ref) | |
263 | { | |
264 | struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt); | |
265 | struct mem_cgroup *memcg; | |
266 | unsigned int nr_bytes; | |
267 | unsigned int nr_pages; | |
268 | unsigned long flags; | |
269 | ||
270 | /* | |
271 | * At this point all allocated objects are freed, and | |
272 | * objcg->nr_charged_bytes can't have an arbitrary byte value. | |
273 | * However, it can be PAGE_SIZE or (x * PAGE_SIZE). | |
274 | * | |
275 | * The following sequence can lead to it: | |
276 | * 1) CPU0: objcg == stock->cached_objcg | |
277 | * 2) CPU1: we do a small allocation (e.g. 92 bytes), | |
278 | * PAGE_SIZE bytes are charged | |
279 | * 3) CPU1: a process from another memcg is allocating something, | |
280 | * the stock if flushed, | |
281 | * objcg->nr_charged_bytes = PAGE_SIZE - 92 | |
282 | * 5) CPU0: we do release this object, | |
283 | * 92 bytes are added to stock->nr_bytes | |
284 | * 6) CPU0: stock is flushed, | |
285 | * 92 bytes are added to objcg->nr_charged_bytes | |
286 | * | |
287 | * In the result, nr_charged_bytes == PAGE_SIZE. | |
288 | * This page will be uncharged in obj_cgroup_release(). | |
289 | */ | |
290 | nr_bytes = atomic_read(&objcg->nr_charged_bytes); | |
291 | WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); | |
292 | nr_pages = nr_bytes >> PAGE_SHIFT; | |
293 | ||
294 | spin_lock_irqsave(&css_set_lock, flags); | |
295 | memcg = obj_cgroup_memcg(objcg); | |
296 | if (nr_pages) | |
297 | __memcg_kmem_uncharge(memcg, nr_pages); | |
298 | list_del(&objcg->list); | |
299 | mem_cgroup_put(memcg); | |
300 | spin_unlock_irqrestore(&css_set_lock, flags); | |
301 | ||
302 | percpu_ref_exit(ref); | |
303 | kfree_rcu(objcg, rcu); | |
304 | } | |
305 | ||
306 | static struct obj_cgroup *obj_cgroup_alloc(void) | |
307 | { | |
308 | struct obj_cgroup *objcg; | |
309 | int ret; | |
310 | ||
311 | objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL); | |
312 | if (!objcg) | |
313 | return NULL; | |
314 | ||
315 | ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, | |
316 | GFP_KERNEL); | |
317 | if (ret) { | |
318 | kfree(objcg); | |
319 | return NULL; | |
320 | } | |
321 | INIT_LIST_HEAD(&objcg->list); | |
322 | return objcg; | |
323 | } | |
324 | ||
325 | static void memcg_reparent_objcgs(struct mem_cgroup *memcg, | |
326 | struct mem_cgroup *parent) | |
327 | { | |
328 | struct obj_cgroup *objcg, *iter; | |
329 | ||
330 | objcg = rcu_replace_pointer(memcg->objcg, NULL, true); | |
331 | ||
332 | spin_lock_irq(&css_set_lock); | |
333 | ||
334 | /* Move active objcg to the parent's list */ | |
335 | xchg(&objcg->memcg, parent); | |
336 | css_get(&parent->css); | |
337 | list_add(&objcg->list, &parent->objcg_list); | |
338 | ||
339 | /* Move already reparented objcgs to the parent's list */ | |
340 | list_for_each_entry(iter, &memcg->objcg_list, list) { | |
341 | css_get(&parent->css); | |
342 | xchg(&iter->memcg, parent); | |
343 | css_put(&memcg->css); | |
344 | } | |
345 | list_splice(&memcg->objcg_list, &parent->objcg_list); | |
346 | ||
347 | spin_unlock_irq(&css_set_lock); | |
348 | ||
349 | percpu_ref_kill(&objcg->refcnt); | |
350 | } | |
351 | ||
55007d84 | 352 | /* |
9855609b | 353 | * This will be used as a shrinker list's index. |
b8627835 LZ |
354 | * The main reason for not using cgroup id for this: |
355 | * this works better in sparse environments, where we have a lot of memcgs, | |
356 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
357 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
358 | * 200 entry array for that. | |
55007d84 | 359 | * |
dbcf73e2 VD |
360 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
361 | * will double each time we have to increase it. | |
55007d84 | 362 | */ |
dbcf73e2 VD |
363 | static DEFINE_IDA(memcg_cache_ida); |
364 | int memcg_nr_cache_ids; | |
749c5415 | 365 | |
05257a1a VD |
366 | /* Protects memcg_nr_cache_ids */ |
367 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
368 | ||
369 | void memcg_get_cache_ids(void) | |
370 | { | |
371 | down_read(&memcg_cache_ids_sem); | |
372 | } | |
373 | ||
374 | void memcg_put_cache_ids(void) | |
375 | { | |
376 | up_read(&memcg_cache_ids_sem); | |
377 | } | |
378 | ||
55007d84 GC |
379 | /* |
380 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
381 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
382 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
383 | * tunable, but that is strictly not necessary. | |
384 | * | |
b8627835 | 385 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
386 | * this constant directly from cgroup, but it is understandable that this is |
387 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 388 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
389 | * increase ours as well if it increases. |
390 | */ | |
391 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 392 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 393 | |
d7f25f8a GC |
394 | /* |
395 | * A lot of the calls to the cache allocation functions are expected to be | |
272911a4 | 396 | * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are |
d7f25f8a GC |
397 | * conditional to this static branch, we'll have to allow modules that does |
398 | * kmem_cache_alloc and the such to see this symbol as well | |
399 | */ | |
ef12947c | 400 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 401 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
0a432dcb | 402 | #endif |
17cc4dfe | 403 | |
0a4465d3 KT |
404 | static int memcg_shrinker_map_size; |
405 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); | |
406 | ||
407 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) | |
408 | { | |
409 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); | |
410 | } | |
411 | ||
412 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, | |
413 | int size, int old_size) | |
414 | { | |
415 | struct memcg_shrinker_map *new, *old; | |
416 | int nid; | |
417 | ||
418 | lockdep_assert_held(&memcg_shrinker_map_mutex); | |
419 | ||
420 | for_each_node(nid) { | |
421 | old = rcu_dereference_protected( | |
422 | mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); | |
423 | /* Not yet online memcg */ | |
424 | if (!old) | |
425 | return 0; | |
426 | ||
86daf94e | 427 | new = kvmalloc_node(sizeof(*new) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
428 | if (!new) |
429 | return -ENOMEM; | |
430 | ||
431 | /* Set all old bits, clear all new bits */ | |
432 | memset(new->map, (int)0xff, old_size); | |
433 | memset((void *)new->map + old_size, 0, size - old_size); | |
434 | ||
435 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); | |
436 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); | |
437 | } | |
438 | ||
439 | return 0; | |
440 | } | |
441 | ||
442 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) | |
443 | { | |
444 | struct mem_cgroup_per_node *pn; | |
445 | struct memcg_shrinker_map *map; | |
446 | int nid; | |
447 | ||
448 | if (mem_cgroup_is_root(memcg)) | |
449 | return; | |
450 | ||
451 | for_each_node(nid) { | |
452 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
453 | map = rcu_dereference_protected(pn->shrinker_map, true); | |
454 | if (map) | |
455 | kvfree(map); | |
456 | rcu_assign_pointer(pn->shrinker_map, NULL); | |
457 | } | |
458 | } | |
459 | ||
460 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
461 | { | |
462 | struct memcg_shrinker_map *map; | |
463 | int nid, size, ret = 0; | |
464 | ||
465 | if (mem_cgroup_is_root(memcg)) | |
466 | return 0; | |
467 | ||
468 | mutex_lock(&memcg_shrinker_map_mutex); | |
469 | size = memcg_shrinker_map_size; | |
470 | for_each_node(nid) { | |
86daf94e | 471 | map = kvzalloc_node(sizeof(*map) + size, GFP_KERNEL, nid); |
0a4465d3 KT |
472 | if (!map) { |
473 | memcg_free_shrinker_maps(memcg); | |
474 | ret = -ENOMEM; | |
475 | break; | |
476 | } | |
477 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); | |
478 | } | |
479 | mutex_unlock(&memcg_shrinker_map_mutex); | |
480 | ||
481 | return ret; | |
482 | } | |
483 | ||
484 | int memcg_expand_shrinker_maps(int new_id) | |
485 | { | |
486 | int size, old_size, ret = 0; | |
487 | struct mem_cgroup *memcg; | |
488 | ||
489 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); | |
490 | old_size = memcg_shrinker_map_size; | |
491 | if (size <= old_size) | |
492 | return 0; | |
493 | ||
494 | mutex_lock(&memcg_shrinker_map_mutex); | |
495 | if (!root_mem_cgroup) | |
496 | goto unlock; | |
497 | ||
498 | for_each_mem_cgroup(memcg) { | |
499 | if (mem_cgroup_is_root(memcg)) | |
500 | continue; | |
501 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); | |
75866af6 VA |
502 | if (ret) { |
503 | mem_cgroup_iter_break(NULL, memcg); | |
0a4465d3 | 504 | goto unlock; |
75866af6 | 505 | } |
0a4465d3 KT |
506 | } |
507 | unlock: | |
508 | if (!ret) | |
509 | memcg_shrinker_map_size = size; | |
510 | mutex_unlock(&memcg_shrinker_map_mutex); | |
511 | return ret; | |
512 | } | |
fae91d6d KT |
513 | |
514 | void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id) | |
515 | { | |
516 | if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) { | |
517 | struct memcg_shrinker_map *map; | |
518 | ||
519 | rcu_read_lock(); | |
520 | map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map); | |
f90280d6 KT |
521 | /* Pairs with smp mb in shrink_slab() */ |
522 | smp_mb__before_atomic(); | |
fae91d6d KT |
523 | set_bit(shrinker_id, map->map); |
524 | rcu_read_unlock(); | |
525 | } | |
526 | } | |
527 | ||
ad7fa852 TH |
528 | /** |
529 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
530 | * @page: page of interest | |
531 | * | |
532 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
533 | * with @page is returned. The returned css remains associated with @page | |
534 | * until it is released. | |
535 | * | |
536 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
537 | * is returned. | |
ad7fa852 TH |
538 | */ |
539 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
540 | { | |
541 | struct mem_cgroup *memcg; | |
542 | ||
ad7fa852 TH |
543 | memcg = page->mem_cgroup; |
544 | ||
9e10a130 | 545 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
546 | memcg = root_mem_cgroup; |
547 | ||
ad7fa852 TH |
548 | return &memcg->css; |
549 | } | |
550 | ||
2fc04524 VD |
551 | /** |
552 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
553 | * @page: the page | |
554 | * | |
555 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
556 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
557 | * is safe to call this function without holding a reference to @page. | |
558 | * | |
559 | * Note, this function is inherently racy, because there is nothing to prevent | |
560 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
561 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
562 | * do not care (such as procfs interfaces). | |
563 | */ | |
564 | ino_t page_cgroup_ino(struct page *page) | |
565 | { | |
566 | struct mem_cgroup *memcg; | |
567 | unsigned long ino = 0; | |
568 | ||
569 | rcu_read_lock(); | |
9855609b | 570 | memcg = page->mem_cgroup; |
286e04b8 | 571 | |
9855609b RG |
572 | /* |
573 | * The lowest bit set means that memcg isn't a valid | |
574 | * memcg pointer, but a obj_cgroups pointer. | |
575 | * In this case the page is shared and doesn't belong | |
576 | * to any specific memory cgroup. | |
577 | */ | |
578 | if ((unsigned long) memcg & 0x1UL) | |
579 | memcg = NULL; | |
286e04b8 | 580 | |
2fc04524 VD |
581 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
582 | memcg = parent_mem_cgroup(memcg); | |
583 | if (memcg) | |
584 | ino = cgroup_ino(memcg->css.cgroup); | |
585 | rcu_read_unlock(); | |
586 | return ino; | |
587 | } | |
588 | ||
ef8f2327 MG |
589 | static struct mem_cgroup_per_node * |
590 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 591 | { |
97a6c37b | 592 | int nid = page_to_nid(page); |
f64c3f54 | 593 | |
ef8f2327 | 594 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
595 | } |
596 | ||
ef8f2327 MG |
597 | static struct mem_cgroup_tree_per_node * |
598 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 599 | { |
ef8f2327 | 600 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
601 | } |
602 | ||
ef8f2327 | 603 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
604 | soft_limit_tree_from_page(struct page *page) |
605 | { | |
606 | int nid = page_to_nid(page); | |
bb4cc1a8 | 607 | |
ef8f2327 | 608 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
609 | } |
610 | ||
ef8f2327 MG |
611 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
612 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 613 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
614 | { |
615 | struct rb_node **p = &mctz->rb_root.rb_node; | |
616 | struct rb_node *parent = NULL; | |
ef8f2327 | 617 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 618 | bool rightmost = true; |
bb4cc1a8 AM |
619 | |
620 | if (mz->on_tree) | |
621 | return; | |
622 | ||
623 | mz->usage_in_excess = new_usage_in_excess; | |
624 | if (!mz->usage_in_excess) | |
625 | return; | |
626 | while (*p) { | |
627 | parent = *p; | |
ef8f2327 | 628 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 629 | tree_node); |
fa90b2fd | 630 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 631 | p = &(*p)->rb_left; |
fa90b2fd DB |
632 | rightmost = false; |
633 | } | |
634 | ||
bb4cc1a8 AM |
635 | /* |
636 | * We can't avoid mem cgroups that are over their soft | |
637 | * limit by the same amount | |
638 | */ | |
639 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
640 | p = &(*p)->rb_right; | |
641 | } | |
fa90b2fd DB |
642 | |
643 | if (rightmost) | |
644 | mctz->rb_rightmost = &mz->tree_node; | |
645 | ||
bb4cc1a8 AM |
646 | rb_link_node(&mz->tree_node, parent, p); |
647 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
648 | mz->on_tree = true; | |
649 | } | |
650 | ||
ef8f2327 MG |
651 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
652 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
653 | { |
654 | if (!mz->on_tree) | |
655 | return; | |
fa90b2fd DB |
656 | |
657 | if (&mz->tree_node == mctz->rb_rightmost) | |
658 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
659 | ||
bb4cc1a8 AM |
660 | rb_erase(&mz->tree_node, &mctz->rb_root); |
661 | mz->on_tree = false; | |
662 | } | |
663 | ||
ef8f2327 MG |
664 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
665 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 666 | { |
0a31bc97 JW |
667 | unsigned long flags; |
668 | ||
669 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 670 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 671 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
672 | } |
673 | ||
3e32cb2e JW |
674 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
675 | { | |
676 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 677 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
678 | unsigned long excess = 0; |
679 | ||
680 | if (nr_pages > soft_limit) | |
681 | excess = nr_pages - soft_limit; | |
682 | ||
683 | return excess; | |
684 | } | |
bb4cc1a8 AM |
685 | |
686 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
687 | { | |
3e32cb2e | 688 | unsigned long excess; |
ef8f2327 MG |
689 | struct mem_cgroup_per_node *mz; |
690 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 691 | |
e231875b | 692 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
693 | if (!mctz) |
694 | return; | |
bb4cc1a8 AM |
695 | /* |
696 | * Necessary to update all ancestors when hierarchy is used. | |
697 | * because their event counter is not touched. | |
698 | */ | |
699 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 700 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 701 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
702 | /* |
703 | * We have to update the tree if mz is on RB-tree or | |
704 | * mem is over its softlimit. | |
705 | */ | |
706 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
707 | unsigned long flags; |
708 | ||
709 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
710 | /* if on-tree, remove it */ |
711 | if (mz->on_tree) | |
cf2c8127 | 712 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
713 | /* |
714 | * Insert again. mz->usage_in_excess will be updated. | |
715 | * If excess is 0, no tree ops. | |
716 | */ | |
cf2c8127 | 717 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 718 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
719 | } |
720 | } | |
721 | } | |
722 | ||
723 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
724 | { | |
ef8f2327 MG |
725 | struct mem_cgroup_tree_per_node *mctz; |
726 | struct mem_cgroup_per_node *mz; | |
727 | int nid; | |
bb4cc1a8 | 728 | |
e231875b | 729 | for_each_node(nid) { |
ef8f2327 MG |
730 | mz = mem_cgroup_nodeinfo(memcg, nid); |
731 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
732 | if (mctz) |
733 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
734 | } |
735 | } | |
736 | ||
ef8f2327 MG |
737 | static struct mem_cgroup_per_node * |
738 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 739 | { |
ef8f2327 | 740 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
741 | |
742 | retry: | |
743 | mz = NULL; | |
fa90b2fd | 744 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
745 | goto done; /* Nothing to reclaim from */ |
746 | ||
fa90b2fd DB |
747 | mz = rb_entry(mctz->rb_rightmost, |
748 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
749 | /* |
750 | * Remove the node now but someone else can add it back, | |
751 | * we will to add it back at the end of reclaim to its correct | |
752 | * position in the tree. | |
753 | */ | |
cf2c8127 | 754 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 755 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 756 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
757 | goto retry; |
758 | done: | |
759 | return mz; | |
760 | } | |
761 | ||
ef8f2327 MG |
762 | static struct mem_cgroup_per_node * |
763 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 764 | { |
ef8f2327 | 765 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 766 | |
0a31bc97 | 767 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 768 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 769 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
770 | return mz; |
771 | } | |
772 | ||
db9adbcb JW |
773 | /** |
774 | * __mod_memcg_state - update cgroup memory statistics | |
775 | * @memcg: the memory cgroup | |
776 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
777 | * @val: delta to add to the counter, can be negative | |
778 | */ | |
779 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
780 | { | |
ea426c2a | 781 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb JW |
782 | |
783 | if (mem_cgroup_disabled()) | |
784 | return; | |
785 | ||
ea426c2a RG |
786 | if (vmstat_item_in_bytes(idx)) |
787 | threshold <<= PAGE_SHIFT; | |
788 | ||
db9adbcb | 789 | x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); |
ea426c2a | 790 | if (unlikely(abs(x) > threshold)) { |
42a30035 JW |
791 | struct mem_cgroup *mi; |
792 | ||
766a4c19 YS |
793 | /* |
794 | * Batch local counters to keep them in sync with | |
795 | * the hierarchical ones. | |
796 | */ | |
797 | __this_cpu_add(memcg->vmstats_local->stat[idx], x); | |
42a30035 JW |
798 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
799 | atomic_long_add(x, &mi->vmstats[idx]); | |
db9adbcb JW |
800 | x = 0; |
801 | } | |
802 | __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); | |
803 | } | |
804 | ||
42a30035 JW |
805 | static struct mem_cgroup_per_node * |
806 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
807 | { | |
808 | struct mem_cgroup *parent; | |
809 | ||
810 | parent = parent_mem_cgroup(pn->memcg); | |
811 | if (!parent) | |
812 | return NULL; | |
813 | return mem_cgroup_nodeinfo(parent, nid); | |
814 | } | |
815 | ||
eedc4e5a RG |
816 | void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, |
817 | int val) | |
db9adbcb JW |
818 | { |
819 | struct mem_cgroup_per_node *pn; | |
42a30035 | 820 | struct mem_cgroup *memcg; |
ea426c2a | 821 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb | 822 | |
db9adbcb | 823 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
42a30035 | 824 | memcg = pn->memcg; |
db9adbcb JW |
825 | |
826 | /* Update memcg */ | |
42a30035 | 827 | __mod_memcg_state(memcg, idx, val); |
db9adbcb | 828 | |
b4c46484 RG |
829 | /* Update lruvec */ |
830 | __this_cpu_add(pn->lruvec_stat_local->count[idx], val); | |
831 | ||
ea426c2a RG |
832 | if (vmstat_item_in_bytes(idx)) |
833 | threshold <<= PAGE_SHIFT; | |
834 | ||
db9adbcb | 835 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); |
ea426c2a | 836 | if (unlikely(abs(x) > threshold)) { |
eedc4e5a | 837 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
42a30035 JW |
838 | struct mem_cgroup_per_node *pi; |
839 | ||
42a30035 JW |
840 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) |
841 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
842 | x = 0; |
843 | } | |
844 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
845 | } | |
846 | ||
eedc4e5a RG |
847 | /** |
848 | * __mod_lruvec_state - update lruvec memory statistics | |
849 | * @lruvec: the lruvec | |
850 | * @idx: the stat item | |
851 | * @val: delta to add to the counter, can be negative | |
852 | * | |
853 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
854 | * function updates the all three counters that are affected by a | |
855 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
856 | */ | |
857 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
858 | int val) | |
859 | { | |
860 | /* Update node */ | |
861 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); | |
862 | ||
863 | /* Update memcg and lruvec */ | |
864 | if (!mem_cgroup_disabled()) | |
865 | __mod_memcg_lruvec_state(lruvec, idx, val); | |
866 | } | |
867 | ||
ec9f0238 RG |
868 | void __mod_lruvec_slab_state(void *p, enum node_stat_item idx, int val) |
869 | { | |
4f103c63 | 870 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
871 | struct mem_cgroup *memcg; |
872 | struct lruvec *lruvec; | |
873 | ||
874 | rcu_read_lock(); | |
4f103c63 | 875 | memcg = mem_cgroup_from_obj(p); |
ec9f0238 RG |
876 | |
877 | /* Untracked pages have no memcg, no lruvec. Update only the node */ | |
878 | if (!memcg || memcg == root_mem_cgroup) { | |
879 | __mod_node_page_state(pgdat, idx, val); | |
880 | } else { | |
867e5e1d | 881 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
ec9f0238 RG |
882 | __mod_lruvec_state(lruvec, idx, val); |
883 | } | |
884 | rcu_read_unlock(); | |
885 | } | |
886 | ||
8380ce47 RG |
887 | void mod_memcg_obj_state(void *p, int idx, int val) |
888 | { | |
889 | struct mem_cgroup *memcg; | |
890 | ||
891 | rcu_read_lock(); | |
892 | memcg = mem_cgroup_from_obj(p); | |
893 | if (memcg) | |
894 | mod_memcg_state(memcg, idx, val); | |
895 | rcu_read_unlock(); | |
896 | } | |
897 | ||
db9adbcb JW |
898 | /** |
899 | * __count_memcg_events - account VM events in a cgroup | |
900 | * @memcg: the memory cgroup | |
901 | * @idx: the event item | |
902 | * @count: the number of events that occured | |
903 | */ | |
904 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
905 | unsigned long count) | |
906 | { | |
907 | unsigned long x; | |
908 | ||
909 | if (mem_cgroup_disabled()) | |
910 | return; | |
911 | ||
912 | x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); | |
913 | if (unlikely(x > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
914 | struct mem_cgroup *mi; |
915 | ||
766a4c19 YS |
916 | /* |
917 | * Batch local counters to keep them in sync with | |
918 | * the hierarchical ones. | |
919 | */ | |
920 | __this_cpu_add(memcg->vmstats_local->events[idx], x); | |
42a30035 JW |
921 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
922 | atomic_long_add(x, &mi->vmevents[idx]); | |
db9adbcb JW |
923 | x = 0; |
924 | } | |
925 | __this_cpu_write(memcg->vmstats_percpu->events[idx], x); | |
926 | } | |
927 | ||
42a30035 | 928 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 929 | { |
871789d4 | 930 | return atomic_long_read(&memcg->vmevents[event]); |
e9f8974f JW |
931 | } |
932 | ||
42a30035 JW |
933 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
934 | { | |
815744d7 JW |
935 | long x = 0; |
936 | int cpu; | |
937 | ||
938 | for_each_possible_cpu(cpu) | |
939 | x += per_cpu(memcg->vmstats_local->events[event], cpu); | |
940 | return x; | |
42a30035 JW |
941 | } |
942 | ||
c0ff4b85 | 943 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 944 | struct page *page, |
3fba69a5 | 945 | int nr_pages) |
d52aa412 | 946 | { |
e401f176 KH |
947 | /* pagein of a big page is an event. So, ignore page size */ |
948 | if (nr_pages > 0) | |
c9019e9b | 949 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 950 | else { |
c9019e9b | 951 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
952 | nr_pages = -nr_pages; /* for event */ |
953 | } | |
e401f176 | 954 | |
871789d4 | 955 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
956 | } |
957 | ||
f53d7ce3 JW |
958 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
959 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
960 | { |
961 | unsigned long val, next; | |
962 | ||
871789d4 CD |
963 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
964 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 965 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 966 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
967 | switch (target) { |
968 | case MEM_CGROUP_TARGET_THRESH: | |
969 | next = val + THRESHOLDS_EVENTS_TARGET; | |
970 | break; | |
bb4cc1a8 AM |
971 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
972 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
973 | break; | |
f53d7ce3 JW |
974 | default: |
975 | break; | |
976 | } | |
871789d4 | 977 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 978 | return true; |
7a159cc9 | 979 | } |
f53d7ce3 | 980 | return false; |
d2265e6f KH |
981 | } |
982 | ||
983 | /* | |
984 | * Check events in order. | |
985 | * | |
986 | */ | |
c0ff4b85 | 987 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
988 | { |
989 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
990 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
991 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 992 | bool do_softlimit; |
f53d7ce3 | 993 | |
bb4cc1a8 AM |
994 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
995 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 996 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
997 | if (unlikely(do_softlimit)) |
998 | mem_cgroup_update_tree(memcg, page); | |
0a31bc97 | 999 | } |
d2265e6f KH |
1000 | } |
1001 | ||
cf475ad2 | 1002 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 1003 | { |
31a78f23 BS |
1004 | /* |
1005 | * mm_update_next_owner() may clear mm->owner to NULL | |
1006 | * if it races with swapoff, page migration, etc. | |
1007 | * So this can be called with p == NULL. | |
1008 | */ | |
1009 | if (unlikely(!p)) | |
1010 | return NULL; | |
1011 | ||
073219e9 | 1012 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 1013 | } |
33398cf2 | 1014 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 1015 | |
d46eb14b SB |
1016 | /** |
1017 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
1018 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
1019 | * | |
1020 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
1021 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
1022 | * returned. | |
1023 | */ | |
1024 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 1025 | { |
d46eb14b SB |
1026 | struct mem_cgroup *memcg; |
1027 | ||
1028 | if (mem_cgroup_disabled()) | |
1029 | return NULL; | |
0b7f569e | 1030 | |
54595fe2 KH |
1031 | rcu_read_lock(); |
1032 | do { | |
6f6acb00 MH |
1033 | /* |
1034 | * Page cache insertions can happen withou an | |
1035 | * actual mm context, e.g. during disk probing | |
1036 | * on boot, loopback IO, acct() writes etc. | |
1037 | */ | |
1038 | if (unlikely(!mm)) | |
df381975 | 1039 | memcg = root_mem_cgroup; |
6f6acb00 MH |
1040 | else { |
1041 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1042 | if (unlikely(!memcg)) | |
1043 | memcg = root_mem_cgroup; | |
1044 | } | |
00d484f3 | 1045 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 1046 | rcu_read_unlock(); |
c0ff4b85 | 1047 | return memcg; |
54595fe2 | 1048 | } |
d46eb14b SB |
1049 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
1050 | ||
f745c6f5 SB |
1051 | /** |
1052 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
1053 | * @page: page from which memcg should be extracted. | |
1054 | * | |
1055 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
1056 | * root_mem_cgroup is returned. | |
1057 | */ | |
1058 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
1059 | { | |
1060 | struct mem_cgroup *memcg = page->mem_cgroup; | |
1061 | ||
1062 | if (mem_cgroup_disabled()) | |
1063 | return NULL; | |
1064 | ||
1065 | rcu_read_lock(); | |
8965aa28 SB |
1066 | /* Page should not get uncharged and freed memcg under us. */ |
1067 | if (!memcg || WARN_ON_ONCE(!css_tryget(&memcg->css))) | |
f745c6f5 SB |
1068 | memcg = root_mem_cgroup; |
1069 | rcu_read_unlock(); | |
1070 | return memcg; | |
1071 | } | |
1072 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
1073 | ||
d46eb14b SB |
1074 | /** |
1075 | * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg. | |
1076 | */ | |
1077 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
1078 | { | |
1079 | if (unlikely(current->active_memcg)) { | |
8965aa28 | 1080 | struct mem_cgroup *memcg; |
d46eb14b SB |
1081 | |
1082 | rcu_read_lock(); | |
8965aa28 SB |
1083 | /* current->active_memcg must hold a ref. */ |
1084 | if (WARN_ON_ONCE(!css_tryget(¤t->active_memcg->css))) | |
1085 | memcg = root_mem_cgroup; | |
1086 | else | |
d46eb14b SB |
1087 | memcg = current->active_memcg; |
1088 | rcu_read_unlock(); | |
1089 | return memcg; | |
1090 | } | |
1091 | return get_mem_cgroup_from_mm(current->mm); | |
1092 | } | |
54595fe2 | 1093 | |
5660048c JW |
1094 | /** |
1095 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1096 | * @root: hierarchy root | |
1097 | * @prev: previously returned memcg, NULL on first invocation | |
1098 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1099 | * | |
1100 | * Returns references to children of the hierarchy below @root, or | |
1101 | * @root itself, or %NULL after a full round-trip. | |
1102 | * | |
1103 | * Caller must pass the return value in @prev on subsequent | |
1104 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1105 | * to cancel a hierarchy walk before the round-trip is complete. | |
1106 | * | |
b213b54f | 1107 | * Reclaimers can specify a node and a priority level in @reclaim to |
5660048c | 1108 | * divide up the memcgs in the hierarchy among all concurrent |
b213b54f | 1109 | * reclaimers operating on the same node and priority. |
5660048c | 1110 | */ |
694fbc0f | 1111 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1112 | struct mem_cgroup *prev, |
694fbc0f | 1113 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1114 | { |
3f649ab7 | 1115 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 1116 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1117 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1118 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1119 | |
694fbc0f AM |
1120 | if (mem_cgroup_disabled()) |
1121 | return NULL; | |
5660048c | 1122 | |
9f3a0d09 JW |
1123 | if (!root) |
1124 | root = root_mem_cgroup; | |
7d74b06f | 1125 | |
9f3a0d09 | 1126 | if (prev && !reclaim) |
5ac8fb31 | 1127 | pos = prev; |
14067bb3 | 1128 | |
9f3a0d09 JW |
1129 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1130 | if (prev) | |
5ac8fb31 | 1131 | goto out; |
694fbc0f | 1132 | return root; |
9f3a0d09 | 1133 | } |
14067bb3 | 1134 | |
542f85f9 | 1135 | rcu_read_lock(); |
5f578161 | 1136 | |
5ac8fb31 | 1137 | if (reclaim) { |
ef8f2327 | 1138 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1139 | |
ef8f2327 | 1140 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
9da83f3f | 1141 | iter = &mz->iter; |
5ac8fb31 JW |
1142 | |
1143 | if (prev && reclaim->generation != iter->generation) | |
1144 | goto out_unlock; | |
1145 | ||
6df38689 | 1146 | while (1) { |
4db0c3c2 | 1147 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1148 | if (!pos || css_tryget(&pos->css)) |
1149 | break; | |
5ac8fb31 | 1150 | /* |
6df38689 VD |
1151 | * css reference reached zero, so iter->position will |
1152 | * be cleared by ->css_released. However, we should not | |
1153 | * rely on this happening soon, because ->css_released | |
1154 | * is called from a work queue, and by busy-waiting we | |
1155 | * might block it. So we clear iter->position right | |
1156 | * away. | |
5ac8fb31 | 1157 | */ |
6df38689 VD |
1158 | (void)cmpxchg(&iter->position, pos, NULL); |
1159 | } | |
5ac8fb31 JW |
1160 | } |
1161 | ||
1162 | if (pos) | |
1163 | css = &pos->css; | |
1164 | ||
1165 | for (;;) { | |
1166 | css = css_next_descendant_pre(css, &root->css); | |
1167 | if (!css) { | |
1168 | /* | |
1169 | * Reclaimers share the hierarchy walk, and a | |
1170 | * new one might jump in right at the end of | |
1171 | * the hierarchy - make sure they see at least | |
