]>
Commit | Line | Data |
---|---|---|
c942fddf | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
8cdea7c0 BS |
2 | /* memcontrol.c - Memory Controller |
3 | * | |
4 | * Copyright IBM Corporation, 2007 | |
5 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
6 | * | |
78fb7466 PE |
7 | * Copyright 2007 OpenVZ SWsoft Inc |
8 | * Author: Pavel Emelianov <xemul@openvz.org> | |
9 | * | |
2e72b634 KS |
10 | * Memory thresholds |
11 | * Copyright (C) 2009 Nokia Corporation | |
12 | * Author: Kirill A. Shutemov | |
13 | * | |
7ae1e1d0 GC |
14 | * Kernel Memory Controller |
15 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
16 | * Authors: Glauber Costa and Suleiman Souhlal | |
17 | * | |
1575e68b JW |
18 | * Native page reclaim |
19 | * Charge lifetime sanitation | |
20 | * Lockless page tracking & accounting | |
21 | * Unified hierarchy configuration model | |
22 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
8cdea7c0 BS |
23 | */ |
24 | ||
3e32cb2e | 25 | #include <linux/page_counter.h> |
8cdea7c0 BS |
26 | #include <linux/memcontrol.h> |
27 | #include <linux/cgroup.h> | |
a520110e | 28 | #include <linux/pagewalk.h> |
6e84f315 | 29 | #include <linux/sched/mm.h> |
3a4f8a0b | 30 | #include <linux/shmem_fs.h> |
4ffef5fe | 31 | #include <linux/hugetlb.h> |
d13d1443 | 32 | #include <linux/pagemap.h> |
1ff9e6e1 | 33 | #include <linux/vm_event_item.h> |
d52aa412 | 34 | #include <linux/smp.h> |
8a9f3ccd | 35 | #include <linux/page-flags.h> |
66e1707b | 36 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
37 | #include <linux/bit_spinlock.h> |
38 | #include <linux/rcupdate.h> | |
e222432b | 39 | #include <linux/limits.h> |
b9e15baf | 40 | #include <linux/export.h> |
8c7c6e34 | 41 | #include <linux/mutex.h> |
bb4cc1a8 | 42 | #include <linux/rbtree.h> |
b6ac57d5 | 43 | #include <linux/slab.h> |
66e1707b | 44 | #include <linux/swap.h> |
02491447 | 45 | #include <linux/swapops.h> |
66e1707b | 46 | #include <linux/spinlock.h> |
2e72b634 | 47 | #include <linux/eventfd.h> |
79bd9814 | 48 | #include <linux/poll.h> |
2e72b634 | 49 | #include <linux/sort.h> |
66e1707b | 50 | #include <linux/fs.h> |
d2ceb9b7 | 51 | #include <linux/seq_file.h> |
70ddf637 | 52 | #include <linux/vmpressure.h> |
b69408e8 | 53 | #include <linux/mm_inline.h> |
5d1ea48b | 54 | #include <linux/swap_cgroup.h> |
cdec2e42 | 55 | #include <linux/cpu.h> |
158e0a2d | 56 | #include <linux/oom.h> |
0056f4e6 | 57 | #include <linux/lockdep.h> |
79bd9814 | 58 | #include <linux/file.h> |
b23afb93 | 59 | #include <linux/tracehook.h> |
0e4b01df | 60 | #include <linux/psi.h> |
c8713d0b | 61 | #include <linux/seq_buf.h> |
08e552c6 | 62 | #include "internal.h" |
d1a4c0b3 | 63 | #include <net/sock.h> |
4bd2c1ee | 64 | #include <net/ip.h> |
f35c3a8e | 65 | #include "slab.h" |
8cdea7c0 | 66 | |
7c0f6ba6 | 67 | #include <linux/uaccess.h> |
8697d331 | 68 | |
cc8e970c KM |
69 | #include <trace/events/vmscan.h> |
70 | ||
073219e9 TH |
71 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
72 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 73 | |
7d828602 JW |
74 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
75 | ||
a181b0e8 | 76 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 77 | |
f7e1cb6e JW |
78 | /* Socket memory accounting disabled? */ |
79 | static bool cgroup_memory_nosocket; | |
80 | ||
04823c83 VD |
81 | /* Kernel memory accounting disabled? */ |
82 | static bool cgroup_memory_nokmem; | |
83 | ||
21afa38e | 84 | /* Whether the swap controller is active */ |
c255a458 | 85 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 86 | int do_swap_account __read_mostly; |
c077719b | 87 | #else |
a0db00fc | 88 | #define do_swap_account 0 |
c077719b KH |
89 | #endif |
90 | ||
97b27821 TH |
91 | #ifdef CONFIG_CGROUP_WRITEBACK |
92 | static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq); | |
93 | #endif | |
94 | ||
7941d214 JW |
95 | /* Whether legacy memory+swap accounting is active */ |
96 | static bool do_memsw_account(void) | |
97 | { | |
98 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; | |
99 | } | |
100 | ||
71cd3113 | 101 | static const char *const mem_cgroup_lru_names[] = { |
58cf188e SZ |
102 | "inactive_anon", |
103 | "active_anon", | |
104 | "inactive_file", | |
105 | "active_file", | |
106 | "unevictable", | |
107 | }; | |
108 | ||
a0db00fc KS |
109 | #define THRESHOLDS_EVENTS_TARGET 128 |
110 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
111 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 112 | |
bb4cc1a8 AM |
113 | /* |
114 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
115 | * their hierarchy representation | |
116 | */ | |
117 | ||
ef8f2327 | 118 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 119 | struct rb_root rb_root; |
fa90b2fd | 120 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
121 | spinlock_t lock; |
122 | }; | |
123 | ||
bb4cc1a8 AM |
124 | struct mem_cgroup_tree { |
125 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
126 | }; | |
127 | ||
128 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
129 | ||
9490ff27 KH |
130 | /* for OOM */ |
131 | struct mem_cgroup_eventfd_list { | |
132 | struct list_head list; | |
133 | struct eventfd_ctx *eventfd; | |
134 | }; | |
2e72b634 | 135 | |
79bd9814 TH |
136 | /* |
137 | * cgroup_event represents events which userspace want to receive. | |
138 | */ | |
3bc942f3 | 139 | struct mem_cgroup_event { |
79bd9814 | 140 | /* |
59b6f873 | 141 | * memcg which the event belongs to. |
79bd9814 | 142 | */ |
59b6f873 | 143 | struct mem_cgroup *memcg; |
79bd9814 TH |
144 | /* |
145 | * eventfd to signal userspace about the event. | |
146 | */ | |
147 | struct eventfd_ctx *eventfd; | |
148 | /* | |
149 | * Each of these stored in a list by the cgroup. | |
150 | */ | |
151 | struct list_head list; | |
fba94807 TH |
152 | /* |
153 | * register_event() callback will be used to add new userspace | |
154 | * waiter for changes related to this event. Use eventfd_signal() | |
155 | * on eventfd to send notification to userspace. | |
156 | */ | |
59b6f873 | 157 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 158 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
159 | /* |
160 | * unregister_event() callback will be called when userspace closes | |
161 | * the eventfd or on cgroup removing. This callback must be set, | |
162 | * if you want provide notification functionality. | |
163 | */ | |
59b6f873 | 164 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 165 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
166 | /* |
167 | * All fields below needed to unregister event when | |
168 | * userspace closes eventfd. | |
169 | */ | |
170 | poll_table pt; | |
171 | wait_queue_head_t *wqh; | |
ac6424b9 | 172 | wait_queue_entry_t wait; |
79bd9814 TH |
173 | struct work_struct remove; |
174 | }; | |
175 | ||
c0ff4b85 R |
176 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
177 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 178 | |
7dc74be0 DN |
179 | /* Stuffs for move charges at task migration. */ |
180 | /* | |
1dfab5ab | 181 | * Types of charges to be moved. |
7dc74be0 | 182 | */ |
1dfab5ab JW |
183 | #define MOVE_ANON 0x1U |
184 | #define MOVE_FILE 0x2U | |
185 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 186 | |
4ffef5fe DN |
187 | /* "mc" and its members are protected by cgroup_mutex */ |
188 | static struct move_charge_struct { | |
b1dd693e | 189 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 190 | struct mm_struct *mm; |
4ffef5fe DN |
191 | struct mem_cgroup *from; |
192 | struct mem_cgroup *to; | |
1dfab5ab | 193 | unsigned long flags; |
4ffef5fe | 194 | unsigned long precharge; |
854ffa8d | 195 | unsigned long moved_charge; |
483c30b5 | 196 | unsigned long moved_swap; |
8033b97c DN |
197 | struct task_struct *moving_task; /* a task moving charges */ |
198 | wait_queue_head_t waitq; /* a waitq for other context */ | |
199 | } mc = { | |
2bd9bb20 | 200 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
201 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
202 | }; | |
4ffef5fe | 203 | |
4e416953 BS |
204 | /* |
205 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
206 | * limit reclaim to prevent infinite loops, if they ever occur. | |
207 | */ | |
a0db00fc | 208 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 209 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 210 | |
217bc319 KH |
211 | enum charge_type { |
212 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 213 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 214 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 215 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
216 | NR_CHARGE_TYPE, |
217 | }; | |
218 | ||
8c7c6e34 | 219 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
220 | enum res_type { |
221 | _MEM, | |
222 | _MEMSWAP, | |
223 | _OOM_TYPE, | |
510fc4e1 | 224 | _KMEM, |
d55f90bf | 225 | _TCP, |
86ae53e1 GC |
226 | }; |
227 | ||
a0db00fc KS |
228 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
229 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 230 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
231 | /* Used for OOM nofiier */ |
232 | #define OOM_CONTROL (0) | |
8c7c6e34 | 233 | |
b05706f1 KT |
234 | /* |
235 | * Iteration constructs for visiting all cgroups (under a tree). If | |
236 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
237 | * be used for reference counting. | |
238 | */ | |
239 | #define for_each_mem_cgroup_tree(iter, root) \ | |
240 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
241 | iter != NULL; \ | |
242 | iter = mem_cgroup_iter(root, iter, NULL)) | |
243 | ||
244 | #define for_each_mem_cgroup(iter) \ | |
245 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
246 | iter != NULL; \ | |
247 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
248 | ||
7775face TH |
249 | static inline bool should_force_charge(void) |
250 | { | |
251 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
252 | (current->flags & PF_EXITING); | |
253 | } | |
254 | ||
70ddf637 AV |
255 | /* Some nice accessors for the vmpressure. */ |
256 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
257 | { | |
258 | if (!memcg) | |
259 | memcg = root_mem_cgroup; | |
260 | return &memcg->vmpressure; | |
261 | } | |
262 | ||
263 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
264 | { | |
265 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
266 | } | |
267 | ||
84c07d11 | 268 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 269 | /* |
f7ce3190 | 270 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
271 | * The main reason for not using cgroup id for this: |
272 | * this works better in sparse environments, where we have a lot of memcgs, | |
273 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
274 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
275 | * 200 entry array for that. | |
55007d84 | 276 | * |
dbcf73e2 VD |
277 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
278 | * will double each time we have to increase it. | |
55007d84 | 279 | */ |
dbcf73e2 VD |
280 | static DEFINE_IDA(memcg_cache_ida); |
281 | int memcg_nr_cache_ids; | |
749c5415 | 282 | |
05257a1a VD |
283 | /* Protects memcg_nr_cache_ids */ |
284 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
285 | ||
286 | void memcg_get_cache_ids(void) | |
287 | { | |
288 | down_read(&memcg_cache_ids_sem); | |
289 | } | |
290 | ||
291 | void memcg_put_cache_ids(void) | |
292 | { | |
293 | up_read(&memcg_cache_ids_sem); | |
294 | } | |
295 | ||
55007d84 GC |
296 | /* |
297 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
298 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
299 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
300 | * tunable, but that is strictly not necessary. | |
301 | * | |
b8627835 | 302 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
303 | * this constant directly from cgroup, but it is understandable that this is |
304 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 305 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
306 | * increase ours as well if it increases. |
307 | */ | |
308 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 309 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 310 | |
d7f25f8a GC |
311 | /* |
312 | * A lot of the calls to the cache allocation functions are expected to be | |
313 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
314 | * conditional to this static branch, we'll have to allow modules that does | |
315 | * kmem_cache_alloc and the such to see this symbol as well | |
316 | */ | |
ef12947c | 317 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 318 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 319 | |
17cc4dfe | 320 | struct workqueue_struct *memcg_kmem_cache_wq; |
0a432dcb | 321 | #endif |
17cc4dfe | 322 | |
0a4465d3 KT |
323 | static int memcg_shrinker_map_size; |
324 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); | |
325 | ||
326 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) | |
327 | { | |
328 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); | |
329 | } | |
330 | ||
331 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, | |
332 | int size, int old_size) | |
333 | { | |
334 | struct memcg_shrinker_map *new, *old; | |
335 | int nid; | |
336 | ||
337 | lockdep_assert_held(&memcg_shrinker_map_mutex); | |
338 | ||
339 | for_each_node(nid) { | |
340 | old = rcu_dereference_protected( | |
341 | mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); | |
342 | /* Not yet online memcg */ | |
343 | if (!old) | |
344 | return 0; | |
345 | ||
346 | new = kvmalloc(sizeof(*new) + size, GFP_KERNEL); | |
347 | if (!new) | |
348 | return -ENOMEM; | |
349 | ||
350 | /* Set all old bits, clear all new bits */ | |
351 | memset(new->map, (int)0xff, old_size); | |
352 | memset((void *)new->map + old_size, 0, size - old_size); | |
353 | ||
354 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); | |
355 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); | |
356 | } | |
357 | ||
358 | return 0; | |
359 | } | |
360 | ||
361 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) | |
362 | { | |
363 | struct mem_cgroup_per_node *pn; | |
364 | struct memcg_shrinker_map *map; | |
365 | int nid; | |
366 | ||
367 | if (mem_cgroup_is_root(memcg)) | |
368 | return; | |
369 | ||
370 | for_each_node(nid) { | |
371 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
372 | map = rcu_dereference_protected(pn->shrinker_map, true); | |
373 | if (map) | |
374 | kvfree(map); | |
375 | rcu_assign_pointer(pn->shrinker_map, NULL); | |
376 | } | |
377 | } | |
378 | ||
379 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
380 | { | |
381 | struct memcg_shrinker_map *map; | |
382 | int nid, size, ret = 0; | |
383 | ||
384 | if (mem_cgroup_is_root(memcg)) | |
385 | return 0; | |