1172 | * one group and restart from the beginning. | |
1173 | */ | |
1174 | if (!prev) | |
1175 | continue; | |
1176 | break; | |
527a5ec9 | 1177 | } |
7d74b06f | 1178 | |
5ac8fb31 JW |
1179 | /* |
1180 | * Verify the css and acquire a reference. The root | |
1181 | * is provided by the caller, so we know it's alive | |
1182 | * and kicking, and don't take an extra reference. | |
1183 | */ | |
1184 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1185 | |
5ac8fb31 JW |
1186 | if (css == &root->css) |
1187 | break; | |
14067bb3 | 1188 | |
0b8f73e1 JW |
1189 | if (css_tryget(css)) |
1190 | break; | |
9f3a0d09 | 1191 | |
5ac8fb31 | 1192 | memcg = NULL; |
9f3a0d09 | 1193 | } |
5ac8fb31 JW |
1194 | |
1195 | if (reclaim) { | |
5ac8fb31 | 1196 | /* |
6df38689 VD |
1197 | * The position could have already been updated by a competing |
1198 | * thread, so check that the value hasn't changed since we read | |
1199 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1200 | */ |
6df38689 VD |
1201 | (void)cmpxchg(&iter->position, pos, memcg); |
1202 | ||
5ac8fb31 JW |
1203 | if (pos) |
1204 | css_put(&pos->css); | |
1205 | ||
1206 | if (!memcg) | |
1207 | iter->generation++; | |
1208 | else if (!prev) | |
1209 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1210 | } |
5ac8fb31 | 1211 | |
542f85f9 MH |
1212 | out_unlock: |
1213 | rcu_read_unlock(); | |
5ac8fb31 | 1214 | out: |
c40046f3 MH |
1215 | if (prev && prev != root) |
1216 | css_put(&prev->css); | |
1217 | ||
9f3a0d09 | 1218 | return memcg; |
14067bb3 | 1219 | } |
7d74b06f | 1220 | |
5660048c JW |
1221 | /** |
1222 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1223 | * @root: hierarchy root | |
1224 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1225 | */ | |
1226 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1227 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1228 | { |
1229 | if (!root) | |
1230 | root = root_mem_cgroup; | |
1231 | if (prev && prev != root) | |
1232 | css_put(&prev->css); | |
1233 | } | |
7d74b06f | 1234 | |
54a83d6b MC |
1235 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1236 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1237 | { |
6df38689 | 1238 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1239 | struct mem_cgroup_per_node *mz; |
1240 | int nid; | |
6df38689 | 1241 | |
54a83d6b MC |
1242 | for_each_node(nid) { |
1243 | mz = mem_cgroup_nodeinfo(from, nid); | |
9da83f3f YS |
1244 | iter = &mz->iter; |
1245 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1246 | } |
1247 | } | |
1248 | ||
54a83d6b MC |
1249 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1250 | { | |
1251 | struct mem_cgroup *memcg = dead_memcg; | |
1252 | struct mem_cgroup *last; | |
1253 | ||
1254 | do { | |
1255 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1256 | last = memcg; | |
1257 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1258 | ||
1259 | /* | |
1260 | * When cgruop1 non-hierarchy mode is used, | |
1261 | * parent_mem_cgroup() does not walk all the way up to the | |
1262 | * cgroup root (root_mem_cgroup). So we have to handle | |
1263 | * dead_memcg from cgroup root separately. | |
1264 | */ | |
1265 | if (last != root_mem_cgroup) | |
1266 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1267 | dead_memcg); | |
1268 | } | |
1269 | ||
7c5f64f8 VD |
1270 | /** |
1271 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1272 | * @memcg: hierarchy root | |
1273 | * @fn: function to call for each task | |
1274 | * @arg: argument passed to @fn | |
1275 | * | |
1276 | * This function iterates over tasks attached to @memcg or to any of its | |
1277 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1278 | * value, the function breaks the iteration loop and returns the value. | |
1279 | * Otherwise, it will iterate over all tasks and return 0. | |
1280 | * | |
1281 | * This function must not be called for the root memory cgroup. | |
1282 | */ | |
1283 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1284 | int (*fn)(struct task_struct *, void *), void *arg) | |
1285 | { | |
1286 | struct mem_cgroup *iter; | |
1287 | int ret = 0; | |
1288 | ||
1289 | BUG_ON(memcg == root_mem_cgroup); | |
1290 | ||
1291 | for_each_mem_cgroup_tree(iter, memcg) { | |
1292 | struct css_task_iter it; | |
1293 | struct task_struct *task; | |
1294 | ||
f168a9a5 | 1295 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1296 | while (!ret && (task = css_task_iter_next(&it))) |
1297 | ret = fn(task, arg); | |
1298 | css_task_iter_end(&it); | |
1299 | if (ret) { | |
1300 | mem_cgroup_iter_break(memcg, iter); | |
1301 | break; | |
1302 | } | |
1303 | } | |
1304 | return ret; | |
1305 | } | |
1306 | ||
925b7673 | 1307 | /** |
dfe0e773 | 1308 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1309 | * @page: the page |
f144c390 | 1310 | * @pgdat: pgdat of the page |
dfe0e773 | 1311 | * |
a0b5b414 JW |
1312 | * This function relies on page->mem_cgroup being stable - see the |
1313 | * access rules in commit_charge(). | |
925b7673 | 1314 | */ |
599d0c95 | 1315 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1316 | { |
ef8f2327 | 1317 | struct mem_cgroup_per_node *mz; |
925b7673 | 1318 | struct mem_cgroup *memcg; |
bea8c150 | 1319 | struct lruvec *lruvec; |
6d12e2d8 | 1320 | |
bea8c150 | 1321 | if (mem_cgroup_disabled()) { |
867e5e1d | 1322 | lruvec = &pgdat->__lruvec; |
bea8c150 HD |
1323 | goto out; |
1324 | } | |
925b7673 | 1325 | |
1306a85a | 1326 | memcg = page->mem_cgroup; |
7512102c | 1327 | /* |
dfe0e773 | 1328 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1329 | * possibly migrated - before they are charged. |
7512102c | 1330 | */ |
29833315 JW |
1331 | if (!memcg) |
1332 | memcg = root_mem_cgroup; | |
7512102c | 1333 | |
ef8f2327 | 1334 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1335 | lruvec = &mz->lruvec; |
1336 | out: | |
1337 | /* | |
1338 | * Since a node can be onlined after the mem_cgroup was created, | |
1339 | * we have to be prepared to initialize lruvec->zone here; | |
1340 | * and if offlined then reonlined, we need to reinitialize it. | |
1341 | */ | |
599d0c95 MG |
1342 | if (unlikely(lruvec->pgdat != pgdat)) |
1343 | lruvec->pgdat = pgdat; | |
bea8c150 | 1344 | return lruvec; |
08e552c6 | 1345 | } |
b69408e8 | 1346 | |
925b7673 | 1347 | /** |
fa9add64 HD |
1348 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1349 | * @lruvec: mem_cgroup per zone lru vector | |
1350 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1351 | * @zid: zone id of the accounted pages |
fa9add64 | 1352 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1353 | * |
ca707239 HD |
1354 | * This function must be called under lru_lock, just before a page is added |
1355 | * to or just after a page is removed from an lru list (that ordering being | |
1356 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1357 | */ |
fa9add64 | 1358 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1359 | int zid, int nr_pages) |
3f58a829 | 1360 | { |
ef8f2327 | 1361 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1362 | unsigned long *lru_size; |
ca707239 | 1363 | long size; |
3f58a829 MK |
1364 | |
1365 | if (mem_cgroup_disabled()) | |
1366 | return; | |
1367 | ||
ef8f2327 | 1368 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1369 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1370 | |
1371 | if (nr_pages < 0) | |
1372 | *lru_size += nr_pages; | |
1373 | ||
1374 | size = *lru_size; | |
b4536f0c MH |
1375 | if (WARN_ONCE(size < 0, |
1376 | "%s(%p, %d, %d): lru_size %ld\n", | |
1377 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1378 | VM_BUG_ON(1); |
1379 | *lru_size = 0; | |
1380 | } | |
1381 | ||
1382 | if (nr_pages > 0) | |
1383 | *lru_size += nr_pages; | |
08e552c6 | 1384 | } |
544122e5 | 1385 | |
19942822 | 1386 | /** |
9d11ea9f | 1387 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1388 | * @memcg: the memory cgroup |
19942822 | 1389 | * |
9d11ea9f | 1390 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1391 | * pages. |
19942822 | 1392 | */ |
c0ff4b85 | 1393 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1394 | { |
3e32cb2e JW |
1395 | unsigned long margin = 0; |
1396 | unsigned long count; | |
1397 | unsigned long limit; | |
9d11ea9f | 1398 | |
3e32cb2e | 1399 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1400 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1401 | if (count < limit) |
1402 | margin = limit - count; | |
1403 | ||
7941d214 | 1404 | if (do_memsw_account()) { |
3e32cb2e | 1405 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1406 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1407 | if (count < limit) |
3e32cb2e | 1408 | margin = min(margin, limit - count); |
cbedbac3 LR |
1409 | else |
1410 | margin = 0; | |
3e32cb2e JW |
1411 | } |
1412 | ||
1413 | return margin; | |
19942822 JW |
1414 | } |
1415 | ||
32047e2a | 1416 | /* |
bdcbb659 | 1417 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1418 | * |
bdcbb659 QH |
1419 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1420 | * moving cgroups. This is for waiting at high-memory pressure | |
1421 | * caused by "move". | |
32047e2a | 1422 | */ |
c0ff4b85 | 1423 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1424 | { |
2bd9bb20 KH |
1425 | struct mem_cgroup *from; |
1426 | struct mem_cgroup *to; | |
4b534334 | 1427 | bool ret = false; |
2bd9bb20 KH |
1428 | /* |
1429 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1430 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1431 | */ | |
1432 | spin_lock(&mc.lock); | |
1433 | from = mc.from; | |
1434 | to = mc.to; | |
1435 | if (!from) | |
1436 | goto unlock; | |
3e92041d | 1437 | |
2314b42d JW |
1438 | ret = mem_cgroup_is_descendant(from, memcg) || |
1439 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1440 | unlock: |
1441 | spin_unlock(&mc.lock); | |
4b534334 KH |
1442 | return ret; |
1443 | } | |
1444 | ||
c0ff4b85 | 1445 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1446 | { |
1447 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1448 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1449 | DEFINE_WAIT(wait); |
1450 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1451 | /* moving charge context might have finished. */ | |
1452 | if (mc.moving_task) | |
1453 | schedule(); | |
1454 | finish_wait(&mc.waitq, &wait); | |
1455 | return true; | |
1456 | } | |
1457 | } | |
1458 | return false; | |
1459 | } | |
1460 | ||
c8713d0b JW |
1461 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1462 | { | |
1463 | struct seq_buf s; | |
1464 | int i; | |
71cd3113 | 1465 | |
c8713d0b JW |
1466 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1467 | if (!s.buffer) | |
1468 | return NULL; | |
1469 | ||
1470 | /* | |
1471 | * Provide statistics on the state of the memory subsystem as | |
1472 | * well as cumulative event counters that show past behavior. | |
1473 | * | |
1474 | * This list is ordered following a combination of these gradients: | |
1475 | * 1) generic big picture -> specifics and details | |
1476 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1477 | * | |
1478 | * Current memory state: | |
1479 | */ | |
1480 | ||
1481 | seq_buf_printf(&s, "anon %llu\n", | |
be5d0a74 | 1482 | (u64)memcg_page_state(memcg, NR_ANON_MAPPED) * |
c8713d0b JW |
1483 | PAGE_SIZE); |
1484 | seq_buf_printf(&s, "file %llu\n", | |
0d1c2072 | 1485 | (u64)memcg_page_state(memcg, NR_FILE_PAGES) * |
c8713d0b JW |
1486 | PAGE_SIZE); |
1487 | seq_buf_printf(&s, "kernel_stack %llu\n", | |
991e7673 | 1488 | (u64)memcg_page_state(memcg, NR_KERNEL_STACK_KB) * |
c8713d0b JW |
1489 | 1024); |
1490 | seq_buf_printf(&s, "slab %llu\n", | |
d42f3245 RG |
1491 | (u64)(memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B) + |
1492 | memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B))); | |
c8713d0b JW |
1493 | seq_buf_printf(&s, "sock %llu\n", |
1494 | (u64)memcg_page_state(memcg, MEMCG_SOCK) * | |
1495 | PAGE_SIZE); | |
1496 | ||
1497 | seq_buf_printf(&s, "shmem %llu\n", | |
1498 | (u64)memcg_page_state(memcg, NR_SHMEM) * | |
1499 | PAGE_SIZE); | |
1500 | seq_buf_printf(&s, "file_mapped %llu\n", | |
1501 | (u64)memcg_page_state(memcg, NR_FILE_MAPPED) * | |
1502 | PAGE_SIZE); | |
1503 | seq_buf_printf(&s, "file_dirty %llu\n", | |
1504 | (u64)memcg_page_state(memcg, NR_FILE_DIRTY) * | |
1505 | PAGE_SIZE); | |
1506 | seq_buf_printf(&s, "file_writeback %llu\n", | |
1507 | (u64)memcg_page_state(memcg, NR_WRITEBACK) * | |
1508 | PAGE_SIZE); | |
1509 | ||
468c3982 | 1510 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 1511 | seq_buf_printf(&s, "anon_thp %llu\n", |
468c3982 JW |
1512 | (u64)memcg_page_state(memcg, NR_ANON_THPS) * |
1513 | HPAGE_PMD_SIZE); | |
1514 | #endif | |
c8713d0b JW |
1515 | |
1516 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 1517 | seq_buf_printf(&s, "%s %llu\n", lru_list_name(i), |
c8713d0b JW |
1518 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
1519 | PAGE_SIZE); | |
1520 | ||
1521 | seq_buf_printf(&s, "slab_reclaimable %llu\n", | |
d42f3245 | 1522 | (u64)memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B)); |
c8713d0b | 1523 | seq_buf_printf(&s, "slab_unreclaimable %llu\n", |
d42f3245 | 1524 | (u64)memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B)); |
c8713d0b JW |
1525 | |
1526 | /* Accumulated memory events */ | |
1527 | ||
ebc5d83d KK |
1528 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT), |
1529 | memcg_events(memcg, PGFAULT)); | |
1530 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT), | |
1531 | memcg_events(memcg, PGMAJFAULT)); | |
c8713d0b JW |
1532 | |
1533 | seq_buf_printf(&s, "workingset_refault %lu\n", | |
1534 | memcg_page_state(memcg, WORKINGSET_REFAULT)); | |
1535 | seq_buf_printf(&s, "workingset_activate %lu\n", | |
1536 | memcg_page_state(memcg, WORKINGSET_ACTIVATE)); | |
a6f5576b YS |
1537 | seq_buf_printf(&s, "workingset_restore %lu\n", |
1538 | memcg_page_state(memcg, WORKINGSET_RESTORE)); | |
c8713d0b JW |
1539 | seq_buf_printf(&s, "workingset_nodereclaim %lu\n", |
1540 | memcg_page_state(memcg, WORKINGSET_NODERECLAIM)); | |
1541 | ||
ebc5d83d KK |
1542 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL), |
1543 | memcg_events(memcg, PGREFILL)); | |
c8713d0b JW |
1544 | seq_buf_printf(&s, "pgscan %lu\n", |
1545 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1546 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1547 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1548 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1549 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
ebc5d83d KK |
1550 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE), |
1551 | memcg_events(memcg, PGACTIVATE)); | |
1552 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE), | |
1553 | memcg_events(memcg, PGDEACTIVATE)); | |
1554 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE), | |
1555 | memcg_events(memcg, PGLAZYFREE)); | |
1556 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED), | |
1557 | memcg_events(memcg, PGLAZYFREED)); | |
c8713d0b JW |
1558 | |
1559 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
ebc5d83d | 1560 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC), |
c8713d0b | 1561 | memcg_events(memcg, THP_FAULT_ALLOC)); |
ebc5d83d | 1562 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC), |
c8713d0b JW |
1563 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1564 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1565 | ||
1566 | /* The above should easily fit into one page */ | |
1567 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1568 | ||
1569 | return s.buffer; | |
1570 | } | |
71cd3113 | 1571 | |
58cf188e | 1572 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1573 | /** |
f0c867d9 | 1574 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1575 | * memory controller. | |
e222432b BS |
1576 | * @memcg: The memory cgroup that went over limit |
1577 | * @p: Task that is going to be killed | |
1578 | * | |
1579 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1580 | * enabled | |
1581 | */ | |
f0c867d9 | 1582 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1583 | { |
e222432b BS |
1584 | rcu_read_lock(); |
1585 | ||
f0c867d9 | 1586 | if (memcg) { |
1587 | pr_cont(",oom_memcg="); | |
1588 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1589 | } else | |
1590 | pr_cont(",global_oom"); | |
2415b9f5 | 1591 | if (p) { |
f0c867d9 | 1592 | pr_cont(",task_memcg="); |
2415b9f5 | 1593 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1594 | } |
e222432b | 1595 | rcu_read_unlock(); |
f0c867d9 | 1596 | } |
1597 | ||
1598 | /** | |
1599 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1600 | * memory controller. | |
1601 | * @memcg: The memory cgroup that went over limit | |
1602 | */ | |
1603 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1604 | { | |
c8713d0b | 1605 | char *buf; |
e222432b | 1606 | |
3e32cb2e JW |
1607 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1608 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1609 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1610 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1611 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1612 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1613 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1614 | else { |
1615 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1616 | K((u64)page_counter_read(&memcg->memsw)), | |
1617 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1618 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1619 | K((u64)page_counter_read(&memcg->kmem)), | |
1620 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1621 | } |
c8713d0b JW |
1622 | |
1623 | pr_info("Memory cgroup stats for "); | |
1624 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1625 | pr_cont(":"); | |
1626 | buf = memory_stat_format(memcg); | |
1627 | if (!buf) | |
1628 | return; | |
1629 | pr_info("%s", buf); | |
1630 | kfree(buf); | |
e222432b BS |
1631 | } |
1632 | ||
a63d83f4 DR |
1633 | /* |
1634 | * Return the memory (and swap, if configured) limit for a memcg. | |
1635 | */ | |
bbec2e15 | 1636 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1637 | { |
bbec2e15 | 1638 | unsigned long max; |
f3e8eb70 | 1639 | |
15b42562 | 1640 | max = READ_ONCE(memcg->memory.max); |
9a5a8f19 | 1641 | if (mem_cgroup_swappiness(memcg)) { |
bbec2e15 RG |
1642 | unsigned long memsw_max; |
1643 | unsigned long swap_max; | |
9a5a8f19 | 1644 | |
bbec2e15 | 1645 | memsw_max = memcg->memsw.max; |
32d087cd | 1646 | swap_max = READ_ONCE(memcg->swap.max); |
bbec2e15 RG |
1647 | swap_max = min(swap_max, (unsigned long)total_swap_pages); |
1648 | max = min(max + swap_max, memsw_max); | |
9a5a8f19 | 1649 | } |
bbec2e15 | 1650 | return max; |
a63d83f4 DR |
1651 | } |
1652 | ||
9783aa99 CD |
1653 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1654 | { | |
1655 | return page_counter_read(&memcg->memory); | |
1656 | } | |
1657 | ||
b6e6edcf | 1658 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1659 | int order) |
9cbb78bb | 1660 | { |
6e0fc46d DR |
1661 | struct oom_control oc = { |
1662 | .zonelist = NULL, | |
1663 | .nodemask = NULL, | |
2a966b77 | 1664 | .memcg = memcg, |
6e0fc46d DR |
1665 | .gfp_mask = gfp_mask, |
1666 | .order = order, | |
6e0fc46d | 1667 | }; |
7c5f64f8 | 1668 | bool ret; |
9cbb78bb | 1669 | |
7775face TH |
1670 | if (mutex_lock_killable(&oom_lock)) |
1671 | return true; | |
1672 | /* | |
1673 | * A few threads which were not waiting at mutex_lock_killable() can | |
1674 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1675 | */ | |
1676 | ret = should_force_charge() || out_of_memory(&oc); | |
dc56401f | 1677 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1678 | return ret; |
9cbb78bb DR |
1679 | } |
1680 | ||
0608f43d | 1681 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1682 | pg_data_t *pgdat, |
0608f43d AM |
1683 | gfp_t gfp_mask, |
1684 | unsigned long *total_scanned) | |
1685 | { | |
1686 | struct mem_cgroup *victim = NULL; | |
1687 | int total = 0; | |
1688 | int loop = 0; | |
1689 | unsigned long excess; | |
1690 | unsigned long nr_scanned; | |
1691 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1692 | .pgdat = pgdat, |
0608f43d AM |
1693 | }; |
1694 | ||
3e32cb2e | 1695 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1696 | |
1697 | while (1) { | |
1698 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1699 | if (!victim) { | |
1700 | loop++; | |
1701 | if (loop >= 2) { | |
1702 | /* | |
1703 | * If we have not been able to reclaim | |
1704 | * anything, it might because there are | |
1705 | * no reclaimable pages under this hierarchy | |
1706 | */ | |
1707 | if (!total) | |
1708 | break; | |
1709 | /* | |
1710 | * We want to do more targeted reclaim. | |
1711 | * excess >> 2 is not to excessive so as to | |
1712 | * reclaim too much, nor too less that we keep | |
1713 | * coming back to reclaim from this cgroup | |
1714 | */ | |
1715 | if (total >= (excess >> 2) || | |
1716 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1717 | break; | |
1718 | } | |
1719 | continue; | |
1720 | } | |
a9dd0a83 | 1721 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1722 | pgdat, &nr_scanned); |
0608f43d | 1723 | *total_scanned += nr_scanned; |
3e32cb2e | 1724 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1725 | break; |
6d61ef40 | 1726 | } |
0608f43d AM |
1727 | mem_cgroup_iter_break(root_memcg, victim); |
1728 | return total; | |
6d61ef40 BS |
1729 | } |
1730 | ||
0056f4e6 JW |
1731 | #ifdef CONFIG_LOCKDEP |
1732 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1733 | .name = "memcg_oom_lock", | |
1734 | }; | |
1735 | #endif | |
1736 | ||
fb2a6fc5 JW |
1737 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1738 | ||
867578cb KH |
1739 | /* |
1740 | * Check OOM-Killer is already running under our hierarchy. | |
1741 | * If someone is running, return false. | |
1742 | */ | |
fb2a6fc5 | 1743 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1744 | { |
79dfdacc | 1745 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1746 | |
fb2a6fc5 JW |
1747 | spin_lock(&memcg_oom_lock); |
1748 | ||
9f3a0d09 | 1749 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1750 | if (iter->oom_lock) { |
79dfdacc MH |
1751 | /* |
1752 | * this subtree of our hierarchy is already locked | |
1753 | * so we cannot give a lock. | |
1754 | */ | |
79dfdacc | 1755 | failed = iter; |
9f3a0d09 JW |
1756 | mem_cgroup_iter_break(memcg, iter); |
1757 | break; | |
23751be0 JW |
1758 | } else |
1759 | iter->oom_lock = true; | |
7d74b06f | 1760 | } |
867578cb | 1761 | |
fb2a6fc5 JW |
1762 | if (failed) { |
1763 | /* | |
1764 | * OK, we failed to lock the whole subtree so we have | |
1765 | * to clean up what we set up to the failing subtree | |
1766 | */ | |
1767 | for_each_mem_cgroup_tree(iter, memcg) { | |
1768 | if (iter == failed) { | |
1769 | mem_cgroup_iter_break(memcg, iter); | |
1770 | break; | |
1771 | } | |
1772 | iter->oom_lock = false; | |
79dfdacc | 1773 | } |
0056f4e6 JW |
1774 | } else |
1775 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1776 | |
1777 | spin_unlock(&memcg_oom_lock); | |
1778 | ||
1779 | return !failed; | |
a636b327 | 1780 | } |
0b7f569e | 1781 | |
fb2a6fc5 | 1782 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1783 | { |
7d74b06f KH |
1784 | struct mem_cgroup *iter; |
1785 | ||
fb2a6fc5 | 1786 | spin_lock(&memcg_oom_lock); |
5facae4f | 1787 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1788 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1789 | iter->oom_lock = false; |
fb2a6fc5 | 1790 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1791 | } |
1792 | ||
c0ff4b85 | 1793 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1794 | { |
1795 | struct mem_cgroup *iter; | |
1796 | ||
c2b42d3c | 1797 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1798 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1799 | iter->under_oom++; |
1800 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1801 | } |
1802 | ||
c0ff4b85 | 1803 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1804 | { |
1805 | struct mem_cgroup *iter; | |
1806 | ||
867578cb KH |
1807 | /* |
1808 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1809 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1810 | */ |
c2b42d3c | 1811 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1812 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1813 | if (iter->under_oom > 0) |
1814 | iter->under_oom--; | |
1815 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1816 | } |
1817 | ||
867578cb KH |
1818 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1819 | ||
dc98df5a | 1820 | struct oom_wait_info { |
d79154bb | 1821 | struct mem_cgroup *memcg; |
ac6424b9 | 1822 | wait_queue_entry_t wait; |
dc98df5a KH |
1823 | }; |
1824 | ||
ac6424b9 | 1825 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1826 | unsigned mode, int sync, void *arg) |
1827 | { | |
d79154bb HD |
1828 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1829 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1830 | struct oom_wait_info *oom_wait_info; |
1831 | ||
1832 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1833 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1834 | |
2314b42d JW |
1835 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1836 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1837 | return 0; |
dc98df5a KH |
1838 | return autoremove_wake_function(wait, mode, sync, arg); |
1839 | } | |
1840 | ||
c0ff4b85 | 1841 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1842 | { |
c2b42d3c TH |
1843 | /* |
1844 | * For the following lockless ->under_oom test, the only required | |
1845 | * guarantee is that it must see the state asserted by an OOM when | |
1846 | * this function is called as a result of userland actions | |
1847 | * triggered by the notification of the OOM. This is trivially | |
1848 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1849 | * triggering notification. | |
1850 | */ | |
1851 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1852 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1853 | } |
1854 | ||
29ef680a MH |
1855 | enum oom_status { |
1856 | OOM_SUCCESS, | |
1857 | OOM_FAILED, | |
1858 | OOM_ASYNC, | |
1859 | OOM_SKIPPED | |
1860 | }; | |
1861 | ||
1862 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1863 | { |
7056d3a3 MH |
1864 | enum oom_status ret; |
1865 | bool locked; | |
1866 | ||
29ef680a MH |
1867 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1868 | return OOM_SKIPPED; | |
1869 | ||
7a1adfdd RG |
1870 | memcg_memory_event(memcg, MEMCG_OOM); |
1871 | ||
867578cb | 1872 | /* |
49426420 JW |
1873 | * We are in the middle of the charge context here, so we |
1874 | * don't want to block when potentially sitting on a callstack | |
1875 | * that holds all kinds of filesystem and mm locks. | |
1876 | * | |
29ef680a MH |
1877 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1878 | * handling until the charge can succeed; remember the context and put | |
1879 | * the task to sleep at the end of the page fault when all locks are | |
1880 | * released. | |
49426420 | 1881 | * |
29ef680a MH |
1882 | * On the other hand, in-kernel OOM killer allows for an async victim |
1883 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1884 | * relying on the oom victim to make a forward progress and we can | |
1885 | * invoke the oom killer here. | |
1886 | * | |
1887 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1888 | * victim and then we have to bail out from the charge path. | |
867578cb | 1889 | */ |
29ef680a MH |
1890 | if (memcg->oom_kill_disable) { |
1891 | if (!current->in_user_fault) | |
1892 | return OOM_SKIPPED; | |
1893 | css_get(&memcg->css); | |
1894 | current->memcg_in_oom = memcg; | |
1895 | current->memcg_oom_gfp_mask = mask; | |
1896 | current->memcg_oom_order = order; | |
1897 | ||
1898 | return OOM_ASYNC; | |
1899 | } | |
1900 | ||
7056d3a3 MH |
1901 | mem_cgroup_mark_under_oom(memcg); |
1902 | ||
1903 | locked = mem_cgroup_oom_trylock(memcg); | |
1904 | ||
1905 | if (locked) | |
1906 | mem_cgroup_oom_notify(memcg); | |
1907 | ||
1908 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1909 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1910 | ret = OOM_SUCCESS; |
1911 | else | |
1912 | ret = OOM_FAILED; | |
1913 | ||
1914 | if (locked) | |
1915 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1916 | |
7056d3a3 | 1917 | return ret; |
3812c8c8 JW |
1918 | } |
1919 | ||
1920 | /** | |
1921 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1922 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1923 | * |
49426420 JW |
1924 | * This has to be called at the end of a page fault if the memcg OOM |
1925 | * handler was enabled. | |
3812c8c8 | 1926 | * |
49426420 | 1927 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1928 | * sleep on a waitqueue until the userspace task resolves the |
1929 | * situation. Sleeping directly in the charge context with all kinds | |
1930 | * of locks held is not a good idea, instead we remember an OOM state | |
1931 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1932 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1933 | * |
1934 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1935 | * completed, %false otherwise. |
3812c8c8 | 1936 | */ |
49426420 | 1937 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1938 | { |
626ebc41 | 1939 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1940 | struct oom_wait_info owait; |
49426420 | 1941 | bool locked; |
3812c8c8 JW |
1942 | |
1943 | /* OOM is global, do not handle */ | |
3812c8c8 | 1944 | if (!memcg) |
49426420 | 1945 | return false; |
3812c8c8 | 1946 | |
7c5f64f8 | 1947 | if (!handle) |
49426420 | 1948 | goto cleanup; |
3812c8c8 JW |
1949 | |
1950 | owait.memcg = memcg; | |
1951 | owait.wait.flags = 0; | |
1952 | owait.wait.func = memcg_oom_wake_function; | |
1953 | owait.wait.private = current; | |
2055da97 | 1954 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1955 | |
3812c8c8 | 1956 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1957 | mem_cgroup_mark_under_oom(memcg); |
1958 | ||
1959 | locked = mem_cgroup_oom_trylock(memcg); | |
1960 | ||
1961 | if (locked) | |
1962 | mem_cgroup_oom_notify(memcg); | |
1963 | ||
1964 | if (locked && !memcg->oom_kill_disable) { | |
1965 | mem_cgroup_unmark_under_oom(memcg); | |
1966 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1967 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1968 | current->memcg_oom_order); | |
49426420 | 1969 | } else { |
3812c8c8 | 1970 | schedule(); |
49426420 JW |
1971 | mem_cgroup_unmark_under_oom(memcg); |
1972 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1973 | } | |
1974 | ||
1975 | if (locked) { | |
fb2a6fc5 JW |
1976 | mem_cgroup_oom_unlock(memcg); |
1977 | /* | |
1978 | * There is no guarantee that an OOM-lock contender | |
1979 | * sees the wakeups triggered by the OOM kill | |
1980 | * uncharges. Wake any sleepers explicitely. | |
1981 | */ | |
1982 | memcg_oom_recover(memcg); | |
1983 | } | |
49426420 | 1984 | cleanup: |
626ebc41 | 1985 | current->memcg_in_oom = NULL; |
3812c8c8 | 1986 | css_put(&memcg->css); |
867578cb | 1987 | return true; |
0b7f569e KH |
1988 | } |
1989 | ||
3d8b38eb RG |
1990 | /** |
1991 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
1992 | * @victim: task to be killed by the OOM killer | |
1993 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
1994 | * | |
1995 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
1996 | * by killing all belonging OOM-killable tasks. | |
1997 | * | |
1998 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
1999 | */ | |
2000 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2001 | struct mem_cgroup *oom_domain) | |
2002 | { | |
2003 | struct mem_cgroup *oom_group = NULL; | |
2004 | struct mem_cgroup *memcg; | |
2005 | ||
2006 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2007 | return NULL; | |
2008 | ||
2009 | if (!oom_domain) | |
2010 | oom_domain = root_mem_cgroup; | |
2011 | ||
2012 | rcu_read_lock(); | |
2013 | ||
2014 | memcg = mem_cgroup_from_task(victim); | |
2015 | if (memcg == root_mem_cgroup) | |
2016 | goto out; | |
2017 | ||
48fe267c RG |
2018 | /* |
2019 | * If the victim task has been asynchronously moved to a different | |
2020 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
2021 | * In this case it's better to ignore memory.group.oom. | |
2022 | */ | |
2023 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
2024 | goto out; | |
2025 | ||
3d8b38eb RG |
2026 | /* |
2027 | * Traverse the memory cgroup hierarchy from the victim task's | |
2028 | * cgroup up to the OOMing cgroup (or root) to find the | |
2029 | * highest-level memory cgroup with oom.group set. | |
2030 | */ | |
2031 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
2032 | if (memcg->oom_group) | |
2033 | oom_group = memcg; | |
2034 | ||
2035 | if (memcg == oom_domain) | |
2036 | break; | |
2037 | } | |
2038 | ||
2039 | if (oom_group) | |
2040 | css_get(&oom_group->css); | |
2041 | out: | |
2042 | rcu_read_unlock(); | |
2043 | ||
2044 | return oom_group; | |
2045 | } | |
2046 | ||
2047 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2048 | { | |
2049 | pr_info("Tasks in "); | |
2050 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2051 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2052 | } | |
2053 | ||
d7365e78 | 2054 | /** |
81f8c3a4 JW |
2055 | * lock_page_memcg - lock a page->mem_cgroup binding |
2056 | * @page: the page | |
32047e2a | 2057 | * |
81f8c3a4 | 2058 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
2059 | * another cgroup. |
2060 | * | |
2061 | * It ensures lifetime of the returned memcg. Caller is responsible | |
2062 | * for the lifetime of the page; __unlock_page_memcg() is available | |
2063 | * when @page might get freed inside the locked section. | |
d69b042f | 2064 | */ |
739f79fc | 2065 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 | 2066 | { |
9da7b521 | 2067 | struct page *head = compound_head(page); /* rmap on tail pages */ |
89c06bd5 | 2068 | struct mem_cgroup *memcg; |
6de22619 | 2069 | unsigned long flags; |
89c06bd5 | 2070 | |
6de22619 JW |
2071 | /* |
2072 | * The RCU lock is held throughout the transaction. The fast | |
2073 | * path can get away without acquiring the memcg->move_lock | |
2074 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
2075 | * |
2076 | * The RCU lock also protects the memcg from being freed when | |
2077 | * the page state that is going to change is the only thing | |
2078 | * preventing the page itself from being freed. E.g. writeback | |
2079 | * doesn't hold a page reference and relies on PG_writeback to | |
2080 | * keep off truncation, migration and so forth. | |
2081 | */ | |
d7365e78 JW |
2082 | rcu_read_lock(); |
2083 | ||
2084 | if (mem_cgroup_disabled()) | |
739f79fc | 2085 | return NULL; |
89c06bd5 | 2086 | again: |
9da7b521 | 2087 | memcg = head->mem_cgroup; |
29833315 | 2088 | if (unlikely(!memcg)) |
739f79fc | 2089 | return NULL; |
d7365e78 | 2090 | |
bdcbb659 | 2091 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 2092 | return memcg; |
89c06bd5 | 2093 | |
6de22619 | 2094 | spin_lock_irqsave(&memcg->move_lock, flags); |
9da7b521 | 2095 | if (memcg != head->mem_cgroup) { |
6de22619 | 2096 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2097 | goto again; |
2098 | } | |
6de22619 JW |
2099 | |
2100 | /* | |
2101 | * When charge migration first begins, we can have locked and | |
2102 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 2103 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
2104 | */ |
2105 | memcg->move_lock_task = current; | |
2106 | memcg->move_lock_flags = flags; | |
d7365e78 | 2107 | |
739f79fc | 2108 | return memcg; |
89c06bd5 | 2109 | } |
81f8c3a4 | 2110 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2111 | |
d7365e78 | 2112 | /** |
739f79fc JW |
2113 | * __unlock_page_memcg - unlock and unpin a memcg |
2114 | * @memcg: the memcg | |
2115 | * | |
2116 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2117 | */ |
739f79fc | 2118 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2119 | { |
6de22619 JW |
2120 | if (memcg && memcg->move_lock_task == current) { |
2121 | unsigned long flags = memcg->move_lock_flags; | |
2122 | ||
2123 | memcg->move_lock_task = NULL; | |
2124 | memcg->move_lock_flags = 0; | |
2125 | ||
2126 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2127 | } | |
89c06bd5 | 2128 | |
d7365e78 | 2129 | rcu_read_unlock(); |
89c06bd5 | 2130 | } |
739f79fc JW |
2131 | |
2132 | /** | |
2133 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
2134 | * @page: the page | |
2135 | */ | |
2136 | void unlock_page_memcg(struct page *page) | |
2137 | { | |
9da7b521 JW |
2138 | struct page *head = compound_head(page); |
2139 | ||
2140 | __unlock_page_memcg(head->mem_cgroup); | |
739f79fc | 2141 | } |
81f8c3a4 | 2142 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2143 | |
cdec2e42 KH |
2144 | struct memcg_stock_pcp { |
2145 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2146 | unsigned int nr_pages; |
bf4f0599 RG |
2147 | |
2148 | #ifdef CONFIG_MEMCG_KMEM | |
2149 | struct obj_cgroup *cached_objcg; | |
2150 | unsigned int nr_bytes; | |
2151 | #endif | |
2152 | ||
cdec2e42 | 2153 | struct work_struct work; |
26fe6168 | 2154 | unsigned long flags; |
a0db00fc | 2155 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2156 | }; |
2157 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2158 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2159 | |
bf4f0599 RG |
2160 | #ifdef CONFIG_MEMCG_KMEM |
2161 | static void drain_obj_stock(struct memcg_stock_pcp *stock); | |
2162 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2163 | struct mem_cgroup *root_memcg); | |
2164 | ||
2165 | #else | |
2166 | static inline void drain_obj_stock(struct memcg_stock_pcp *stock) | |
2167 | { | |
2168 | } | |
2169 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2170 | struct mem_cgroup *root_memcg) | |
2171 | { | |
2172 | return false; | |
2173 | } | |
2174 | #endif | |
2175 | ||
a0956d54 SS |
2176 | /** |
2177 | * consume_stock: Try to consume stocked charge on this cpu. | |
2178 | * @memcg: memcg to consume from. | |
2179 | * @nr_pages: how many pages to charge. | |
2180 | * | |
2181 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2182 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2183 | * service an allocation will refill the stock. | |
2184 | * | |
2185 | * returns true if successful, false otherwise. | |
cdec2e42 | 2186 | */ |
a0956d54 | 2187 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2188 | { |
2189 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2190 | unsigned long flags; |
3e32cb2e | 2191 | bool ret = false; |
cdec2e42 | 2192 | |
a983b5eb | 2193 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2194 | return ret; |
a0956d54 | 2195 | |
db2ba40c JW |
2196 | local_irq_save(flags); |
2197 | ||
2198 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2199 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2200 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2201 | ret = true; |
2202 | } | |
db2ba40c JW |
2203 | |
2204 | local_irq_restore(flags); | |
2205 | ||
cdec2e42 KH |
2206 | return ret; |
2207 | } | |
2208 | ||
2209 | /* | |
3e32cb2e | 2210 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2211 | */ |
2212 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2213 | { | |
2214 | struct mem_cgroup *old = stock->cached; | |
2215 | ||
1a3e1f40 JW |
2216 | if (!old) |
2217 | return; | |
2218 | ||
11c9ea4e | 2219 | if (stock->nr_pages) { |
3e32cb2e | 2220 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2221 | if (do_memsw_account()) |
3e32cb2e | 2222 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2223 | stock->nr_pages = 0; |
cdec2e42 | 2224 | } |
1a3e1f40 JW |
2225 | |
2226 | css_put(&old->css); | |
cdec2e42 | 2227 | stock->cached = NULL; |
cdec2e42 KH |
2228 | } |
2229 | ||
cdec2e42 KH |
2230 | static void drain_local_stock(struct work_struct *dummy) |
2231 | { | |
db2ba40c JW |
2232 | struct memcg_stock_pcp *stock; |
2233 | unsigned long flags; | |
2234 | ||
72f0184c MH |
2235 | /* |
2236 | * The only protection from memory hotplug vs. drain_stock races is | |
2237 | * that we always operate on local CPU stock here with IRQ disabled | |
2238 | */ | |
db2ba40c JW |
2239 | local_irq_save(flags); |
2240 | ||
2241 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 | 2242 | drain_obj_stock(stock); |
cdec2e42 | 2243 | drain_stock(stock); |
26fe6168 | 2244 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2245 | |
2246 | local_irq_restore(flags); | |
cdec2e42 KH |
2247 | } |
2248 | ||
2249 | /* | |
3e32cb2e | 2250 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2251 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2252 | */ |
c0ff4b85 | 2253 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2254 | { |
db2ba40c JW |
2255 | struct memcg_stock_pcp *stock; |
2256 | unsigned long flags; | |
2257 | ||
2258 | local_irq_save(flags); | |
cdec2e42 | 2259 | |
db2ba40c | 2260 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2261 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2262 | drain_stock(stock); |
1a3e1f40 | 2263 | css_get(&memcg->css); |
c0ff4b85 | 2264 | stock->cached = memcg; |
cdec2e42 | 2265 | } |
11c9ea4e | 2266 | stock->nr_pages += nr_pages; |
db2ba40c | 2267 | |
a983b5eb | 2268 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2269 | drain_stock(stock); |
2270 | ||
db2ba40c | 2271 | local_irq_restore(flags); |
cdec2e42 KH |
2272 | } |
2273 | ||
2274 | /* | |
c0ff4b85 | 2275 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2276 | * of the hierarchy under it. |
cdec2e42 | 2277 | */ |
6d3d6aa2 | 2278 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2279 | { |
26fe6168 | 2280 | int cpu, curcpu; |
d38144b7 | 2281 | |
6d3d6aa2 JW |
2282 | /* If someone's already draining, avoid adding running more workers. */ |
2283 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2284 | return; | |
72f0184c MH |
2285 | /* |
2286 | * Notify other cpus that system-wide "drain" is running | |
2287 | * We do not care about races with the cpu hotplug because cpu down | |
2288 | * as well as workers from this path always operate on the local | |
2289 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2290 | */ | |
5af12d0e | 2291 | curcpu = get_cpu(); |
cdec2e42 KH |
2292 | for_each_online_cpu(cpu) { |
2293 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2294 | struct mem_cgroup *memcg; |
e1a366be | 2295 | bool flush = false; |
26fe6168 | 2296 | |
e1a366be | 2297 | rcu_read_lock(); |
c0ff4b85 | 2298 | memcg = stock->cached; |
e1a366be RG |
2299 | if (memcg && stock->nr_pages && |
2300 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2301 | flush = true; | |
bf4f0599 RG |
2302 | if (obj_stock_flush_required(stock, root_memcg)) |
2303 | flush = true; | |
e1a366be RG |
2304 | rcu_read_unlock(); |
2305 | ||
2306 | if (flush && | |
2307 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2308 | if (cpu == curcpu) |
2309 | drain_local_stock(&stock->work); | |
2310 | else | |
2311 | schedule_work_on(cpu, &stock->work); | |
2312 | } | |
cdec2e42 | 2313 | } |
5af12d0e | 2314 | put_cpu(); |
9f50fad6 | 2315 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2316 | } |
2317 | ||
308167fc | 2318 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2319 | { |
cdec2e42 | 2320 | struct memcg_stock_pcp *stock; |
42a30035 | 2321 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2322 | |
cdec2e42 KH |
2323 | stock = &per_cpu(memcg_stock, cpu); |
2324 | drain_stock(stock); | |
a983b5eb JW |
2325 | |
2326 | for_each_mem_cgroup(memcg) { | |
2327 | int i; | |
2328 | ||
2329 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2330 | int nid; | |
2331 | long x; | |
2332 | ||
871789d4 | 2333 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
815744d7 | 2334 | if (x) |
42a30035 JW |
2335 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2336 | atomic_long_add(x, &memcg->vmstats[i]); | |
a983b5eb JW |
2337 | |
2338 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2339 | continue; | |
2340 | ||
2341 | for_each_node(nid) { | |
2342 | struct mem_cgroup_per_node *pn; | |
2343 | ||
2344 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2345 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
815744d7 | 2346 | if (x) |
42a30035 JW |
2347 | do { |
2348 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2349 | } while ((pn = parent_nodeinfo(pn, nid))); | |
a983b5eb JW |
2350 | } |
2351 | } | |
2352 | ||
e27be240 | 2353 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2354 | long x; |
2355 | ||
871789d4 | 2356 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
815744d7 | 2357 | if (x) |
42a30035 JW |
2358 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2359 | atomic_long_add(x, &memcg->vmevents[i]); | |
a983b5eb JW |
2360 | } |
2361 | } | |
2362 | ||
308167fc | 2363 | return 0; |
cdec2e42 KH |
2364 | } |
2365 | ||
f7e1cb6e JW |
2366 | static void reclaim_high(struct mem_cgroup *memcg, |
2367 | unsigned int nr_pages, | |
2368 | gfp_t gfp_mask) | |
2369 | { | |
2370 | do { | |
d1663a90 JK |
2371 | if (page_counter_read(&memcg->memory) <= |
2372 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2373 | continue; |
e27be240 | 2374 | memcg_memory_event(memcg, MEMCG_HIGH); |
f7e1cb6e | 2375 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
4bf17307 CD |
2376 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2377 | !mem_cgroup_is_root(memcg)); | |
f7e1cb6e JW |
2378 | } |
2379 | ||
2380 | static void high_work_func(struct work_struct *work) | |
2381 | { | |
2382 | struct mem_cgroup *memcg; | |
2383 | ||
2384 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2385 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2386 | } |
2387 | ||
0e4b01df CD |
2388 | /* |
2389 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2390 | * enough to still cause a significant slowdown in most cases, while still | |
2391 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2392 | */ | |
2393 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2394 | ||
2395 | /* | |
2396 | * When calculating the delay, we use these either side of the exponentiation to | |
2397 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2398 | * below. | |
2399 | * | |
2400 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2401 | * overage ratio to a delay. | |
2402 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down down the | |
2403 | * proposed penalty in order to reduce to a reasonable number of jiffies, and | |
2404 | * to produce a reasonable delay curve. | |
2405 | * | |
2406 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2407 | * reasonable delay curve compared to precision-adjusted overage, not | |
2408 | * penalising heavily at first, but still making sure that growth beyond the | |
2409 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2410 | * example, with a high of 100 megabytes: | |
2411 | * | |
2412 | * +-------+------------------------+ | |
2413 | * | usage | time to allocate in ms | | |
2414 | * +-------+------------------------+ | |
2415 | * | 100M | 0 | | |
2416 | * | 101M | 6 | | |
2417 | * | 102M | 25 | | |
2418 | * | 103M | 57 | | |
2419 | * | 104M | 102 | | |
2420 | * | 105M | 159 | | |
2421 | * | 106M | 230 | | |
2422 | * | 107M | 313 | | |
2423 | * | 108M | 409 | | |
2424 | * | 109M | 518 | | |
2425 | * | 110M | 639 | | |
2426 | * | 111M | 774 | | |
2427 | * | 112M | 921 | | |
2428 | * | 113M | 1081 | | |
2429 | * | 114M | 1254 | | |
2430 | * | 115M | 1439 | | |
2431 | * | 116M | 1638 | | |
2432 | * | 117M | 1849 | | |
2433 | * | 118M | 2000 | | |
2434 | * | 119M | 2000 | | |
2435 | * | 120M | 2000 | | |
2436 | * +-------+------------------------+ | |
2437 | */ | |
2438 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2439 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2440 | ||
8a5dbc65 | 2441 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2442 | { |
8a5dbc65 | 2443 | u64 overage; |
b23afb93 | 2444 | |
8a5dbc65 JK |
2445 | if (usage <= high) |
2446 | return 0; | |
e26733e0 | 2447 | |
8a5dbc65 JK |
2448 | /* |
2449 | * Prevent division by 0 in overage calculation by acting as if | |
2450 | * it was a threshold of 1 page | |
2451 | */ | |
2452 | high = max(high, 1UL); | |
9b8b1754 | 2453 | |
8a5dbc65 JK |
2454 | overage = usage - high; |
2455 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2456 | return div64_u64(overage, high); | |
2457 | } | |
e26733e0 | 2458 | |
8a5dbc65 JK |
2459 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2460 | { | |
2461 | u64 overage, max_overage = 0; | |
e26733e0 | 2462 | |
8a5dbc65 JK |
2463 | do { |
2464 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2465 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2466 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2467 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2468 | !mem_cgroup_is_root(memcg)); | |
2469 | ||
8a5dbc65 JK |
2470 | return max_overage; |
2471 | } | |
2472 | ||
4b82ab4f JK |
2473 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2474 | { | |
2475 | u64 overage, max_overage = 0; | |
2476 | ||
2477 | do { | |
2478 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2479 | READ_ONCE(memcg->swap.high)); | |
2480 | if (overage) | |
2481 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2482 | max_overage = max(overage, max_overage); | |
2483 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2484 | !mem_cgroup_is_root(memcg)); | |
2485 | ||
2486 | return max_overage; | |
2487 | } | |
2488 | ||
8a5dbc65 JK |
2489 | /* |
2490 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2491 | * is exceeding its memory.high by checking both it and its ancestors. | |
2492 | */ | |
2493 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2494 | unsigned int nr_pages, | |
2495 | u64 max_overage) | |
2496 | { | |
2497 | unsigned long penalty_jiffies; | |
2498 | ||
e26733e0 CD |
2499 | if (!max_overage) |
2500 | return 0; | |
0e4b01df CD |
2501 | |
2502 | /* | |
0e4b01df CD |
2503 | * We use overage compared to memory.high to calculate the number of |
2504 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2505 | * fairly lenient on small overages, and increasingly harsh when the | |
2506 | * memcg in question makes it clear that it has no intention of stopping | |
2507 | * its crazy behaviour, so we exponentially increase the delay based on | |
2508 | * overage amount. | |
2509 | */ | |
e26733e0 CD |
2510 | penalty_jiffies = max_overage * max_overage * HZ; |
2511 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2512 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2513 | |
2514 | /* | |
2515 | * Factor in the task's own contribution to the overage, such that four | |
2516 | * N-sized allocations are throttled approximately the same as one | |
2517 | * 4N-sized allocation. | |
2518 | * | |
2519 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2520 | * larger the current charge patch is than that. | |
2521 | */ | |
ff144e69 | 2522 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2523 | } |
2524 | ||
2525 | /* | |
2526 | * Scheduled by try_charge() to be executed from the userland return path | |
2527 | * and reclaims memory over the high limit. | |
2528 | */ | |
2529 | void mem_cgroup_handle_over_high(void) | |
2530 | { | |
2531 | unsigned long penalty_jiffies; | |
2532 | unsigned long pflags; | |
2533 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
2534 | struct mem_cgroup *memcg; | |
2535 | ||
2536 | if (likely(!nr_pages)) | |
2537 | return; | |
2538 | ||
2539 | memcg = get_mem_cgroup_from_mm(current->mm); | |
2540 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
2541 | current->memcg_nr_pages_over_high = 0; | |
2542 | ||
2543 | /* | |
2544 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2545 | * allocators proactively to slow down excessive growth. | |
2546 | */ | |
8a5dbc65 JK |
2547 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2548 | mem_find_max_overage(memcg)); | |
0e4b01df | 2549 | |
4b82ab4f JK |
2550 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2551 | swap_find_max_overage(memcg)); | |
2552 | ||
ff144e69 JK |
2553 | /* |
2554 | * Clamp the max delay per usermode return so as to still keep the | |
2555 | * application moving forwards and also permit diagnostics, albeit | |
2556 | * extremely slowly. | |
2557 | */ | |
2558 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2559 | ||
0e4b01df CD |
2560 | /* |
2561 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2562 | * that it's not even worth doing, in an attempt to be nice to those who | |
2563 | * go only a small amount over their memory.high value and maybe haven't | |
2564 | * been aggressively reclaimed enough yet. | |
2565 | */ | |
2566 | if (penalty_jiffies <= HZ / 100) | |
2567 | goto out; | |
2568 | ||
2569 | /* | |
2570 | * If we exit early, we're guaranteed to die (since | |
2571 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2572 | * need to account for any ill-begotten jiffies to pay them off later. | |
2573 | */ | |
2574 | psi_memstall_enter(&pflags); | |
2575 | schedule_timeout_killable(penalty_jiffies); | |
2576 | psi_memstall_leave(&pflags); | |
2577 | ||
2578 | out: | |
2579 | css_put(&memcg->css); | |
b23afb93 TH |
2580 | } |
2581 | ||
00501b53 JW |
2582 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2583 | unsigned int nr_pages) | |
8a9f3ccd | 2584 | { |
a983b5eb | 2585 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
9b130619 | 2586 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2587 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2588 | struct page_counter *counter; |
6539cc05 | 2589 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2590 | bool may_swap = true; |
2591 | bool drained = false; | |
29ef680a | 2592 | enum oom_status oom_status; |
a636b327 | 2593 | |
ce00a967 | 2594 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2595 | return 0; |
6539cc05 | 2596 | retry: |
b6b6cc72 | 2597 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2598 | return 0; |
8a9f3ccd | 2599 | |
7941d214 | 2600 | if (!do_memsw_account() || |
6071ca52 JW |
2601 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2602 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2603 | goto done_restock; |
7941d214 | 2604 | if (do_memsw_account()) |
3e32cb2e JW |
2605 | page_counter_uncharge(&memcg->memsw, batch); |
2606 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2607 | } else { |
3e32cb2e | 2608 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2609 | may_swap = false; |
3fbe7244 | 2610 | } |
7a81b88c | 2611 | |
6539cc05 JW |
2612 | if (batch > nr_pages) { |
2613 | batch = nr_pages; | |
2614 | goto retry; | |
2615 | } | |
6d61ef40 | 2616 | |
869712fd JW |
2617 | /* |
2618 | * Memcg doesn't have a dedicated reserve for atomic | |
2619 | * allocations. But like the global atomic pool, we need to | |
2620 | * put the burden of reclaim on regular allocation requests | |
2621 | * and let these go through as privileged allocations. | |
2622 | */ | |
2623 | if (gfp_mask & __GFP_ATOMIC) | |
2624 | goto force; | |
2625 | ||
06b078fc JW |
2626 | /* |
2627 | * Unlike in global OOM situations, memcg is not in a physical | |
2628 | * memory shortage. Allow dying and OOM-killed tasks to | |
2629 | * bypass the last charges so that they can exit quickly and | |
2630 | * free their memory. | |
2631 | */ | |
7775face | 2632 | if (unlikely(should_force_charge())) |
10d53c74 | 2633 | goto force; |
06b078fc | 2634 | |
89a28483 JW |
2635 | /* |
2636 | * Prevent unbounded recursion when reclaim operations need to | |
2637 | * allocate memory. This might exceed the limits temporarily, | |
2638 | * but we prefer facilitating memory reclaim and getting back | |
2639 | * under the limit over triggering OOM kills in these cases. | |
2640 | */ | |
2641 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2642 | goto force; | |
2643 | ||
06b078fc JW |
2644 | if (unlikely(task_in_memcg_oom(current))) |
2645 | goto nomem; | |
2646 | ||
d0164adc | 2647 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2648 | goto nomem; |
4b534334 | 2649 | |
e27be240 | 2650 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2651 | |
b70a2a21 JW |
2652 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2653 | gfp_mask, may_swap); | |
6539cc05 | 2654 | |
61e02c74 | 2655 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2656 | goto retry; |
28c34c29 | 2657 | |
b70a2a21 | 2658 | if (!drained) { |
6d3d6aa2 | 2659 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2660 | drained = true; |
2661 | goto retry; | |
2662 | } | |
2663 | ||
28c34c29 JW |
2664 | if (gfp_mask & __GFP_NORETRY) |
2665 | goto nomem; | |
6539cc05 JW |
2666 | /* |
2667 | * Even though the limit is exceeded at this point, reclaim | |
2668 | * may have been able to free some pages. Retry the charge | |
2669 | * before killing the task. | |
2670 | * | |
2671 | * Only for regular pages, though: huge pages are rather | |
2672 | * unlikely to succeed so close to the limit, and we fall back | |
2673 | * to regular pages anyway in case of failure. | |
2674 | */ | |
61e02c74 | 2675 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2676 | goto retry; |
2677 | /* | |
2678 | * At task move, charge accounts can be doubly counted. So, it's | |
2679 | * better to wait until the end of task_move if something is going on. | |
2680 | */ | |
2681 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2682 | goto retry; | |
2683 | ||
9b130619 JW |
2684 | if (nr_retries--) |
2685 | goto retry; | |
2686 | ||
38d38493 | 2687 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2688 | goto nomem; |
2689 | ||
06b078fc | 2690 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2691 | goto force; |
06b078fc | 2692 | |
6539cc05 | 2693 | if (fatal_signal_pending(current)) |
10d53c74 | 2694 | goto force; |
6539cc05 | 2695 | |
29ef680a MH |
2696 | /* |
2697 | * keep retrying as long as the memcg oom killer is able to make | |
2698 | * a forward progress or bypass the charge if the oom killer | |
2699 | * couldn't make any progress. | |
2700 | */ | |
2701 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2702 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2703 | switch (oom_status) { |
2704 | case OOM_SUCCESS: | |
2705 | nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
29ef680a MH |
2706 | goto retry; |
2707 | case OOM_FAILED: | |
2708 | goto force; | |
2709 | default: | |
2710 | goto nomem; | |
2711 | } | |
7a81b88c | 2712 | nomem: |
6d1fdc48 | 2713 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2714 | return -ENOMEM; |
10d53c74 TH |
2715 | force: |
2716 | /* | |
2717 | * The allocation either can't fail or will lead to more memory | |
2718 | * being freed very soon. Allow memory usage go over the limit | |
2719 | * temporarily by force charging it. | |
2720 | */ | |
2721 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2722 | if (do_memsw_account()) |
10d53c74 | 2723 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
2724 | |
2725 | return 0; | |
6539cc05 JW |
2726 | |
2727 | done_restock: | |
2728 | if (batch > nr_pages) | |
2729 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2730 | |
241994ed | 2731 | /* |
b23afb93 TH |
2732 | * If the hierarchy is above the normal consumption range, schedule |
2733 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2734 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2735 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2736 | * not recorded as it most likely matches current's and won't | |
2737 | * change in the meantime. As high limit is checked again before | |
2738 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2739 | */ |
2740 | do { | |
4b82ab4f JK |
2741 | bool mem_high, swap_high; |
2742 | ||
2743 | mem_high = page_counter_read(&memcg->memory) > | |
2744 | READ_ONCE(memcg->memory.high); | |
2745 | swap_high = page_counter_read(&memcg->swap) > | |
2746 | READ_ONCE(memcg->swap.high); | |
2747 | ||
2748 | /* Don't bother a random interrupted task */ | |
2749 | if (in_interrupt()) { | |
2750 | if (mem_high) { | |
f7e1cb6e JW |
2751 | schedule_work(&memcg->high_work); |
2752 | break; | |
2753 | } | |
4b82ab4f JK |
2754 | continue; |
2755 | } | |
2756 | ||
2757 | if (mem_high || swap_high) { | |
2758 | /* | |
2759 | * The allocating tasks in this cgroup will need to do | |
2760 | * reclaim or be throttled to prevent further growth | |
2761 | * of the memory or swap footprints. | |
2762 | * | |
2763 | * Target some best-effort fairness between the tasks, | |
2764 | * and distribute reclaim work and delay penalties | |
2765 | * based on how much each task is actually allocating. | |
2766 | */ | |
9516a18a | 2767 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2768 | set_notify_resume(current); |
2769 | break; | |
2770 | } | |
241994ed | 2771 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2772 | |
2773 | return 0; | |
7a81b88c | 2774 | } |
8a9f3ccd | 2775 | |
f0e45fb4 | 2776 | #if defined(CONFIG_MEMCG_KMEM) || defined(CONFIG_MMU) |
00501b53 | 2777 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2778 | { |
ce00a967 JW |
2779 | if (mem_cgroup_is_root(memcg)) |
2780 | return; | |
2781 | ||
3e32cb2e | 2782 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2783 | if (do_memsw_account()) |
3e32cb2e | 2784 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f | 2785 | } |
f0e45fb4 | 2786 | #endif |
d01dd17f | 2787 | |
d9eb1ea2 | 2788 | static void commit_charge(struct page *page, struct mem_cgroup *memcg) |
0a31bc97 | 2789 | { |
1306a85a | 2790 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
0a31bc97 | 2791 | /* |
a0b5b414 | 2792 | * Any of the following ensures page->mem_cgroup stability: |
0a31bc97 | 2793 | * |
a0b5b414 JW |
2794 | * - the page lock |
2795 | * - LRU isolation | |
2796 | * - lock_page_memcg() | |
2797 | * - exclusive reference | |
0a31bc97 | 2798 | */ |
1306a85a | 2799 | page->mem_cgroup = memcg; |
7a81b88c | 2800 | } |
66e1707b | 2801 | |
84c07d11 | 2802 | #ifdef CONFIG_MEMCG_KMEM |
10befea9 RG |
2803 | int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s, |
2804 | gfp_t gfp) | |
2805 | { | |
2806 | unsigned int objects = objs_per_slab_page(s, page); | |
2807 | void *vec; | |
2808 | ||
2809 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, | |
2810 | page_to_nid(page)); | |
2811 | if (!vec) | |
2812 | return -ENOMEM; | |
2813 | ||
2814 | if (cmpxchg(&page->obj_cgroups, NULL, | |
2815 | (struct obj_cgroup **) ((unsigned long)vec | 0x1UL))) | |
2816 | kfree(vec); | |
2817 | else | |
2818 | kmemleak_not_leak(vec); | |
2819 | ||
2820 | return 0; | |
2821 | } | |
2822 | ||
8380ce47 RG |
2823 | /* |
2824 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2825 | * | |
2826 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
2827 | * cgroup_mutex, etc. | |
2828 | */ | |
2829 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2830 | { | |
2831 | struct page *page; | |
2832 | ||
2833 | if (mem_cgroup_disabled()) | |
2834 | return NULL; | |
2835 | ||
2836 | page = virt_to_head_page(p); | |
2837 | ||
2838 | /* | |
9855609b RG |
2839 | * Slab objects are accounted individually, not per-page. |
2840 | * Memcg membership data for each individual object is saved in | |
2841 | * the page->obj_cgroups. | |
8380ce47 | 2842 | */ |
9855609b RG |
2843 | if (page_has_obj_cgroups(page)) { |
2844 | struct obj_cgroup *objcg; | |
2845 | unsigned int off; | |
2846 | ||
2847 | off = obj_to_index(page->slab_cache, page, p); | |
2848 | objcg = page_obj_cgroups(page)[off]; | |
10befea9 RG |
2849 | if (objcg) |
2850 | return obj_cgroup_memcg(objcg); | |
2851 | ||
2852 | return NULL; | |
9855609b | 2853 | } |
8380ce47 RG |
2854 | |
2855 | /* All other pages use page->mem_cgroup */ | |
2856 | return page->mem_cgroup; | |
2857 | } | |
2858 | ||
bf4f0599 RG |
2859 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
2860 | { | |
2861 | struct obj_cgroup *objcg = NULL; | |
2862 | struct mem_cgroup *memcg; | |
2863 | ||
2864 | if (unlikely(!current->mm && !current->active_memcg)) | |
2865 | return NULL; | |
2866 | ||
2867 | rcu_read_lock(); | |
2868 | if (unlikely(current->active_memcg)) | |
2869 | memcg = rcu_dereference(current->active_memcg); | |
2870 | else | |
2871 | memcg = mem_cgroup_from_task(current); | |
2872 | ||
2873 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { | |
2874 | objcg = rcu_dereference(memcg->objcg); | |
2875 | if (objcg && obj_cgroup_tryget(objcg)) | |
2876 | break; | |
2877 | } | |
2878 | rcu_read_unlock(); | |
2879 | ||
2880 | return objcg; | |
2881 | } | |
2882 | ||
f3bb3043 | 2883 | static int memcg_alloc_cache_id(void) |
55007d84 | 2884 | { |
f3bb3043 VD |
2885 | int id, size; |
2886 | int err; | |
2887 | ||
dbcf73e2 | 2888 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2889 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2890 | if (id < 0) | |
2891 | return id; | |
55007d84 | 2892 | |
dbcf73e2 | 2893 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2894 | return id; |
2895 | ||
2896 | /* | |
2897 | * There's no space for the new id in memcg_caches arrays, | |
2898 | * so we have to grow them. | |
2899 | */ | |
05257a1a | 2900 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2901 | |
2902 | size = 2 * (id + 1); | |
55007d84 GC |
2903 | if (size < MEMCG_CACHES_MIN_SIZE) |
2904 | size = MEMCG_CACHES_MIN_SIZE; | |
2905 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2906 | size = MEMCG_CACHES_MAX_SIZE; | |
2907 | ||
9855609b | 2908 | err = memcg_update_all_list_lrus(size); |
05257a1a VD |
2909 | if (!err) |
2910 | memcg_nr_cache_ids = size; | |
2911 | ||
2912 | up_write(&memcg_cache_ids_sem); | |
2913 | ||
f3bb3043 | 2914 | if (err) { |
dbcf73e2 | 2915 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2916 | return err; |
2917 | } | |
2918 | return id; | |
2919 | } | |
2920 | ||
2921 | static void memcg_free_cache_id(int id) | |
2922 | { | |
dbcf73e2 | 2923 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2924 | } |
2925 | ||
45264778 | 2926 | /** |
4b13f64d | 2927 | * __memcg_kmem_charge: charge a number of kernel pages to a memcg |
10eaec2f | 2928 | * @memcg: memory cgroup to charge |
45264778 | 2929 | * @gfp: reclaim mode |
92d0510c | 2930 | * @nr_pages: number of pages to charge |
45264778 VD |
2931 | * |
2932 | * Returns 0 on success, an error code on failure. | |
2933 | */ | |
4b13f64d RG |