386 | ||
387 | mutex_lock(&memcg_shrinker_map_mutex); | |
388 | size = memcg_shrinker_map_size; | |
389 | for_each_node(nid) { | |
390 | map = kvzalloc(sizeof(*map) + size, GFP_KERNEL); | |
391 | if (!map) { | |
392 | memcg_free_shrinker_maps(memcg); | |
393 | ret = -ENOMEM; | |
394 | break; | |
395 | } | |
396 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); | |
397 | } | |
398 | mutex_unlock(&memcg_shrinker_map_mutex); | |
399 | ||
400 | return ret; | |
401 | } | |
402 | ||
403 | int memcg_expand_shrinker_maps(int new_id) | |
404 | { | |
405 | int size, old_size, ret = 0; | |
406 | struct mem_cgroup *memcg; | |
407 | ||
408 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); | |
409 | old_size = memcg_shrinker_map_size; | |
410 | if (size <= old_size) | |
411 | return 0; | |
412 | ||
413 | mutex_lock(&memcg_shrinker_map_mutex); | |
414 | if (!root_mem_cgroup) | |
415 | goto unlock; | |
416 | ||
417 | for_each_mem_cgroup(memcg) { | |
418 | if (mem_cgroup_is_root(memcg)) | |
419 | continue; | |
420 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); | |
7c936e36 VA |
421 | if (ret) { |
422 | mem_cgroup_iter_break(NULL, memcg); | |
0a4465d3 | 423 | goto unlock; |
7c936e36 | 424 | } |
0a4465d3 KT |
425 | } |
426 | unlock: | |
427 | if (!ret) | |
428 | memcg_shrinker_map_size = size; | |
429 | mutex_unlock(&memcg_shrinker_map_mutex); | |
430 | return ret; | |
431 | } | |
fae91d6d KT |
432 | |
433 | void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id) | |
434 | { | |
435 | if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) { | |
436 | struct memcg_shrinker_map *map; | |
437 | ||
438 | rcu_read_lock(); | |
439 | map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map); | |
f90280d6 KT |
440 | /* Pairs with smp mb in shrink_slab() */ |
441 | smp_mb__before_atomic(); | |
fae91d6d KT |
442 | set_bit(shrinker_id, map->map); |
443 | rcu_read_unlock(); | |
444 | } | |
445 | } | |
446 | ||
ad7fa852 TH |
447 | /** |
448 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
449 | * @page: page of interest | |
450 | * | |
451 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
452 | * with @page is returned. The returned css remains associated with @page | |
453 | * until it is released. | |
454 | * | |
455 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
456 | * is returned. | |
ad7fa852 TH |
457 | */ |
458 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
459 | { | |
460 | struct mem_cgroup *memcg; | |
461 | ||
ad7fa852 TH |
462 | memcg = page->mem_cgroup; |
463 | ||
9e10a130 | 464 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
465 | memcg = root_mem_cgroup; |
466 | ||
ad7fa852 TH |
467 | return &memcg->css; |
468 | } | |
469 | ||
2fc04524 VD |
470 | /** |
471 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
472 | * @page: the page | |
473 | * | |
474 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
475 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
476 | * is safe to call this function without holding a reference to @page. | |
477 | * | |
478 | * Note, this function is inherently racy, because there is nothing to prevent | |
479 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
480 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
481 | * do not care (such as procfs interfaces). | |
482 | */ | |
483 | ino_t page_cgroup_ino(struct page *page) | |
484 | { | |
485 | struct mem_cgroup *memcg; | |
486 | unsigned long ino = 0; | |
487 | ||
488 | rcu_read_lock(); | |
221ec5c0 | 489 | if (PageSlab(page) && !PageTail(page)) |
4d96ba35 RG |
490 | memcg = memcg_from_slab_page(page); |
491 | else | |
492 | memcg = READ_ONCE(page->mem_cgroup); | |
2fc04524 VD |
493 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
494 | memcg = parent_mem_cgroup(memcg); | |
495 | if (memcg) | |
496 | ino = cgroup_ino(memcg->css.cgroup); | |
497 | rcu_read_unlock(); | |
498 | return ino; | |
499 | } | |
500 | ||
ef8f2327 MG |
501 | static struct mem_cgroup_per_node * |
502 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 503 | { |
97a6c37b | 504 | int nid = page_to_nid(page); |
f64c3f54 | 505 | |
ef8f2327 | 506 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
507 | } |
508 | ||
ef8f2327 MG |
509 | static struct mem_cgroup_tree_per_node * |
510 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 511 | { |
ef8f2327 | 512 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
513 | } |
514 | ||
ef8f2327 | 515 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
516 | soft_limit_tree_from_page(struct page *page) |
517 | { | |
518 | int nid = page_to_nid(page); | |
bb4cc1a8 | 519 | |
ef8f2327 | 520 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
521 | } |
522 | ||
ef8f2327 MG |
523 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
524 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 525 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
526 | { |
527 | struct rb_node **p = &mctz->rb_root.rb_node; | |
528 | struct rb_node *parent = NULL; | |
ef8f2327 | 529 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 530 | bool rightmost = true; |
bb4cc1a8 AM |
531 | |
532 | if (mz->on_tree) | |
533 | return; | |
534 | ||
535 | mz->usage_in_excess = new_usage_in_excess; | |
536 | if (!mz->usage_in_excess) | |
537 | return; | |
538 | while (*p) { | |
539 | parent = *p; | |
ef8f2327 | 540 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 541 | tree_node); |
fa90b2fd | 542 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 543 | p = &(*p)->rb_left; |
fa90b2fd DB |
544 | rightmost = false; |
545 | } | |
546 | ||
bb4cc1a8 AM |
547 | /* |
548 | * We can't avoid mem cgroups that are over their soft | |
549 | * limit by the same amount | |
550 | */ | |
551 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
552 | p = &(*p)->rb_right; | |
553 | } | |
fa90b2fd DB |
554 | |
555 | if (rightmost) | |
556 | mctz->rb_rightmost = &mz->tree_node; | |
557 | ||
bb4cc1a8 AM |
558 | rb_link_node(&mz->tree_node, parent, p); |
559 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
560 | mz->on_tree = true; | |
561 | } | |
562 | ||
ef8f2327 MG |
563 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
564 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
565 | { |
566 | if (!mz->on_tree) | |
567 | return; | |
fa90b2fd DB |
568 | |
569 | if (&mz->tree_node == mctz->rb_rightmost) | |
570 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
571 | ||
bb4cc1a8 AM |
572 | rb_erase(&mz->tree_node, &mctz->rb_root); |
573 | mz->on_tree = false; | |
574 | } | |
575 | ||
ef8f2327 MG |
576 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
577 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 578 | { |
0a31bc97 JW |
579 | unsigned long flags; |
580 | ||
581 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 582 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 583 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
584 | } |
585 | ||
3e32cb2e JW |
586 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
587 | { | |
588 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 589 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
590 | unsigned long excess = 0; |
591 | ||
592 | if (nr_pages > soft_limit) | |
593 | excess = nr_pages - soft_limit; | |
594 | ||
595 | return excess; | |
596 | } | |
bb4cc1a8 AM |
597 | |
598 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
599 | { | |
3e32cb2e | 600 | unsigned long excess; |
ef8f2327 MG |
601 | struct mem_cgroup_per_node *mz; |
602 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 603 | |
e231875b | 604 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
605 | if (!mctz) |
606 | return; | |
bb4cc1a8 AM |
607 | /* |
608 | * Necessary to update all ancestors when hierarchy is used. | |
609 | * because their event counter is not touched. | |
610 | */ | |
611 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 612 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 613 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
614 | /* |
615 | * We have to update the tree if mz is on RB-tree or | |
616 | * mem is over its softlimit. | |
617 | */ | |
618 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
619 | unsigned long flags; |
620 | ||
621 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
622 | /* if on-tree, remove it */ |
623 | if (mz->on_tree) | |
cf2c8127 | 624 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
625 | /* |
626 | * Insert again. mz->usage_in_excess will be updated. | |
627 | * If excess is 0, no tree ops. | |
628 | */ | |
cf2c8127 | 629 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 630 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
631 | } |
632 | } | |
633 | } | |
634 | ||
635 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
636 | { | |
ef8f2327 MG |
637 | struct mem_cgroup_tree_per_node *mctz; |
638 | struct mem_cgroup_per_node *mz; | |
639 | int nid; | |
bb4cc1a8 | 640 | |
e231875b | 641 | for_each_node(nid) { |
ef8f2327 MG |
642 | mz = mem_cgroup_nodeinfo(memcg, nid); |
643 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
644 | if (mctz) |
645 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
646 | } |
647 | } | |
648 | ||
ef8f2327 MG |
649 | static struct mem_cgroup_per_node * |
650 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 651 | { |
ef8f2327 | 652 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
653 | |
654 | retry: | |
655 | mz = NULL; | |
fa90b2fd | 656 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
657 | goto done; /* Nothing to reclaim from */ |
658 | ||
fa90b2fd DB |
659 | mz = rb_entry(mctz->rb_rightmost, |
660 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
661 | /* |
662 | * Remove the node now but someone else can add it back, | |
663 | * we will to add it back at the end of reclaim to its correct | |
664 | * position in the tree. | |
665 | */ | |
cf2c8127 | 666 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 667 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 668 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
669 | goto retry; |
670 | done: | |
671 | return mz; | |
672 | } | |
673 | ||
ef8f2327 MG |
674 | static struct mem_cgroup_per_node * |
675 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 676 | { |
ef8f2327 | 677 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 678 | |
0a31bc97 | 679 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 680 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 681 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
682 | return mz; |
683 | } | |
684 | ||
db9adbcb JW |
685 | /** |
686 | * __mod_memcg_state - update cgroup memory statistics | |
687 | * @memcg: the memory cgroup | |
688 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
689 | * @val: delta to add to the counter, can be negative | |
690 | */ | |
691 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
692 | { | |
693 | long x; | |
694 | ||
695 | if (mem_cgroup_disabled()) | |
696 | return; | |
697 | ||
698 | x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); | |
699 | if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
700 | struct mem_cgroup *mi; |
701 | ||
766a4c19 YS |
702 | /* |
703 | * Batch local counters to keep them in sync with | |
704 | * the hierarchical ones. | |
705 | */ | |
706 | __this_cpu_add(memcg->vmstats_local->stat[idx], x); | |
42a30035 JW |
707 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
708 | atomic_long_add(x, &mi->vmstats[idx]); | |
db9adbcb JW |
709 | x = 0; |
710 | } | |
711 | __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); | |
712 | } | |
713 | ||
42a30035 JW |
714 | static struct mem_cgroup_per_node * |
715 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
716 | { | |
717 | struct mem_cgroup *parent; | |
718 | ||
719 | parent = parent_mem_cgroup(pn->memcg); | |
720 | if (!parent) | |
721 | return NULL; | |
722 | return mem_cgroup_nodeinfo(parent, nid); | |
723 | } | |
724 | ||
db9adbcb JW |
725 | /** |
726 | * __mod_lruvec_state - update lruvec memory statistics | |
727 | * @lruvec: the lruvec | |
728 | * @idx: the stat item | |
729 | * @val: delta to add to the counter, can be negative | |
730 | * | |
731 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
732 | * function updates the all three counters that are affected by a | |
733 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
734 | */ | |
735 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
736 | int val) | |
737 | { | |
42a30035 | 738 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
db9adbcb | 739 | struct mem_cgroup_per_node *pn; |
42a30035 | 740 | struct mem_cgroup *memcg; |
db9adbcb JW |
741 | long x; |
742 | ||
743 | /* Update node */ | |
42a30035 | 744 | __mod_node_page_state(pgdat, idx, val); |
db9adbcb JW |
745 | |
746 | if (mem_cgroup_disabled()) | |
747 | return; | |
748 | ||
749 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); | |
42a30035 | 750 | memcg = pn->memcg; |
db9adbcb JW |
751 | |
752 | /* Update memcg */ | |
42a30035 | 753 | __mod_memcg_state(memcg, idx, val); |
db9adbcb | 754 | |
b4c46484 RG |
755 | /* Update lruvec */ |
756 | __this_cpu_add(pn->lruvec_stat_local->count[idx], val); | |
757 | ||
db9adbcb JW |
758 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); |
759 | if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
760 | struct mem_cgroup_per_node *pi; |
761 | ||
42a30035 JW |
762 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) |
763 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
764 | x = 0; |
765 | } | |
766 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
767 | } | |
768 | ||
ec9f0238 RG |
769 | void __mod_lruvec_slab_state(void *p, enum node_stat_item idx, int val) |
770 | { | |
771 | struct page *page = virt_to_head_page(p); | |
772 | pg_data_t *pgdat = page_pgdat(page); | |
773 | struct mem_cgroup *memcg; | |
774 | struct lruvec *lruvec; | |
775 | ||
776 | rcu_read_lock(); | |
777 | memcg = memcg_from_slab_page(page); | |
778 | ||
00e81d19 MS |
779 | /* |
780 | * Untracked pages have no memcg, no lruvec. Update only the | |
781 | * node. If we reparent the slab objects to the root memcg, | |
782 | * when we free the slab object, we need to update the per-memcg | |
783 | * vmstats to keep it correct for the root memcg. | |
784 | */ | |
785 | if (!memcg) { | |
ec9f0238 RG |
786 | __mod_node_page_state(pgdat, idx, val); |
787 | } else { | |
788 | lruvec = mem_cgroup_lruvec(pgdat, memcg); | |
789 | __mod_lruvec_state(lruvec, idx, val); | |
790 | } | |
791 | rcu_read_unlock(); | |
792 | } | |
793 | ||
ec0bed29 RG |
794 | void mod_memcg_obj_state(void *p, int idx, int val) |
795 | { | |
796 | struct mem_cgroup *memcg; | |
797 | ||
798 | rcu_read_lock(); | |
799 | memcg = mem_cgroup_from_obj(p); | |
800 | if (memcg) | |
801 | mod_memcg_state(memcg, idx, val); | |
802 | rcu_read_unlock(); | |
803 | } | |
804 | ||
db9adbcb JW |
805 | /** |
806 | * __count_memcg_events - account VM events in a cgroup | |
807 | * @memcg: the memory cgroup | |
808 | * @idx: the event item | |
809 | * @count: the number of events that occured | |
810 | */ | |
811 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
812 | unsigned long count) | |
813 | { | |
814 | unsigned long x; | |
815 | ||
816 | if (mem_cgroup_disabled()) | |
817 | return; | |
818 | ||
819 | x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); | |
820 | if (unlikely(x > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
821 | struct mem_cgroup *mi; |
822 | ||
766a4c19 YS |
823 | /* |
824 | * Batch local counters to keep them in sync with | |
825 | * the hierarchical ones. | |
826 | */ | |
827 | __this_cpu_add(memcg->vmstats_local->events[idx], x); | |
42a30035 JW |
828 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
829 | atomic_long_add(x, &mi->vmevents[idx]); | |
db9adbcb JW |
830 | x = 0; |
831 | } | |
832 | __this_cpu_write(memcg->vmstats_percpu->events[idx], x); | |
833 | } | |
834 | ||
42a30035 | 835 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 836 | { |
871789d4 | 837 | return atomic_long_read(&memcg->vmevents[event]); |
e9f8974f JW |
838 | } |
839 | ||
42a30035 JW |
840 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
841 | { | |
815744d7 JW |
842 | long x = 0; |
843 | int cpu; | |
844 | ||
845 | for_each_possible_cpu(cpu) | |
846 | x += per_cpu(memcg->vmstats_local->events[event], cpu); | |
847 | return x; | |
42a30035 JW |
848 | } |
849 | ||
c0ff4b85 | 850 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 851 | struct page *page, |
f627c2f5 | 852 | bool compound, int nr_pages) |
d52aa412 | 853 | { |
b2402857 KH |
854 | /* |
855 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
856 | * counted as CACHE even if it's on ANON LRU. | |
857 | */ | |
0a31bc97 | 858 | if (PageAnon(page)) |
c9019e9b | 859 | __mod_memcg_state(memcg, MEMCG_RSS, nr_pages); |
9a4caf1e | 860 | else { |
c9019e9b | 861 | __mod_memcg_state(memcg, MEMCG_CACHE, nr_pages); |
9a4caf1e | 862 | if (PageSwapBacked(page)) |
c9019e9b | 863 | __mod_memcg_state(memcg, NR_SHMEM, nr_pages); |
9a4caf1e | 864 | } |
55e462b0 | 865 | |
f627c2f5 KS |
866 | if (compound) { |
867 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
c9019e9b | 868 | __mod_memcg_state(memcg, MEMCG_RSS_HUGE, nr_pages); |
f627c2f5 | 869 | } |
b070e65c | 870 | |
e401f176 KH |
871 | /* pagein of a big page is an event. So, ignore page size */ |
872 | if (nr_pages > 0) | |
c9019e9b | 873 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 874 | else { |
c9019e9b | 875 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
876 | nr_pages = -nr_pages; /* for event */ |
877 | } | |
e401f176 | 878 | |
871789d4 | 879 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
880 | } |
881 | ||
f53d7ce3 JW |
882 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
883 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
884 | { |
885 | unsigned long val, next; | |
886 | ||
871789d4 CD |
887 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
888 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 889 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 890 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
891 | switch (target) { |
892 | case MEM_CGROUP_TARGET_THRESH: | |
893 | next = val + THRESHOLDS_EVENTS_TARGET; | |
894 | break; | |
bb4cc1a8 AM |
895 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
896 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
897 | break; | |
f53d7ce3 JW |
898 | case MEM_CGROUP_TARGET_NUMAINFO: |
899 | next = val + NUMAINFO_EVENTS_TARGET; | |
900 | break; | |
901 | default: | |
902 | break; | |
903 | } | |
871789d4 | 904 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 905 | return true; |
7a159cc9 | 906 | } |
f53d7ce3 | 907 | return false; |
d2265e6f KH |
908 | } |
909 | ||
910 | /* | |
911 | * Check events in order. | |
912 | * | |
913 | */ | |
c0ff4b85 | 914 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
915 | { |
916 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
917 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
918 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 919 | bool do_softlimit; |
82b3f2a7 | 920 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 921 | |
bb4cc1a8 AM |
922 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
923 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
924 | #if MAX_NUMNODES > 1 |
925 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
926 | MEM_CGROUP_TARGET_NUMAINFO); | |
927 | #endif | |
c0ff4b85 | 928 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
929 | if (unlikely(do_softlimit)) |
930 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 931 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 932 | if (unlikely(do_numainfo)) |
c0ff4b85 | 933 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 934 | #endif |
0a31bc97 | 935 | } |
d2265e6f KH |
936 | } |
937 | ||
cf475ad2 | 938 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 939 | { |
31a78f23 BS |
940 | /* |
941 | * mm_update_next_owner() may clear mm->owner to NULL | |
942 | * if it races with swapoff, page migration, etc. | |
943 | * So this can be called with p == NULL. | |
944 | */ | |
945 | if (unlikely(!p)) | |
946 | return NULL; | |
947 | ||
073219e9 | 948 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 949 | } |
33398cf2 | 950 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 951 | |
d46eb14b SB |
952 | /** |
953 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
954 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
955 | * | |
956 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
957 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
958 | * returned. | |
959 | */ | |
960 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 961 | { |
d46eb14b SB |
962 | struct mem_cgroup *memcg; |
963 | ||
964 | if (mem_cgroup_disabled()) | |
965 | return NULL; | |
0b7f569e | 966 | |
54595fe2 KH |
967 | rcu_read_lock(); |
968 | do { | |
6f6acb00 MH |
969 | /* |
970 | * Page cache insertions can happen withou an | |
971 | * actual mm context, e.g. during disk probing | |
972 | * on boot, loopback IO, acct() writes etc. | |
973 | */ | |
974 | if (unlikely(!mm)) | |
df381975 | 975 | memcg = root_mem_cgroup; |
6f6acb00 MH |
976 | else { |
977 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
978 | if (unlikely(!memcg)) | |
979 | memcg = root_mem_cgroup; | |
980 | } | |
00d484f3 | 981 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 982 | rcu_read_unlock(); |
c0ff4b85 | 983 | return memcg; |
54595fe2 | 984 | } |
d46eb14b SB |
985 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
986 | ||
f745c6f5 SB |
987 | /** |
988 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
989 | * @page: page from which memcg should be extracted. | |
990 | * | |
991 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
992 | * root_mem_cgroup is returned. | |
993 | */ | |
994 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
995 | { | |
996 | struct mem_cgroup *memcg = page->mem_cgroup; | |
997 | ||
998 | if (mem_cgroup_disabled()) | |
999 | return NULL; | |
1000 | ||
1001 | rcu_read_lock(); | |
1002 | if (!memcg || !css_tryget_online(&memcg->css)) | |
1003 | memcg = root_mem_cgroup; | |
1004 | rcu_read_unlock(); | |
1005 | return memcg; | |
1006 | } | |
1007 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
1008 | ||
d46eb14b SB |
1009 | /** |
1010 | * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg. | |
1011 | */ | |
1012 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
1013 | { | |
1014 | if (unlikely(current->active_memcg)) { | |
1015 | struct mem_cgroup *memcg = root_mem_cgroup; | |
1016 | ||
1017 | rcu_read_lock(); | |
1018 | if (css_tryget_online(¤t->active_memcg->css)) | |
1019 | memcg = current->active_memcg; | |
1020 | rcu_read_unlock(); | |
1021 | return memcg; | |
1022 | } | |
1023 | return get_mem_cgroup_from_mm(current->mm); | |
1024 | } | |
54595fe2 | 1025 | |
5660048c JW |
1026 | /** |
1027 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
1028 | * @root: hierarchy root | |
1029 | * @prev: previously returned memcg, NULL on first invocation | |
1030 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
1031 | * | |
1032 | * Returns references to children of the hierarchy below @root, or | |
1033 | * @root itself, or %NULL after a full round-trip. | |
1034 | * | |
1035 | * Caller must pass the return value in @prev on subsequent | |
1036 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
1037 | * to cancel a hierarchy walk before the round-trip is complete. | |
1038 | * | |
b213b54f | 1039 | * Reclaimers can specify a node and a priority level in @reclaim to |
5660048c | 1040 | * divide up the memcgs in the hierarchy among all concurrent |
b213b54f | 1041 | * reclaimers operating on the same node and priority. |
5660048c | 1042 | */ |
694fbc0f | 1043 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1044 | struct mem_cgroup *prev, |
694fbc0f | 1045 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1046 | { |
33398cf2 | 1047 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 1048 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1049 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1050 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1051 | |
694fbc0f AM |
1052 | if (mem_cgroup_disabled()) |
1053 | return NULL; | |
5660048c | 1054 | |
9f3a0d09 JW |
1055 | if (!root) |
1056 | root = root_mem_cgroup; | |
7d74b06f | 1057 | |
9f3a0d09 | 1058 | if (prev && !reclaim) |
5ac8fb31 | 1059 | pos = prev; |
14067bb3 | 1060 | |
9f3a0d09 JW |
1061 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1062 | if (prev) | |
5ac8fb31 | 1063 | goto out; |
694fbc0f | 1064 | return root; |
9f3a0d09 | 1065 | } |
14067bb3 | 1066 | |
542f85f9 | 1067 | rcu_read_lock(); |
5f578161 | 1068 | |
5ac8fb31 | 1069 | if (reclaim) { |
ef8f2327 | 1070 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1071 | |
ef8f2327 | 1072 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
5ac8fb31 JW |
1073 | iter = &mz->iter[reclaim->priority]; |
1074 | ||
1075 | if (prev && reclaim->generation != iter->generation) | |
1076 | goto out_unlock; | |
1077 | ||
6df38689 | 1078 | while (1) { |
4db0c3c2 | 1079 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1080 | if (!pos || css_tryget(&pos->css)) |
1081 | break; | |
5ac8fb31 | 1082 | /* |
6df38689 VD |
1083 | * css reference reached zero, so iter->position will |
1084 | * be cleared by ->css_released. However, we should not | |
1085 | * rely on this happening soon, because ->css_released | |
1086 | * is called from a work queue, and by busy-waiting we | |
1087 | * might block it. So we clear iter->position right | |
1088 | * away. | |
5ac8fb31 | 1089 | */ |
6df38689 VD |
1090 | (void)cmpxchg(&iter->position, pos, NULL); |
1091 | } | |
5ac8fb31 JW |
1092 | } |
1093 | ||
1094 | if (pos) | |
1095 | css = &pos->css; | |
1096 | ||
1097 | for (;;) { | |
1098 | css = css_next_descendant_pre(css, &root->css); | |
1099 | if (!css) { | |
1100 | /* | |
1101 | * Reclaimers share the hierarchy walk, and a | |
1102 | * new one might jump in right at the end of | |
1103 | * the hierarchy - make sure they see at least | |
1104 | * one group and restart from the beginning. | |
1105 | */ | |
1106 | if (!prev) | |
1107 | continue; | |
1108 | break; | |
527a5ec9 | 1109 | } |
7d74b06f | 1110 | |
5ac8fb31 JW |
1111 | /* |
1112 | * Verify the css and acquire a reference. The root | |
1113 | * is provided by the caller, so we know it's alive | |
1114 | * and kicking, and don't take an extra reference. | |
1115 | */ | |
1116 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1117 | |
5ac8fb31 JW |
1118 | if (css == &root->css) |
1119 | break; | |
14067bb3 | 1120 | |
0b8f73e1 JW |
1121 | if (css_tryget(css)) |
1122 | break; | |
9f3a0d09 | 1123 | |
5ac8fb31 | 1124 | memcg = NULL; |
9f3a0d09 | 1125 | } |
5ac8fb31 JW |
1126 | |
1127 | if (reclaim) { | |
5ac8fb31 | 1128 | /* |
6df38689 VD |
1129 | * The position could have already been updated by a competing |
1130 | * thread, so check that the value hasn't changed since we read | |
1131 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1132 | */ |
6df38689 VD |
1133 | (void)cmpxchg(&iter->position, pos, memcg); |
1134 | ||
5ac8fb31 JW |
1135 | if (pos) |
1136 | css_put(&pos->css); | |
1137 | ||
1138 | if (!memcg) | |
1139 | iter->generation++; | |
1140 | else if (!prev) | |
1141 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1142 | } |
5ac8fb31 | 1143 | |
542f85f9 MH |
1144 | out_unlock: |
1145 | rcu_read_unlock(); | |
5ac8fb31 | 1146 | out: |
c40046f3 MH |
1147 | if (prev && prev != root) |
1148 | css_put(&prev->css); | |
1149 | ||
9f3a0d09 | 1150 | return memcg; |
14067bb3 | 1151 | } |
7d74b06f | 1152 | |
5660048c JW |
1153 | /** |
1154 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1155 | * @root: hierarchy root | |
1156 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1157 | */ | |
1158 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1159 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1160 | { |
1161 | if (!root) | |
1162 | root = root_mem_cgroup; | |
1163 | if (prev && prev != root) | |
1164 | css_put(&prev->css); | |
1165 | } | |
7d74b06f | 1166 | |
54a83d6b MC |
1167 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1168 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1169 | { |
6df38689 | 1170 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1171 | struct mem_cgroup_per_node *mz; |
1172 | int nid; | |
6df38689 VD |
1173 | int i; |
1174 | ||
54a83d6b MC |
1175 | for_each_node(nid) { |
1176 | mz = mem_cgroup_nodeinfo(from, nid); | |
1177 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
1178 | iter = &mz->iter[i]; | |
1179 | cmpxchg(&iter->position, | |
1180 | dead_memcg, NULL); | |
6df38689 VD |
1181 | } |
1182 | } | |
1183 | } | |
1184 | ||
54a83d6b MC |
1185 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1186 | { | |
1187 | struct mem_cgroup *memcg = dead_memcg; | |
1188 | struct mem_cgroup *last; | |
1189 | ||
1190 | do { | |
1191 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1192 | last = memcg; | |
1193 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1194 | ||
1195 | /* | |
1196 | * When cgruop1 non-hierarchy mode is used, | |
1197 | * parent_mem_cgroup() does not walk all the way up to the | |
1198 | * cgroup root (root_mem_cgroup). So we have to handle | |
1199 | * dead_memcg from cgroup root separately. | |
1200 | */ | |
1201 | if (last != root_mem_cgroup) | |
1202 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1203 | dead_memcg); | |
1204 | } | |
1205 | ||
7c5f64f8 VD |
1206 | /** |