2934 | int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp, |
2935 | unsigned int nr_pages) | |
7ae1e1d0 | 2936 | { |
f3ccb2c4 | 2937 | struct page_counter *counter; |
7ae1e1d0 GC |
2938 | int ret; |
2939 | ||
f3ccb2c4 | 2940 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2941 | if (ret) |
f3ccb2c4 | 2942 | return ret; |
52c29b04 JW |
2943 | |
2944 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2945 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
e55d9d9b MH |
2946 | |
2947 | /* | |
2948 | * Enforce __GFP_NOFAIL allocation because callers are not | |
2949 | * prepared to see failures and likely do not have any failure | |
2950 | * handling code. | |
2951 | */ | |
2952 | if (gfp & __GFP_NOFAIL) { | |
2953 | page_counter_charge(&memcg->kmem, nr_pages); | |
2954 | return 0; | |
2955 | } | |
52c29b04 JW |
2956 | cancel_charge(memcg, nr_pages); |
2957 | return -ENOMEM; | |
7ae1e1d0 | 2958 | } |
f3ccb2c4 | 2959 | return 0; |
7ae1e1d0 GC |
2960 | } |
2961 | ||
4b13f64d RG |
2962 | /** |
2963 | * __memcg_kmem_uncharge: uncharge a number of kernel pages from a memcg | |
2964 | * @memcg: memcg to uncharge | |
2965 | * @nr_pages: number of pages to uncharge | |
2966 | */ | |
2967 | void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages) | |
2968 | { | |
2969 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2970 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2971 | ||
2972 | page_counter_uncharge(&memcg->memory, nr_pages); | |
2973 | if (do_memsw_account()) | |
2974 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
2975 | } | |
2976 | ||
45264778 | 2977 | /** |
f4b00eab | 2978 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
2979 | * @page: page to charge |
2980 | * @gfp: reclaim mode | |
2981 | * @order: allocation order | |
2982 | * | |
2983 | * Returns 0 on success, an error code on failure. | |
2984 | */ | |
f4b00eab | 2985 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2986 | { |
f3ccb2c4 | 2987 | struct mem_cgroup *memcg; |
fcff7d7e | 2988 | int ret = 0; |
7ae1e1d0 | 2989 | |
60cd4bcd | 2990 | if (memcg_kmem_bypass()) |
45264778 VD |
2991 | return 0; |
2992 | ||
d46eb14b | 2993 | memcg = get_mem_cgroup_from_current(); |
c4159a75 | 2994 | if (!mem_cgroup_is_root(memcg)) { |
4b13f64d | 2995 | ret = __memcg_kmem_charge(memcg, gfp, 1 << order); |
4d96ba35 RG |
2996 | if (!ret) { |
2997 | page->mem_cgroup = memcg; | |
c4159a75 | 2998 | __SetPageKmemcg(page); |
1a3e1f40 | 2999 | return 0; |
4d96ba35 | 3000 | } |
c4159a75 | 3001 | } |
7ae1e1d0 | 3002 | css_put(&memcg->css); |
d05e83a6 | 3003 | return ret; |
7ae1e1d0 | 3004 | } |
49a18eae | 3005 | |
45264778 | 3006 | /** |
f4b00eab | 3007 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
3008 | * @page: page to uncharge |
3009 | * @order: allocation order | |
3010 | */ | |
f4b00eab | 3011 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 3012 | { |
1306a85a | 3013 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 3014 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3015 | |
7ae1e1d0 GC |
3016 | if (!memcg) |
3017 | return; | |
3018 | ||
309381fe | 3019 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
4b13f64d | 3020 | __memcg_kmem_uncharge(memcg, nr_pages); |
1306a85a | 3021 | page->mem_cgroup = NULL; |
1a3e1f40 | 3022 | css_put(&memcg->css); |
c4159a75 VD |
3023 | |
3024 | /* slab pages do not have PageKmemcg flag set */ | |
3025 | if (PageKmemcg(page)) | |
3026 | __ClearPageKmemcg(page); | |
60d3fd32 | 3027 | } |
bf4f0599 RG |
3028 | |
3029 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3030 | { | |
3031 | struct memcg_stock_pcp *stock; | |
3032 | unsigned long flags; | |
3033 | bool ret = false; | |
3034 | ||
3035 | local_irq_save(flags); | |
3036 | ||
3037 | stock = this_cpu_ptr(&memcg_stock); | |
3038 | if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { | |
3039 | stock->nr_bytes -= nr_bytes; | |
3040 | ret = true; | |
3041 | } | |
3042 | ||
3043 | local_irq_restore(flags); | |
3044 | ||
3045 | return ret; | |
3046 | } | |
3047 | ||
3048 | static void drain_obj_stock(struct memcg_stock_pcp *stock) | |
3049 | { | |
3050 | struct obj_cgroup *old = stock->cached_objcg; | |
3051 | ||
3052 | if (!old) | |
3053 | return; | |
3054 | ||
3055 | if (stock->nr_bytes) { | |
3056 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3057 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3058 | ||
3059 | if (nr_pages) { | |
3060 | rcu_read_lock(); | |
3061 | __memcg_kmem_uncharge(obj_cgroup_memcg(old), nr_pages); | |
3062 | rcu_read_unlock(); | |
3063 | } | |
3064 | ||
3065 | /* | |
3066 | * The leftover is flushed to the centralized per-memcg value. | |
3067 | * On the next attempt to refill obj stock it will be moved | |
3068 | * to a per-cpu stock (probably, on an other CPU), see | |
3069 | * refill_obj_stock(). | |
3070 | * | |
3071 | * How often it's flushed is a trade-off between the memory | |
3072 | * limit enforcement accuracy and potential CPU contention, | |
3073 | * so it might be changed in the future. | |
3074 | */ | |
3075 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3076 | stock->nr_bytes = 0; | |
3077 | } | |
3078 | ||
3079 | obj_cgroup_put(old); | |
3080 | stock->cached_objcg = NULL; | |
3081 | } | |
3082 | ||
3083 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3084 | struct mem_cgroup *root_memcg) | |
3085 | { | |
3086 | struct mem_cgroup *memcg; | |
3087 | ||
3088 | if (stock->cached_objcg) { | |
3089 | memcg = obj_cgroup_memcg(stock->cached_objcg); | |
3090 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) | |
3091 | return true; | |
3092 | } | |
3093 | ||
3094 | return false; | |
3095 | } | |
3096 | ||
3097 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3098 | { | |
3099 | struct memcg_stock_pcp *stock; | |
3100 | unsigned long flags; | |
3101 | ||
3102 | local_irq_save(flags); | |
3103 | ||
3104 | stock = this_cpu_ptr(&memcg_stock); | |
3105 | if (stock->cached_objcg != objcg) { /* reset if necessary */ | |
3106 | drain_obj_stock(stock); | |
3107 | obj_cgroup_get(objcg); | |
3108 | stock->cached_objcg = objcg; | |
3109 | stock->nr_bytes = atomic_xchg(&objcg->nr_charged_bytes, 0); | |
3110 | } | |
3111 | stock->nr_bytes += nr_bytes; | |
3112 | ||
3113 | if (stock->nr_bytes > PAGE_SIZE) | |
3114 | drain_obj_stock(stock); | |
3115 | ||
3116 | local_irq_restore(flags); | |
3117 | } | |
3118 | ||
3119 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3120 | { | |
3121 | struct mem_cgroup *memcg; | |
3122 | unsigned int nr_pages, nr_bytes; | |
3123 | int ret; | |
3124 | ||
3125 | if (consume_obj_stock(objcg, size)) | |
3126 | return 0; | |
3127 | ||
3128 | /* | |
3129 | * In theory, memcg->nr_charged_bytes can have enough | |
3130 | * pre-charged bytes to satisfy the allocation. However, | |
3131 | * flushing memcg->nr_charged_bytes requires two atomic | |
3132 | * operations, and memcg->nr_charged_bytes can't be big, | |
3133 | * so it's better to ignore it and try grab some new pages. | |
3134 | * memcg->nr_charged_bytes will be flushed in | |
3135 | * refill_obj_stock(), called from this function or | |
3136 | * independently later. | |
3137 | */ | |
3138 | rcu_read_lock(); | |
3139 | memcg = obj_cgroup_memcg(objcg); | |
3140 | css_get(&memcg->css); | |
3141 | rcu_read_unlock(); | |
3142 | ||
3143 | nr_pages = size >> PAGE_SHIFT; | |
3144 | nr_bytes = size & (PAGE_SIZE - 1); | |
3145 | ||
3146 | if (nr_bytes) | |
3147 | nr_pages += 1; | |
3148 | ||
3149 | ret = __memcg_kmem_charge(memcg, gfp, nr_pages); | |
3150 | if (!ret && nr_bytes) | |
3151 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes); | |
3152 | ||
3153 | css_put(&memcg->css); | |
3154 | return ret; | |
3155 | } | |
3156 | ||
3157 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3158 | { | |
3159 | refill_obj_stock(objcg, size); | |
3160 | } | |
3161 | ||
84c07d11 | 3162 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3163 | |
ca3e0214 KH |
3164 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3165 | ||
ca3e0214 KH |
3166 | /* |
3167 | * Because tail pages are not marked as "used", set it. We're under | |
f4b7e272 | 3168 | * pgdat->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 3169 | */ |
e94c8a9c | 3170 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 3171 | { |
1a3e1f40 | 3172 | struct mem_cgroup *memcg = head->mem_cgroup; |
e94c8a9c | 3173 | int i; |
ca3e0214 | 3174 | |
3d37c4a9 KH |
3175 | if (mem_cgroup_disabled()) |
3176 | return; | |
b070e65c | 3177 | |
1a3e1f40 JW |
3178 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
3179 | css_get(&memcg->css); | |
3180 | head[i].mem_cgroup = memcg; | |
3181 | } | |
ca3e0214 | 3182 | } |
12d27107 | 3183 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3184 | |
c255a458 | 3185 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
3186 | /** |
3187 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3188 | * @entry: swap entry to be moved | |
3189 | * @from: mem_cgroup which the entry is moved from | |
3190 | * @to: mem_cgroup which the entry is moved to | |
3191 | * | |
3192 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3193 | * as the mem_cgroup's id of @from. | |
3194 | * | |
3195 | * Returns 0 on success, -EINVAL on failure. | |
3196 | * | |
3e32cb2e | 3197 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3198 | * both res and memsw, and called css_get(). |
3199 | */ | |
3200 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3201 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3202 | { |
3203 | unsigned short old_id, new_id; | |
3204 | ||
34c00c31 LZ |
3205 | old_id = mem_cgroup_id(from); |
3206 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3207 | |
3208 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3209 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3210 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3211 | return 0; |
3212 | } | |
3213 | return -EINVAL; | |
3214 | } | |
3215 | #else | |
3216 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3217 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3218 | { |
3219 | return -EINVAL; | |
3220 | } | |
8c7c6e34 | 3221 | #endif |
d13d1443 | 3222 | |
bbec2e15 | 3223 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3224 | |
bbec2e15 RG |
3225 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3226 | unsigned long max, bool memsw) | |
628f4235 | 3227 | { |
3e32cb2e | 3228 | bool enlarge = false; |
bb4a7ea2 | 3229 | bool drained = false; |
3e32cb2e | 3230 | int ret; |
c054a78c YZ |
3231 | bool limits_invariant; |
3232 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3233 | |
3e32cb2e | 3234 | do { |
628f4235 KH |
3235 | if (signal_pending(current)) { |
3236 | ret = -EINTR; | |
3237 | break; | |
3238 | } | |
3e32cb2e | 3239 | |
bbec2e15 | 3240 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3241 | /* |
3242 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3243 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3244 | */ |
15b42562 | 3245 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3246 | max <= memcg->memsw.max; |
c054a78c | 3247 | if (!limits_invariant) { |
bbec2e15 | 3248 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3249 | ret = -EINVAL; |
8c7c6e34 KH |
3250 | break; |
3251 | } | |
bbec2e15 | 3252 | if (max > counter->max) |
3e32cb2e | 3253 | enlarge = true; |
bbec2e15 RG |
3254 | ret = page_counter_set_max(counter, max); |
3255 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3256 | |
3257 | if (!ret) | |
3258 | break; | |
3259 | ||
bb4a7ea2 SB |
3260 | if (!drained) { |
3261 | drain_all_stock(memcg); | |
3262 | drained = true; | |
3263 | continue; | |
3264 | } | |
3265 | ||
1ab5c056 AR |
3266 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
3267 | GFP_KERNEL, !memsw)) { | |
3268 | ret = -EBUSY; | |
3269 | break; | |
3270 | } | |
3271 | } while (true); | |
3e32cb2e | 3272 | |
3c11ecf4 KH |
3273 | if (!ret && enlarge) |
3274 | memcg_oom_recover(memcg); | |
3e32cb2e | 3275 | |
628f4235 KH |
3276 | return ret; |
3277 | } | |
3278 | ||
ef8f2327 | 3279 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3280 | gfp_t gfp_mask, |
3281 | unsigned long *total_scanned) | |
3282 | { | |
3283 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3284 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3285 | unsigned long reclaimed; |
3286 | int loop = 0; | |
ef8f2327 | 3287 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3288 | unsigned long excess; |
0608f43d AM |
3289 | unsigned long nr_scanned; |
3290 | ||
3291 | if (order > 0) | |
3292 | return 0; | |
3293 | ||
ef8f2327 | 3294 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
3295 | |
3296 | /* | |
3297 | * Do not even bother to check the largest node if the root | |
3298 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3299 | * are acceptable as soft limit is best effort anyway. | |
3300 | */ | |
bfc7228b | 3301 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3302 | return 0; |
3303 | ||
0608f43d AM |
3304 | /* |
3305 | * This loop can run a while, specially if mem_cgroup's continuously | |
3306 | * keep exceeding their soft limit and putting the system under | |
3307 | * pressure | |
3308 | */ | |
3309 | do { | |
3310 | if (next_mz) | |
3311 | mz = next_mz; | |
3312 | else | |
3313 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3314 | if (!mz) | |
3315 | break; | |
3316 | ||
3317 | nr_scanned = 0; | |
ef8f2327 | 3318 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
3319 | gfp_mask, &nr_scanned); |
3320 | nr_reclaimed += reclaimed; | |
3321 | *total_scanned += nr_scanned; | |
0a31bc97 | 3322 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3323 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3324 | |
3325 | /* | |
3326 | * If we failed to reclaim anything from this memory cgroup | |
3327 | * it is time to move on to the next cgroup | |
3328 | */ | |
3329 | next_mz = NULL; | |
bc2f2e7f VD |
3330 | if (!reclaimed) |
3331 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3332 | ||
3e32cb2e | 3333 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3334 | /* |
3335 | * One school of thought says that we should not add | |
3336 | * back the node to the tree if reclaim returns 0. | |
3337 | * But our reclaim could return 0, simply because due | |
3338 | * to priority we are exposing a smaller subset of | |
3339 | * memory to reclaim from. Consider this as a longer | |
3340 | * term TODO. | |
3341 | */ | |
3342 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3343 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3344 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3345 | css_put(&mz->memcg->css); |
3346 | loop++; | |
3347 | /* | |
3348 | * Could not reclaim anything and there are no more | |
3349 | * mem cgroups to try or we seem to be looping without | |
3350 | * reclaiming anything. | |
3351 | */ | |
3352 | if (!nr_reclaimed && | |
3353 | (next_mz == NULL || | |
3354 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3355 | break; | |
3356 | } while (!nr_reclaimed); | |
3357 | if (next_mz) | |
3358 | css_put(&next_mz->memcg->css); | |
3359 | return nr_reclaimed; | |
3360 | } | |
3361 | ||
ea280e7b TH |
3362 | /* |
3363 | * Test whether @memcg has children, dead or alive. Note that this | |
3364 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3365 | * returns %true if there are child csses according to the cgroup | |
b8f2935f | 3366 | * hierarchy. Testing use_hierarchy is the caller's responsibility. |
ea280e7b | 3367 | */ |
b5f99b53 GC |
3368 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3369 | { | |
ea280e7b TH |
3370 | bool ret; |
3371 | ||
ea280e7b TH |
3372 | rcu_read_lock(); |
3373 | ret = css_next_child(NULL, &memcg->css); | |
3374 | rcu_read_unlock(); | |
3375 | return ret; | |
b5f99b53 GC |
3376 | } |
3377 | ||
c26251f9 | 3378 | /* |
51038171 | 3379 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3380 | * |
3381 | * Caller is responsible for holding css reference for memcg. | |
3382 | */ | |
3383 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3384 | { | |
3385 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3386 | |
c1e862c1 KH |
3387 | /* we call try-to-free pages for make this cgroup empty */ |
3388 | lru_add_drain_all(); | |
d12c60f6 JS |
3389 | |
3390 | drain_all_stock(memcg); | |
3391 | ||
f817ed48 | 3392 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3393 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3394 | int progress; |
c1e862c1 | 3395 | |
c26251f9 MH |
3396 | if (signal_pending(current)) |
3397 | return -EINTR; | |
3398 | ||
b70a2a21 JW |
3399 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3400 | GFP_KERNEL, true); | |
c1e862c1 | 3401 | if (!progress) { |
f817ed48 | 3402 | nr_retries--; |
c1e862c1 | 3403 | /* maybe some writeback is necessary */ |
8aa7e847 | 3404 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3405 | } |
f817ed48 KH |
3406 | |
3407 | } | |
ab5196c2 MH |
3408 | |
3409 | return 0; | |
cc847582 KH |
3410 | } |
3411 | ||
6770c64e TH |
3412 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3413 | char *buf, size_t nbytes, | |
3414 | loff_t off) | |
c1e862c1 | 3415 | { |
6770c64e | 3416 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3417 | |
d8423011 MH |
3418 | if (mem_cgroup_is_root(memcg)) |
3419 | return -EINVAL; | |
6770c64e | 3420 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3421 | } |
3422 | ||
182446d0 TH |
3423 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3424 | struct cftype *cft) | |
18f59ea7 | 3425 | { |
182446d0 | 3426 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3427 | } |
3428 | ||
182446d0 TH |
3429 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3430 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3431 | { |
3432 | int retval = 0; | |
182446d0 | 3433 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3434 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3435 | |
567fb435 | 3436 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 3437 | return 0; |
567fb435 | 3438 | |
18f59ea7 | 3439 | /* |
af901ca1 | 3440 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3441 | * in the child subtrees. If it is unset, then the change can |
3442 | * occur, provided the current cgroup has no children. | |
3443 | * | |
3444 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3445 | * set if there are no children. | |
3446 | */ | |
c0ff4b85 | 3447 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3448 | (val == 1 || val == 0)) { |
ea280e7b | 3449 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3450 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3451 | else |
3452 | retval = -EBUSY; | |
3453 | } else | |
3454 | retval = -EINVAL; | |
567fb435 | 3455 | |
18f59ea7 BS |
3456 | return retval; |
3457 | } | |
3458 | ||
6f646156 | 3459 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3460 | { |
42a30035 | 3461 | unsigned long val; |
ce00a967 | 3462 | |
3e32cb2e | 3463 | if (mem_cgroup_is_root(memcg)) { |
0d1c2072 | 3464 | val = memcg_page_state(memcg, NR_FILE_PAGES) + |
be5d0a74 | 3465 | memcg_page_state(memcg, NR_ANON_MAPPED); |
42a30035 JW |
3466 | if (swap) |
3467 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3468 | } else { |
ce00a967 | 3469 | if (!swap) |
3e32cb2e | 3470 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3471 | else |
3e32cb2e | 3472 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3473 | } |
c12176d3 | 3474 | return val; |
ce00a967 JW |
3475 | } |
3476 | ||
3e32cb2e JW |
3477 | enum { |
3478 | RES_USAGE, | |
3479 | RES_LIMIT, | |
3480 | RES_MAX_USAGE, | |
3481 | RES_FAILCNT, | |
3482 | RES_SOFT_LIMIT, | |
3483 | }; | |
ce00a967 | 3484 | |
791badbd | 3485 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3486 | struct cftype *cft) |
8cdea7c0 | 3487 | { |
182446d0 | 3488 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3489 | struct page_counter *counter; |
af36f906 | 3490 | |
3e32cb2e | 3491 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3492 | case _MEM: |
3e32cb2e JW |
3493 | counter = &memcg->memory; |
3494 | break; | |
8c7c6e34 | 3495 | case _MEMSWAP: |
3e32cb2e JW |
3496 | counter = &memcg->memsw; |
3497 | break; | |
510fc4e1 | 3498 | case _KMEM: |
3e32cb2e | 3499 | counter = &memcg->kmem; |
510fc4e1 | 3500 | break; |
d55f90bf | 3501 | case _TCP: |
0db15298 | 3502 | counter = &memcg->tcpmem; |
d55f90bf | 3503 | break; |
8c7c6e34 KH |
3504 | default: |
3505 | BUG(); | |
8c7c6e34 | 3506 | } |
3e32cb2e JW |
3507 | |
3508 | switch (MEMFILE_ATTR(cft->private)) { | |
3509 | case RES_USAGE: | |
3510 | if (counter == &memcg->memory) | |
c12176d3 | 3511 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3512 | if (counter == &memcg->memsw) |
c12176d3 | 3513 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3514 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3515 | case RES_LIMIT: | |
bbec2e15 | 3516 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3517 | case RES_MAX_USAGE: |
3518 | return (u64)counter->watermark * PAGE_SIZE; | |
3519 | case RES_FAILCNT: | |
3520 | return counter->failcnt; | |
3521 | case RES_SOFT_LIMIT: | |
3522 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3523 | default: | |
3524 | BUG(); | |
3525 | } | |
8cdea7c0 | 3526 | } |
510fc4e1 | 3527 | |
4a87e2a2 | 3528 | static void memcg_flush_percpu_vmstats(struct mem_cgroup *memcg) |
c350a99e | 3529 | { |
4a87e2a2 | 3530 | unsigned long stat[MEMCG_NR_STAT] = {0}; |
c350a99e RG |
3531 | struct mem_cgroup *mi; |
3532 | int node, cpu, i; | |
c350a99e RG |
3533 | |
3534 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3535 | for (i = 0; i < MEMCG_NR_STAT; i++) |
6c1c2808 | 3536 | stat[i] += per_cpu(memcg->vmstats_percpu->stat[i], cpu); |
c350a99e RG |
3537 | |
3538 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
4a87e2a2 | 3539 | for (i = 0; i < MEMCG_NR_STAT; i++) |
c350a99e RG |
3540 | atomic_long_add(stat[i], &mi->vmstats[i]); |
3541 | ||
3542 | for_each_node(node) { | |
3543 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; | |
3544 | struct mem_cgroup_per_node *pi; | |
3545 | ||
4a87e2a2 | 3546 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3547 | stat[i] = 0; |
3548 | ||
3549 | for_each_online_cpu(cpu) | |
4a87e2a2 | 3550 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
6c1c2808 SB |
3551 | stat[i] += per_cpu( |
3552 | pn->lruvec_stat_cpu->count[i], cpu); | |
c350a99e RG |
3553 | |
3554 | for (pi = pn; pi; pi = parent_nodeinfo(pi, node)) | |
4a87e2a2 | 3555 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3556 | atomic_long_add(stat[i], &pi->lruvec_stat[i]); |
3557 | } | |
3558 | } | |
3559 | ||
bb65f89b RG |
3560 | static void memcg_flush_percpu_vmevents(struct mem_cgroup *memcg) |
3561 | { | |
3562 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
3563 | struct mem_cgroup *mi; | |
3564 | int cpu, i; | |
3565 | ||
3566 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3567 | events[i] = 0; | |
3568 | ||
3569 | for_each_online_cpu(cpu) | |
3570 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
6c1c2808 SB |
3571 | events[i] += per_cpu(memcg->vmstats_percpu->events[i], |
3572 | cpu); | |
bb65f89b RG |
3573 | |
3574 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
3575 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3576 | atomic_long_add(events[i], &mi->vmevents[i]); | |
3577 | } | |
3578 | ||
84c07d11 | 3579 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3580 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3581 | { |
bf4f0599 | 3582 | struct obj_cgroup *objcg; |
d6441637 VD |
3583 | int memcg_id; |
3584 | ||
b313aeee VD |
3585 | if (cgroup_memory_nokmem) |
3586 | return 0; | |
3587 | ||
2a4db7eb | 3588 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3589 | BUG_ON(memcg->kmem_state); |
d6441637 | 3590 | |
f3bb3043 | 3591 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3592 | if (memcg_id < 0) |
3593 | return memcg_id; | |
d6441637 | 3594 | |
bf4f0599 RG |
3595 | objcg = obj_cgroup_alloc(); |
3596 | if (!objcg) { | |
3597 | memcg_free_cache_id(memcg_id); | |
3598 | return -ENOMEM; | |
3599 | } | |
3600 | objcg->memcg = memcg; | |
3601 | rcu_assign_pointer(memcg->objcg, objcg); | |
3602 | ||
d648bcc7 RG |
3603 | static_branch_enable(&memcg_kmem_enabled_key); |
3604 | ||
d6441637 | 3605 | /* |
567e9ab2 | 3606 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 3607 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
3608 | * guarantee no one starts accounting before all call sites are |
3609 | * patched. | |
3610 | */ | |
900a38f0 | 3611 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3612 | memcg->kmem_state = KMEM_ONLINE; |
0b8f73e1 JW |
3613 | |
3614 | return 0; | |
d6441637 VD |
3615 | } |
3616 | ||
8e0a8912 JW |
3617 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3618 | { | |
3619 | struct cgroup_subsys_state *css; | |
3620 | struct mem_cgroup *parent, *child; | |
3621 | int kmemcg_id; | |
3622 | ||
3623 | if (memcg->kmem_state != KMEM_ONLINE) | |
3624 | return; | |
9855609b | 3625 | |
8e0a8912 JW |
3626 | memcg->kmem_state = KMEM_ALLOCATED; |
3627 | ||
8e0a8912 JW |
3628 | parent = parent_mem_cgroup(memcg); |
3629 | if (!parent) | |
3630 | parent = root_mem_cgroup; | |
3631 | ||
bf4f0599 | 3632 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a RG |
3633 | |
3634 | kmemcg_id = memcg->kmemcg_id; | |
3635 | BUG_ON(kmemcg_id < 0); | |
3636 | ||
8e0a8912 JW |
3637 | /* |
3638 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3639 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3640 | * to ones of the parent. After we have finished, all list_lrus | |
3641 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3642 | * ordering is imposed by list_lru_node->lock taken by | |
3643 | * memcg_drain_all_list_lrus(). | |
3644 | */ | |
3a06bb78 | 3645 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3646 | css_for_each_descendant_pre(css, &memcg->css) { |
3647 | child = mem_cgroup_from_css(css); | |
3648 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3649 | child->kmemcg_id = parent->kmemcg_id; | |
3650 | if (!memcg->use_hierarchy) | |
3651 | break; | |
3652 | } | |
3a06bb78 TH |
3653 | rcu_read_unlock(); |
3654 | ||
9bec5c35 | 3655 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3656 | |
3657 | memcg_free_cache_id(kmemcg_id); | |
3658 | } | |
3659 | ||
3660 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3661 | { | |
0b8f73e1 JW |
3662 | /* css_alloc() failed, offlining didn't happen */ |
3663 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3664 | memcg_offline_kmem(memcg); | |
8e0a8912 | 3665 | } |
d6441637 | 3666 | #else |
0b8f73e1 | 3667 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3668 | { |
3669 | return 0; | |
3670 | } | |
3671 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3672 | { | |
3673 | } | |
3674 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3675 | { | |
3676 | } | |
84c07d11 | 3677 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3678 | |
bbec2e15 RG |
3679 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3680 | unsigned long max) | |
d6441637 | 3681 | { |
b313aeee | 3682 | int ret; |
127424c8 | 3683 | |
bbec2e15 RG |
3684 | mutex_lock(&memcg_max_mutex); |
3685 | ret = page_counter_set_max(&memcg->kmem, max); | |
3686 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3687 | return ret; |
d6441637 | 3688 | } |
510fc4e1 | 3689 | |
bbec2e15 | 3690 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3691 | { |
3692 | int ret; | |
3693 | ||
bbec2e15 | 3694 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3695 | |
bbec2e15 | 3696 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3697 | if (ret) |
3698 | goto out; | |
3699 | ||
0db15298 | 3700 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3701 | /* |
3702 | * The active flag needs to be written after the static_key | |
3703 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3704 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3705 | * for details, and note that we don't mark any socket as | |
3706 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3707 | * |
3708 | * We need to do this, because static_keys will span multiple | |
3709 | * sites, but we can't control their order. If we mark a socket | |
3710 | * as accounted, but the accounting functions are not patched in | |
3711 | * yet, we'll lose accounting. | |
3712 | * | |
2d758073 | 3713 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3714 | * because when this value change, the code to process it is not |
3715 | * patched in yet. | |
3716 | */ | |
3717 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3718 | memcg->tcpmem_active = true; |
d55f90bf VD |
3719 | } |
3720 | out: | |
bbec2e15 | 3721 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3722 | return ret; |
3723 | } | |
d55f90bf | 3724 | |
628f4235 KH |
3725 | /* |
3726 | * The user of this function is... | |
3727 | * RES_LIMIT. | |
3728 | */ | |
451af504 TH |
3729 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3730 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3731 | { |
451af504 | 3732 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3733 | unsigned long nr_pages; |
628f4235 KH |
3734 | int ret; |
3735 | ||
451af504 | 3736 | buf = strstrip(buf); |
650c5e56 | 3737 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3738 | if (ret) |
3739 | return ret; | |
af36f906 | 3740 | |
3e32cb2e | 3741 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3742 | case RES_LIMIT: |
4b3bde4c BS |
3743 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3744 | ret = -EINVAL; | |
3745 | break; | |
3746 | } | |
3e32cb2e JW |
3747 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3748 | case _MEM: | |
bbec2e15 | 3749 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3750 | break; |
3e32cb2e | 3751 | case _MEMSWAP: |
bbec2e15 | 3752 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3753 | break; |
3e32cb2e | 3754 | case _KMEM: |
0158115f MH |
3755 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " |
3756 | "Please report your usecase to linux-mm@kvack.org if you " | |
3757 | "depend on this functionality.\n"); | |
bbec2e15 | 3758 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3759 | break; |
d55f90bf | 3760 | case _TCP: |
bbec2e15 | 3761 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3762 | break; |
3e32cb2e | 3763 | } |
296c81d8 | 3764 | break; |
3e32cb2e JW |
3765 | case RES_SOFT_LIMIT: |
3766 | memcg->soft_limit = nr_pages; | |
3767 | ret = 0; | |
628f4235 KH |
3768 | break; |
3769 | } | |
451af504 | 3770 | return ret ?: nbytes; |
8cdea7c0 BS |
3771 | } |
3772 | ||
6770c64e TH |
3773 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3774 | size_t nbytes, loff_t off) | |
c84872e1 | 3775 | { |
6770c64e | 3776 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3777 | struct page_counter *counter; |
c84872e1 | 3778 | |
3e32cb2e JW |
3779 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3780 | case _MEM: | |
3781 | counter = &memcg->memory; | |
3782 | break; | |
3783 | case _MEMSWAP: | |
3784 | counter = &memcg->memsw; | |
3785 | break; | |
3786 | case _KMEM: | |
3787 | counter = &memcg->kmem; | |
3788 | break; | |
d55f90bf | 3789 | case _TCP: |
0db15298 | 3790 | counter = &memcg->tcpmem; |
d55f90bf | 3791 | break; |
3e32cb2e JW |
3792 | default: |
3793 | BUG(); | |
3794 | } | |
af36f906 | 3795 | |
3e32cb2e | 3796 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3797 | case RES_MAX_USAGE: |
3e32cb2e | 3798 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3799 | break; |
3800 | case RES_FAILCNT: | |
3e32cb2e | 3801 | counter->failcnt = 0; |
29f2a4da | 3802 | break; |
3e32cb2e JW |
3803 | default: |
3804 | BUG(); | |
29f2a4da | 3805 | } |
f64c3f54 | 3806 | |
6770c64e | 3807 | return nbytes; |
c84872e1 PE |
3808 | } |
3809 | ||
182446d0 | 3810 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3811 | struct cftype *cft) |
3812 | { | |
182446d0 | 3813 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3814 | } |
3815 | ||
02491447 | 3816 | #ifdef CONFIG_MMU |
182446d0 | 3817 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3818 | struct cftype *cft, u64 val) |
3819 | { | |
182446d0 | 3820 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3821 | |
1dfab5ab | 3822 | if (val & ~MOVE_MASK) |
7dc74be0 | 3823 | return -EINVAL; |
ee5e8472 | 3824 | |
7dc74be0 | 3825 | /* |
ee5e8472 GC |
3826 | * No kind of locking is needed in here, because ->can_attach() will |
3827 | * check this value once in the beginning of the process, and then carry | |
3828 | * on with stale data. This means that changes to this value will only | |
3829 | * affect task migrations starting after the change. | |
7dc74be0 | 3830 | */ |
c0ff4b85 | 3831 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3832 | return 0; |
3833 | } | |
02491447 | 3834 | #else |
182446d0 | 3835 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3836 | struct cftype *cft, u64 val) |
3837 | { | |
3838 | return -ENOSYS; | |
3839 | } | |
3840 | #endif | |
7dc74be0 | 3841 | |
406eb0c9 | 3842 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3843 | |
3844 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3845 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3846 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3847 | ||
3848 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 3849 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 3850 | { |