1207 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1208 | * @memcg: hierarchy root | |
1209 | * @fn: function to call for each task | |
1210 | * @arg: argument passed to @fn | |
1211 | * | |
1212 | * This function iterates over tasks attached to @memcg or to any of its | |
1213 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1214 | * value, the function breaks the iteration loop and returns the value. | |
1215 | * Otherwise, it will iterate over all tasks and return 0. | |
1216 | * | |
1217 | * This function must not be called for the root memory cgroup. | |
1218 | */ | |
1219 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1220 | int (*fn)(struct task_struct *, void *), void *arg) | |
1221 | { | |
1222 | struct mem_cgroup *iter; | |
1223 | int ret = 0; | |
1224 | ||
1225 | BUG_ON(memcg == root_mem_cgroup); | |
1226 | ||
1227 | for_each_mem_cgroup_tree(iter, memcg) { | |
1228 | struct css_task_iter it; | |
1229 | struct task_struct *task; | |
1230 | ||
f168a9a5 | 1231 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1232 | while (!ret && (task = css_task_iter_next(&it))) |
1233 | ret = fn(task, arg); | |
1234 | css_task_iter_end(&it); | |
1235 | if (ret) { | |
1236 | mem_cgroup_iter_break(memcg, iter); | |
1237 | break; | |
1238 | } | |
1239 | } | |
1240 | return ret; | |
1241 | } | |
1242 | ||
925b7673 | 1243 | /** |
dfe0e773 | 1244 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1245 | * @page: the page |
f144c390 | 1246 | * @pgdat: pgdat of the page |
dfe0e773 JW |
1247 | * |
1248 | * This function is only safe when following the LRU page isolation | |
1249 | * and putback protocol: the LRU lock must be held, and the page must | |
1250 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1251 | */ |
599d0c95 | 1252 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1253 | { |
ef8f2327 | 1254 | struct mem_cgroup_per_node *mz; |
925b7673 | 1255 | struct mem_cgroup *memcg; |
bea8c150 | 1256 | struct lruvec *lruvec; |
6d12e2d8 | 1257 | |
bea8c150 | 1258 | if (mem_cgroup_disabled()) { |
599d0c95 | 1259 | lruvec = &pgdat->lruvec; |
bea8c150 HD |
1260 | goto out; |
1261 | } | |
925b7673 | 1262 | |
1306a85a | 1263 | memcg = page->mem_cgroup; |
7512102c | 1264 | /* |
dfe0e773 | 1265 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1266 | * possibly migrated - before they are charged. |
7512102c | 1267 | */ |
29833315 JW |
1268 | if (!memcg) |
1269 | memcg = root_mem_cgroup; | |
7512102c | 1270 | |
ef8f2327 | 1271 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1272 | lruvec = &mz->lruvec; |
1273 | out: | |
1274 | /* | |
1275 | * Since a node can be onlined after the mem_cgroup was created, | |
1276 | * we have to be prepared to initialize lruvec->zone here; | |
1277 | * and if offlined then reonlined, we need to reinitialize it. | |
1278 | */ | |
599d0c95 MG |
1279 | if (unlikely(lruvec->pgdat != pgdat)) |
1280 | lruvec->pgdat = pgdat; | |
bea8c150 | 1281 | return lruvec; |
08e552c6 | 1282 | } |
b69408e8 | 1283 | |
925b7673 | 1284 | /** |
fa9add64 HD |
1285 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1286 | * @lruvec: mem_cgroup per zone lru vector | |
1287 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1288 | * @zid: zone id of the accounted pages |
fa9add64 | 1289 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1290 | * |
ca707239 HD |
1291 | * This function must be called under lru_lock, just before a page is added |
1292 | * to or just after a page is removed from an lru list (that ordering being | |
1293 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1294 | */ |
fa9add64 | 1295 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1296 | int zid, int nr_pages) |
3f58a829 | 1297 | { |
ef8f2327 | 1298 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1299 | unsigned long *lru_size; |
ca707239 | 1300 | long size; |
3f58a829 MK |
1301 | |
1302 | if (mem_cgroup_disabled()) | |
1303 | return; | |
1304 | ||
ef8f2327 | 1305 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1306 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1307 | |
1308 | if (nr_pages < 0) | |
1309 | *lru_size += nr_pages; | |
1310 | ||
1311 | size = *lru_size; | |
b4536f0c MH |
1312 | if (WARN_ONCE(size < 0, |
1313 | "%s(%p, %d, %d): lru_size %ld\n", | |
1314 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1315 | VM_BUG_ON(1); |
1316 | *lru_size = 0; | |
1317 | } | |
1318 | ||
1319 | if (nr_pages > 0) | |
1320 | *lru_size += nr_pages; | |
08e552c6 | 1321 | } |
544122e5 | 1322 | |
19942822 | 1323 | /** |
9d11ea9f | 1324 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1325 | * @memcg: the memory cgroup |
19942822 | 1326 | * |
9d11ea9f | 1327 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1328 | * pages. |
19942822 | 1329 | */ |
c0ff4b85 | 1330 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1331 | { |
3e32cb2e JW |
1332 | unsigned long margin = 0; |
1333 | unsigned long count; | |
1334 | unsigned long limit; | |
9d11ea9f | 1335 | |
3e32cb2e | 1336 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1337 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1338 | if (count < limit) |
1339 | margin = limit - count; | |
1340 | ||
7941d214 | 1341 | if (do_memsw_account()) { |
3e32cb2e | 1342 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1343 | limit = READ_ONCE(memcg->memsw.max); |
3e32cb2e JW |
1344 | if (count <= limit) |
1345 | margin = min(margin, limit - count); | |
cbedbac3 LR |
1346 | else |
1347 | margin = 0; | |
3e32cb2e JW |
1348 | } |
1349 | ||
1350 | return margin; | |
19942822 JW |
1351 | } |
1352 | ||
32047e2a | 1353 | /* |
bdcbb659 | 1354 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1355 | * |
bdcbb659 QH |
1356 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1357 | * moving cgroups. This is for waiting at high-memory pressure | |
1358 | * caused by "move". | |
32047e2a | 1359 | */ |
c0ff4b85 | 1360 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1361 | { |
2bd9bb20 KH |
1362 | struct mem_cgroup *from; |
1363 | struct mem_cgroup *to; | |
4b534334 | 1364 | bool ret = false; |
2bd9bb20 KH |
1365 | /* |
1366 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1367 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1368 | */ | |
1369 | spin_lock(&mc.lock); | |
1370 | from = mc.from; | |
1371 | to = mc.to; | |
1372 | if (!from) | |
1373 | goto unlock; | |
3e92041d | 1374 | |
2314b42d JW |
1375 | ret = mem_cgroup_is_descendant(from, memcg) || |
1376 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1377 | unlock: |
1378 | spin_unlock(&mc.lock); | |
4b534334 KH |
1379 | return ret; |
1380 | } | |
1381 | ||
c0ff4b85 | 1382 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1383 | { |
1384 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1385 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1386 | DEFINE_WAIT(wait); |
1387 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1388 | /* moving charge context might have finished. */ | |
1389 | if (mc.moving_task) | |
1390 | schedule(); | |
1391 | finish_wait(&mc.waitq, &wait); | |
1392 | return true; | |
1393 | } | |
1394 | } | |
1395 | return false; | |
1396 | } | |
1397 | ||
c8713d0b JW |
1398 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1399 | { | |
1400 | struct seq_buf s; | |
1401 | int i; | |
71cd3113 | 1402 | |
c8713d0b JW |
1403 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1404 | if (!s.buffer) | |
1405 | return NULL; | |
1406 | ||
1407 | /* | |
1408 | * Provide statistics on the state of the memory subsystem as | |
1409 | * well as cumulative event counters that show past behavior. | |
1410 | * | |
1411 | * This list is ordered following a combination of these gradients: | |
1412 | * 1) generic big picture -> specifics and details | |
1413 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1414 | * | |
1415 | * Current memory state: | |
1416 | */ | |
1417 | ||
1418 | seq_buf_printf(&s, "anon %llu\n", | |
1419 | (u64)memcg_page_state(memcg, MEMCG_RSS) * | |
1420 | PAGE_SIZE); | |
1421 | seq_buf_printf(&s, "file %llu\n", | |
1422 | (u64)memcg_page_state(memcg, MEMCG_CACHE) * | |
1423 | PAGE_SIZE); | |
1424 | seq_buf_printf(&s, "kernel_stack %llu\n", | |
1425 | (u64)memcg_page_state(memcg, MEMCG_KERNEL_STACK_KB) * | |
1426 | 1024); | |
1427 | seq_buf_printf(&s, "slab %llu\n", | |
1428 | (u64)(memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) + | |
1429 | memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE)) * | |
1430 | PAGE_SIZE); | |
1431 | seq_buf_printf(&s, "sock %llu\n", | |
1432 | (u64)memcg_page_state(memcg, MEMCG_SOCK) * | |
1433 | PAGE_SIZE); | |
1434 | ||
1435 | seq_buf_printf(&s, "shmem %llu\n", | |
1436 | (u64)memcg_page_state(memcg, NR_SHMEM) * | |
1437 | PAGE_SIZE); | |
1438 | seq_buf_printf(&s, "file_mapped %llu\n", | |
1439 | (u64)memcg_page_state(memcg, NR_FILE_MAPPED) * | |
1440 | PAGE_SIZE); | |
1441 | seq_buf_printf(&s, "file_dirty %llu\n", | |
1442 | (u64)memcg_page_state(memcg, NR_FILE_DIRTY) * | |
1443 | PAGE_SIZE); | |
1444 | seq_buf_printf(&s, "file_writeback %llu\n", | |
1445 | (u64)memcg_page_state(memcg, NR_WRITEBACK) * | |
1446 | PAGE_SIZE); | |
1447 | ||
1448 | /* | |
1449 | * TODO: We should eventually replace our own MEMCG_RSS_HUGE counter | |
1450 | * with the NR_ANON_THP vm counter, but right now it's a pain in the | |
1451 | * arse because it requires migrating the work out of rmap to a place | |
1452 | * where the page->mem_cgroup is set up and stable. | |
1453 | */ | |
1454 | seq_buf_printf(&s, "anon_thp %llu\n", | |
1455 | (u64)memcg_page_state(memcg, MEMCG_RSS_HUGE) * | |
1456 | PAGE_SIZE); | |
1457 | ||
1458 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1459 | seq_buf_printf(&s, "%s %llu\n", mem_cgroup_lru_names[i], | |
1460 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * | |
1461 | PAGE_SIZE); | |
1462 | ||
1463 | seq_buf_printf(&s, "slab_reclaimable %llu\n", | |
1464 | (u64)memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) * | |
1465 | PAGE_SIZE); | |
1466 | seq_buf_printf(&s, "slab_unreclaimable %llu\n", | |
1467 | (u64)memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE) * | |
1468 | PAGE_SIZE); | |
1469 | ||
1470 | /* Accumulated memory events */ | |
1471 | ||
1472 | seq_buf_printf(&s, "pgfault %lu\n", memcg_events(memcg, PGFAULT)); | |
1473 | seq_buf_printf(&s, "pgmajfault %lu\n", memcg_events(memcg, PGMAJFAULT)); | |
1474 | ||
1475 | seq_buf_printf(&s, "workingset_refault %lu\n", | |
1476 | memcg_page_state(memcg, WORKINGSET_REFAULT)); | |
1477 | seq_buf_printf(&s, "workingset_activate %lu\n", | |
1478 | memcg_page_state(memcg, WORKINGSET_ACTIVATE)); | |
1479 | seq_buf_printf(&s, "workingset_nodereclaim %lu\n", | |
1480 | memcg_page_state(memcg, WORKINGSET_NODERECLAIM)); | |
1481 | ||
1482 | seq_buf_printf(&s, "pgrefill %lu\n", memcg_events(memcg, PGREFILL)); | |
1483 | seq_buf_printf(&s, "pgscan %lu\n", | |
1484 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1485 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1486 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1487 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1488 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
1489 | seq_buf_printf(&s, "pgactivate %lu\n", memcg_events(memcg, PGACTIVATE)); | |
1490 | seq_buf_printf(&s, "pgdeactivate %lu\n", memcg_events(memcg, PGDEACTIVATE)); | |
1491 | seq_buf_printf(&s, "pglazyfree %lu\n", memcg_events(memcg, PGLAZYFREE)); | |
1492 | seq_buf_printf(&s, "pglazyfreed %lu\n", memcg_events(memcg, PGLAZYFREED)); | |
1493 | ||
1494 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1495 | seq_buf_printf(&s, "thp_fault_alloc %lu\n", | |
1496 | memcg_events(memcg, THP_FAULT_ALLOC)); | |
1497 | seq_buf_printf(&s, "thp_collapse_alloc %lu\n", | |
1498 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); | |
1499 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1500 | ||
1501 | /* The above should easily fit into one page */ | |
1502 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1503 | ||
1504 | return s.buffer; | |
1505 | } | |
71cd3113 | 1506 | |
58cf188e | 1507 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1508 | /** |
f0c867d9 | 1509 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1510 | * memory controller. | |
e222432b BS |
1511 | * @memcg: The memory cgroup that went over limit |
1512 | * @p: Task that is going to be killed | |
1513 | * | |
1514 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1515 | * enabled | |
1516 | */ | |
f0c867d9 | 1517 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1518 | { |
e222432b BS |
1519 | rcu_read_lock(); |
1520 | ||
f0c867d9 | 1521 | if (memcg) { |
1522 | pr_cont(",oom_memcg="); | |
1523 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1524 | } else | |
1525 | pr_cont(",global_oom"); | |
2415b9f5 | 1526 | if (p) { |
f0c867d9 | 1527 | pr_cont(",task_memcg="); |
2415b9f5 | 1528 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1529 | } |
e222432b | 1530 | rcu_read_unlock(); |
f0c867d9 | 1531 | } |
1532 | ||
1533 | /** | |
1534 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1535 | * memory controller. | |
1536 | * @memcg: The memory cgroup that went over limit | |
1537 | */ | |
1538 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1539 | { | |
c8713d0b | 1540 | char *buf; |
e222432b | 1541 | |
3e32cb2e JW |
1542 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1543 | K((u64)page_counter_read(&memcg->memory)), | |
bbec2e15 | 1544 | K((u64)memcg->memory.max), memcg->memory.failcnt); |
c8713d0b JW |
1545 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1546 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1547 | K((u64)page_counter_read(&memcg->swap)), | |
1548 | K((u64)memcg->swap.max), memcg->swap.failcnt); | |
1549 | else { | |
1550 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1551 | K((u64)page_counter_read(&memcg->memsw)), | |
1552 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1553 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1554 | K((u64)page_counter_read(&memcg->kmem)), | |
1555 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1556 | } |
c8713d0b JW |
1557 | |
1558 | pr_info("Memory cgroup stats for "); | |
1559 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1560 | pr_cont(":"); | |
1561 | buf = memory_stat_format(memcg); | |
1562 | if (!buf) | |
1563 | return; | |
1564 | pr_info("%s", buf); | |
1565 | kfree(buf); | |
e222432b BS |
1566 | } |
1567 | ||
a63d83f4 DR |
1568 | /* |
1569 | * Return the memory (and swap, if configured) limit for a memcg. | |
1570 | */ | |
bbec2e15 | 1571 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1572 | { |
bbec2e15 | 1573 | unsigned long max; |
f3e8eb70 | 1574 | |
bbec2e15 | 1575 | max = memcg->memory.max; |
9a5a8f19 | 1576 | if (mem_cgroup_swappiness(memcg)) { |
bbec2e15 RG |
1577 | unsigned long memsw_max; |
1578 | unsigned long swap_max; | |
9a5a8f19 | 1579 | |
bbec2e15 RG |
1580 | memsw_max = memcg->memsw.max; |
1581 | swap_max = memcg->swap.max; | |
1582 | swap_max = min(swap_max, (unsigned long)total_swap_pages); | |
1583 | max = min(max + swap_max, memsw_max); | |
9a5a8f19 | 1584 | } |
bbec2e15 | 1585 | return max; |
a63d83f4 DR |
1586 | } |
1587 | ||
9783aa99 CD |
1588 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1589 | { | |
1590 | return page_counter_read(&memcg->memory); | |
1591 | } | |
1592 | ||
b6e6edcf | 1593 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1594 | int order) |
9cbb78bb | 1595 | { |
6e0fc46d DR |
1596 | struct oom_control oc = { |
1597 | .zonelist = NULL, | |
1598 | .nodemask = NULL, | |
2a966b77 | 1599 | .memcg = memcg, |
6e0fc46d DR |
1600 | .gfp_mask = gfp_mask, |
1601 | .order = order, | |
6e0fc46d | 1602 | }; |
7c5f64f8 | 1603 | bool ret; |
9cbb78bb | 1604 | |
7775face TH |
1605 | if (mutex_lock_killable(&oom_lock)) |
1606 | return true; | |
1607 | /* | |
1608 | * A few threads which were not waiting at mutex_lock_killable() can | |
1609 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1610 | */ | |
1611 | ret = should_force_charge() || out_of_memory(&oc); | |
dc56401f | 1612 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1613 | return ret; |
9cbb78bb DR |
1614 | } |
1615 | ||
ae6e71d3 MC |
1616 | #if MAX_NUMNODES > 1 |
1617 | ||
4d0c066d KH |
1618 | /** |
1619 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1620 | * @memcg: the target memcg |
4d0c066d KH |
1621 | * @nid: the node ID to be checked. |
1622 | * @noswap : specify true here if the user wants flle only information. | |
1623 | * | |
1624 | * This function returns whether the specified memcg contains any | |
1625 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1626 | * pages in the node. | |
1627 | */ | |
c0ff4b85 | 1628 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1629 | int nid, bool noswap) |
1630 | { | |
2b487e59 JW |
1631 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); |
1632 | ||
def0fdae JW |
1633 | if (lruvec_page_state(lruvec, NR_INACTIVE_FILE) || |
1634 | lruvec_page_state(lruvec, NR_ACTIVE_FILE)) | |
4d0c066d KH |
1635 | return true; |
1636 | if (noswap || !total_swap_pages) | |
1637 | return false; | |
def0fdae JW |
1638 | if (lruvec_page_state(lruvec, NR_INACTIVE_ANON) || |
1639 | lruvec_page_state(lruvec, NR_ACTIVE_ANON)) | |
4d0c066d KH |
1640 | return true; |
1641 | return false; | |
1642 | ||
1643 | } | |
889976db YH |
1644 | |
1645 | /* | |
1646 | * Always updating the nodemask is not very good - even if we have an empty | |
1647 | * list or the wrong list here, we can start from some node and traverse all | |
1648 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1649 | * | |
1650 | */ | |
c0ff4b85 | 1651 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1652 | { |
1653 | int nid; | |
453a9bf3 KH |
1654 | /* |
1655 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1656 | * pagein/pageout changes since the last update. | |
1657 | */ | |
c0ff4b85 | 1658 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1659 | return; |
c0ff4b85 | 1660 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1661 | return; |
1662 | ||
889976db | 1663 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1664 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1665 | |
31aaea4a | 1666 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1667 | |
c0ff4b85 R |
1668 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1669 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1670 | } |
453a9bf3 | 1671 | |
c0ff4b85 R |
1672 | atomic_set(&memcg->numainfo_events, 0); |
1673 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1674 | } |
1675 | ||
1676 | /* | |
1677 | * Selecting a node where we start reclaim from. Because what we need is just | |
1678 | * reducing usage counter, start from anywhere is O,K. Considering | |
1679 | * memory reclaim from current node, there are pros. and cons. | |
1680 | * | |
1681 | * Freeing memory from current node means freeing memory from a node which | |
1682 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1683 | * hit limits, it will see a contention on a node. But freeing from remote | |
1684 | * node means more costs for memory reclaim because of memory latency. | |
1685 | * | |
1686 | * Now, we use round-robin. Better algorithm is welcomed. | |
1687 | */ | |
c0ff4b85 | 1688 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1689 | { |
1690 | int node; | |
1691 | ||
c0ff4b85 R |
1692 | mem_cgroup_may_update_nodemask(memcg); |
1693 | node = memcg->last_scanned_node; | |
889976db | 1694 | |
0edaf86c | 1695 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1696 | /* |
fda3d69b MH |
1697 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1698 | * last time it really checked all the LRUs due to rate limiting. | |
1699 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1700 | */ |
1701 | if (unlikely(node == MAX_NUMNODES)) | |
1702 | node = numa_node_id(); | |
1703 | ||
c0ff4b85 | 1704 | memcg->last_scanned_node = node; |
889976db YH |
1705 | return node; |
1706 | } | |
889976db | 1707 | #else |
c0ff4b85 | 1708 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1709 | { |
1710 | return 0; | |
1711 | } | |
1712 | #endif | |
1713 | ||
0608f43d | 1714 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1715 | pg_data_t *pgdat, |
0608f43d AM |
1716 | gfp_t gfp_mask, |
1717 | unsigned long *total_scanned) | |
1718 | { | |
1719 | struct mem_cgroup *victim = NULL; | |
1720 | int total = 0; | |
1721 | int loop = 0; | |
1722 | unsigned long excess; | |
1723 | unsigned long nr_scanned; | |
1724 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1725 | .pgdat = pgdat, |
0608f43d AM |
1726 | .priority = 0, |
1727 | }; | |
1728 | ||
3e32cb2e | 1729 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1730 | |
1731 | while (1) { | |
1732 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1733 | if (!victim) { | |
1734 | loop++; | |
1735 | if (loop >= 2) { | |
1736 | /* | |
1737 | * If we have not been able to reclaim | |
1738 | * anything, it might because there are | |
1739 | * no reclaimable pages under this hierarchy | |
1740 | */ | |
1741 | if (!total) | |
1742 | break; | |
1743 | /* | |
1744 | * We want to do more targeted reclaim. | |
1745 | * excess >> 2 is not to excessive so as to | |
1746 | * reclaim too much, nor too less that we keep | |
1747 | * coming back to reclaim from this cgroup | |
1748 | */ | |
1749 | if (total >= (excess >> 2) || | |
1750 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1751 | break; | |
1752 | } | |
1753 | continue; | |
1754 | } | |
a9dd0a83 | 1755 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1756 | pgdat, &nr_scanned); |
0608f43d | 1757 | *total_scanned += nr_scanned; |
3e32cb2e | 1758 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1759 | break; |
6d61ef40 | 1760 | } |
0608f43d AM |
1761 | mem_cgroup_iter_break(root_memcg, victim); |
1762 | return total; | |
6d61ef40 BS |
1763 | } |
1764 | ||
0056f4e6 JW |
1765 | #ifdef CONFIG_LOCKDEP |
1766 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1767 | .name = "memcg_oom_lock", | |
1768 | }; | |
1769 | #endif | |
1770 | ||
fb2a6fc5 JW |
1771 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1772 | ||
867578cb KH |
1773 | /* |
1774 | * Check OOM-Killer is already running under our hierarchy. | |
1775 | * If someone is running, return false. | |
1776 | */ | |
fb2a6fc5 | 1777 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1778 | { |
79dfdacc | 1779 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1780 | |
fb2a6fc5 JW |
1781 | spin_lock(&memcg_oom_lock); |
1782 | ||
9f3a0d09 | 1783 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1784 | if (iter->oom_lock) { |
79dfdacc MH |
1785 | /* |
1786 | * this subtree of our hierarchy is already locked | |
1787 | * so we cannot give a lock. | |
1788 | */ | |
79dfdacc | 1789 | failed = iter; |
9f3a0d09 JW |
1790 | mem_cgroup_iter_break(memcg, iter); |
1791 | break; | |
23751be0 JW |
1792 | } else |
1793 | iter->oom_lock = true; | |
7d74b06f | 1794 | } |
867578cb | 1795 | |
fb2a6fc5 JW |
1796 | if (failed) { |
1797 | /* | |
1798 | * OK, we failed to lock the whole subtree so we have | |
1799 | * to clean up what we set up to the failing subtree | |
1800 | */ | |
1801 | for_each_mem_cgroup_tree(iter, memcg) { | |
1802 | if (iter == failed) { | |
1803 | mem_cgroup_iter_break(memcg, iter); | |
1804 | break; | |
1805 | } | |
1806 | iter->oom_lock = false; | |
79dfdacc | 1807 | } |
0056f4e6 JW |
1808 | } else |
1809 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1810 | |
1811 | spin_unlock(&memcg_oom_lock); | |
1812 | ||
1813 | return !failed; | |
a636b327 | 1814 | } |
0b7f569e | 1815 | |
fb2a6fc5 | 1816 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1817 | { |
7d74b06f KH |
1818 | struct mem_cgroup *iter; |
1819 | ||
fb2a6fc5 | 1820 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1821 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1822 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1823 | iter->oom_lock = false; |
fb2a6fc5 | 1824 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1825 | } |
1826 | ||
c0ff4b85 | 1827 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1828 | { |
1829 | struct mem_cgroup *iter; | |
1830 | ||
c2b42d3c | 1831 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1832 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1833 | iter->under_oom++; |
1834 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1835 | } |
1836 | ||
c0ff4b85 | 1837 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1838 | { |
1839 | struct mem_cgroup *iter; | |
1840 | ||
867578cb KH |
1841 | /* |
1842 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1843 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1844 | */ |
c2b42d3c | 1845 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1846 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1847 | if (iter->under_oom > 0) |
1848 | iter->under_oom--; | |
1849 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1850 | } |
1851 | ||
867578cb KH |
1852 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1853 | ||
dc98df5a | 1854 | struct oom_wait_info { |
d79154bb | 1855 | struct mem_cgroup *memcg; |
ac6424b9 | 1856 | wait_queue_entry_t wait; |
dc98df5a KH |
1857 | }; |
1858 | ||
ac6424b9 | 1859 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1860 | unsigned mode, int sync, void *arg) |
1861 | { | |
d79154bb HD |
1862 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1863 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1864 | struct oom_wait_info *oom_wait_info; |
1865 | ||
1866 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1867 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1868 | |
2314b42d JW |
1869 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1870 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1871 | return 0; |
dc98df5a KH |
1872 | return autoremove_wake_function(wait, mode, sync, arg); |
1873 | } | |
1874 | ||
c0ff4b85 | 1875 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1876 | { |
c2b42d3c TH |
1877 | /* |
1878 | * For the following lockless ->under_oom test, the only required | |
1879 | * guarantee is that it must see the state asserted by an OOM when | |
1880 | * this function is called as a result of userland actions | |
1881 | * triggered by the notification of the OOM. This is trivially | |
1882 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1883 | * triggering notification. | |
1884 | */ | |
1885 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1886 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1887 | } |
1888 | ||
29ef680a MH |
1889 | enum oom_status { |
1890 | OOM_SUCCESS, | |
1891 | OOM_FAILED, | |
1892 | OOM_ASYNC, | |
1893 | OOM_SKIPPED | |
1894 | }; | |
1895 | ||
1896 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1897 | { |
7056d3a3 MH |
1898 | enum oom_status ret; |
1899 | bool locked; | |
1900 | ||
29ef680a MH |
1901 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1902 | return OOM_SKIPPED; | |
1903 | ||
7a1adfdd RG |
1904 | memcg_memory_event(memcg, MEMCG_OOM); |
1905 | ||
867578cb | 1906 | /* |
49426420 JW |
1907 | * We are in the middle of the charge context here, so we |
1908 | * don't want to block when potentially sitting on a callstack | |
1909 | * that holds all kinds of filesystem and mm locks. | |
1910 | * | |
29ef680a MH |
1911 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1912 | * handling until the charge can succeed; remember the context and put | |
1913 | * the task to sleep at the end of the page fault when all locks are | |
1914 | * released. | |
49426420 | 1915 | * |
29ef680a MH |
1916 | * On the other hand, in-kernel OOM killer allows for an async victim |
1917 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1918 | * relying on the oom victim to make a forward progress and we can | |
1919 | * invoke the oom killer here. | |
1920 | * | |
1921 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1922 | * victim and then we have to bail out from the charge path. | |
867578cb | 1923 | */ |
29ef680a MH |
1924 | if (memcg->oom_kill_disable) { |
1925 | if (!current->in_user_fault) | |
1926 | return OOM_SKIPPED; | |
1927 | css_get(&memcg->css); | |
1928 | current->memcg_in_oom = memcg; | |
1929 | current->memcg_oom_gfp_mask = mask; | |
1930 | current->memcg_oom_order = order; | |
1931 | ||
1932 | return OOM_ASYNC; | |
1933 | } | |
1934 | ||
7056d3a3 MH |
1935 | mem_cgroup_mark_under_oom(memcg); |
1936 | ||
1937 | locked = mem_cgroup_oom_trylock(memcg); | |
1938 | ||
1939 | if (locked) | |
1940 | mem_cgroup_oom_notify(memcg); | |
1941 | ||
1942 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1943 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1944 | ret = OOM_SUCCESS; |
1945 | else | |
1946 | ret = OOM_FAILED; | |
1947 | ||
1948 | if (locked) | |
1949 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1950 | |
7056d3a3 | 1951 | return ret; |
3812c8c8 JW |
1952 | } |
1953 | ||
1954 | /** | |
1955 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1956 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1957 | * |
49426420 JW |
1958 | * This has to be called at the end of a page fault if the memcg OOM |
1959 | * handler was enabled. | |
3812c8c8 | 1960 | * |
49426420 | 1961 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1962 | * sleep on a waitqueue until the userspace task resolves the |
1963 | * situation. Sleeping directly in the charge context with all kinds | |
1964 | * of locks held is not a good idea, instead we remember an OOM state | |
1965 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1966 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1967 | * |