867e5e1d | 3851 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
3852 | unsigned long nr = 0; |
3853 | enum lru_list lru; | |
3854 | ||
3855 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3856 | ||
3857 | for_each_lru(lru) { | |
3858 | if (!(BIT(lru) & lru_mask)) | |
3859 | continue; | |
dd8657b6 SB |
3860 | if (tree) |
3861 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
3862 | else | |
3863 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
3864 | } |
3865 | return nr; | |
3866 | } | |
3867 | ||
3868 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
3869 | unsigned int lru_mask, |
3870 | bool tree) | |
113b7dfd JW |
3871 | { |
3872 | unsigned long nr = 0; | |
3873 | enum lru_list lru; | |
3874 | ||
3875 | for_each_lru(lru) { | |
3876 | if (!(BIT(lru) & lru_mask)) | |
3877 | continue; | |
dd8657b6 SB |
3878 | if (tree) |
3879 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
3880 | else | |
3881 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
3882 | } |
3883 | return nr; | |
3884 | } | |
3885 | ||
2da8ca82 | 3886 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3887 | { |
25485de6 GT |
3888 | struct numa_stat { |
3889 | const char *name; | |
3890 | unsigned int lru_mask; | |
3891 | }; | |
3892 | ||
3893 | static const struct numa_stat stats[] = { | |
3894 | { "total", LRU_ALL }, | |
3895 | { "file", LRU_ALL_FILE }, | |
3896 | { "anon", LRU_ALL_ANON }, | |
3897 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3898 | }; | |
3899 | const struct numa_stat *stat; | |
406eb0c9 | 3900 | int nid; |
aa9694bb | 3901 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3902 | |
25485de6 | 3903 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3904 | seq_printf(m, "%s=%lu", stat->name, |
3905 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3906 | false)); | |
3907 | for_each_node_state(nid, N_MEMORY) | |
3908 | seq_printf(m, " N%d=%lu", nid, | |
3909 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3910 | stat->lru_mask, false)); | |
25485de6 | 3911 | seq_putc(m, '\n'); |
406eb0c9 | 3912 | } |
406eb0c9 | 3913 | |
071aee13 | 3914 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3915 | |
3916 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
3917 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3918 | true)); | |
3919 | for_each_node_state(nid, N_MEMORY) | |
3920 | seq_printf(m, " N%d=%lu", nid, | |
3921 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3922 | stat->lru_mask, true)); | |
071aee13 | 3923 | seq_putc(m, '\n'); |
406eb0c9 | 3924 | } |
406eb0c9 | 3925 | |
406eb0c9 YH |
3926 | return 0; |
3927 | } | |
3928 | #endif /* CONFIG_NUMA */ | |
3929 | ||
c8713d0b | 3930 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 3931 | NR_FILE_PAGES, |
be5d0a74 | 3932 | NR_ANON_MAPPED, |
468c3982 JW |
3933 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3934 | NR_ANON_THPS, | |
3935 | #endif | |
c8713d0b JW |
3936 | NR_SHMEM, |
3937 | NR_FILE_MAPPED, | |
3938 | NR_FILE_DIRTY, | |
3939 | NR_WRITEBACK, | |
3940 | MEMCG_SWAP, | |
3941 | }; | |
3942 | ||
3943 | static const char *const memcg1_stat_names[] = { | |
3944 | "cache", | |
3945 | "rss", | |
468c3982 | 3946 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 3947 | "rss_huge", |
468c3982 | 3948 | #endif |
c8713d0b JW |
3949 | "shmem", |
3950 | "mapped_file", | |
3951 | "dirty", | |
3952 | "writeback", | |
3953 | "swap", | |
3954 | }; | |
3955 | ||
df0e53d0 | 3956 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3957 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
3958 | PGPGIN, |
3959 | PGPGOUT, | |
3960 | PGFAULT, | |
3961 | PGMAJFAULT, | |
3962 | }; | |
3963 | ||
2da8ca82 | 3964 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3965 | { |
aa9694bb | 3966 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 3967 | unsigned long memory, memsw; |
af7c4b0e JW |
3968 | struct mem_cgroup *mi; |
3969 | unsigned int i; | |
406eb0c9 | 3970 | |
71cd3113 | 3971 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 3972 | |
71cd3113 | 3973 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
3974 | unsigned long nr; |
3975 | ||
71cd3113 | 3976 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3977 | continue; |
468c3982 JW |
3978 | nr = memcg_page_state_local(memcg, memcg1_stats[i]); |
3979 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
3980 | if (memcg1_stats[i] == NR_ANON_THPS) | |
3981 | nr *= HPAGE_PMD_NR; | |
3982 | #endif | |
3983 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE); | |
1dd3a273 | 3984 | } |
7b854121 | 3985 | |
df0e53d0 | 3986 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d | 3987 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
205b20cc | 3988 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
3989 | |
3990 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 3991 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
205b20cc | 3992 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 3993 | PAGE_SIZE); |
af7c4b0e | 3994 | |
14067bb3 | 3995 | /* Hierarchical information */ |
3e32cb2e JW |
3996 | memory = memsw = PAGE_COUNTER_MAX; |
3997 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
3998 | memory = min(memory, READ_ONCE(mi->memory.max)); |
3999 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 4000 | } |
3e32cb2e JW |
4001 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
4002 | (u64)memory * PAGE_SIZE); | |
7941d214 | 4003 | if (do_memsw_account()) |
3e32cb2e JW |
4004 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
4005 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 4006 | |
8de7ecc6 | 4007 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
71cd3113 | 4008 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 4009 | continue; |
8de7ecc6 | 4010 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
dd923990 YS |
4011 | (u64)memcg_page_state(memcg, memcg1_stats[i]) * |
4012 | PAGE_SIZE); | |
af7c4b0e JW |
4013 | } |
4014 | ||
8de7ecc6 | 4015 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d KK |
4016 | seq_printf(m, "total_%s %llu\n", |
4017 | vm_event_name(memcg1_events[i]), | |
dd923990 | 4018 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 4019 | |
8de7ecc6 | 4020 | for (i = 0; i < NR_LRU_LISTS; i++) |
ebc5d83d | 4021 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
42a30035 JW |
4022 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
4023 | PAGE_SIZE); | |
14067bb3 | 4024 | |
7f016ee8 | 4025 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 4026 | { |
ef8f2327 MG |
4027 | pg_data_t *pgdat; |
4028 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
4029 | unsigned long anon_cost = 0; |
4030 | unsigned long file_cost = 0; | |
7f016ee8 | 4031 | |
ef8f2327 MG |
4032 | for_each_online_pgdat(pgdat) { |
4033 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
7f016ee8 | 4034 | |
1431d4d1 JW |
4035 | anon_cost += mz->lruvec.anon_cost; |
4036 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 4037 | } |
1431d4d1 JW |
4038 | seq_printf(m, "anon_cost %lu\n", anon_cost); |
4039 | seq_printf(m, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
4040 | } |
4041 | #endif | |
4042 | ||
d2ceb9b7 KH |
4043 | return 0; |
4044 | } | |
4045 | ||
182446d0 TH |
4046 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4047 | struct cftype *cft) | |
a7885eb8 | 4048 | { |
182446d0 | 4049 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4050 | |
1f4c025b | 4051 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4052 | } |
4053 | ||
182446d0 TH |
4054 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4055 | struct cftype *cft, u64 val) | |
a7885eb8 | 4056 | { |
182446d0 | 4057 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4058 | |
3dae7fec | 4059 | if (val > 100) |
a7885eb8 KM |
4060 | return -EINVAL; |
4061 | ||
14208b0e | 4062 | if (css->parent) |
3dae7fec JW |
4063 | memcg->swappiness = val; |
4064 | else | |
4065 | vm_swappiness = val; | |
068b38c1 | 4066 | |
a7885eb8 KM |
4067 | return 0; |
4068 | } | |
4069 | ||
2e72b634 KS |
4070 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4071 | { | |
4072 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4073 | unsigned long usage; |
2e72b634 KS |
4074 | int i; |
4075 | ||
4076 | rcu_read_lock(); | |
4077 | if (!swap) | |
2c488db2 | 4078 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4079 | else |
2c488db2 | 4080 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4081 | |
4082 | if (!t) | |
4083 | goto unlock; | |
4084 | ||
ce00a967 | 4085 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4086 | |
4087 | /* | |
748dad36 | 4088 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4089 | * If it's not true, a threshold was crossed after last |
4090 | * call of __mem_cgroup_threshold(). | |
4091 | */ | |
5407a562 | 4092 | i = t->current_threshold; |
2e72b634 KS |
4093 | |
4094 | /* | |
4095 | * Iterate backward over array of thresholds starting from | |
4096 | * current_threshold and check if a threshold is crossed. | |
4097 | * If none of thresholds below usage is crossed, we read | |
4098 | * only one element of the array here. | |
4099 | */ | |
4100 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4101 | eventfd_signal(t->entries[i].eventfd, 1); | |
4102 | ||
4103 | /* i = current_threshold + 1 */ | |
4104 | i++; | |
4105 | ||
4106 | /* | |
4107 | * Iterate forward over array of thresholds starting from | |
4108 | * current_threshold+1 and check if a threshold is crossed. | |
4109 | * If none of thresholds above usage is crossed, we read | |
4110 | * only one element of the array here. | |
4111 | */ | |
4112 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4113 | eventfd_signal(t->entries[i].eventfd, 1); | |
4114 | ||
4115 | /* Update current_threshold */ | |
5407a562 | 4116 | t->current_threshold = i - 1; |
2e72b634 KS |
4117 | unlock: |
4118 | rcu_read_unlock(); | |
4119 | } | |
4120 | ||
4121 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4122 | { | |
ad4ca5f4 KS |
4123 | while (memcg) { |
4124 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4125 | if (do_memsw_account()) |
ad4ca5f4 KS |
4126 | __mem_cgroup_threshold(memcg, true); |
4127 | ||
4128 | memcg = parent_mem_cgroup(memcg); | |
4129 | } | |
2e72b634 KS |
4130 | } |
4131 | ||
4132 | static int compare_thresholds(const void *a, const void *b) | |
4133 | { | |
4134 | const struct mem_cgroup_threshold *_a = a; | |
4135 | const struct mem_cgroup_threshold *_b = b; | |
4136 | ||
2bff24a3 GT |
4137 | if (_a->threshold > _b->threshold) |
4138 | return 1; | |
4139 | ||
4140 | if (_a->threshold < _b->threshold) | |
4141 | return -1; | |
4142 | ||
4143 | return 0; | |
2e72b634 KS |
4144 | } |
4145 | ||
c0ff4b85 | 4146 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4147 | { |
4148 | struct mem_cgroup_eventfd_list *ev; | |
4149 | ||
2bcf2e92 MH |
4150 | spin_lock(&memcg_oom_lock); |
4151 | ||
c0ff4b85 | 4152 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4153 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4154 | |
4155 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4156 | return 0; |
4157 | } | |
4158 | ||
c0ff4b85 | 4159 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4160 | { |
7d74b06f KH |
4161 | struct mem_cgroup *iter; |
4162 | ||
c0ff4b85 | 4163 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4164 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4165 | } |
4166 | ||
59b6f873 | 4167 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4168 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4169 | { |
2c488db2 KS |
4170 | struct mem_cgroup_thresholds *thresholds; |
4171 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4172 | unsigned long threshold; |
4173 | unsigned long usage; | |
2c488db2 | 4174 | int i, size, ret; |
2e72b634 | 4175 | |
650c5e56 | 4176 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4177 | if (ret) |
4178 | return ret; | |
4179 | ||
4180 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4181 | |
05b84301 | 4182 | if (type == _MEM) { |
2c488db2 | 4183 | thresholds = &memcg->thresholds; |
ce00a967 | 4184 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4185 | } else if (type == _MEMSWAP) { |
2c488db2 | 4186 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4187 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4188 | } else |
2e72b634 KS |
4189 | BUG(); |
4190 | ||
2e72b634 | 4191 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4192 | if (thresholds->primary) |
2e72b634 KS |
4193 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4194 | ||
2c488db2 | 4195 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4196 | |
4197 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4198 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4199 | if (!new) { |
2e72b634 KS |
4200 | ret = -ENOMEM; |
4201 | goto unlock; | |
4202 | } | |
2c488db2 | 4203 | new->size = size; |
2e72b634 KS |
4204 | |
4205 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4206 | if (thresholds->primary) { |
4207 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4208 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4209 | } |
4210 | ||
2e72b634 | 4211 | /* Add new threshold */ |
2c488db2 KS |
4212 | new->entries[size - 1].eventfd = eventfd; |
4213 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4214 | |
4215 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4216 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4217 | compare_thresholds, NULL); |
4218 | ||
4219 | /* Find current threshold */ | |
2c488db2 | 4220 | new->current_threshold = -1; |
2e72b634 | 4221 | for (i = 0; i < size; i++) { |
748dad36 | 4222 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4223 | /* |
2c488db2 KS |
4224 | * new->current_threshold will not be used until |
4225 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4226 | * it here. |
4227 | */ | |
2c488db2 | 4228 | ++new->current_threshold; |
748dad36 SZ |
4229 | } else |
4230 | break; | |
2e72b634 KS |
4231 | } |
4232 | ||
2c488db2 KS |
4233 | /* Free old spare buffer and save old primary buffer as spare */ |
4234 | kfree(thresholds->spare); | |
4235 | thresholds->spare = thresholds->primary; | |
4236 | ||
4237 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4238 | |
907860ed | 4239 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4240 | synchronize_rcu(); |
4241 | ||
2e72b634 KS |
4242 | unlock: |
4243 | mutex_unlock(&memcg->thresholds_lock); | |
4244 | ||
4245 | return ret; | |
4246 | } | |
4247 | ||
59b6f873 | 4248 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4249 | struct eventfd_ctx *eventfd, const char *args) |
4250 | { | |
59b6f873 | 4251 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4252 | } |
4253 | ||
59b6f873 | 4254 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4255 | struct eventfd_ctx *eventfd, const char *args) |
4256 | { | |
59b6f873 | 4257 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4258 | } |
4259 | ||
59b6f873 | 4260 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4261 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4262 | { |
2c488db2 KS |
4263 | struct mem_cgroup_thresholds *thresholds; |
4264 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4265 | unsigned long usage; |
7d36665a | 4266 | int i, j, size, entries; |
2e72b634 KS |
4267 | |
4268 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4269 | |
4270 | if (type == _MEM) { | |
2c488db2 | 4271 | thresholds = &memcg->thresholds; |
ce00a967 | 4272 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4273 | } else if (type == _MEMSWAP) { |
2c488db2 | 4274 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4275 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4276 | } else |
2e72b634 KS |
4277 | BUG(); |
4278 | ||
371528ca AV |
4279 | if (!thresholds->primary) |
4280 | goto unlock; | |
4281 | ||
2e72b634 KS |
4282 | /* Check if a threshold crossed before removing */ |
4283 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4284 | ||
4285 | /* Calculate new number of threshold */ | |
7d36665a | 4286 | size = entries = 0; |
2c488db2 KS |
4287 | for (i = 0; i < thresholds->primary->size; i++) { |
4288 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4289 | size++; |
7d36665a CX |
4290 | else |
4291 | entries++; | |
2e72b634 KS |
4292 | } |
4293 | ||
2c488db2 | 4294 | new = thresholds->spare; |
907860ed | 4295 | |
7d36665a CX |
4296 | /* If no items related to eventfd have been cleared, nothing to do */ |
4297 | if (!entries) | |
4298 | goto unlock; | |
4299 | ||
2e72b634 KS |
4300 | /* Set thresholds array to NULL if we don't have thresholds */ |
4301 | if (!size) { | |
2c488db2 KS |
4302 | kfree(new); |
4303 | new = NULL; | |
907860ed | 4304 | goto swap_buffers; |
2e72b634 KS |
4305 | } |
4306 | ||
2c488db2 | 4307 | new->size = size; |
2e72b634 KS |
4308 | |
4309 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4310 | new->current_threshold = -1; |
4311 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4312 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4313 | continue; |
4314 | ||
2c488db2 | 4315 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4316 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4317 | /* |
2c488db2 | 4318 | * new->current_threshold will not be used |
2e72b634 KS |
4319 | * until rcu_assign_pointer(), so it's safe to increment |
4320 | * it here. | |
4321 | */ | |
2c488db2 | 4322 | ++new->current_threshold; |
2e72b634 KS |
4323 | } |
4324 | j++; | |
4325 | } | |
4326 | ||
907860ed | 4327 | swap_buffers: |
2c488db2 KS |
4328 | /* Swap primary and spare array */ |
4329 | thresholds->spare = thresholds->primary; | |
8c757763 | 4330 | |
2c488db2 | 4331 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4332 | |
907860ed | 4333 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4334 | synchronize_rcu(); |
6611d8d7 MC |
4335 | |
4336 | /* If all events are unregistered, free the spare array */ | |
4337 | if (!new) { | |
4338 | kfree(thresholds->spare); | |
4339 | thresholds->spare = NULL; | |
4340 | } | |
371528ca | 4341 | unlock: |
2e72b634 | 4342 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4343 | } |
c1e862c1 | 4344 | |
59b6f873 | 4345 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4346 | struct eventfd_ctx *eventfd) |
4347 | { | |
59b6f873 | 4348 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4349 | } |
4350 | ||
59b6f873 | 4351 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4352 | struct eventfd_ctx *eventfd) |
4353 | { | |
59b6f873 | 4354 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4355 | } |
4356 | ||
59b6f873 | 4357 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4358 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4359 | { |
9490ff27 | 4360 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4361 | |
9490ff27 KH |
4362 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4363 | if (!event) | |
4364 | return -ENOMEM; | |
4365 | ||
1af8efe9 | 4366 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4367 | |
4368 | event->eventfd = eventfd; | |
4369 | list_add(&event->list, &memcg->oom_notify); | |
4370 | ||
4371 | /* already in OOM ? */ | |
c2b42d3c | 4372 | if (memcg->under_oom) |
9490ff27 | 4373 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4374 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4375 | |
4376 | return 0; | |
4377 | } | |
4378 | ||
59b6f873 | 4379 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4380 | struct eventfd_ctx *eventfd) |
9490ff27 | 4381 | { |
9490ff27 | 4382 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4383 | |
1af8efe9 | 4384 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4385 | |
c0ff4b85 | 4386 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4387 | if (ev->eventfd == eventfd) { |
4388 | list_del(&ev->list); | |
4389 | kfree(ev); | |
4390 | } | |
4391 | } | |
4392 | ||
1af8efe9 | 4393 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4394 | } |
4395 | ||
2da8ca82 | 4396 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4397 | { |
aa9694bb | 4398 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4399 | |
791badbd | 4400 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4401 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4402 | seq_printf(sf, "oom_kill %lu\n", |
4403 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4404 | return 0; |
4405 | } | |
4406 | ||
182446d0 | 4407 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4408 | struct cftype *cft, u64 val) |
4409 | { | |
182446d0 | 4410 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4411 | |
4412 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4413 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4414 | return -EINVAL; |
4415 | ||
c0ff4b85 | 4416 | memcg->oom_kill_disable = val; |
4d845ebf | 4417 | if (!val) |
c0ff4b85 | 4418 | memcg_oom_recover(memcg); |
3dae7fec | 4419 | |
3c11ecf4 KH |
4420 | return 0; |
4421 | } | |
4422 | ||
52ebea74 TH |
4423 | #ifdef CONFIG_CGROUP_WRITEBACK |
4424 | ||
3a8e9ac8 TH |
4425 | #include <trace/events/writeback.h> |
4426 | ||
841710aa TH |
4427 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4428 | { | |
4429 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4430 | } | |
4431 | ||
4432 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4433 | { | |
4434 | wb_domain_exit(&memcg->cgwb_domain); | |
4435 | } | |
4436 | ||
2529bb3a TH |
4437 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4438 | { | |
4439 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4440 | } | |
4441 | ||
841710aa TH |
4442 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4443 | { | |
4444 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4445 | ||
4446 | if (!memcg->css.parent) | |
4447 | return NULL; | |
4448 | ||
4449 | return &memcg->cgwb_domain; | |
4450 | } | |
4451 | ||
0b3d6e6f GT |
4452 | /* |
4453 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
4454 | * Keep in sync with memcg_exact_page(). | |
4455 | */ | |
4456 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
4457 | { | |
871789d4 | 4458 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
4459 | int cpu; |
4460 | ||
4461 | for_each_online_cpu(cpu) | |
871789d4 | 4462 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
4463 | if (x < 0) |
4464 | x = 0; | |
4465 | return x; | |
4466 | } | |
4467 | ||
c2aa723a TH |
4468 | /** |
4469 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4470 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4471 | * @pfilepages: out parameter for number of file pages |
4472 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4473 | * @pdirty: out parameter for number of dirty pages |
4474 | * @pwriteback: out parameter for number of pages under writeback | |
4475 | * | |
c5edf9cd TH |
4476 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4477 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4478 | * is a bit more involved. | |
c2aa723a | 4479 | * |
c5edf9cd TH |
4480 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4481 | * headroom is calculated as the lowest headroom of itself and the | |
4482 | * ancestors. Note that this doesn't consider the actual amount of | |
4483 | * available memory in the system. The caller should further cap | |
4484 | * *@pheadroom accordingly. | |
c2aa723a | 4485 | */ |
c5edf9cd TH |
4486 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4487 | unsigned long *pheadroom, unsigned long *pdirty, | |
4488 | unsigned long *pwriteback) | |
c2aa723a TH |
4489 | { |
4490 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4491 | struct mem_cgroup *parent; | |
c2aa723a | 4492 | |
0b3d6e6f | 4493 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a | 4494 | |
0b3d6e6f | 4495 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4496 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4497 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4498 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4499 | |
c2aa723a | 4500 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 4501 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 4502 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
4503 | unsigned long used = page_counter_read(&memcg->memory); |
4504 | ||
c5edf9cd | 4505 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4506 | memcg = parent; |
4507 | } | |
c2aa723a TH |
4508 | } |
4509 | ||
97b27821 TH |
4510 | /* |
4511 | * Foreign dirty flushing | |
4512 | * | |
4513 | * There's an inherent mismatch between memcg and writeback. The former | |
4514 | * trackes ownership per-page while the latter per-inode. This was a | |
4515 | * deliberate design decision because honoring per-page ownership in the | |
4516 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4517 | * and deemed unnecessary given that write-sharing an inode across | |
4518 | * different cgroups isn't a common use-case. | |
4519 | * | |
4520 | * Combined with inode majority-writer ownership switching, this works well | |
4521 | * enough in most cases but there are some pathological cases. For | |
4522 | * example, let's say there are two cgroups A and B which keep writing to | |
4523 | * different but confined parts of the same inode. B owns the inode and | |
4524 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4525 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4526 | * triggering background writeback. A will be slowed down without a way to | |
4527 | * make writeback of the dirty pages happen. | |
4528 | * | |
4529 | * Conditions like the above can lead to a cgroup getting repatedly and | |
4530 | * severely throttled after making some progress after each | |
4531 | * dirty_expire_interval while the underyling IO device is almost | |
4532 | * completely idle. | |
4533 | * | |
4534 | * Solving this problem completely requires matching the ownership tracking | |
4535 | * granularities between memcg and writeback in either direction. However, | |
4536 | * the more egregious behaviors can be avoided by simply remembering the | |
4537 | * most recent foreign dirtying events and initiating remote flushes on | |
4538 | * them when local writeback isn't enough to keep the memory clean enough. | |
4539 | * | |
4540 | * The following two functions implement such mechanism. When a foreign | |
4541 | * page - a page whose memcg and writeback ownerships don't match - is | |
4542 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4543 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4544 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4545 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4546 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4547 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4548 | * limited to MEMCG_CGWB_FRN_CNT. | |
4549 | * | |
4550 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4551 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4552 | * records are lockless and racy. | |
4553 | */ | |
4554 | void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, | |
4555 | struct bdi_writeback *wb) | |
4556 | { | |
4557 | struct mem_cgroup *memcg = page->mem_cgroup; | |
4558 | struct memcg_cgwb_frn *frn; | |
4559 | u64 now = get_jiffies_64(); | |
4560 | u64 oldest_at = now; | |
4561 | int oldest = -1; | |
4562 | int i; | |
4563 | ||
3a8e9ac8 TH |
4564 | trace_track_foreign_dirty(page, wb); |
4565 | ||
97b27821 TH |
4566 | /* |
4567 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4568 | * using it. If not replace the oldest one which isn't being | |
4569 | * written out. | |
4570 | */ | |
4571 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4572 | frn = &memcg->cgwb_frn[i]; | |
4573 | if (frn->bdi_id == wb->bdi->id && | |
4574 | frn->memcg_id == wb->memcg_css->id) | |
4575 | break; | |
4576 | if (time_before64(frn->at, oldest_at) && | |
4577 | atomic_read(&frn->done.cnt) == 1) { | |
4578 | oldest = i; | |
4579 | oldest_at = frn->at; | |
4580 | } | |
4581 | } | |
4582 | ||
4583 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4584 | /* | |
4585 | * Re-using an existing one. Update timestamp lazily to | |
4586 | * avoid making the cacheline hot. We want them to be | |
4587 | * reasonably up-to-date and significantly shorter than | |
4588 | * dirty_expire_interval as that's what expires the record. | |
4589 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4590 | */ | |
4591 | unsigned long update_intv = | |
4592 | min_t(unsigned long, HZ, | |
4593 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4594 | ||
4595 | if (time_before64(frn->at, now - update_intv)) | |
4596 | frn->at = now; | |
4597 | } else if (oldest >= 0) { | |
4598 | /* replace the oldest free one */ | |
4599 | frn = &memcg->cgwb_frn[oldest]; | |
4600 | frn->bdi_id = wb->bdi->id; | |
4601 | frn->memcg_id = wb->memcg_css->id; | |
4602 | frn->at = now; | |
4603 | } | |
4604 | } | |
4605 | ||
4606 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4607 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4608 | { | |
4609 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4610 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4611 | u64 now = jiffies_64; | |
4612 | int i; | |
4613 | ||
4614 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4615 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4616 | ||
4617 | /* | |
4618 | * If the record is older than dirty_expire_interval, | |
4619 | * writeback on it has already started. No need to kick it | |
4620 | * off again. Also, don't start a new one if there's | |
4621 | * already one in flight. | |
4622 | */ | |
4623 | if (time_after64(frn->at, now - intv) && | |
4624 | atomic_read(&frn->done.cnt) == 1) { | |
4625 | frn->at = 0; | |
3a8e9ac8 | 4626 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
97b27821 TH |
4627 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0, |
4628 | WB_REASON_FOREIGN_FLUSH, | |
4629 | &frn->done); | |
4630 | } | |
4631 | } | |
4632 | } | |
4633 | ||
841710aa TH |
4634 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4635 | ||
4636 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4637 | { | |
4638 | return 0; | |
4639 | } | |
4640 | ||
4641 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4642 | { | |
4643 | } | |
4644 | ||
2529bb3a TH |
4645 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4646 | { | |
4647 | } | |
4648 | ||
52ebea74 TH |
4649 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4650 | ||
3bc942f3 TH |
4651 | /* |
4652 | * DO NOT USE IN NEW FILES. | |
4653 | * | |
4654 | * "cgroup.event_control" implementation. | |
4655 | * | |
4656 | * This is way over-engineered. It tries to support fully configurable | |
4657 | * events for each user. Such level of flexibility is completely | |
4658 | * unnecessary especially in the light of the planned unified hierarchy. | |
4659 | * | |
4660 | * Please deprecate this and replace with something simpler if at all | |
4661 | * possible. | |
4662 | */ | |
4663 | ||
79bd9814 TH |
4664 | /* |
4665 | * Unregister event and free resources. | |
4666 | * | |
4667 | * Gets called from workqueue. | |
4668 | */ | |
3bc942f3 | 4669 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4670 | { |
3bc942f3 TH |
4671 | struct mem_cgroup_event *event = |
4672 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4673 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4674 | |
4675 | remove_wait_queue(event->wqh, &event->wait); | |
4676 | ||
59b6f873 | 4677 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4678 | |
4679 | /* Notify userspace the event is going away. */ | |
4680 | eventfd_signal(event->eventfd, 1); | |
4681 | ||
4682 | eventfd_ctx_put(event->eventfd); | |
4683 | kfree(event); | |
59b6f873 | 4684 | css_put(&memcg->css); |
79bd9814 TH |
4685 | } |
4686 | ||
4687 | /* | |
a9a08845 | 4688 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4689 | * |
4690 | * Called with wqh->lock held and interrupts disabled. | |
4691 | */ | |
ac6424b9 | 4692 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4693 | int sync, void *key) |
79bd9814 | 4694 | { |
3bc942f3 TH |
4695 | struct mem_cgroup_event *event = |
4696 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4697 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4698 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4699 | |
a9a08845 | 4700 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4701 | /* |
4702 | * If the event has been detached at cgroup removal, we | |
4703 | * can simply return knowing the other side will cleanup | |
4704 | * for us. | |
4705 | * | |
4706 | * We can't race against event freeing since the other | |
4707 | * side will require wqh->lock via remove_wait_queue(), | |
4708 | * which we hold. | |
4709 | */ | |
fba94807 | 4710 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4711 | if (!list_empty(&event->list)) { |
4712 | list_del_init(&event->list); | |
4713 | /* | |
4714 | * We are in atomic context, but cgroup_event_remove() | |
4715 | * may sleep, so we have to call it in workqueue. | |
4716 | */ | |
4717 | schedule_work(&event->remove); | |
4718 | } | |
fba94807 | 4719 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4720 | } |
4721 | ||
4722 | return 0; | |
4723 | } | |
4724 | ||
3bc942f3 | 4725 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4726 | wait_queue_head_t *wqh, poll_table *pt) |
4727 | { | |
3bc942f3 TH |
4728 | struct mem_cgroup_event *event = |
4729 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4730 | |
4731 | event->wqh = wqh; | |
4732 | add_wait_queue(wqh, &event->wait); | |
4733 | } | |
4734 | ||
4735 | /* | |
3bc942f3 TH |
4736 | * DO NOT USE IN NEW FILES. |
4737 | * | |
79bd9814 TH |
4738 | * Parse input and register new cgroup event handler. |
4739 | * | |
4740 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4741 | * Interpretation of args is defined by control file implementation. | |
4742 | */ | |
451af504 TH |
4743 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4744 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4745 | { |
451af504 | 4746 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4747 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4748 | struct mem_cgroup_event *event; |
79bd9814 TH |
4749 | struct cgroup_subsys_state *cfile_css; |
4750 | unsigned int efd, cfd; | |
4751 | struct fd efile; | |
4752 | struct fd cfile; | |
fba94807 | 4753 | const char *name; |
79bd9814 TH |
4754 | char *endp; |
4755 | int ret; | |
4756 | ||
451af504 TH |
4757 | buf = strstrip(buf); |
4758 | ||
4759 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4760 | if (*endp != ' ') |
4761 | return -EINVAL; | |
451af504 | 4762 | buf = endp + 1; |
79bd9814 | 4763 | |
451af504 | 4764 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4765 | if ((*endp != ' ') && (*endp != '\0')) |
4766 | return -EINVAL; | |
451af504 | 4767 | buf = endp + 1; |
79bd9814 TH |
4768 | |
4769 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4770 | if (!event) | |
4771 | return -ENOMEM; | |
4772 | ||
59b6f873 | 4773 | event->memcg = memcg; |
79bd9814 | 4774 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4775 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4776 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4777 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4778 | |
4779 | efile = fdget(efd); | |
4780 | if (!efile.file) { | |
4781 | ret = -EBADF; | |
4782 | goto out_kfree; | |
4783 | } | |
4784 | ||
4785 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4786 | if (IS_ERR(event->eventfd)) { | |
4787 | ret = PTR_ERR(event->eventfd); | |
4788 | goto out_put_efile; | |
4789 | } | |
4790 | ||
4791 | cfile = fdget(cfd); | |
4792 | if (!cfile.file) { | |
4793 | ret = -EBADF; | |
4794 | goto out_put_eventfd; | |
4795 | } | |
4796 | ||
4797 | /* the process need read permission on control file */ | |
4798 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4799 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4800 | if (ret < 0) | |
4801 | goto out_put_cfile; | |
4802 | ||
fba94807 TH |
4803 | /* |
4804 | * Determine the event callbacks and set them in @event. This used | |
4805 | * to be done via struct cftype but cgroup core no longer knows | |
4806 | * about these events. The following is crude but the whole thing | |
4807 | * is for compatibility anyway. | |
3bc942f3 TH |
4808 | * |
4809 | * DO NOT ADD NEW FILES. | |
fba94807 | 4810 | */ |
b583043e | 4811 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4812 | |
4813 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4814 | event->register_event = mem_cgroup_usage_register_event; | |
4815 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4816 | } else if (!strcmp(name, "memory.oom_control")) { | |
4817 | event->register_event = mem_cgroup_oom_register_event; | |
4818 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4819 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4820 | event->register_event = vmpressure_register_event; | |
4821 | event->unregister_event = vmpressure_unregister_event; | |
4822 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4823 | event->register_event = memsw_cgroup_usage_register_event; |
4824 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4825 | } else { |
4826 | ret = -EINVAL; | |
4827 | goto out_put_cfile; | |
4828 | } | |
4829 | ||
79bd9814 | 4830 | /* |
b5557c4c TH |
4831 | * Verify @cfile should belong to @css. Also, remaining events are |
4832 | * automatically removed on cgroup destruction but the removal is | |
4833 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4834 | */ |
b583043e | 4835 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4836 | &memory_cgrp_subsys); |
79bd9814 | 4837 | ret = -EINVAL; |
5a17f543 | 4838 | if (IS_ERR(cfile_css)) |
79bd9814 | 4839 | goto out_put_cfile; |
5a17f543 TH |
4840 | if (cfile_css != css) { |
4841 | css_put(cfile_css); | |
79bd9814 | 4842 | goto out_put_cfile; |
5a17f543 | 4843 | } |
79bd9814 | 4844 | |
451af504 | 4845 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4846 | if (ret) |
4847 | goto out_put_css; | |
4848 | ||
9965ed17 | 4849 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4850 | |
fba94807 TH |
4851 | spin_lock(&memcg->event_list_lock); |
4852 | list_add(&event->list, &memcg->event_list); | |
4853 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4854 | |
4855 | fdput(cfile); | |
4856 | fdput(efile); | |
4857 | ||
451af504 | 4858 | return nbytes; |
79bd9814 TH |
4859 | |
4860 | out_put_css: | |
b5557c4c | 4861 | css_put(css); |
79bd9814 TH |
4862 | out_put_cfile: |
4863 | fdput(cfile); | |
4864 | out_put_eventfd: | |
4865 | eventfd_ctx_put(event->eventfd); | |
4866 | out_put_efile: | |
4867 | fdput(efile); | |
4868 | out_kfree: | |
4869 | kfree(event); | |
4870 | ||
4871 | return ret; | |
4872 | } | |
4873 | ||
241994ed | 4874 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4875 | { |
0eea1030 | 4876 | .name = "usage_in_bytes", |
8c7c6e34 | 4877 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4878 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4879 | }, |
c84872e1 PE |
4880 | { |
4881 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4882 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4883 | .write = mem_cgroup_reset, |
791badbd | 4884 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4885 | }, |
8cdea7c0 | 4886 | { |
0eea1030 | 4887 | .name = "limit_in_bytes", |
8c7c6e34 | 4888 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4889 | .write = mem_cgroup_write, |
791badbd | 4890 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4891 | }, |
296c81d8 BS |
4892 | { |
4893 | .name = "soft_limit_in_bytes", | |
4894 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4895 | .write = mem_cgroup_write, |
791badbd | 4896 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4897 | }, |
8cdea7c0 BS |
4898 | { |
4899 | .name = "failcnt", | |
8c7c6e34 | 4900 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4901 | .write = mem_cgroup_reset, |
791badbd | 4902 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4903 | }, |
d2ceb9b7 KH |
4904 | { |
4905 | .name = "stat", | |
2da8ca82 | 4906 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4907 | }, |
c1e862c1 KH |
4908 | { |
4909 | .name = "force_empty", | |
6770c64e | 4910 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4911 | }, |
18f59ea7 BS |
4912 | { |
4913 | .name = "use_hierarchy", | |
4914 | .write_u64 = mem_cgroup_hierarchy_write, | |
4915 | .read_u64 = mem_cgroup_hierarchy_read, | |
4916 | }, | |
79bd9814 | 4917 | { |
3bc942f3 | 4918 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4919 | .write = memcg_write_event_control, |
7dbdb199 | 4920 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4921 | }, |
a7885eb8 KM |
4922 | { |
4923 | .name = "swappiness", | |
4924 | .read_u64 = mem_cgroup_swappiness_read, | |
4925 | .write_u64 = mem_cgroup_swappiness_write, | |
4926 | }, | |
7dc74be0 DN |
4927 | { |
4928 | .name = "move_charge_at_immigrate", | |
4929 | .read_u64 = mem_cgroup_move_charge_read, | |
4930 | .write_u64 = mem_cgroup_move_charge_write, | |
4931 | }, | |
9490ff27 KH |
4932 | { |
4933 | .name = "oom_control", | |
2da8ca82 | 4934 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4935 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4936 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4937 | }, | |
70ddf637 AV |
4938 | { |
4939 | .name = "pressure_level", | |
70ddf637 | 4940 | }, |
406eb0c9 YH |
4941 | #ifdef CONFIG_NUMA |
4942 | { | |
4943 | .name = "numa_stat", | |
2da8ca82 | 4944 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4945 | }, |
4946 | #endif | |
510fc4e1 GC |
4947 | { |
4948 | .name = "kmem.limit_in_bytes", | |
4949 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4950 | .write = mem_cgroup_write, |
791badbd | 4951 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4952 | }, |
4953 | { | |
4954 | .name = "kmem.usage_in_bytes", | |
4955 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4956 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4957 | }, |
4958 | { | |
4959 | .name = "kmem.failcnt", | |
4960 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4961 | .write = mem_cgroup_reset, |
791badbd | 4962 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4963 | }, |
4964 | { | |
4965 | .name = "kmem.max_usage_in_bytes", | |
4966 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4967 | .write = mem_cgroup_reset, |
791badbd | 4968 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4969 | }, |
a87425a3 YS |
4970 | #if defined(CONFIG_MEMCG_KMEM) && \ |
4971 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) | |
749c5415 GC |
4972 | { |
4973 | .name = "kmem.slabinfo", | |
b047501c | 4974 | .seq_show = memcg_slab_show, |
749c5415 GC |
4975 | }, |
4976 | #endif | |
d55f90bf VD |
4977 | { |
4978 | .name = "kmem.tcp.limit_in_bytes", | |
4979 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4980 | .write = mem_cgroup_write, | |
4981 | .read_u64 = mem_cgroup_read_u64, | |
4982 | }, | |
4983 | { | |
4984 | .name = "kmem.tcp.usage_in_bytes", | |
4985 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4986 | .read_u64 = mem_cgroup_read_u64, | |
4987 | }, | |
4988 | { | |
4989 | .name = "kmem.tcp.failcnt", | |
4990 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4991 | .write = mem_cgroup_reset, | |
4992 | .read_u64 = mem_cgroup_read_u64, | |
4993 | }, | |
4994 | { | |
4995 | .name = "kmem.tcp.max_usage_in_bytes", | |
4996 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4997 | .write = mem_cgroup_reset, | |
4998 | .read_u64 = mem_cgroup_read_u64, | |
4999 | }, | |
6bc10349 | 5000 | { }, /* terminate */ |
af36f906 | 5001 | }; |
8c7c6e34 | 5002 | |
73f576c0 JW |
5003 | /* |
5004 | * Private memory cgroup IDR | |
5005 | * | |
5006 | * Swap-out records and page cache shadow entries need to store memcg | |
5007 | * references in constrained space, so we maintain an ID space that is | |
5008 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
5009 | * memory-controlled cgroups to 64k. | |
5010 | * | |
b8f2935f | 5011 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
5012 | * the cgroup has been destroyed, such as page cache or reclaimable |
5013 | * slab objects, that don't need to hang on to the ID. We want to keep | |
5014 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
5015 | * relatively small ID space and prevent the creation of new cgroups | |
5016 | * even when there are much fewer than 64k cgroups - possibly none. | |
5017 | * | |
5018 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
5019 | * be freed and recycled when it's no longer needed, which is usually | |
5020 | * when the CSS is offlined. | |
5021 | * | |
5022 | * The only exception to that are records of swapped out tmpfs/shmem | |
5023 | * pages that need to be attributed to live ancestors on swapin. But | |
5024 | * those references are manageable from userspace. | |
5025 | */ | |
5026 | ||
5027 | static DEFINE_IDR(mem_cgroup_idr); | |
5028 | ||
7e97de0b KT |
5029 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
5030 | { | |
5031 | if (memcg->id.id > 0) { | |
5032 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
5033 | memcg->id.id = 0; | |
5034 | } | |
5035 | } | |
5036 | ||
c1514c0a VF |
5037 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
5038 | unsigned int n) | |
73f576c0 | 5039 | { |
1c2d479a | 5040 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5041 | } |
5042 | ||
615d66c3 | 5043 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5044 | { |
1c2d479a | 5045 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5046 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5047 | |
5048 | /* Memcg ID pins CSS */ | |
5049 | css_put(&memcg->css); | |
5050 | } | |
5051 | } | |
5052 | ||
615d66c3 VD |
5053 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5054 | { | |
5055 | mem_cgroup_id_put_many(memcg, 1); | |
5056 | } | |
5057 | ||
73f576c0 JW |
5058 | /** |
5059 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5060 | * @id: the memcg id to look up | |
5061 | * | |
5062 | * Caller must hold rcu_read_lock(). | |
5063 | */ | |
5064 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5065 | { | |
5066 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5067 | return idr_find(&mem_cgroup_idr, id); | |
5068 | } | |
5069 | ||
ef8f2327 | 5070 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5071 | { |
5072 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 5073 | int tmp = node; |
1ecaab2b KH |
5074 | /* |
5075 | * This routine is called against possible nodes. | |
5076 | * But it's BUG to call kmalloc() against offline node. | |
5077 | * | |
5078 | * TODO: this routine can waste much memory for nodes which will | |
5079 | * never be onlined. It's better to use memory hotplug callback | |
5080 | * function. | |
5081 | */ | |
41e3355d KH |
5082 | if (!node_state(node, N_NORMAL_MEMORY)) |
5083 | tmp = -1; | |
17295c88 | 5084 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5085 | if (!pn) |
5086 | return 1; | |
1ecaab2b | 5087 | |
815744d7 JW |
5088 | pn->lruvec_stat_local = alloc_percpu(struct lruvec_stat); |
5089 | if (!pn->lruvec_stat_local) { | |
5090 | kfree(pn); | |
5091 | return 1; | |
5092 | } | |
5093 | ||
a983b5eb JW |
5094 | pn->lruvec_stat_cpu = alloc_percpu(struct lruvec_stat); |
5095 | if (!pn->lruvec_stat_cpu) { | |
815744d7 | 5096 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c JW |
5097 | kfree(pn); |
5098 | return 1; | |
5099 | } | |
5100 | ||
ef8f2327 MG |
5101 | lruvec_init(&pn->lruvec); |
5102 | pn->usage_in_excess = 0; | |
5103 | pn->on_tree = false; | |
5104 | pn->memcg = memcg; | |
5105 | ||
54f72fe0 | 5106 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5107 | return 0; |
5108 | } | |
5109 | ||
ef8f2327 | 5110 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5111 | { |
00f3ca2c JW |
5112 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5113 | ||
4eaf431f MH |
5114 | if (!pn) |
5115 | return; | |
5116 | ||
a983b5eb | 5117 | free_percpu(pn->lruvec_stat_cpu); |
815744d7 | 5118 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c | 5119 | kfree(pn); |
1ecaab2b KH |
5120 | } |
5121 | ||
40e952f9 | 5122 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5123 | { |
c8b2a36f | 5124 | int node; |
59927fb9 | 5125 | |
c8b2a36f | 5126 | for_each_node(node) |
ef8f2327 | 5127 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 5128 | free_percpu(memcg->vmstats_percpu); |
815744d7 | 5129 | free_percpu(memcg->vmstats_local); |
8ff69e2c | 5130 | kfree(memcg); |
59927fb9 | 5131 | } |
3afe36b1 | 5132 | |
40e952f9 TE |
5133 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5134 | { | |
5135 | memcg_wb_domain_exit(memcg); | |
7961eee3 SB |
5136 | /* |
5137 | * Flush percpu vmstats and vmevents to guarantee the value correctness | |
5138 | * on parent's and all ancestor levels. | |
5139 | */ | |
4a87e2a2 | 5140 | memcg_flush_percpu_vmstats(memcg); |
7961eee3 | 5141 | memcg_flush_percpu_vmevents(memcg); |
40e952f9 TE |
5142 | __mem_cgroup_free(memcg); |
5143 | } | |
5144 | ||
0b8f73e1 | 5145 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 5146 | { |
d142e3e6 | 5147 | struct mem_cgroup *memcg; |
b9726c26 | 5148 | unsigned int size; |
6d12e2d8 | 5149 | int node; |
97b27821 | 5150 | int __maybe_unused i; |
11d67612 | 5151 | long error = -ENOMEM; |
8cdea7c0 | 5152 | |
0b8f73e1 JW |
5153 | size = sizeof(struct mem_cgroup); |
5154 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
5155 | ||
5156 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 5157 | if (!memcg) |
11d67612 | 5158 | return ERR_PTR(error); |
0b8f73e1 | 5159 | |
73f576c0 JW |
5160 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
5161 | 1, MEM_CGROUP_ID_MAX, | |
5162 | GFP_KERNEL); | |
11d67612 YS |
5163 | if (memcg->id.id < 0) { |
5164 | error = memcg->id.id; | |
73f576c0 | 5165 | goto fail; |
11d67612 | 5166 | } |
73f576c0 | 5167 | |
815744d7 JW |
5168 | memcg->vmstats_local = alloc_percpu(struct memcg_vmstats_percpu); |
5169 | if (!memcg->vmstats_local) | |
5170 | goto fail; | |
5171 | ||
871789d4 CD |
5172 | memcg->vmstats_percpu = alloc_percpu(struct memcg_vmstats_percpu); |
5173 | if (!memcg->vmstats_percpu) | |
0b8f73e1 | 5174 | goto fail; |
78fb7466 | 5175 | |
3ed28fa1 | 5176 | for_each_node(node) |
ef8f2327 | 5177 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5178 | goto fail; |
f64c3f54 | 5179 | |
0b8f73e1 JW |
5180 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5181 | goto fail; | |
28dbc4b6 | 5182 | |
f7e1cb6e | 5183 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5184 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5185 | mutex_init(&memcg->thresholds_lock); |
5186 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5187 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5188 | INIT_LIST_HEAD(&memcg->event_list); |
5189 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5190 | memcg->socket_pressure = jiffies; |
84c07d11 | 5191 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5192 | memcg->kmemcg_id = -1; |
bf4f0599 | 5193 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5194 | #endif |
52ebea74 TH |
5195 | #ifdef CONFIG_CGROUP_WRITEBACK |
5196 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5197 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5198 | memcg->cgwb_frn[i].done = | |
5199 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5200 | #endif |
5201 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5202 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5203 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5204 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5205 | #endif |
73f576c0 | 5206 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
5207 | return memcg; |
5208 | fail: | |
7e97de0b | 5209 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5210 | __mem_cgroup_free(memcg); |
11d67612 | 5211 | return ERR_PTR(error); |
d142e3e6 GC |
5212 | } |
5213 | ||
0b8f73e1 JW |
5214 | static struct cgroup_subsys_state * __ref |
5215 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5216 | { |
0b8f73e1 JW |
5217 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
5218 | struct mem_cgroup *memcg; | |
5219 | long error = -ENOMEM; | |
d142e3e6 | 5220 | |
0b8f73e1 | 5221 | memcg = mem_cgroup_alloc(); |
11d67612 YS |
5222 | if (IS_ERR(memcg)) |
5223 | return ERR_CAST(memcg); | |
d142e3e6 | 5224 | |
d1663a90 | 5225 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
0b8f73e1 | 5226 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5227 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 JW |
5228 | if (parent) { |
5229 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5230 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
5231 | } | |
5232 | if (parent && parent->use_hierarchy) { | |
5233 | memcg->use_hierarchy = true; | |
3e32cb2e | 5234 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5235 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
5236 | page_counter_init(&memcg->memsw, &parent->memsw); |
5237 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 5238 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5239 | } else { |
3e32cb2e | 5240 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 5241 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
5242 | page_counter_init(&memcg->memsw, NULL); |
5243 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 5244 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
5245 | /* |
5246 | * Deeper hierachy with use_hierarchy == false doesn't make | |
5247 | * much sense so let cgroup subsystem know about this | |
5248 | * unfortunate state in our controller. | |
5249 | */ | |
d142e3e6 | 5250 | if (parent != root_mem_cgroup) |
073219e9 | 5251 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 5252 | } |
d6441637 | 5253 | |
0b8f73e1 JW |
5254 | /* The following stuff does not apply to the root */ |
5255 | if (!parent) { | |
5256 | root_mem_cgroup = memcg; | |
5257 | return &memcg->css; | |
5258 | } | |
5259 | ||
b313aeee | 5260 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
5261 | if (error) |
5262 | goto fail; | |
127424c8 | 5263 | |
f7e1cb6e | 5264 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5265 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5266 | |
0b8f73e1 JW |
5267 | return &memcg->css; |
5268 | fail: | |
7e97de0b | 5269 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 5270 | mem_cgroup_free(memcg); |
11d67612 | 5271 | return ERR_PTR(error); |
0b8f73e1 JW |
5272 | } |
5273 | ||
73f576c0 | 5274 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5275 | { |
58fa2a55 VD |
5276 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5277 | ||
0a4465d3 KT |
5278 | /* |
5279 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
5280 | * by the time the maps are allocated. So, we allocate maps | |
5281 | * here, when for_each_mem_cgroup() can't skip it. | |
5282 | */ | |
5283 | if (memcg_alloc_shrinker_maps(memcg)) { | |
5284 | mem_cgroup_id_remove(memcg); | |
5285 | return -ENOMEM; | |
5286 | } | |
5287 | ||
73f576c0 | 5288 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5289 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5290 | css_get(css); |
2f7dd7a4 | 5291 | return 0; |
8cdea7c0 BS |
5292 | } |
5293 | ||
eb95419b | 5294 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5295 | { |
eb95419b | 5296 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5297 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5298 | |
5299 | /* | |
5300 | * Unregister events and notify userspace. | |
5301 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5302 | * directory to avoid race between userspace and kernelspace. | |
5303 | */ | |
fba94807 TH |
5304 | spin_lock(&memcg->event_list_lock); |
5305 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5306 | list_del_init(&event->list); |
5307 | schedule_work(&event->remove); | |
5308 | } | |
fba94807 | 5309 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5310 | |
bf8d5d52 | 5311 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5312 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5313 | |
567e9ab2 | 5314 | memcg_offline_kmem(memcg); |
52ebea74 | 5315 | wb_memcg_offline(memcg); |
73f576c0 | 5316 | |
591edfb1 RG |
5317 | drain_all_stock(memcg); |
5318 | ||
73f576c0 | 5319 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5320 | } |
5321 | ||
6df38689 VD |
5322 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5323 | { | |
5324 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5325 | ||
5326 | invalidate_reclaim_iterators(memcg); | |
5327 | } | |
5328 | ||
eb95419b | 5329 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5330 | { |
eb95419b | 5331 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5332 | int __maybe_unused i; |
c268e994 | 5333 | |
97b27821 TH |
5334 | #ifdef CONFIG_CGROUP_WRITEBACK |
5335 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5336 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5337 | #endif | |
f7e1cb6e | 5338 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5339 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5340 | |
0db15298 | 5341 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5342 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5343 | |
0b8f73e1 JW |
5344 | vmpressure_cleanup(&memcg->vmpressure); |
5345 | cancel_work_sync(&memcg->high_work); | |
5346 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 5347 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 5348 | memcg_free_kmem(memcg); |
0b8f73e1 | 5349 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5350 | } |
5351 | ||
1ced953b TH |
5352 | /** |
5353 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5354 | * @css: the target css | |
5355 | * | |
5356 | * Reset the states of the mem_cgroup associated with @css. This is | |
5357 | * invoked when the userland requests disabling on the default hierarchy | |
5358 | * but the memcg is pinned through dependency. The memcg should stop | |
5359 | * applying policies and should revert to the vanilla state as it may be | |
5360 | * made visible again. | |
5361 | * | |
5362 | * The current implementation only resets the essential configurations. | |
5363 | * This needs to be expanded to cover all the visible parts. | |
5364 | */ | |
5365 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5366 | { | |
5367 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5368 | ||
bbec2e15 RG |
5369 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5370 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
5371 | page_counter_set_max(&memcg->memsw, PAGE_COUNTER_MAX); | |
5372 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); | |
5373 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5374 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5375 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5376 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
24d404dc | 5377 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5378 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5379 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5380 | } |
5381 | ||
02491447 | 5382 | #ifdef CONFIG_MMU |
7dc74be0 | 5383 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5384 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5385 | { |
05b84301 | 5386 | int ret; |
9476db97 | 5387 | |
d0164adc MG |
5388 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5389 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5390 | if (!ret) { |
854ffa8d | 5391 | mc.precharge += count; |
854ffa8d DN |
5392 | return ret; |
5393 | } | |
9476db97 | 5394 | |
3674534b | 5395 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5396 | while (count--) { |
3674534b | 5397 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5398 | if (ret) |
38c5d72f | 5399 | return ret; |
854ffa8d | 5400 | mc.precharge++; |
9476db97 | 5401 | cond_resched(); |
854ffa8d | 5402 | } |
9476db97 | 5403 | return 0; |
4ffef5fe DN |
5404 | } |
5405 | ||
4ffef5fe DN |
5406 | union mc_target { |
5407 | struct page *page; | |
02491447 | 5408 | swp_entry_t ent; |
4ffef5fe DN |
5409 | }; |
5410 | ||
4ffef5fe | 5411 | enum mc_target_type { |
8d32ff84 | 5412 | MC_TARGET_NONE = 0, |
4ffef5fe | 5413 | MC_TARGET_PAGE, |
02491447 | 5414 | MC_TARGET_SWAP, |
c733a828 | 5415 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5416 | }; |
5417 | ||
90254a65 DN |
5418 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5419 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5420 | { |
25b2995a | 5421 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5422 | |
90254a65 DN |
5423 | if (!page || !page_mapped(page)) |
5424 | return NULL; | |
5425 | if (PageAnon(page)) { | |
1dfab5ab | 5426 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5427 | return NULL; |
1dfab5ab JW |
5428 | } else { |
5429 | if (!(mc.flags & MOVE_FILE)) | |
5430 | return NULL; | |
5431 | } | |
90254a65 DN |
5432 | if (!get_page_unless_zero(page)) |
5433 | return NULL; | |
5434 | ||
5435 | return page; | |
5436 | } | |
5437 | ||
c733a828 | 5438 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5439 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5440 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5441 | { |
90254a65 DN |
5442 | struct page *page = NULL; |
5443 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5444 | ||
1dfab5ab | 5445 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 5446 | return NULL; |
c733a828 JG |
5447 | |
5448 | /* | |
5449 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
5450 | * a device and because they are not accessible by CPU they are store | |
5451 | * as special swap entry in the CPU page table. | |
5452 | */ | |
5453 | if (is_device_private_entry(ent)) { | |
5454 | page = device_private_entry_to_page(ent); | |
5455 | /* | |
5456 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
5457 | * a refcount of 1 when free (unlike normal page) | |
5458 | */ | |
5459 | if (!page_ref_add_unless(page, 1, 1)) | |
5460 | return NULL; | |
5461 | return page; | |
5462 | } | |
5463 | ||
4b91355e KH |
5464 | /* |
5465 | * Because lookup_swap_cache() updates some statistics counter, | |
5466 | * we call find_get_page() with swapper_space directly. | |
5467 | */ | |
f6ab1f7f | 5468 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 5469 | entry->val = ent.val; |
90254a65 DN |
5470 | |
5471 | return page; | |
5472 | } | |
4b91355e KH |
5473 | #else |
5474 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5475 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5476 | { |
5477 | return NULL; | |
5478 | } | |
5479 | #endif | |
90254a65 | 5480 | |
87946a72 DN |
5481 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5482 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5483 | { | |
5484 | struct page *page = NULL; | |
87946a72 DN |
5485 | struct address_space *mapping; |
5486 | pgoff_t pgoff; | |
5487 | ||
5488 | if (!vma->vm_file) /* anonymous vma */ | |
5489 | return NULL; | |
1dfab5ab | 5490 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5491 | return NULL; |
5492 | ||
87946a72 | 5493 | mapping = vma->vm_file->f_mapping; |
0661a336 | 5494 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
5495 | |
5496 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5497 | #ifdef CONFIG_SWAP |
5498 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
5499 | if (shmem_mapping(mapping)) { |
5500 | page = find_get_entry(mapping, pgoff); | |
3159f943 | 5501 | if (xa_is_value(page)) { |
139b6a6f | 5502 | swp_entry_t swp = radix_to_swp_entry(page); |
2d1c4980 | 5503 | *entry = swp; |
f6ab1f7f HY |
5504 | page = find_get_page(swap_address_space(swp), |
5505 | swp_offset(swp)); | |
139b6a6f JW |
5506 | } |
5507 | } else | |
5508 | page = find_get_page(mapping, pgoff); | |
5509 | #else | |
5510 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 5511 | #endif |
87946a72 DN |
5512 | return page; |
5513 | } | |
5514 | ||
b1b0deab CG |
5515 | /** |
5516 | * mem_cgroup_move_account - move account of the page | |
5517 | * @page: the page | |
25843c2b | 5518 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5519 | * @from: mem_cgroup which the page is moved from. |
5520 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5521 | * | |
3ac808fd | 5522 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5523 | * |
5524 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5525 | * from old cgroup. | |
5526 | */ | |
5527 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5528 | bool compound, |
b1b0deab CG |
5529 | struct mem_cgroup *from, |
5530 | struct mem_cgroup *to) | |
5531 | { | |
ae8af438 KK |
5532 | struct lruvec *from_vec, *to_vec; |
5533 | struct pglist_data *pgdat; | |
f627c2f5 | 5534 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab CG |
5535 | int ret; |
5536 | ||
5537 | VM_BUG_ON(from == to); | |
5538 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5539 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5540 | |
5541 | /* | |
6a93ca8f | 5542 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 5543 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 5544 | */ |
f627c2f5 | 5545 | ret = -EBUSY; |
b1b0deab CG |
5546 | if (!trylock_page(page)) |
5547 | goto out; | |
5548 | ||
5549 | ret = -EINVAL; | |
5550 | if (page->mem_cgroup != from) | |
5551 | goto out_unlock; | |
5552 | ||
ae8af438 | 5553 | pgdat = page_pgdat(page); |
867e5e1d JW |
5554 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5555 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5556 | |
abb242f5 | 5557 | lock_page_memcg(page); |
b1b0deab | 5558 | |
be5d0a74 JW |
5559 | if (PageAnon(page)) { |
5560 | if (page_mapped(page)) { | |
5561 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); | |
5562 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
468c3982 JW |
5563 | if (PageTransHuge(page)) { |
5564 | __mod_lruvec_state(from_vec, NR_ANON_THPS, | |
5565 | -nr_pages); | |
5566 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
5567 | nr_pages); | |
5568 | } | |
5569 | ||
be5d0a74 JW |
5570 | } |
5571 | } else { | |
0d1c2072 JW |
5572 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
5573 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
5574 | ||
5575 | if (PageSwapBacked(page)) { | |
5576 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); | |
5577 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
5578 | } | |
5579 | ||
49e50d27 JW |
5580 | if (page_mapped(page)) { |
5581 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); | |
5582 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
5583 | } | |
b1b0deab | 5584 | |
49e50d27 JW |
5585 | if (PageDirty(page)) { |