1968 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1969 | * completed, %false otherwise. |
3812c8c8 | 1970 | */ |
49426420 | 1971 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1972 | { |
626ebc41 | 1973 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1974 | struct oom_wait_info owait; |
49426420 | 1975 | bool locked; |
3812c8c8 JW |
1976 | |
1977 | /* OOM is global, do not handle */ | |
3812c8c8 | 1978 | if (!memcg) |
49426420 | 1979 | return false; |
3812c8c8 | 1980 | |
7c5f64f8 | 1981 | if (!handle) |
49426420 | 1982 | goto cleanup; |
3812c8c8 JW |
1983 | |
1984 | owait.memcg = memcg; | |
1985 | owait.wait.flags = 0; | |
1986 | owait.wait.func = memcg_oom_wake_function; | |
1987 | owait.wait.private = current; | |
2055da97 | 1988 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1989 | |
3812c8c8 | 1990 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1991 | mem_cgroup_mark_under_oom(memcg); |
1992 | ||
1993 | locked = mem_cgroup_oom_trylock(memcg); | |
1994 | ||
1995 | if (locked) | |
1996 | mem_cgroup_oom_notify(memcg); | |
1997 | ||
1998 | if (locked && !memcg->oom_kill_disable) { | |
1999 | mem_cgroup_unmark_under_oom(memcg); | |
2000 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
2001 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
2002 | current->memcg_oom_order); | |
49426420 | 2003 | } else { |
3812c8c8 | 2004 | schedule(); |
49426420 JW |
2005 | mem_cgroup_unmark_under_oom(memcg); |
2006 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
2007 | } | |
2008 | ||
2009 | if (locked) { | |
fb2a6fc5 JW |
2010 | mem_cgroup_oom_unlock(memcg); |
2011 | /* | |
2012 | * There is no guarantee that an OOM-lock contender | |
2013 | * sees the wakeups triggered by the OOM kill | |
2014 | * uncharges. Wake any sleepers explicitely. | |
2015 | */ | |
2016 | memcg_oom_recover(memcg); | |
2017 | } | |
49426420 | 2018 | cleanup: |
626ebc41 | 2019 | current->memcg_in_oom = NULL; |
3812c8c8 | 2020 | css_put(&memcg->css); |
867578cb | 2021 | return true; |
0b7f569e KH |
2022 | } |
2023 | ||
3d8b38eb RG |
2024 | /** |
2025 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
2026 | * @victim: task to be killed by the OOM killer | |
2027 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
2028 | * | |
2029 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
2030 | * by killing all belonging OOM-killable tasks. | |
2031 | * | |
2032 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
2033 | */ | |
2034 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
2035 | struct mem_cgroup *oom_domain) | |
2036 | { | |
2037 | struct mem_cgroup *oom_group = NULL; | |
2038 | struct mem_cgroup *memcg; | |
2039 | ||
2040 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2041 | return NULL; | |
2042 | ||
2043 | if (!oom_domain) | |
2044 | oom_domain = root_mem_cgroup; | |
2045 | ||
2046 | rcu_read_lock(); | |
2047 | ||
2048 | memcg = mem_cgroup_from_task(victim); | |
2049 | if (memcg == root_mem_cgroup) | |
2050 | goto out; | |
2051 | ||
2052 | /* | |
2053 | * Traverse the memory cgroup hierarchy from the victim task's | |
2054 | * cgroup up to the OOMing cgroup (or root) to find the | |
2055 | * highest-level memory cgroup with oom.group set. | |
2056 | */ | |
2057 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
2058 | if (memcg->oom_group) | |
2059 | oom_group = memcg; | |
2060 | ||
2061 | if (memcg == oom_domain) | |
2062 | break; | |
2063 | } | |
2064 | ||
2065 | if (oom_group) | |
2066 | css_get(&oom_group->css); | |
2067 | out: | |
2068 | rcu_read_unlock(); | |
2069 | ||
2070 | return oom_group; | |
2071 | } | |
2072 | ||
2073 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2074 | { | |
2075 | pr_info("Tasks in "); | |
2076 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2077 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2078 | } | |
2079 | ||
d7365e78 | 2080 | /** |
81f8c3a4 JW |
2081 | * lock_page_memcg - lock a page->mem_cgroup binding |
2082 | * @page: the page | |
32047e2a | 2083 | * |
81f8c3a4 | 2084 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
2085 | * another cgroup. |
2086 | * | |
2087 | * It ensures lifetime of the returned memcg. Caller is responsible | |
2088 | * for the lifetime of the page; __unlock_page_memcg() is available | |
2089 | * when @page might get freed inside the locked section. | |
d69b042f | 2090 | */ |
739f79fc | 2091 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 KH |
2092 | { |
2093 | struct mem_cgroup *memcg; | |
6de22619 | 2094 | unsigned long flags; |
89c06bd5 | 2095 | |
6de22619 JW |
2096 | /* |
2097 | * The RCU lock is held throughout the transaction. The fast | |
2098 | * path can get away without acquiring the memcg->move_lock | |
2099 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
2100 | * |
2101 | * The RCU lock also protects the memcg from being freed when | |
2102 | * the page state that is going to change is the only thing | |
2103 | * preventing the page itself from being freed. E.g. writeback | |
2104 | * doesn't hold a page reference and relies on PG_writeback to | |
2105 | * keep off truncation, migration and so forth. | |
2106 | */ | |
d7365e78 JW |
2107 | rcu_read_lock(); |
2108 | ||
2109 | if (mem_cgroup_disabled()) | |
739f79fc | 2110 | return NULL; |
89c06bd5 | 2111 | again: |
1306a85a | 2112 | memcg = page->mem_cgroup; |
29833315 | 2113 | if (unlikely(!memcg)) |
739f79fc | 2114 | return NULL; |
d7365e78 | 2115 | |
bdcbb659 | 2116 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 2117 | return memcg; |
89c06bd5 | 2118 | |
6de22619 | 2119 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 2120 | if (memcg != page->mem_cgroup) { |
6de22619 | 2121 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2122 | goto again; |
2123 | } | |
6de22619 JW |
2124 | |
2125 | /* | |
2126 | * When charge migration first begins, we can have locked and | |
2127 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 2128 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
2129 | */ |
2130 | memcg->move_lock_task = current; | |
2131 | memcg->move_lock_flags = flags; | |
d7365e78 | 2132 | |
739f79fc | 2133 | return memcg; |
89c06bd5 | 2134 | } |
81f8c3a4 | 2135 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2136 | |
d7365e78 | 2137 | /** |
739f79fc JW |
2138 | * __unlock_page_memcg - unlock and unpin a memcg |
2139 | * @memcg: the memcg | |
2140 | * | |
2141 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2142 | */ |
739f79fc | 2143 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2144 | { |
6de22619 JW |
2145 | if (memcg && memcg->move_lock_task == current) { |
2146 | unsigned long flags = memcg->move_lock_flags; | |
2147 | ||
2148 | memcg->move_lock_task = NULL; | |
2149 | memcg->move_lock_flags = 0; | |
2150 | ||
2151 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2152 | } | |
89c06bd5 | 2153 | |
d7365e78 | 2154 | rcu_read_unlock(); |
89c06bd5 | 2155 | } |
739f79fc JW |
2156 | |
2157 | /** | |
2158 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
2159 | * @page: the page | |
2160 | */ | |
2161 | void unlock_page_memcg(struct page *page) | |
2162 | { | |
2163 | __unlock_page_memcg(page->mem_cgroup); | |
2164 | } | |
81f8c3a4 | 2165 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2166 | |
cdec2e42 KH |
2167 | struct memcg_stock_pcp { |
2168 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2169 | unsigned int nr_pages; |
cdec2e42 | 2170 | struct work_struct work; |
26fe6168 | 2171 | unsigned long flags; |
a0db00fc | 2172 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2173 | }; |
2174 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2175 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2176 | |
a0956d54 SS |
2177 | /** |
2178 | * consume_stock: Try to consume stocked charge on this cpu. | |
2179 | * @memcg: memcg to consume from. | |
2180 | * @nr_pages: how many pages to charge. | |
2181 | * | |
2182 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2183 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2184 | * service an allocation will refill the stock. | |
2185 | * | |
2186 | * returns true if successful, false otherwise. | |
cdec2e42 | 2187 | */ |
a0956d54 | 2188 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2189 | { |
2190 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2191 | unsigned long flags; |
3e32cb2e | 2192 | bool ret = false; |
cdec2e42 | 2193 | |
a983b5eb | 2194 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2195 | return ret; |
a0956d54 | 2196 | |
db2ba40c JW |
2197 | local_irq_save(flags); |
2198 | ||
2199 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2200 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2201 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2202 | ret = true; |
2203 | } | |
db2ba40c JW |
2204 | |
2205 | local_irq_restore(flags); | |
2206 | ||
cdec2e42 KH |
2207 | return ret; |
2208 | } | |
2209 | ||
2210 | /* | |
3e32cb2e | 2211 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2212 | */ |
2213 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2214 | { | |
2215 | struct mem_cgroup *old = stock->cached; | |
2216 | ||
11c9ea4e | 2217 | if (stock->nr_pages) { |
3e32cb2e | 2218 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2219 | if (do_memsw_account()) |
3e32cb2e | 2220 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2221 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2222 | stock->nr_pages = 0; |
cdec2e42 KH |
2223 | } |
2224 | stock->cached = NULL; | |
cdec2e42 KH |
2225 | } |
2226 | ||
cdec2e42 KH |
2227 | static void drain_local_stock(struct work_struct *dummy) |
2228 | { | |
db2ba40c JW |
2229 | struct memcg_stock_pcp *stock; |
2230 | unsigned long flags; | |
2231 | ||
72f0184c MH |
2232 | /* |
2233 | * The only protection from memory hotplug vs. drain_stock races is | |
2234 | * that we always operate on local CPU stock here with IRQ disabled | |
2235 | */ | |
db2ba40c JW |
2236 | local_irq_save(flags); |
2237 | ||
2238 | stock = this_cpu_ptr(&memcg_stock); | |
cdec2e42 | 2239 | drain_stock(stock); |
26fe6168 | 2240 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2241 | |
2242 | local_irq_restore(flags); | |
cdec2e42 KH |
2243 | } |
2244 | ||
2245 | /* | |
3e32cb2e | 2246 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2247 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2248 | */ |
c0ff4b85 | 2249 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2250 | { |
db2ba40c JW |
2251 | struct memcg_stock_pcp *stock; |
2252 | unsigned long flags; | |
2253 | ||
2254 | local_irq_save(flags); | |
cdec2e42 | 2255 | |
db2ba40c | 2256 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2257 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2258 | drain_stock(stock); |
c0ff4b85 | 2259 | stock->cached = memcg; |
cdec2e42 | 2260 | } |
11c9ea4e | 2261 | stock->nr_pages += nr_pages; |
db2ba40c | 2262 | |
a983b5eb | 2263 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2264 | drain_stock(stock); |
2265 | ||
db2ba40c | 2266 | local_irq_restore(flags); |
cdec2e42 KH |
2267 | } |
2268 | ||
2269 | /* | |
c0ff4b85 | 2270 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2271 | * of the hierarchy under it. |
cdec2e42 | 2272 | */ |
6d3d6aa2 | 2273 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2274 | { |
26fe6168 | 2275 | int cpu, curcpu; |
d38144b7 | 2276 | |
6d3d6aa2 JW |
2277 | /* If someone's already draining, avoid adding running more workers. */ |
2278 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2279 | return; | |
72f0184c MH |
2280 | /* |
2281 | * Notify other cpus that system-wide "drain" is running | |
2282 | * We do not care about races with the cpu hotplug because cpu down | |
2283 | * as well as workers from this path always operate on the local | |
2284 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2285 | */ | |
5af12d0e | 2286 | curcpu = get_cpu(); |
cdec2e42 KH |
2287 | for_each_online_cpu(cpu) { |
2288 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2289 | struct mem_cgroup *memcg; |
e1a366be | 2290 | bool flush = false; |
26fe6168 | 2291 | |
e1a366be | 2292 | rcu_read_lock(); |
c0ff4b85 | 2293 | memcg = stock->cached; |
e1a366be RG |
2294 | if (memcg && stock->nr_pages && |
2295 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2296 | flush = true; | |
2297 | rcu_read_unlock(); | |
2298 | ||
2299 | if (flush && | |
2300 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2301 | if (cpu == curcpu) |
2302 | drain_local_stock(&stock->work); | |
2303 | else | |
2304 | schedule_work_on(cpu, &stock->work); | |
2305 | } | |
cdec2e42 | 2306 | } |
5af12d0e | 2307 | put_cpu(); |
9f50fad6 | 2308 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2309 | } |
2310 | ||
308167fc | 2311 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2312 | { |
cdec2e42 | 2313 | struct memcg_stock_pcp *stock; |
42a30035 | 2314 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2315 | |
cdec2e42 KH |
2316 | stock = &per_cpu(memcg_stock, cpu); |
2317 | drain_stock(stock); | |
a983b5eb JW |
2318 | |
2319 | for_each_mem_cgroup(memcg) { | |
2320 | int i; | |
2321 | ||
2322 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2323 | int nid; | |
2324 | long x; | |
2325 | ||
871789d4 | 2326 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
815744d7 | 2327 | if (x) |
42a30035 JW |
2328 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2329 | atomic_long_add(x, &memcg->vmstats[i]); | |
a983b5eb JW |
2330 | |
2331 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2332 | continue; | |
2333 | ||
2334 | for_each_node(nid) { | |
2335 | struct mem_cgroup_per_node *pn; | |
2336 | ||
2337 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2338 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
815744d7 | 2339 | if (x) |
42a30035 JW |
2340 | do { |
2341 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2342 | } while ((pn = parent_nodeinfo(pn, nid))); | |
a983b5eb JW |
2343 | } |
2344 | } | |
2345 | ||
e27be240 | 2346 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2347 | long x; |
2348 | ||
871789d4 | 2349 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
815744d7 | 2350 | if (x) |
42a30035 JW |
2351 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2352 | atomic_long_add(x, &memcg->vmevents[i]); | |
a983b5eb JW |
2353 | } |
2354 | } | |
2355 | ||
308167fc | 2356 | return 0; |
cdec2e42 KH |
2357 | } |
2358 | ||
f7e1cb6e JW |
2359 | static void reclaim_high(struct mem_cgroup *memcg, |
2360 | unsigned int nr_pages, | |
2361 | gfp_t gfp_mask) | |
2362 | { | |
2363 | do { | |
2364 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
2365 | continue; | |
e27be240 | 2366 | memcg_memory_event(memcg, MEMCG_HIGH); |
f7e1cb6e JW |
2367 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
2368 | } while ((memcg = parent_mem_cgroup(memcg))); | |
2369 | } | |