5586 | struct address_space *mapping = page_mapping(page); | |
c4843a75 | 5587 | |
49e50d27 JW |
5588 | if (mapping_cap_account_dirty(mapping)) { |
5589 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, | |
5590 | -nr_pages); | |
5591 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
5592 | nr_pages); | |
5593 | } | |
c4843a75 GT |
5594 | } |
5595 | } | |
5596 | ||
b1b0deab | 5597 | if (PageWriteback(page)) { |
ae8af438 KK |
5598 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5599 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5600 | } |
5601 | ||
5602 | /* | |
abb242f5 JW |
5603 | * All state has been migrated, let's switch to the new memcg. |
5604 | * | |
b1b0deab | 5605 | * It is safe to change page->mem_cgroup here because the page |
abb242f5 JW |
5606 | * is referenced, charged, isolated, and locked: we can't race |
5607 | * with (un)charging, migration, LRU putback, or anything else | |
5608 | * that would rely on a stable page->mem_cgroup. | |
5609 | * | |
5610 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
5611 | * to save space. As soon as we switch page->mem_cgroup to a | |
5612 | * new memcg that isn't locked, the above state can change | |
5613 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5614 | */ |
abb242f5 | 5615 | smp_mb(); |
b1b0deab | 5616 | |
1a3e1f40 JW |
5617 | css_get(&to->css); |
5618 | css_put(&from->css); | |
5619 | ||
5620 | page->mem_cgroup = to; | |
87eaceb3 | 5621 | |
abb242f5 | 5622 | __unlock_page_memcg(from); |
b1b0deab CG |
5623 | |
5624 | ret = 0; | |
5625 | ||
5626 | local_irq_disable(); | |
3fba69a5 | 5627 | mem_cgroup_charge_statistics(to, page, nr_pages); |
b1b0deab | 5628 | memcg_check_events(to, page); |
3fba69a5 | 5629 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
b1b0deab CG |
5630 | memcg_check_events(from, page); |
5631 | local_irq_enable(); | |
5632 | out_unlock: | |
5633 | unlock_page(page); | |
5634 | out: | |
5635 | return ret; | |
5636 | } | |
5637 | ||
7cf7806c LR |
5638 | /** |
5639 | * get_mctgt_type - get target type of moving charge | |
5640 | * @vma: the vma the pte to be checked belongs | |
5641 | * @addr: the address corresponding to the pte to be checked | |
5642 | * @ptent: the pte to be checked | |
5643 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5644 | * | |
5645 | * Returns | |
5646 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5647 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5648 | * move charge. if @target is not NULL, the page is stored in target->page | |
5649 | * with extra refcnt got(Callers should handle it). | |
5650 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5651 | * target for charge migration. if @target is not NULL, the entry is stored | |
5652 | * in target->ent. | |
25b2995a CH |
5653 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
5654 | * (so ZONE_DEVICE page and thus not on the lru). | |
df6ad698 JG |
5655 | * For now we such page is charge like a regular page would be as for all |
5656 | * intent and purposes it is just special memory taking the place of a | |
5657 | * regular page. | |
c733a828 JG |
5658 | * |
5659 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5660 | * |
5661 | * Called with pte lock held. | |
5662 | */ | |
5663 | ||
8d32ff84 | 5664 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5665 | unsigned long addr, pte_t ptent, union mc_target *target) |
5666 | { | |
5667 | struct page *page = NULL; | |
8d32ff84 | 5668 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5669 | swp_entry_t ent = { .val = 0 }; |
5670 | ||
5671 | if (pte_present(ptent)) | |
5672 | page = mc_handle_present_pte(vma, addr, ptent); | |
5673 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5674 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5675 | else if (pte_none(ptent)) |
87946a72 | 5676 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5677 | |
5678 | if (!page && !ent.val) | |
8d32ff84 | 5679 | return ret; |
02491447 | 5680 | if (page) { |
02491447 | 5681 | /* |
0a31bc97 | 5682 | * Do only loose check w/o serialization. |
1306a85a | 5683 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5684 | * not under LRU exclusion. |
02491447 | 5685 | */ |
1306a85a | 5686 | if (page->mem_cgroup == mc.from) { |
02491447 | 5687 | ret = MC_TARGET_PAGE; |
25b2995a | 5688 | if (is_device_private_page(page)) |
c733a828 | 5689 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5690 | if (target) |
5691 | target->page = page; | |
5692 | } | |
5693 | if (!ret || !target) | |
5694 | put_page(page); | |
5695 | } | |
3e14a57b HY |
5696 | /* |
5697 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5698 | * But we cannot move a tail-page in a THP. | |
5699 | */ | |
5700 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5701 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5702 | ret = MC_TARGET_SWAP; |
5703 | if (target) | |
5704 | target->ent = ent; | |
4ffef5fe | 5705 | } |
4ffef5fe DN |
5706 | return ret; |
5707 | } | |
5708 | ||
12724850 NH |
5709 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5710 | /* | |
d6810d73 HY |
5711 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5712 | * not support them for now. | |
12724850 NH |
5713 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5714 | */ | |
5715 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5716 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5717 | { | |
5718 | struct page *page = NULL; | |
12724850 NH |
5719 | enum mc_target_type ret = MC_TARGET_NONE; |
5720 | ||
84c3fc4e ZY |
5721 | if (unlikely(is_swap_pmd(pmd))) { |
5722 | VM_BUG_ON(thp_migration_supported() && | |
5723 | !is_pmd_migration_entry(pmd)); | |
5724 | return ret; | |
5725 | } | |
12724850 | 5726 | page = pmd_page(pmd); |
309381fe | 5727 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5728 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5729 | return ret; |
1306a85a | 5730 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
5731 | ret = MC_TARGET_PAGE; |
5732 | if (target) { | |
5733 | get_page(page); | |
5734 | target->page = page; | |
5735 | } | |
5736 | } | |
5737 | return ret; | |
5738 | } | |
5739 | #else | |
5740 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5741 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5742 | { | |
5743 | return MC_TARGET_NONE; | |
5744 | } | |
5745 | #endif | |
5746 | ||
4ffef5fe DN |
5747 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5748 | unsigned long addr, unsigned long end, | |
5749 | struct mm_walk *walk) | |
5750 | { | |
26bcd64a | 5751 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5752 | pte_t *pte; |
5753 | spinlock_t *ptl; | |
5754 | ||
b6ec57f4 KS |
5755 | ptl = pmd_trans_huge_lock(pmd, vma); |
5756 | if (ptl) { | |
c733a828 JG |
5757 | /* |
5758 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5759 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5760 | * this might change. | |
c733a828 | 5761 | */ |
12724850 NH |
5762 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5763 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5764 | spin_unlock(ptl); |
1a5a9906 | 5765 | return 0; |
12724850 | 5766 | } |
03319327 | 5767 | |
45f83cef AA |
5768 | if (pmd_trans_unstable(pmd)) |
5769 | return 0; | |
4ffef5fe DN |
5770 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5771 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5772 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5773 | mc.precharge++; /* increment precharge temporarily */ |
5774 | pte_unmap_unlock(pte - 1, ptl); | |
5775 | cond_resched(); | |
5776 | ||
7dc74be0 DN |
5777 | return 0; |
5778 | } | |
5779 | ||
7b86ac33 CH |
5780 | static const struct mm_walk_ops precharge_walk_ops = { |
5781 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5782 | }; | |
5783 | ||
4ffef5fe DN |
5784 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5785 | { | |
5786 | unsigned long precharge; | |
4ffef5fe | 5787 | |
d8ed45c5 | 5788 | mmap_read_lock(mm); |
7b86ac33 | 5789 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
d8ed45c5 | 5790 | mmap_read_unlock(mm); |
4ffef5fe DN |
5791 | |
5792 | precharge = mc.precharge; | |
5793 | mc.precharge = 0; | |
5794 | ||
5795 | return precharge; | |
5796 | } | |
5797 | ||
4ffef5fe DN |
5798 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5799 | { | |
dfe076b0 DN |
5800 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5801 | ||
5802 | VM_BUG_ON(mc.moving_task); | |
5803 | mc.moving_task = current; | |
5804 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5805 | } |
5806 | ||
dfe076b0 DN |
5807 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5808 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5809 | { |
2bd9bb20 KH |
5810 | struct mem_cgroup *from = mc.from; |
5811 | struct mem_cgroup *to = mc.to; | |
5812 | ||
4ffef5fe | 5813 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5814 | if (mc.precharge) { |
00501b53 | 5815 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5816 | mc.precharge = 0; |
5817 | } | |
5818 | /* | |
5819 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5820 | * we must uncharge here. | |
5821 | */ | |
5822 | if (mc.moved_charge) { | |
00501b53 | 5823 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5824 | mc.moved_charge = 0; |
4ffef5fe | 5825 | } |
483c30b5 DN |
5826 | /* we must fixup refcnts and charges */ |
5827 | if (mc.moved_swap) { | |
483c30b5 | 5828 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5829 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5830 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5831 | |
615d66c3 VD |
5832 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5833 | ||
05b84301 | 5834 | /* |
3e32cb2e JW |
5835 | * we charged both to->memory and to->memsw, so we |
5836 | * should uncharge to->memory. | |
05b84301 | 5837 | */ |
ce00a967 | 5838 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5839 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5840 | ||
483c30b5 DN |
5841 | mc.moved_swap = 0; |
5842 | } | |
dfe076b0 DN |
5843 | memcg_oom_recover(from); |
5844 | memcg_oom_recover(to); | |
5845 | wake_up_all(&mc.waitq); | |
5846 | } | |
5847 | ||
5848 | static void mem_cgroup_clear_mc(void) | |
5849 | { | |
264a0ae1 TH |
5850 | struct mm_struct *mm = mc.mm; |
5851 | ||
dfe076b0 DN |
5852 | /* |
5853 | * we must clear moving_task before waking up waiters at the end of | |
5854 | * task migration. | |
5855 | */ | |
5856 | mc.moving_task = NULL; | |
5857 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5858 | spin_lock(&mc.lock); |
4ffef5fe DN |
5859 | mc.from = NULL; |
5860 | mc.to = NULL; | |
264a0ae1 | 5861 | mc.mm = NULL; |
2bd9bb20 | 5862 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5863 | |
5864 | mmput(mm); | |
4ffef5fe DN |
5865 | } |
5866 | ||
1f7dd3e5 | 5867 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5868 | { |
1f7dd3e5 | 5869 | struct cgroup_subsys_state *css; |
eed67d75 | 5870 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5871 | struct mem_cgroup *from; |
4530eddb | 5872 | struct task_struct *leader, *p; |
9f2115f9 | 5873 | struct mm_struct *mm; |
1dfab5ab | 5874 | unsigned long move_flags; |
9f2115f9 | 5875 | int ret = 0; |
7dc74be0 | 5876 | |
1f7dd3e5 TH |
5877 | /* charge immigration isn't supported on the default hierarchy */ |
5878 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5879 | return 0; |
5880 | ||
4530eddb TH |
5881 | /* |
5882 | * Multi-process migrations only happen on the default hierarchy | |
5883 | * where charge immigration is not used. Perform charge | |
5884 | * immigration if @tset contains a leader and whine if there are | |
5885 | * multiple. | |
5886 | */ | |
5887 | p = NULL; | |
1f7dd3e5 | 5888 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5889 | WARN_ON_ONCE(p); |
5890 | p = leader; | |
1f7dd3e5 | 5891 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5892 | } |
5893 | if (!p) | |
5894 | return 0; | |
5895 | ||
1f7dd3e5 TH |
5896 | /* |
5897 | * We are now commited to this value whatever it is. Changes in this | |
5898 | * tunable will only affect upcoming migrations, not the current one. | |
5899 | * So we need to save it, and keep it going. | |
5900 | */ | |
5901 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5902 | if (!move_flags) | |
5903 | return 0; | |
5904 | ||
9f2115f9 TH |
5905 | from = mem_cgroup_from_task(p); |
5906 | ||
5907 | VM_BUG_ON(from == memcg); | |
5908 | ||
5909 | mm = get_task_mm(p); | |
5910 | if (!mm) | |
5911 | return 0; | |
5912 | /* We move charges only when we move a owner of the mm */ | |
5913 | if (mm->owner == p) { | |
5914 | VM_BUG_ON(mc.from); | |
5915 | VM_BUG_ON(mc.to); | |
5916 | VM_BUG_ON(mc.precharge); | |
5917 | VM_BUG_ON(mc.moved_charge); | |
5918 | VM_BUG_ON(mc.moved_swap); | |
5919 | ||
5920 | spin_lock(&mc.lock); | |
264a0ae1 | 5921 | mc.mm = mm; |
9f2115f9 TH |
5922 | mc.from = from; |
5923 | mc.to = memcg; | |
5924 | mc.flags = move_flags; | |
5925 | spin_unlock(&mc.lock); | |
5926 | /* We set mc.moving_task later */ | |
5927 | ||
5928 | ret = mem_cgroup_precharge_mc(mm); | |
5929 | if (ret) | |
5930 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5931 | } else { |
5932 | mmput(mm); | |
7dc74be0 DN |
5933 | } |
5934 | return ret; | |
5935 | } | |
5936 | ||
1f7dd3e5 | 5937 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5938 | { |
4e2f245d JW |
5939 | if (mc.to) |
5940 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5941 | } |
5942 | ||
4ffef5fe DN |
5943 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5944 | unsigned long addr, unsigned long end, | |
5945 | struct mm_walk *walk) | |
7dc74be0 | 5946 | { |
4ffef5fe | 5947 | int ret = 0; |
26bcd64a | 5948 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5949 | pte_t *pte; |
5950 | spinlock_t *ptl; | |
12724850 NH |
5951 | enum mc_target_type target_type; |
5952 | union mc_target target; | |
5953 | struct page *page; | |
4ffef5fe | 5954 | |
b6ec57f4 KS |
5955 | ptl = pmd_trans_huge_lock(pmd, vma); |
5956 | if (ptl) { | |
62ade86a | 5957 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5958 | spin_unlock(ptl); |
12724850 NH |
5959 | return 0; |
5960 | } | |
5961 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5962 | if (target_type == MC_TARGET_PAGE) { | |
5963 | page = target.page; | |
5964 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5965 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5966 | mc.from, mc.to)) { |
12724850 NH |
5967 | mc.precharge -= HPAGE_PMD_NR; |
5968 | mc.moved_charge += HPAGE_PMD_NR; | |
5969 | } | |
5970 | putback_lru_page(page); | |
5971 | } | |
5972 | put_page(page); | |
c733a828 JG |
5973 | } else if (target_type == MC_TARGET_DEVICE) { |
5974 | page = target.page; | |
5975 | if (!mem_cgroup_move_account(page, true, | |
5976 | mc.from, mc.to)) { | |
5977 | mc.precharge -= HPAGE_PMD_NR; | |
5978 | mc.moved_charge += HPAGE_PMD_NR; | |
5979 | } | |
5980 | put_page(page); | |
12724850 | 5981 | } |
bf929152 | 5982 | spin_unlock(ptl); |
1a5a9906 | 5983 | return 0; |
12724850 NH |
5984 | } |
5985 | ||
45f83cef AA |
5986 | if (pmd_trans_unstable(pmd)) |
5987 | return 0; | |
4ffef5fe DN |
5988 | retry: |
5989 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5990 | for (; addr != end; addr += PAGE_SIZE) { | |
5991 | pte_t ptent = *(pte++); | |
c733a828 | 5992 | bool device = false; |
02491447 | 5993 | swp_entry_t ent; |
4ffef5fe DN |
5994 | |
5995 | if (!mc.precharge) | |
5996 | break; | |
5997 | ||
8d32ff84 | 5998 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
5999 | case MC_TARGET_DEVICE: |
6000 | device = true; | |
e4a9bc58 | 6001 | fallthrough; |
4ffef5fe DN |
6002 | case MC_TARGET_PAGE: |
6003 | page = target.page; | |
53f9263b KS |
6004 | /* |
6005 | * We can have a part of the split pmd here. Moving it | |
6006 | * can be done but it would be too convoluted so simply | |
6007 | * ignore such a partial THP and keep it in original | |
6008 | * memcg. There should be somebody mapping the head. | |
6009 | */ | |
6010 | if (PageTransCompound(page)) | |
6011 | goto put; | |
c733a828 | 6012 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 6013 | goto put; |
f627c2f5 KS |
6014 | if (!mem_cgroup_move_account(page, false, |
6015 | mc.from, mc.to)) { | |
4ffef5fe | 6016 | mc.precharge--; |
854ffa8d DN |
6017 | /* we uncharge from mc.from later. */ |
6018 | mc.moved_charge++; | |
4ffef5fe | 6019 | } |
c733a828 JG |
6020 | if (!device) |
6021 | putback_lru_page(page); | |
8d32ff84 | 6022 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
6023 | put_page(page); |
6024 | break; | |
02491447 DN |
6025 | case MC_TARGET_SWAP: |
6026 | ent = target.ent; | |
e91cbb42 | 6027 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 6028 | mc.precharge--; |
8d22a935 HD |
6029 | mem_cgroup_id_get_many(mc.to, 1); |
6030 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
6031 | mc.moved_swap++; |
6032 | } | |
02491447 | 6033 | break; |
4ffef5fe DN |
6034 | default: |
6035 | break; | |
6036 | } | |
6037 | } | |
6038 | pte_unmap_unlock(pte - 1, ptl); | |
6039 | cond_resched(); | |
6040 | ||
6041 | if (addr != end) { | |
6042 | /* | |
6043 | * We have consumed all precharges we got in can_attach(). | |
6044 | * We try charge one by one, but don't do any additional | |
6045 | * charges to mc.to if we have failed in charge once in attach() | |
6046 | * phase. | |
6047 | */ | |
854ffa8d | 6048 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
6049 | if (!ret) |
6050 | goto retry; | |
6051 | } | |
6052 | ||
6053 | return ret; | |
6054 | } | |
6055 | ||
7b86ac33 CH |
6056 | static const struct mm_walk_ops charge_walk_ops = { |
6057 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6058 | }; | |
6059 | ||
264a0ae1 | 6060 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6061 | { |
4ffef5fe | 6062 | lru_add_drain_all(); |
312722cb | 6063 | /* |
81f8c3a4 JW |
6064 | * Signal lock_page_memcg() to take the memcg's move_lock |
6065 | * while we're moving its pages to another memcg. Then wait | |
6066 | * for already started RCU-only updates to finish. | |
312722cb JW |
6067 | */ |
6068 | atomic_inc(&mc.from->moving_account); | |
6069 | synchronize_rcu(); | |
dfe076b0 | 6070 | retry: |
d8ed45c5 | 6071 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 6072 | /* |
c1e8d7c6 | 6073 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
6074 | * waitq. So we cancel all extra charges, wake up all waiters, |
6075 | * and retry. Because we cancel precharges, we might not be able | |
6076 | * to move enough charges, but moving charge is a best-effort | |
6077 | * feature anyway, so it wouldn't be a big problem. | |
6078 | */ | |
6079 | __mem_cgroup_clear_mc(); | |
6080 | cond_resched(); | |
6081 | goto retry; | |
6082 | } | |
26bcd64a NH |
6083 | /* |
6084 | * When we have consumed all precharges and failed in doing | |
6085 | * additional charge, the page walk just aborts. | |
6086 | */ | |
7b86ac33 CH |
6087 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
6088 | NULL); | |
0247f3f4 | 6089 | |
d8ed45c5 | 6090 | mmap_read_unlock(mc.mm); |
312722cb | 6091 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6092 | } |
6093 | ||
264a0ae1 | 6094 | static void mem_cgroup_move_task(void) |
67e465a7 | 6095 | { |
264a0ae1 TH |
6096 | if (mc.to) { |
6097 | mem_cgroup_move_charge(); | |
a433658c | 6098 | mem_cgroup_clear_mc(); |
264a0ae1 | 6099 | } |
67e465a7 | 6100 | } |
5cfb80a7 | 6101 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6102 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6103 | { |
6104 | return 0; | |
6105 | } | |
1f7dd3e5 | 6106 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6107 | { |
6108 | } | |
264a0ae1 | 6109 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6110 | { |
6111 | } | |
6112 | #endif | |
67e465a7 | 6113 | |
f00baae7 TH |
6114 | /* |
6115 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
6116 | * to verify whether we're attached to the default hierarchy on each mount |
6117 | * attempt. | |
f00baae7 | 6118 | */ |
eb95419b | 6119 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
6120 | { |
6121 | /* | |
aa6ec29b | 6122 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
6123 | * guarantees that @root doesn't have any children, so turning it |
6124 | * on for the root memcg is enough. | |
6125 | */ | |
9e10a130 | 6126 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
6127 | root_mem_cgroup->use_hierarchy = true; |
6128 | else | |
6129 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
6130 | } |
6131 | ||
677dc973 CD |
6132 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6133 | { | |
6134 | if (value == PAGE_COUNTER_MAX) | |
6135 | seq_puts(m, "max\n"); | |
6136 | else | |
6137 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6138 | ||
6139 | return 0; | |
6140 | } | |
6141 | ||
241994ed JW |
6142 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6143 | struct cftype *cft) | |
6144 | { | |
f5fc3c5d JW |
6145 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6146 | ||
6147 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6148 | } |
6149 | ||
bf8d5d52 RG |
6150 | static int memory_min_show(struct seq_file *m, void *v) |
6151 | { | |
677dc973 CD |
6152 | return seq_puts_memcg_tunable(m, |
6153 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6154 | } |
6155 | ||
6156 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6157 | char *buf, size_t nbytes, loff_t off) | |
6158 | { | |
6159 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6160 | unsigned long min; | |
6161 | int err; | |
6162 | ||
6163 | buf = strstrip(buf); | |
6164 | err = page_counter_memparse(buf, "max", &min); | |
6165 | if (err) | |
6166 | return err; | |
6167 | ||
6168 | page_counter_set_min(&memcg->memory, min); | |
6169 | ||
6170 | return nbytes; | |
6171 | } | |
6172 | ||
241994ed JW |
6173 | static int memory_low_show(struct seq_file *m, void *v) |
6174 | { | |
677dc973 CD |
6175 | return seq_puts_memcg_tunable(m, |
6176 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6177 | } |
6178 | ||
6179 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6180 | char *buf, size_t nbytes, loff_t off) | |
6181 | { | |
6182 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6183 | unsigned long low; | |
6184 | int err; | |
6185 | ||
6186 | buf = strstrip(buf); | |
d2973697 | 6187 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6188 | if (err) |
6189 | return err; | |
6190 | ||
23067153 | 6191 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6192 | |
6193 | return nbytes; | |
6194 | } | |
6195 | ||
6196 | static int memory_high_show(struct seq_file *m, void *v) | |
6197 | { | |
d1663a90 JK |
6198 | return seq_puts_memcg_tunable(m, |
6199 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6200 | } |
6201 | ||
6202 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6203 | char *buf, size_t nbytes, loff_t off) | |
6204 | { | |
6205 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
8c8c383c JW |
6206 | unsigned int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6207 | bool drained = false; | |
241994ed JW |
6208 | unsigned long high; |
6209 | int err; | |
6210 | ||
6211 | buf = strstrip(buf); | |
d2973697 | 6212 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6213 | if (err) |
6214 | return err; | |
6215 | ||
8c8c383c JW |
6216 | for (;;) { |
6217 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6218 | unsigned long reclaimed; | |
6219 | ||
6220 | if (nr_pages <= high) | |
6221 | break; | |
6222 | ||
6223 | if (signal_pending(current)) | |
6224 | break; | |
6225 | ||
6226 | if (!drained) { | |
6227 | drain_all_stock(memcg); | |
6228 | drained = true; | |
6229 | continue; | |
6230 | } | |
6231 | ||
6232 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
6233 | GFP_KERNEL, true); | |
6234 | ||
6235 | if (!reclaimed && !nr_retries--) | |
6236 | break; | |
6237 | } | |
588083bb | 6238 | |
536d3bf2 RG |
6239 | page_counter_set_high(&memcg->memory, high); |
6240 | ||
241994ed JW |
6241 | return nbytes; |
6242 | } | |
6243 | ||
6244 | static int memory_max_show(struct seq_file *m, void *v) | |
6245 | { | |
677dc973 CD |
6246 | return seq_puts_memcg_tunable(m, |
6247 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6248 | } |
6249 | ||
6250 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6251 | char *buf, size_t nbytes, loff_t off) | |
6252 | { | |
6253 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
6254 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
6255 | bool drained = false; | |
241994ed JW |
6256 | unsigned long max; |
6257 | int err; | |
6258 | ||
6259 | buf = strstrip(buf); | |
d2973697 | 6260 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6261 | if (err) |
6262 | return err; | |
6263 | ||
bbec2e15 | 6264 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6265 | |
6266 | for (;;) { | |
6267 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6268 | ||
6269 | if (nr_pages <= max) | |
6270 | break; | |
6271 | ||
7249c9f0 | 6272 | if (signal_pending(current)) |
b6e6edcf | 6273 | break; |
b6e6edcf JW |
6274 | |
6275 | if (!drained) { | |
6276 | drain_all_stock(memcg); | |
6277 | drained = true; | |
6278 | continue; | |
6279 | } | |
6280 | ||
6281 | if (nr_reclaims) { | |
6282 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
6283 | GFP_KERNEL, true)) | |
6284 | nr_reclaims--; | |
6285 | continue; | |
6286 | } | |
6287 | ||
e27be240 | 6288 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6289 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6290 | break; | |
6291 | } | |
241994ed | 6292 | |
2529bb3a | 6293 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6294 | return nbytes; |
6295 | } | |
6296 | ||
1e577f97 SB |
6297 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6298 | { | |
6299 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6300 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6301 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6302 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6303 | seq_printf(m, "oom_kill %lu\n", | |
6304 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
6305 | } | |
6306 | ||
241994ed JW |
6307 | static int memory_events_show(struct seq_file *m, void *v) |
6308 | { | |
aa9694bb | 6309 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6310 | |
1e577f97 SB |
6311 | __memory_events_show(m, memcg->memory_events); |
6312 | return 0; | |
6313 | } | |
6314 | ||
6315 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6316 | { | |
6317 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6318 | |
1e577f97 | 6319 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6320 | return 0; |
6321 | } | |
6322 | ||
587d9f72 JW |
6323 | static int memory_stat_show(struct seq_file *m, void *v) |
6324 | { | |
aa9694bb | 6325 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 6326 | char *buf; |
1ff9e6e1 | 6327 | |
c8713d0b JW |
6328 | buf = memory_stat_format(memcg); |
6329 | if (!buf) | |
6330 | return -ENOMEM; | |
6331 | seq_puts(m, buf); | |
6332 | kfree(buf); | |
587d9f72 JW |
6333 | return 0; |
6334 | } | |
6335 | ||
3d8b38eb RG |
6336 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6337 | { | |
aa9694bb | 6338 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6339 | |
6340 | seq_printf(m, "%d\n", memcg->oom_group); | |
6341 | ||
6342 | return 0; | |
6343 | } | |
6344 | ||
6345 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6346 | char *buf, size_t nbytes, loff_t off) | |
6347 | { | |
6348 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6349 | int ret, oom_group; | |
6350 | ||
6351 | buf = strstrip(buf); | |
6352 | if (!buf) | |
6353 | return -EINVAL; | |
6354 | ||
6355 | ret = kstrtoint(buf, 0, &oom_group); | |
6356 | if (ret) | |
6357 | return ret; | |
6358 | ||
6359 | if (oom_group != 0 && oom_group != 1) | |
6360 | return -EINVAL; | |
6361 | ||
6362 | memcg->oom_group = oom_group; | |
6363 | ||
6364 | return nbytes; | |
6365 | } | |
6366 | ||
241994ed JW |
6367 | static struct cftype memory_files[] = { |
6368 | { | |
6369 | .name = "current", | |
f5fc3c5d | 6370 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6371 | .read_u64 = memory_current_read, |
6372 | }, | |
bf8d5d52 RG |
6373 | { |
6374 | .name = "min", | |
6375 | .flags = CFTYPE_NOT_ON_ROOT, | |
6376 | .seq_show = memory_min_show, | |
6377 | .write = memory_min_write, | |
6378 | }, | |
241994ed JW |
6379 | { |
6380 | .name = "low", | |
6381 | .flags = CFTYPE_NOT_ON_ROOT, | |
6382 | .seq_show = memory_low_show, | |
6383 | .write = memory_low_write, | |
6384 | }, | |
6385 | { | |
6386 | .name = "high", | |
6387 | .flags = CFTYPE_NOT_ON_ROOT, | |
6388 | .seq_show = memory_high_show, | |
6389 | .write = memory_high_write, | |
6390 | }, | |
6391 | { | |
6392 | .name = "max", | |
6393 | .flags = CFTYPE_NOT_ON_ROOT, | |
6394 | .seq_show = memory_max_show, | |
6395 | .write = memory_max_write, | |
6396 | }, | |
6397 | { | |
6398 | .name = "events", | |
6399 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6400 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6401 | .seq_show = memory_events_show, |
6402 | }, | |
1e577f97 SB |
6403 | { |
6404 | .name = "events.local", | |
6405 | .flags = CFTYPE_NOT_ON_ROOT, | |
6406 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6407 | .seq_show = memory_events_local_show, | |
6408 | }, | |
587d9f72 JW |
6409 | { |
6410 | .name = "stat", | |
587d9f72 JW |
6411 | .seq_show = memory_stat_show, |
6412 | }, | |
3d8b38eb RG |
6413 | { |
6414 | .name = "oom.group", | |
6415 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6416 | .seq_show = memory_oom_group_show, | |
6417 | .write = memory_oom_group_write, | |
6418 | }, | |
241994ed JW |
6419 | { } /* terminate */ |
6420 | }; | |
6421 | ||
073219e9 | 6422 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6423 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6424 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6425 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6426 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6427 | .css_free = mem_cgroup_css_free, |
1ced953b | 6428 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6429 | .can_attach = mem_cgroup_can_attach, |
6430 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 6431 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 6432 | .bind = mem_cgroup_bind, |
241994ed JW |
6433 | .dfl_cftypes = memory_files, |
6434 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6435 | .early_init = 0, |
8cdea7c0 | 6436 | }; |
c077719b | 6437 | |
bc50bcc6 JW |
6438 | /* |
6439 | * This function calculates an individual cgroup's effective | |
6440 | * protection which is derived from its own memory.min/low, its | |
6441 | * parent's and siblings' settings, as well as the actual memory | |
6442 | * distribution in the tree. | |
6443 | * | |
6444 | * The following rules apply to the effective protection values: | |
6445 | * | |
6446 | * 1. At the first level of reclaim, effective protection is equal to | |
6447 | * the declared protection in memory.min and memory.low. | |
6448 | * | |
6449 | * 2. To enable safe delegation of the protection configuration, at | |
6450 | * subsequent levels the effective protection is capped to the | |
6451 | * parent's effective protection. | |
6452 | * | |
6453 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6454 | * user is allowed to overcommit the declared protection at a given | |
6455 | * level. If that is the case, the parent's effective protection is | |
6456 | * distributed to the children in proportion to how much protection | |
6457 | * they have declared and how much of it they are utilizing. | |
6458 | * | |
6459 | * This makes distribution proportional, but also work-conserving: | |
6460 | * if one cgroup claims much more protection than it uses memory, | |
6461 | * the unused remainder is available to its siblings. | |
6462 | * | |
6463 | * 4. Conversely, when the declared protection is undercommitted at a | |
6464 | * given level, the distribution of the larger parental protection | |
6465 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6466 | * is capped to its own memory.min/low setting. | |
6467 | * | |
8a931f80 JW |
6468 | * 5. However, to allow protecting recursive subtrees from each other |
6469 | * without having to declare each individual cgroup's fixed share | |
6470 | * of the ancestor's claim to protection, any unutilized - | |
6471 | * "floating" - protection from up the tree is distributed in | |
6472 | * proportion to each cgroup's *usage*. This makes the protection | |
6473 | * neutral wrt sibling cgroups and lets them compete freely over | |
6474 | * the shared parental protection budget, but it protects the | |
6475 | * subtree as a whole from neighboring subtrees. | |
6476 | * | |
6477 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6478 | * against immediate siblings whereas 5. is about protecting against | |
6479 | * neighboring subtrees. | |
bc50bcc6 JW |
6480 | */ |
6481 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6482 | unsigned long parent_usage, |
bc50bcc6 JW |
6483 | unsigned long setting, |
6484 | unsigned long parent_effective, | |
6485 | unsigned long siblings_protected) | |
6486 | { | |
6487 | unsigned long protected; | |
8a931f80 | 6488 | unsigned long ep; |
bc50bcc6 JW |
6489 | |
6490 | protected = min(usage, setting); | |