2370 | ||
2371 | static void high_work_func(struct work_struct *work) | |
2372 | { | |
2373 | struct mem_cgroup *memcg; | |
2374 | ||
2375 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2376 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2377 | } |
2378 | ||
0e4b01df CD |
2379 | /* |
2380 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2381 | * enough to still cause a significant slowdown in most cases, while still | |
2382 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2383 | */ | |
2384 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2385 | ||
2386 | /* | |
2387 | * When calculating the delay, we use these either side of the exponentiation to | |
2388 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2389 | * below. | |
2390 | * | |
2391 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2392 | * overage ratio to a delay. | |
2393 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down down the | |
2394 | * proposed penalty in order to reduce to a reasonable number of jiffies, and | |
2395 | * to produce a reasonable delay curve. | |
2396 | * | |
2397 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2398 | * reasonable delay curve compared to precision-adjusted overage, not | |
2399 | * penalising heavily at first, but still making sure that growth beyond the | |
2400 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2401 | * example, with a high of 100 megabytes: | |
2402 | * | |
2403 | * +-------+------------------------+ | |
2404 | * | usage | time to allocate in ms | | |
2405 | * +-------+------------------------+ | |
2406 | * | 100M | 0 | | |
2407 | * | 101M | 6 | | |
2408 | * | 102M | 25 | | |
2409 | * | 103M | 57 | | |
2410 | * | 104M | 102 | | |
2411 | * | 105M | 159 | | |
2412 | * | 106M | 230 | | |
2413 | * | 107M | 313 | | |
2414 | * | 108M | 409 | | |
2415 | * | 109M | 518 | | |
2416 | * | 110M | 639 | | |
2417 | * | 111M | 774 | | |
2418 | * | 112M | 921 | | |
2419 | * | 113M | 1081 | | |
2420 | * | 114M | 1254 | | |
2421 | * | 115M | 1439 | | |
2422 | * | 116M | 1638 | | |
2423 | * | 117M | 1849 | | |
2424 | * | 118M | 2000 | | |
2425 | * | 119M | 2000 | | |
2426 | * | 120M | 2000 | | |
2427 | * +-------+------------------------+ | |
2428 | */ | |
2429 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2430 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2431 | ||
b23afb93 | 2432 | /* |
86c8665d CD |
2433 | * Get the number of jiffies that we should penalise a mischievous cgroup which |
2434 | * is exceeding its memory.high by checking both it and its ancestors. | |
b23afb93 | 2435 | */ |
86c8665d CD |
2436 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, |
2437 | unsigned int nr_pages) | |
b23afb93 | 2438 | { |
86c8665d CD |
2439 | unsigned long penalty_jiffies; |
2440 | u64 max_overage = 0; | |
b23afb93 | 2441 | |
86c8665d CD |
2442 | do { |
2443 | unsigned long usage, high; | |
2444 | u64 overage; | |
b23afb93 | 2445 | |
86c8665d CD |
2446 | usage = page_counter_read(&memcg->memory); |
2447 | high = READ_ONCE(memcg->high); | |
2448 | ||
673e5c5c JK |
2449 | if (usage <= high) |
2450 | continue; | |
2451 | ||
86c8665d CD |
2452 | /* |
2453 | * Prevent division by 0 in overage calculation by acting as if | |
2454 | * it was a threshold of 1 page | |
2455 | */ | |
2456 | high = max(high, 1UL); | |
2457 | ||
2458 | overage = usage - high; | |
2459 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2460 | overage = div64_u64(overage, high); | |
2461 | ||
2462 | if (overage > max_overage) | |
2463 | max_overage = overage; | |
2464 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2465 | !mem_cgroup_is_root(memcg)); | |
2466 | ||
2467 | if (!max_overage) | |
2468 | return 0; | |
0e4b01df CD |
2469 | |
2470 | /* | |
0e4b01df CD |
2471 | * We use overage compared to memory.high to calculate the number of |
2472 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2473 | * fairly lenient on small overages, and increasingly harsh when the | |
2474 | * memcg in question makes it clear that it has no intention of stopping | |
2475 | * its crazy behaviour, so we exponentially increase the delay based on | |
2476 | * overage amount. | |
2477 | */ | |
86c8665d CD |
2478 | penalty_jiffies = max_overage * max_overage * HZ; |
2479 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2480 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2481 | |
2482 | /* | |
2483 | * Factor in the task's own contribution to the overage, such that four | |
2484 | * N-sized allocations are throttled approximately the same as one | |
2485 | * 4N-sized allocation. | |
2486 | * | |
2487 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2488 | * larger the current charge patch is than that. | |
2489 | */ | |
2490 | penalty_jiffies = penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; | |
2491 | ||
2492 | /* | |
2493 | * Clamp the max delay per usermode return so as to still keep the | |
2494 | * application moving forwards and also permit diagnostics, albeit | |
2495 | * extremely slowly. | |
2496 | */ | |
86c8665d CD |
2497 | return min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); |
2498 | } | |
2499 | ||
2500 | /* | |
2501 | * Scheduled by try_charge() to be executed from the userland return path | |
2502 | * and reclaims memory over the high limit. | |
2503 | */ | |
2504 | void mem_cgroup_handle_over_high(void) | |
2505 | { | |
2506 | unsigned long penalty_jiffies; | |
2507 | unsigned long pflags; | |
2508 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
2509 | struct mem_cgroup *memcg; | |
2510 | ||
2511 | if (likely(!nr_pages)) | |
2512 | return; | |
2513 | ||
2514 | memcg = get_mem_cgroup_from_mm(current->mm); | |
2515 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
2516 | current->memcg_nr_pages_over_high = 0; | |
2517 | ||
2518 | /* | |
2519 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2520 | * allocators proactively to slow down excessive growth. | |
2521 | */ | |
2522 | penalty_jiffies = calculate_high_delay(memcg, nr_pages); | |
0e4b01df CD |
2523 | |
2524 | /* | |
2525 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2526 | * that it's not even worth doing, in an attempt to be nice to those who | |
2527 | * go only a small amount over their memory.high value and maybe haven't | |
2528 | * been aggressively reclaimed enough yet. | |
2529 | */ | |
2530 | if (penalty_jiffies <= HZ / 100) | |
2531 | goto out; | |
2532 | ||
2533 | /* | |
2534 | * If we exit early, we're guaranteed to die (since | |
2535 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2536 | * need to account for any ill-begotten jiffies to pay them off later. | |
2537 | */ | |
2538 | psi_memstall_enter(&pflags); | |
2539 | schedule_timeout_killable(penalty_jiffies); | |
2540 | psi_memstall_leave(&pflags); | |
2541 | ||
2542 | out: | |
2543 | css_put(&memcg->css); | |
b23afb93 TH |
2544 | } |
2545 | ||
00501b53 JW |
2546 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2547 | unsigned int nr_pages) | |
8a9f3ccd | 2548 | { |
a983b5eb | 2549 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
9b130619 | 2550 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2551 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2552 | struct page_counter *counter; |
6539cc05 | 2553 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2554 | bool may_swap = true; |
2555 | bool drained = false; | |
29ef680a | 2556 | enum oom_status oom_status; |
a636b327 | 2557 | |
ce00a967 | 2558 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2559 | return 0; |
6539cc05 | 2560 | retry: |
b6b6cc72 | 2561 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2562 | return 0; |
8a9f3ccd | 2563 | |
7941d214 | 2564 | if (!do_memsw_account() || |
6071ca52 JW |
2565 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2566 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2567 | goto done_restock; |
7941d214 | 2568 | if (do_memsw_account()) |
3e32cb2e JW |
2569 | page_counter_uncharge(&memcg->memsw, batch); |
2570 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2571 | } else { |
3e32cb2e | 2572 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2573 | may_swap = false; |
3fbe7244 | 2574 | } |
7a81b88c | 2575 | |
6539cc05 JW |
2576 | if (batch > nr_pages) { |
2577 | batch = nr_pages; | |
2578 | goto retry; | |
2579 | } | |
6d61ef40 | 2580 | |
869712fd JW |
2581 | /* |
2582 | * Memcg doesn't have a dedicated reserve for atomic | |
2583 | * allocations. But like the global atomic pool, we need to | |
2584 | * put the burden of reclaim on regular allocation requests | |
2585 | * and let these go through as privileged allocations. | |
2586 | */ | |
2587 | if (gfp_mask & __GFP_ATOMIC) | |
2588 | goto force; | |
2589 | ||
06b078fc JW |
2590 | /* |
2591 | * Unlike in global OOM situations, memcg is not in a physical | |
2592 | * memory shortage. Allow dying and OOM-killed tasks to | |
2593 | * bypass the last charges so that they can exit quickly and | |
2594 | * free their memory. | |
2595 | */ | |
7775face | 2596 | if (unlikely(should_force_charge())) |
10d53c74 | 2597 | goto force; |
06b078fc | 2598 | |
89a28483 JW |
2599 | /* |
2600 | * Prevent unbounded recursion when reclaim operations need to | |
2601 | * allocate memory. This might exceed the limits temporarily, | |
2602 | * but we prefer facilitating memory reclaim and getting back | |
2603 | * under the limit over triggering OOM kills in these cases. | |
2604 | */ | |
2605 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2606 | goto force; | |
2607 | ||
06b078fc JW |
2608 | if (unlikely(task_in_memcg_oom(current))) |
2609 | goto nomem; | |
2610 | ||
d0164adc | 2611 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2612 | goto nomem; |
4b534334 | 2613 | |
e27be240 | 2614 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2615 | |
b70a2a21 JW |
2616 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2617 | gfp_mask, may_swap); | |
6539cc05 | 2618 | |
61e02c74 | 2619 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2620 | goto retry; |
28c34c29 | 2621 | |
b70a2a21 | 2622 | if (!drained) { |
6d3d6aa2 | 2623 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2624 | drained = true; |
2625 | goto retry; | |
2626 | } | |
2627 | ||
28c34c29 JW |
2628 | if (gfp_mask & __GFP_NORETRY) |
2629 | goto nomem; | |
6539cc05 JW |
2630 | /* |
2631 | * Even though the limit is exceeded at this point, reclaim | |
2632 | * may have been able to free some pages. Retry the charge | |
2633 | * before killing the task. | |
2634 | * | |
2635 | * Only for regular pages, though: huge pages are rather | |
2636 | * unlikely to succeed so close to the limit, and we fall back | |
2637 | * to regular pages anyway in case of failure. | |
2638 | */ | |
61e02c74 | 2639 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2640 | goto retry; |
2641 | /* | |
2642 | * At task move, charge accounts can be doubly counted. So, it's | |
2643 | * better to wait until the end of task_move if something is going on. | |
2644 | */ | |
2645 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2646 | goto retry; | |
2647 | ||
9b130619 JW |
2648 | if (nr_retries--) |
2649 | goto retry; | |
2650 | ||
38d38493 | 2651 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2652 | goto nomem; |
2653 | ||
06b078fc | 2654 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2655 | goto force; |
06b078fc | 2656 | |
6539cc05 | 2657 | if (fatal_signal_pending(current)) |
10d53c74 | 2658 | goto force; |
6539cc05 | 2659 | |
29ef680a MH |
2660 | /* |
2661 | * keep retrying as long as the memcg oom killer is able to make | |
2662 | * a forward progress or bypass the charge if the oom killer | |
2663 | * couldn't make any progress. | |
2664 | */ | |
2665 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2666 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2667 | switch (oom_status) { |
2668 | case OOM_SUCCESS: | |
2669 | nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
29ef680a MH |
2670 | goto retry; |
2671 | case OOM_FAILED: | |
2672 | goto force; | |
2673 | default: | |
2674 | goto nomem; | |
2675 | } | |
7a81b88c | 2676 | nomem: |
6d1fdc48 | 2677 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2678 | return -ENOMEM; |
10d53c74 TH |
2679 | force: |
2680 | /* | |
2681 | * The allocation either can't fail or will lead to more memory | |
2682 | * being freed very soon. Allow memory usage go over the limit | |
2683 | * temporarily by force charging it. | |
2684 | */ | |
2685 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2686 | if (do_memsw_account()) |
10d53c74 TH |
2687 | page_counter_charge(&memcg->memsw, nr_pages); |
2688 | css_get_many(&memcg->css, nr_pages); | |
2689 | ||
2690 | return 0; | |
6539cc05 JW |
2691 | |
2692 | done_restock: | |
e8ea14cc | 2693 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2694 | if (batch > nr_pages) |
2695 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2696 | |
241994ed | 2697 | /* |
b23afb93 TH |
2698 | * If the hierarchy is above the normal consumption range, schedule |
2699 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2700 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2701 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2702 | * not recorded as it most likely matches current's and won't | |
2703 | * change in the meantime. As high limit is checked again before | |
2704 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2705 | */ |
2706 | do { | |
b23afb93 | 2707 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2708 | /* Don't bother a random interrupted task */ |
2709 | if (in_interrupt()) { | |
2710 | schedule_work(&memcg->high_work); | |
2711 | break; | |
2712 | } | |
9516a18a | 2713 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2714 | set_notify_resume(current); |
2715 | break; | |
2716 | } | |
241994ed | 2717 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2718 | |
2719 | return 0; | |
7a81b88c | 2720 | } |
8a9f3ccd | 2721 | |
00501b53 | 2722 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2723 | { |
ce00a967 JW |
2724 | if (mem_cgroup_is_root(memcg)) |
2725 | return; | |
2726 | ||
3e32cb2e | 2727 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2728 | if (do_memsw_account()) |
3e32cb2e | 2729 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2730 | |
e8ea14cc | 2731 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2732 | } |
2733 | ||
0a31bc97 JW |
2734 | static void lock_page_lru(struct page *page, int *isolated) |
2735 | { | |
f4b7e272 | 2736 | pg_data_t *pgdat = page_pgdat(page); |
0a31bc97 | 2737 | |
f4b7e272 | 2738 | spin_lock_irq(&pgdat->lru_lock); |
0a31bc97 JW |
2739 | if (PageLRU(page)) { |
2740 | struct lruvec *lruvec; | |
2741 | ||
f4b7e272 | 2742 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
0a31bc97 JW |
2743 | ClearPageLRU(page); |