6491 | /* | |
6492 | * If all cgroups at this level combined claim and use more | |
6493 | * protection then what the parent affords them, distribute | |
6494 | * shares in proportion to utilization. | |
6495 | * | |
6496 | * We are using actual utilization rather than the statically | |
6497 | * claimed protection in order to be work-conserving: claimed | |
6498 | * but unused protection is available to siblings that would | |
6499 | * otherwise get a smaller chunk than what they claimed. | |
6500 | */ | |
6501 | if (siblings_protected > parent_effective) | |
6502 | return protected * parent_effective / siblings_protected; | |
6503 | ||
6504 | /* | |
6505 | * Ok, utilized protection of all children is within what the | |
6506 | * parent affords them, so we know whatever this child claims | |
6507 | * and utilizes is effectively protected. | |
6508 | * | |
6509 | * If there is unprotected usage beyond this value, reclaim | |
6510 | * will apply pressure in proportion to that amount. | |
6511 | * | |
6512 | * If there is unutilized protection, the cgroup will be fully | |
6513 | * shielded from reclaim, but we do return a smaller value for | |
6514 | * protection than what the group could enjoy in theory. This | |
6515 | * is okay. With the overcommit distribution above, effective | |
6516 | * protection is always dependent on how memory is actually | |
6517 | * consumed among the siblings anyway. | |
6518 | */ | |
8a931f80 JW |
6519 | ep = protected; |
6520 | ||
6521 | /* | |
6522 | * If the children aren't claiming (all of) the protection | |
6523 | * afforded to them by the parent, distribute the remainder in | |
6524 | * proportion to the (unprotected) memory of each cgroup. That | |
6525 | * way, cgroups that aren't explicitly prioritized wrt each | |
6526 | * other compete freely over the allowance, but they are | |
6527 | * collectively protected from neighboring trees. | |
6528 | * | |
6529 | * We're using unprotected memory for the weight so that if | |
6530 | * some cgroups DO claim explicit protection, we don't protect | |
6531 | * the same bytes twice. | |
cd324edc JW |
6532 | * |
6533 | * Check both usage and parent_usage against the respective | |
6534 | * protected values. One should imply the other, but they | |
6535 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
6536 | */ |
6537 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6538 | return ep; | |
cd324edc JW |
6539 | if (parent_effective > siblings_protected && |
6540 | parent_usage > siblings_protected && | |
6541 | usage > protected) { | |
8a931f80 JW |
6542 | unsigned long unclaimed; |
6543 | ||
6544 | unclaimed = parent_effective - siblings_protected; | |
6545 | unclaimed *= usage - protected; | |
6546 | unclaimed /= parent_usage - siblings_protected; | |
6547 | ||
6548 | ep += unclaimed; | |
6549 | } | |
6550 | ||
6551 | return ep; | |
bc50bcc6 JW |
6552 | } |
6553 | ||
241994ed | 6554 | /** |
bf8d5d52 | 6555 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 6556 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6557 | * @memcg: the memory cgroup to check |
6558 | * | |
23067153 RG |
6559 | * WARNING: This function is not stateless! It can only be used as part |
6560 | * of a top-down tree iteration, not for isolated queries. | |
34c81057 | 6561 | * |
bf8d5d52 RG |
6562 | * Returns one of the following: |
6563 | * MEMCG_PROT_NONE: cgroup memory is not protected | |
6564 | * MEMCG_PROT_LOW: cgroup memory is protected as long there is | |
6565 | * an unprotected supply of reclaimable memory from other cgroups. | |
6566 | * MEMCG_PROT_MIN: cgroup memory is protected | |
241994ed | 6567 | */ |
bf8d5d52 RG |
6568 | enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root, |
6569 | struct mem_cgroup *memcg) | |
241994ed | 6570 | { |
8a931f80 | 6571 | unsigned long usage, parent_usage; |
23067153 RG |
6572 | struct mem_cgroup *parent; |
6573 | ||
241994ed | 6574 | if (mem_cgroup_disabled()) |
bf8d5d52 | 6575 | return MEMCG_PROT_NONE; |
241994ed | 6576 | |
34c81057 SC |
6577 | if (!root) |
6578 | root = root_mem_cgroup; | |
6579 | if (memcg == root) | |
bf8d5d52 | 6580 | return MEMCG_PROT_NONE; |
241994ed | 6581 | |
23067153 | 6582 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 RG |
6583 | if (!usage) |
6584 | return MEMCG_PROT_NONE; | |
6585 | ||
bf8d5d52 | 6586 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
6587 | /* No parent means a non-hierarchical mode on v1 memcg */ |
6588 | if (!parent) | |
6589 | return MEMCG_PROT_NONE; | |
6590 | ||
bc50bcc6 | 6591 | if (parent == root) { |
c3d53200 | 6592 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 6593 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
bc50bcc6 | 6594 | goto out; |
bf8d5d52 RG |
6595 | } |
6596 | ||
8a931f80 JW |
6597 | parent_usage = page_counter_read(&parent->memory); |
6598 | ||
b3a7822e | 6599 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6600 | READ_ONCE(memcg->memory.min), |
6601 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6602 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6603 | |
b3a7822e | 6604 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
6605 | READ_ONCE(memcg->memory.low), |
6606 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 6607 | atomic_long_read(&parent->memory.children_low_usage))); |
23067153 | 6608 | |
bc50bcc6 JW |
6609 | out: |
6610 | if (usage <= memcg->memory.emin) | |
bf8d5d52 | 6611 | return MEMCG_PROT_MIN; |
bc50bcc6 | 6612 | else if (usage <= memcg->memory.elow) |
bf8d5d52 RG |
6613 | return MEMCG_PROT_LOW; |
6614 | else | |
6615 | return MEMCG_PROT_NONE; | |
241994ed JW |
6616 | } |
6617 | ||
00501b53 | 6618 | /** |
f0e45fb4 | 6619 | * mem_cgroup_charge - charge a newly allocated page to a cgroup |
00501b53 JW |
6620 | * @page: page to charge |
6621 | * @mm: mm context of the victim | |
6622 | * @gfp_mask: reclaim mode | |
00501b53 JW |
6623 | * |
6624 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6625 | * pages according to @gfp_mask if necessary. | |
6626 | * | |
f0e45fb4 | 6627 | * Returns 0 on success. Otherwise, an error code is returned. |
00501b53 | 6628 | */ |
d9eb1ea2 | 6629 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
00501b53 | 6630 | { |
3fba69a5 | 6631 | unsigned int nr_pages = hpage_nr_pages(page); |
00501b53 | 6632 | struct mem_cgroup *memcg = NULL; |
00501b53 JW |
6633 | int ret = 0; |
6634 | ||
6635 | if (mem_cgroup_disabled()) | |
6636 | goto out; | |
6637 | ||
6638 | if (PageSwapCache(page)) { | |
2d1c4980 JW |
6639 | swp_entry_t ent = { .val = page_private(page), }; |
6640 | unsigned short id; | |
6641 | ||
00501b53 JW |
6642 | /* |
6643 | * Every swap fault against a single page tries to charge the | |
6644 | * page, bail as early as possible. shmem_unuse() encounters | |
eccb52e7 JW |
6645 | * already charged pages, too. page->mem_cgroup is protected |
6646 | * by the page lock, which serializes swap cache removal, which | |
00501b53 JW |
6647 | * in turn serializes uncharging. |
6648 | */ | |
e993d905 | 6649 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 6650 | if (compound_head(page)->mem_cgroup) |
00501b53 | 6651 | goto out; |
e993d905 | 6652 | |
2d1c4980 JW |
6653 | id = lookup_swap_cgroup_id(ent); |
6654 | rcu_read_lock(); | |
6655 | memcg = mem_cgroup_from_id(id); | |
6656 | if (memcg && !css_tryget_online(&memcg->css)) | |
6657 | memcg = NULL; | |
6658 | rcu_read_unlock(); | |
00501b53 JW |
6659 | } |
6660 | ||
00501b53 JW |
6661 | if (!memcg) |
6662 | memcg = get_mem_cgroup_from_mm(mm); | |
6663 | ||
6664 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
f0e45fb4 JW |
6665 | if (ret) |
6666 | goto out_put; | |
00501b53 | 6667 | |
1a3e1f40 | 6668 | css_get(&memcg->css); |
d9eb1ea2 | 6669 | commit_charge(page, memcg); |
6abb5a86 | 6670 | |
6abb5a86 | 6671 | local_irq_disable(); |
3fba69a5 | 6672 | mem_cgroup_charge_statistics(memcg, page, nr_pages); |
6abb5a86 JW |
6673 | memcg_check_events(memcg, page); |
6674 | local_irq_enable(); | |
00501b53 | 6675 | |
2d1c4980 | 6676 | if (PageSwapCache(page)) { |
00501b53 JW |
6677 | swp_entry_t entry = { .val = page_private(page) }; |
6678 | /* | |
6679 | * The swap entry might not get freed for a long time, | |
6680 | * let's not wait for it. The page already received a | |
6681 | * memory+swap charge, drop the swap entry duplicate. | |
6682 | */ | |
38d8b4e6 | 6683 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 | 6684 | } |
00501b53 | 6685 | |
f0e45fb4 JW |
6686 | out_put: |
6687 | css_put(&memcg->css); | |
6688 | out: | |
6689 | return ret; | |
3fea5a49 JW |
6690 | } |
6691 | ||
a9d5adee JG |
6692 | struct uncharge_gather { |
6693 | struct mem_cgroup *memcg; | |
9f762dbe | 6694 | unsigned long nr_pages; |
a9d5adee | 6695 | unsigned long pgpgout; |
a9d5adee | 6696 | unsigned long nr_kmem; |
a9d5adee JG |
6697 | struct page *dummy_page; |
6698 | }; | |
6699 | ||
6700 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6701 | { |
a9d5adee JG |
6702 | memset(ug, 0, sizeof(*ug)); |
6703 | } | |
6704 | ||
6705 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6706 | { | |
747db954 JW |
6707 | unsigned long flags; |
6708 | ||
a9d5adee | 6709 | if (!mem_cgroup_is_root(ug->memcg)) { |
9f762dbe | 6710 | page_counter_uncharge(&ug->memcg->memory, ug->nr_pages); |
7941d214 | 6711 | if (do_memsw_account()) |
9f762dbe | 6712 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_pages); |
a9d5adee JG |
6713 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) |
6714 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6715 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6716 | } |
747db954 JW |
6717 | |
6718 | local_irq_save(flags); | |
c9019e9b | 6719 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
9f762dbe | 6720 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_pages); |
a9d5adee | 6721 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6722 | local_irq_restore(flags); |
a9d5adee JG |
6723 | } |
6724 | ||
6725 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6726 | { | |
9f762dbe JW |
6727 | unsigned long nr_pages; |
6728 | ||
a9d5adee | 6729 | VM_BUG_ON_PAGE(PageLRU(page), page); |
a9d5adee JG |
6730 | |
6731 | if (!page->mem_cgroup) | |
6732 | return; | |
6733 | ||
6734 | /* | |
6735 | * Nobody should be changing or seriously looking at | |
6736 | * page->mem_cgroup at this point, we have fully | |
6737 | * exclusive access to the page. | |
6738 | */ | |
6739 | ||
6740 | if (ug->memcg != page->mem_cgroup) { | |
6741 | if (ug->memcg) { | |
6742 | uncharge_batch(ug); | |
6743 | uncharge_gather_clear(ug); | |
6744 | } | |
6745 | ug->memcg = page->mem_cgroup; | |
6746 | } | |
6747 | ||
9f762dbe JW |
6748 | nr_pages = compound_nr(page); |
6749 | ug->nr_pages += nr_pages; | |
a9d5adee | 6750 | |
9f762dbe | 6751 | if (!PageKmemcg(page)) { |
a9d5adee JG |
6752 | ug->pgpgout++; |
6753 | } else { | |
9f762dbe | 6754 | ug->nr_kmem += nr_pages; |
a9d5adee JG |
6755 | __ClearPageKmemcg(page); |
6756 | } | |
6757 | ||
6758 | ug->dummy_page = page; | |
6759 | page->mem_cgroup = NULL; | |
1a3e1f40 | 6760 | css_put(&ug->memcg->css); |
747db954 JW |
6761 | } |
6762 | ||
6763 | static void uncharge_list(struct list_head *page_list) | |
6764 | { | |
a9d5adee | 6765 | struct uncharge_gather ug; |
747db954 | 6766 | struct list_head *next; |
a9d5adee JG |
6767 | |
6768 | uncharge_gather_clear(&ug); | |
747db954 | 6769 | |
8b592656 JW |
6770 | /* |
6771 | * Note that the list can be a single page->lru; hence the | |
6772 | * do-while loop instead of a simple list_for_each_entry(). | |
6773 | */ | |
747db954 JW |
6774 | next = page_list->next; |
6775 | do { | |
a9d5adee JG |
6776 | struct page *page; |
6777 | ||
747db954 JW |
6778 | page = list_entry(next, struct page, lru); |
6779 | next = page->lru.next; | |
6780 | ||
a9d5adee | 6781 | uncharge_page(page, &ug); |
747db954 JW |
6782 | } while (next != page_list); |
6783 | ||
a9d5adee JG |
6784 | if (ug.memcg) |
6785 | uncharge_batch(&ug); | |
747db954 JW |
6786 | } |
6787 | ||
0a31bc97 JW |
6788 | /** |
6789 | * mem_cgroup_uncharge - uncharge a page | |
6790 | * @page: page to uncharge | |
6791 | * | |
f0e45fb4 | 6792 | * Uncharge a page previously charged with mem_cgroup_charge(). |
0a31bc97 JW |
6793 | */ |
6794 | void mem_cgroup_uncharge(struct page *page) | |
6795 | { | |
a9d5adee JG |
6796 | struct uncharge_gather ug; |
6797 | ||
0a31bc97 JW |
6798 | if (mem_cgroup_disabled()) |
6799 | return; | |
6800 | ||
747db954 | 6801 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6802 | if (!page->mem_cgroup) |
0a31bc97 JW |
6803 | return; |
6804 | ||
a9d5adee JG |
6805 | uncharge_gather_clear(&ug); |
6806 | uncharge_page(page, &ug); | |
6807 | uncharge_batch(&ug); | |
747db954 | 6808 | } |
0a31bc97 | 6809 | |
747db954 JW |
6810 | /** |
6811 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6812 | * @page_list: list of pages to uncharge | |
6813 | * | |
6814 | * Uncharge a list of pages previously charged with | |
f0e45fb4 | 6815 | * mem_cgroup_charge(). |
747db954 JW |
6816 | */ |
6817 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6818 | { | |
6819 | if (mem_cgroup_disabled()) | |
6820 | return; | |
0a31bc97 | 6821 | |
747db954 JW |
6822 | if (!list_empty(page_list)) |
6823 | uncharge_list(page_list); | |
0a31bc97 JW |
6824 | } |
6825 | ||
6826 | /** | |
6a93ca8f JW |
6827 | * mem_cgroup_migrate - charge a page's replacement |
6828 | * @oldpage: currently circulating page | |
6829 | * @newpage: replacement page | |
0a31bc97 | 6830 | * |
6a93ca8f JW |
6831 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6832 | * be uncharged upon free. | |
0a31bc97 JW |
6833 | * |
6834 | * Both pages must be locked, @newpage->mapping must be set up. | |
6835 | */ | |
6a93ca8f | 6836 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6837 | { |
29833315 | 6838 | struct mem_cgroup *memcg; |
44b7a8d3 | 6839 | unsigned int nr_pages; |
d93c4130 | 6840 | unsigned long flags; |
0a31bc97 JW |
6841 | |
6842 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6843 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6844 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6845 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6846 | newpage); | |
0a31bc97 JW |
6847 | |
6848 | if (mem_cgroup_disabled()) | |
6849 | return; | |
6850 | ||
6851 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6852 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6853 | return; |
6854 | ||
45637bab | 6855 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6856 | memcg = oldpage->mem_cgroup; |
29833315 | 6857 | if (!memcg) |
0a31bc97 JW |
6858 | return; |
6859 | ||
44b7a8d3 | 6860 | /* Force-charge the new page. The old one will be freed soon */ |
92855270 | 6861 | nr_pages = hpage_nr_pages(newpage); |
44b7a8d3 JW |
6862 | |
6863 | page_counter_charge(&memcg->memory, nr_pages); | |
6864 | if (do_memsw_account()) | |
6865 | page_counter_charge(&memcg->memsw, nr_pages); | |
0a31bc97 | 6866 | |
1a3e1f40 | 6867 | css_get(&memcg->css); |
d9eb1ea2 | 6868 | commit_charge(newpage, memcg); |
44b7a8d3 | 6869 | |
d93c4130 | 6870 | local_irq_save(flags); |
3fba69a5 | 6871 | mem_cgroup_charge_statistics(memcg, newpage, nr_pages); |
44b7a8d3 | 6872 | memcg_check_events(memcg, newpage); |
d93c4130 | 6873 | local_irq_restore(flags); |
0a31bc97 JW |
6874 | } |
6875 | ||
ef12947c | 6876 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6877 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6878 | ||
2d758073 | 6879 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6880 | { |
6881 | struct mem_cgroup *memcg; | |
6882 | ||
2d758073 JW |
6883 | if (!mem_cgroup_sockets_enabled) |
6884 | return; | |
6885 | ||
e876ecc6 SB |
6886 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
6887 | if (in_interrupt()) | |
6888 | return; | |
6889 | ||
11092087 JW |
6890 | rcu_read_lock(); |
6891 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6892 | if (memcg == root_mem_cgroup) |
6893 | goto out; | |
0db15298 | 6894 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6895 | goto out; |
8965aa28 | 6896 | if (css_tryget(&memcg->css)) |
11092087 | 6897 | sk->sk_memcg = memcg; |
f7e1cb6e | 6898 | out: |
11092087 JW |
6899 | rcu_read_unlock(); |
6900 | } | |
11092087 | 6901 | |
2d758073 | 6902 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6903 | { |
2d758073 JW |
6904 | if (sk->sk_memcg) |
6905 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6906 | } |
6907 | ||
6908 | /** | |
6909 | * mem_cgroup_charge_skmem - charge socket memory | |
6910 | * @memcg: memcg to charge | |
6911 | * @nr_pages: number of pages to charge | |
6912 | * | |
6913 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6914 | * @memcg's configured limit, %false if the charge had to be forced. | |
6915 | */ | |
6916 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6917 | { | |
f7e1cb6e | 6918 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6919 | |
f7e1cb6e | 6920 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6921 | struct page_counter *fail; |
f7e1cb6e | 6922 | |
0db15298 JW |
6923 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6924 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6925 | return true; |
6926 | } | |
0db15298 JW |
6927 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6928 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6929 | return false; |
11092087 | 6930 | } |
d886f4e4 | 6931 | |
f7e1cb6e JW |
6932 | /* Don't block in the packet receive path */ |
6933 | if (in_softirq()) | |
6934 | gfp_mask = GFP_NOWAIT; | |
6935 | ||
c9019e9b | 6936 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6937 | |
f7e1cb6e JW |
6938 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6939 | return true; | |
6940 | ||
6941 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6942 | return false; |
6943 | } | |
6944 | ||
6945 | /** | |
6946 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
6947 | * @memcg: memcg to uncharge |
6948 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
6949 | */ |
6950 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6951 | { | |
f7e1cb6e | 6952 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6953 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
6954 | return; |
6955 | } | |
d886f4e4 | 6956 | |
c9019e9b | 6957 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 6958 | |
475d0487 | 6959 | refill_stock(memcg, nr_pages); |
11092087 JW |
6960 | } |
6961 | ||
f7e1cb6e JW |
6962 | static int __init cgroup_memory(char *s) |
6963 | { | |
6964 | char *token; | |
6965 | ||
6966 | while ((token = strsep(&s, ",")) != NULL) { | |
6967 | if (!*token) | |
6968 | continue; | |
6969 | if (!strcmp(token, "nosocket")) | |
6970 | cgroup_memory_nosocket = true; | |
04823c83 VD |
6971 | if (!strcmp(token, "nokmem")) |
6972 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
6973 | } |
6974 | return 0; | |
6975 | } | |
6976 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 6977 | |
2d11085e | 6978 | /* |
1081312f MH |
6979 | * subsys_initcall() for memory controller. |
6980 | * | |
308167fc SAS |
6981 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
6982 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
6983 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
6984 | * should be initialized from here. | |
2d11085e MH |
6985 | */ |
6986 | static int __init mem_cgroup_init(void) | |
6987 | { | |
95a045f6 JW |
6988 | int cpu, node; |
6989 | ||
308167fc SAS |
6990 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
6991 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
6992 | |
6993 | for_each_possible_cpu(cpu) | |
6994 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
6995 | drain_local_stock); | |
6996 | ||
6997 | for_each_node(node) { | |
6998 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
6999 | |
7000 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
7001 | node_online(node) ? node : NUMA_NO_NODE); | |
7002 | ||
ef8f2327 | 7003 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7004 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7005 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7006 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7007 | } | |
7008 | ||
2d11085e MH |
7009 | return 0; |
7010 | } | |
7011 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
7012 | |
7013 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
7014 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7015 | { | |
1c2d479a | 7016 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7017 | /* |
7018 | * The root cgroup cannot be destroyed, so it's refcount must | |
7019 | * always be >= 1. | |
7020 | */ | |
7021 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
7022 | VM_BUG_ON(1); | |
7023 | break; | |
7024 | } | |
7025 | memcg = parent_mem_cgroup(memcg); | |
7026 | if (!memcg) | |
7027 | memcg = root_mem_cgroup; | |
7028 | } | |
7029 | return memcg; | |
7030 | } | |
7031 | ||
21afa38e JW |
7032 | /** |
7033 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
7034 | * @page: page whose memsw charge to transfer | |
7035 | * @entry: swap entry to move the charge to | |
7036 | * | |
7037 | * Transfer the memsw charge of @page to @entry. | |
7038 | */ | |
7039 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
7040 | { | |
1f47b61f | 7041 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 7042 | unsigned int nr_entries; |
21afa38e JW |
7043 | unsigned short oldid; |
7044 | ||
7045 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
7046 | VM_BUG_ON_PAGE(page_count(page), page); | |
7047 | ||
2d1c4980 | 7048 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
21afa38e JW |
7049 | return; |
7050 | ||
7051 | memcg = page->mem_cgroup; | |
7052 | ||
7053 | /* Readahead page, never charged */ | |
7054 | if (!memcg) | |
7055 | return; | |
7056 | ||
1f47b61f VD |
7057 | /* |
7058 | * In case the memcg owning these pages has been offlined and doesn't | |
7059 | * have an ID allocated to it anymore, charge the closest online | |
7060 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7061 | */ | |
7062 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
d6810d73 HY |
7063 | nr_entries = hpage_nr_pages(page); |
7064 | /* Get references for the tail pages, too */ | |
7065 | if (nr_entries > 1) | |
7066 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7067 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7068 | nr_entries); | |
21afa38e | 7069 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7070 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
7071 | |
7072 | page->mem_cgroup = NULL; | |
7073 | ||
7074 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7075 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7076 | |
2d1c4980 | 7077 | if (!cgroup_memory_noswap && memcg != swap_memcg) { |
1f47b61f | 7078 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
7079 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7080 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7081 | } |
7082 | ||
ce9ce665 SAS |
7083 | /* |
7084 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7085 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7086 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7087 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
7088 | */ |
7089 | VM_BUG_ON(!irqs_disabled()); | |
3fba69a5 | 7090 | mem_cgroup_charge_statistics(memcg, page, -nr_entries); |
21afa38e | 7091 | memcg_check_events(memcg, page); |
73f576c0 | 7092 | |
1a3e1f40 | 7093 | css_put(&memcg->css); |
21afa38e JW |
7094 | } |
7095 | ||
38d8b4e6 HY |
7096 | /** |
7097 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
7098 | * @page: page being added to swap |
7099 | * @entry: swap entry to charge | |
7100 | * | |
38d8b4e6 | 7101 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
7102 | * |
7103 | * Returns 0 on success, -ENOMEM on failure. | |
7104 | */ | |
7105 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
7106 | { | |
38d8b4e6 | 7107 | unsigned int nr_pages = hpage_nr_pages(page); |
37e84351 | 7108 | struct page_counter *counter; |
38d8b4e6 | 7109 | struct mem_cgroup *memcg; |
37e84351 VD |
7110 | unsigned short oldid; |
7111 | ||
2d1c4980 | 7112 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
37e84351 VD |
7113 | return 0; |
7114 | ||
7115 | memcg = page->mem_cgroup; | |
7116 | ||
7117 | /* Readahead page, never charged */ | |
7118 | if (!memcg) | |
7119 | return 0; | |
7120 | ||
f3a53a3a TH |
7121 | if (!entry.val) { |
7122 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7123 | return 0; |
f3a53a3a | 7124 | } |
bb98f2c5 | 7125 | |
1f47b61f VD |
7126 | memcg = mem_cgroup_id_get_online(memcg); |
7127 | ||
2d1c4980 | 7128 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7129 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7130 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7131 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7132 | mem_cgroup_id_put(memcg); |
37e84351 | 7133 | return -ENOMEM; |
1f47b61f | 7134 | } |
37e84351 | 7135 | |
38d8b4e6 HY |
7136 | /* Get references for the tail pages, too */ |
7137 | if (nr_pages > 1) | |
7138 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7139 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 7140 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7141 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7142 | |
37e84351 VD |
7143 | return 0; |
7144 | } | |
7145 | ||
21afa38e | 7146 | /** |
38d8b4e6 | 7147 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7148 | * @entry: swap entry to uncharge |
38d8b4e6 | 7149 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7150 | */ |
38d8b4e6 | 7151 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7152 | { |
7153 | struct mem_cgroup *memcg; | |
7154 | unsigned short id; | |
7155 | ||
38d8b4e6 | 7156 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7157 | rcu_read_lock(); |
adbe427b | 7158 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7159 | if (memcg) { |
2d1c4980 | 7160 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) { |
37e84351 | 7161 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
38d8b4e6 | 7162 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7163 | else |
38d8b4e6 | 7164 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7165 | } |
c9019e9b | 7166 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7167 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7168 | } |
7169 | rcu_read_unlock(); | |
7170 | } | |
7171 | ||
d8b38438 VD |
7172 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7173 | { | |
7174 | long nr_swap_pages = get_nr_swap_pages(); | |
7175 | ||
eccb52e7 | 7176 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
d8b38438 VD |
7177 | return nr_swap_pages; |
7178 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7179 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7180 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7181 | page_counter_read(&memcg->swap)); |
7182 | return nr_swap_pages; | |
7183 | } | |
7184 | ||
5ccc5aba VD |
7185 | bool mem_cgroup_swap_full(struct page *page) |
7186 | { | |
7187 | struct mem_cgroup *memcg; | |
7188 | ||
7189 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
7190 | ||
7191 | if (vm_swap_full()) | |
7192 | return true; | |
eccb52e7 | 7193 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
5ccc5aba VD |
7194 | return false; |
7195 | ||
7196 | memcg = page->mem_cgroup; | |
7197 | if (!memcg) | |
7198 | return false; | |
7199 | ||
4b82ab4f JK |
7200 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
7201 | unsigned long usage = page_counter_read(&memcg->swap); | |
7202 | ||
7203 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
7204 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 7205 | return true; |
4b82ab4f | 7206 | } |
5ccc5aba VD |
7207 | |
7208 | return false; | |
7209 | } | |
7210 | ||
eccb52e7 | 7211 | static int __init setup_swap_account(char *s) |
21afa38e JW |
7212 | { |
7213 | if (!strcmp(s, "1")) | |
eccb52e7 | 7214 | cgroup_memory_noswap = 0; |
21afa38e | 7215 | else if (!strcmp(s, "0")) |
eccb52e7 | 7216 | cgroup_memory_noswap = 1; |
21afa38e JW |
7217 | return 1; |
7218 | } | |
eccb52e7 | 7219 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 7220 | |
37e84351 VD |
7221 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7222 | struct cftype *cft) | |
7223 | { | |
7224 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7225 | ||
7226 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7227 | } | |
7228 | ||
4b82ab4f JK |
7229 | static int swap_high_show(struct seq_file *m, void *v) |
7230 | { | |
7231 | return seq_puts_memcg_tunable(m, | |
7232 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
7233 | } | |
7234 | ||
7235 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
7236 | char *buf, size_t nbytes, loff_t off) | |
7237 | { | |
7238 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7239 | unsigned long high; | |
7240 | int err; | |
7241 | ||
7242 | buf = strstrip(buf); | |
7243 | err = page_counter_memparse(buf, "max", &high); | |
7244 | if (err) | |
7245 | return err; | |
7246 | ||
7247 | page_counter_set_high(&memcg->swap, high); | |
7248 | ||
7249 | return nbytes; | |
7250 | } | |
7251 | ||
37e84351 VD |
7252 | static int swap_max_show(struct seq_file *m, void *v) |
7253 | { | |
677dc973 CD |
7254 | return seq_puts_memcg_tunable(m, |
7255 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7256 | } |
7257 | ||
7258 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7259 | char *buf, size_t nbytes, loff_t off) | |
7260 | { | |
7261 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7262 | unsigned long max; | |
7263 | int err; | |
7264 | ||
7265 | buf = strstrip(buf); | |
7266 | err = page_counter_memparse(buf, "max", &max); | |
7267 | if (err) | |
7268 | return err; | |
7269 | ||
be09102b | 7270 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7271 | |
7272 | return nbytes; | |
7273 | } | |
7274 | ||
f3a53a3a TH |
7275 | static int swap_events_show(struct seq_file *m, void *v) |
7276 | { | |
aa9694bb | 7277 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 7278 | |
4b82ab4f JK |
7279 | seq_printf(m, "high %lu\n", |
7280 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
7281 | seq_printf(m, "max %lu\n", |
7282 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7283 | seq_printf(m, "fail %lu\n", | |
7284 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7285 | ||
7286 | return 0; | |
7287 | } | |
7288 | ||
37e84351 VD |
7289 | static struct cftype swap_files[] = { |
7290 | { | |
7291 | .name = "swap.current", | |
7292 | .flags = CFTYPE_NOT_ON_ROOT, | |
7293 | .read_u64 = swap_current_read, | |
7294 | }, | |
4b82ab4f JK |
7295 | { |
7296 | .name = "swap.high", | |
7297 | .flags = CFTYPE_NOT_ON_ROOT, | |
7298 | .seq_show = swap_high_show, | |
7299 | .write = swap_high_write, | |
7300 | }, | |
37e84351 VD |
7301 | { |
7302 | .name = "swap.max", | |
7303 | .flags = CFTYPE_NOT_ON_ROOT, | |
7304 | .seq_show = swap_max_show, | |
7305 | .write = swap_max_write, | |
7306 | }, | |
f3a53a3a TH |
7307 | { |
7308 | .name = "swap.events", | |
7309 | .flags = CFTYPE_NOT_ON_ROOT, | |
7310 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7311 | .seq_show = swap_events_show, | |
7312 | }, | |
37e84351 VD |
7313 | { } /* terminate */ |
7314 | }; | |
7315 | ||
eccb52e7 | 7316 | static struct cftype memsw_files[] = { |
21afa38e JW |
7317 | { |
7318 | .name = "memsw.usage_in_bytes", | |
7319 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7320 | .read_u64 = mem_cgroup_read_u64, | |
7321 | }, | |
7322 | { | |
7323 | .name = "memsw.max_usage_in_bytes", | |
7324 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7325 | .write = mem_cgroup_reset, | |
7326 | .read_u64 = mem_cgroup_read_u64, | |
7327 | }, | |
7328 | { | |
7329 | .name = "memsw.limit_in_bytes", | |
7330 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7331 | .write = mem_cgroup_write, | |
7332 | .read_u64 = mem_cgroup_read_u64, | |
7333 | }, | |
7334 | { | |
7335 | .name = "memsw.failcnt", | |
7336 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7337 | .write = mem_cgroup_reset, | |
7338 | .read_u64 = mem_cgroup_read_u64, | |
7339 | }, | |
7340 | { }, /* terminate */ | |
7341 | }; | |
7342 | ||
82ff165c BS |
7343 | /* |
7344 | * If mem_cgroup_swap_init() is implemented as a subsys_initcall() | |
7345 | * instead of a core_initcall(), this could mean cgroup_memory_noswap still | |
7346 | * remains set to false even when memcg is disabled via "cgroup_disable=memory" | |
7347 | * boot parameter. This may result in premature OOPS inside | |
7348 | * mem_cgroup_get_nr_swap_pages() function in corner cases. | |
7349 | */ | |
21afa38e JW |
7350 | static int __init mem_cgroup_swap_init(void) |
7351 | { | |
2d1c4980 JW |
7352 | /* No memory control -> no swap control */ |
7353 | if (mem_cgroup_disabled()) | |
7354 | cgroup_memory_noswap = true; | |
7355 | ||
7356 | if (cgroup_memory_noswap) | |
eccb52e7 JW |
7357 | return 0; |
7358 | ||
7359 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
7360 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
7361 | ||
21afa38e JW |
7362 | return 0; |
7363 | } | |
82ff165c | 7364 | core_initcall(mem_cgroup_swap_init); |
21afa38e JW |
7365 | |
7366 | #endif /* CONFIG_MEMCG_SWAP */ |