2744 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2745 | *isolated = 1; | |
2746 | } else | |
2747 | *isolated = 0; | |
2748 | } | |
2749 | ||
2750 | static void unlock_page_lru(struct page *page, int isolated) | |
2751 | { | |
f4b7e272 | 2752 | pg_data_t *pgdat = page_pgdat(page); |
0a31bc97 JW |
2753 | |
2754 | if (isolated) { | |
2755 | struct lruvec *lruvec; | |
2756 | ||
f4b7e272 | 2757 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
0a31bc97 JW |
2758 | VM_BUG_ON_PAGE(PageLRU(page), page); |
2759 | SetPageLRU(page); | |
2760 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2761 | } | |
f4b7e272 | 2762 | spin_unlock_irq(&pgdat->lru_lock); |
0a31bc97 JW |
2763 | } |
2764 | ||
00501b53 | 2765 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2766 | bool lrucare) |
7a81b88c | 2767 | { |
0a31bc97 | 2768 | int isolated; |
9ce70c02 | 2769 | |
1306a85a | 2770 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2771 | |
2772 | /* | |
2773 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2774 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2775 | */ | |
0a31bc97 JW |
2776 | if (lrucare) |
2777 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2778 | |
0a31bc97 JW |
2779 | /* |
2780 | * Nobody should be changing or seriously looking at | |
1306a85a | 2781 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2782 | * |
2783 | * - the page is uncharged | |
2784 | * | |
2785 | * - the page is off-LRU | |
2786 | * | |
2787 | * - an anonymous fault has exclusive page access, except for | |
2788 | * a locked page table | |
2789 | * | |
2790 | * - a page cache insertion, a swapin fault, or a migration | |
2791 | * have the page locked | |
2792 | */ | |
1306a85a | 2793 | page->mem_cgroup = memcg; |
9ce70c02 | 2794 | |
0a31bc97 JW |
2795 | if (lrucare) |
2796 | unlock_page_lru(page, isolated); | |
7a81b88c | 2797 | } |
66e1707b | 2798 | |
84c07d11 | 2799 | #ifdef CONFIG_MEMCG_KMEM |
ec0bed29 RG |
2800 | /* |
2801 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2802 | * | |
2803 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), | |
2804 | * cgroup_mutex, etc. | |
2805 | */ | |
2806 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2807 | { | |
2808 | struct page *page; | |
2809 | ||
2810 | if (mem_cgroup_disabled()) | |
2811 | return NULL; | |
2812 | ||
2813 | page = virt_to_head_page(p); | |
2814 | ||
2815 | /* | |
2816 | * Slab pages don't have page->mem_cgroup set because corresponding | |
2817 | * kmem caches can be reparented during the lifetime. That's why | |
2818 | * memcg_from_slab_page() should be used instead. | |
2819 | */ | |
2820 | if (PageSlab(page)) | |
2821 | return memcg_from_slab_page(page); | |
2822 | ||
2823 | /* All other pages use page->mem_cgroup */ | |
2824 | return page->mem_cgroup; | |
2825 | } | |
2826 | ||
f3bb3043 | 2827 | static int memcg_alloc_cache_id(void) |
55007d84 | 2828 | { |
f3bb3043 VD |
2829 | int id, size; |
2830 | int err; | |
2831 | ||
dbcf73e2 | 2832 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2833 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2834 | if (id < 0) | |
2835 | return id; | |
55007d84 | 2836 | |
dbcf73e2 | 2837 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2838 | return id; |
2839 | ||
2840 | /* | |
2841 | * There's no space for the new id in memcg_caches arrays, | |
2842 | * so we have to grow them. | |
2843 | */ | |
05257a1a | 2844 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2845 | |
2846 | size = 2 * (id + 1); | |
55007d84 GC |
2847 | if (size < MEMCG_CACHES_MIN_SIZE) |
2848 | size = MEMCG_CACHES_MIN_SIZE; | |
2849 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2850 | size = MEMCG_CACHES_MAX_SIZE; | |
2851 | ||
f3bb3043 | 2852 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2853 | if (!err) |
2854 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2855 | if (!err) |
2856 | memcg_nr_cache_ids = size; | |
2857 | ||
2858 | up_write(&memcg_cache_ids_sem); | |
2859 | ||
f3bb3043 | 2860 | if (err) { |
dbcf73e2 | 2861 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2862 | return err; |
2863 | } | |
2864 | return id; | |
2865 | } | |
2866 | ||
2867 | static void memcg_free_cache_id(int id) | |
2868 | { | |
dbcf73e2 | 2869 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2870 | } |
2871 | ||
d5b3cf71 | 2872 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2873 | struct mem_cgroup *memcg; |
2874 | struct kmem_cache *cachep; | |
2875 | struct work_struct work; | |
2876 | }; | |
2877 | ||
d5b3cf71 | 2878 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2879 | { |
d5b3cf71 VD |
2880 | struct memcg_kmem_cache_create_work *cw = |
2881 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2882 | struct mem_cgroup *memcg = cw->memcg; |
2883 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2884 | |
d5b3cf71 | 2885 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2886 | |
5722d094 | 2887 | css_put(&memcg->css); |
d7f25f8a GC |
2888 | kfree(cw); |
2889 | } | |
2890 | ||
2891 | /* | |
2892 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2893 | */ |
85cfb245 | 2894 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
d5b3cf71 | 2895 | struct kmem_cache *cachep) |
d7f25f8a | 2896 | { |
d5b3cf71 | 2897 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2898 | |
f0a3a24b RG |
2899 | if (!css_tryget_online(&memcg->css)) |
2900 | return; | |
2901 | ||
c892fd82 | 2902 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT | __GFP_NOWARN); |
9168ee5f MS |
2903 | if (!cw) { |
2904 | css_put(&memcg->css); | |
d7f25f8a | 2905 | return; |
9168ee5f | 2906 | } |
8135be5a | 2907 | |
d7f25f8a GC |
2908 | cw->memcg = memcg; |
2909 | cw->cachep = cachep; | |
d5b3cf71 | 2910 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2911 | |
17cc4dfe | 2912 | queue_work(memcg_kmem_cache_wq, &cw->work); |
d7f25f8a GC |
2913 | } |
2914 | ||
45264778 VD |
2915 | static inline bool memcg_kmem_bypass(void) |
2916 | { | |
2917 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
2918 | return true; | |
2919 | return false; | |
2920 | } | |
2921 | ||
2922 | /** | |
2923 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation | |
2924 | * @cachep: the original global kmem cache | |
2925 | * | |
d7f25f8a GC |
2926 | * Return the kmem_cache we're supposed to use for a slab allocation. |
2927 | * We try to use the current memcg's version of the cache. | |
2928 | * | |
45264778 VD |
2929 | * If the cache does not exist yet, if we are the first user of it, we |
2930 | * create it asynchronously in a workqueue and let the current allocation | |
2931 | * go through with the original cache. | |
d7f25f8a | 2932 | * |
45264778 VD |
2933 | * This function takes a reference to the cache it returns to assure it |
2934 | * won't get destroyed while we are working with it. Once the caller is | |
2935 | * done with it, memcg_kmem_put_cache() must be called to release the | |
2936 | * reference. | |
d7f25f8a | 2937 | */ |
45264778 | 2938 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2939 | { |
2940 | struct mem_cgroup *memcg; | |
959c8963 | 2941 | struct kmem_cache *memcg_cachep; |
f0a3a24b | 2942 | struct memcg_cache_array *arr; |
2a4db7eb | 2943 | int kmemcg_id; |
d7f25f8a | 2944 | |
f7ce3190 | 2945 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2946 | |
45264778 | 2947 | if (memcg_kmem_bypass()) |
230e9fc2 VD |
2948 | return cachep; |
2949 | ||
f0a3a24b RG |
2950 | rcu_read_lock(); |
2951 | ||
2952 | if (unlikely(current->active_memcg)) | |
2953 | memcg = current->active_memcg; | |
2954 | else | |
2955 | memcg = mem_cgroup_from_task(current); | |
2956 | ||
2957 | if (!memcg || memcg == root_mem_cgroup) | |
2958 | goto out_unlock; | |
2959 | ||
4db0c3c2 | 2960 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2961 | if (kmemcg_id < 0) |
f0a3a24b | 2962 | goto out_unlock; |
d7f25f8a | 2963 | |
f0a3a24b RG |
2964 | arr = rcu_dereference(cachep->memcg_params.memcg_caches); |
2965 | ||
2966 | /* | |
2967 | * Make sure we will access the up-to-date value. The code updating | |
2968 | * memcg_caches issues a write barrier to match the data dependency | |
2969 | * barrier inside READ_ONCE() (see memcg_create_kmem_cache()). | |
2970 | */ | |
2971 | memcg_cachep = READ_ONCE(arr->entries[kmemcg_id]); | |
ca0dde97 LZ |
2972 | |
2973 | /* | |
2974 | * If we are in a safe context (can wait, and not in interrupt | |
2975 | * context), we could be be predictable and return right away. | |
2976 | * This would guarantee that the allocation being performed | |
2977 | * already belongs in the new cache. | |
2978 | * | |
2979 | * However, there are some clashes that can arrive from locking. | |
2980 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2981 | * memcg_create_kmem_cache, this means no further allocation |
2982 | * could happen with the slab_mutex held. So it's better to | |
2983 | * defer everything. | |
f0a3a24b RG |
2984 | * |
2985 | * If the memcg is dying or memcg_cache is about to be released, | |
2986 | * don't bother creating new kmem_caches. Because memcg_cachep | |
2987 | * is ZEROed as the fist step of kmem offlining, we don't need | |
2988 | * percpu_ref_tryget_live() here. css_tryget_online() check in | |
2989 | * memcg_schedule_kmem_cache_create() will prevent us from | |
2990 | * creation of a new kmem_cache. | |
ca0dde97 | 2991 | */ |
f0a3a24b RG |
2992 | if (unlikely(!memcg_cachep)) |
2993 | memcg_schedule_kmem_cache_create(memcg, cachep); | |
2994 | else if (percpu_ref_tryget(&memcg_cachep->memcg_params.refcnt)) | |
2995 | cachep = memcg_cachep; | |
2996 | out_unlock: | |
2997 | rcu_read_unlock(); | |
ca0dde97 | 2998 | return cachep; |
d7f25f8a | 2999 | } |
d7f25f8a | 3000 | |
45264778 VD |
3001 | /** |
3002 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache | |
3003 | * @cachep: the cache returned by memcg_kmem_get_cache | |
3004 | */ | |
3005 | void memcg_kmem_put_cache(struct kmem_cache *cachep) | |
8135be5a VD |
3006 | { |
3007 | if (!is_root_cache(cachep)) | |
f0a3a24b | 3008 | percpu_ref_put(&cachep->memcg_params.refcnt); |
8135be5a VD |
3009 | } |
3010 | ||
45264778 | 3011 | /** |
60cd4bcd | 3012 | * __memcg_kmem_charge_memcg: charge a kmem page |
45264778 VD |
3013 | * @page: page to charge |
3014 | * @gfp: reclaim mode | |
3015 | * @order: allocation order | |
3016 | * @memcg: memory cgroup to charge | |
3017 | * | |
3018 | * Returns 0 on success, an error code on failure. | |
3019 | */ | |
60cd4bcd | 3020 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
45264778 | 3021 | struct mem_cgroup *memcg) |
7ae1e1d0 | 3022 | { |
f3ccb2c4 VD |
3023 | unsigned int nr_pages = 1 << order; |
3024 | struct page_counter *counter; | |
7ae1e1d0 GC |
3025 | int ret; |
3026 | ||
f3ccb2c4 | 3027 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 3028 | if (ret) |
f3ccb2c4 | 3029 | return ret; |
52c29b04 JW |
3030 | |
3031 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
3032 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
e55d9d9b MH |
3033 | |
3034 | /* | |
3035 | * Enforce __GFP_NOFAIL allocation because callers are not | |
3036 | * prepared to see failures and likely do not have any failure | |
3037 | * handling code. | |
3038 | */ | |
3039 | if (gfp & __GFP_NOFAIL) { | |
3040 | page_counter_charge(&memcg->kmem, nr_pages); | |
3041 | return 0; | |
3042 | } | |
52c29b04 JW |
3043 | cancel_charge(memcg, nr_pages); |
3044 | return -ENOMEM; | |
7ae1e1d0 | 3045 | } |
f3ccb2c4 | 3046 | return 0; |
7ae1e1d0 GC |
3047 | } |
3048 | ||
45264778 | 3049 | /** |
60cd4bcd | 3050 | * __memcg_kmem_charge: charge a kmem page to the current memory cgroup |
45264778 VD |
3051 | * @page: page to charge |
3052 | * @gfp: reclaim mode | |
3053 | * @order: allocation order | |
3054 | * | |
3055 | * Returns 0 on success, an error code on failure. | |
3056 | */ | |
60cd4bcd | 3057 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 3058 | { |
f3ccb2c4 | 3059 | struct mem_cgroup *memcg; |
fcff7d7e | 3060 | int ret = 0; |
7ae1e1d0 | 3061 | |
60cd4bcd | 3062 | if (memcg_kmem_bypass()) |
45264778 VD |
3063 | return 0; |
3064 | ||
d46eb14b | 3065 | memcg = get_mem_cgroup_from_current(); |
c4159a75 | 3066 | if (!mem_cgroup_is_root(memcg)) { |
60cd4bcd | 3067 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); |
4d96ba35 RG |
3068 | if (!ret) { |
3069 | page->mem_cgroup = memcg; | |
c4159a75 | 3070 | __SetPageKmemcg(page); |
4d96ba35 | 3071 | } |
c4159a75 | 3072 | } |
7ae1e1d0 | 3073 | css_put(&memcg->css); |
d05e83a6 | 3074 | return ret; |
7ae1e1d0 | 3075 | } |
49a18eae RG |
3076 | |
3077 | /** | |
3078 | * __memcg_kmem_uncharge_memcg: uncharge a kmem page | |
3079 | * @memcg: memcg to uncharge | |
3080 | * @nr_pages: number of pages to uncharge | |
3081 | */ | |
3082 | void __memcg_kmem_uncharge_memcg(struct mem_cgroup *memcg, | |
3083 | unsigned int nr_pages) | |
3084 | { | |
3085 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
3086 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
3087 | ||
3088 | page_counter_uncharge(&memcg->memory, nr_pages); | |
3089 | if (do_memsw_account()) | |
3090 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
3091 | } | |
45264778 | 3092 | /** |
60cd4bcd | 3093 | * __memcg_kmem_uncharge: uncharge a kmem page |
45264778 VD |
3094 | * @page: page to uncharge |
3095 | * @order: allocation order | |
3096 | */ | |
60cd4bcd | 3097 | void __memcg_kmem_uncharge(struct page *page, int order) |
7ae1e1d0 | 3098 | { |
1306a85a | 3099 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 3100 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 3101 | |
7ae1e1d0 GC |
3102 | if (!memcg) |
3103 | return; | |
3104 | ||
309381fe | 3105 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
49a18eae | 3106 | __memcg_kmem_uncharge_memcg(memcg, nr_pages); |
1306a85a | 3107 | page->mem_cgroup = NULL; |
c4159a75 VD |
3108 | |
3109 | /* slab pages do not have PageKmemcg flag set */ | |
3110 | if (PageKmemcg(page)) | |
3111 | __ClearPageKmemcg(page); | |
3112 | ||
f3ccb2c4 | 3113 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 3114 | } |
84c07d11 | 3115 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3116 | |
ca3e0214 KH |
3117 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3118 | ||
ca3e0214 KH |
3119 | /* |
3120 | * Because tail pages are not marked as "used", set it. We're under | |
f4b7e272 | 3121 | * pgdat->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 3122 | */ |
e94c8a9c | 3123 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 3124 | { |
e94c8a9c | 3125 | int i; |
ca3e0214 | 3126 | |
3d37c4a9 KH |
3127 | if (mem_cgroup_disabled()) |
3128 | return; | |
b070e65c | 3129 | |
29833315 | 3130 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 3131 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 3132 | |
c9019e9b | 3133 | __mod_memcg_state(head->mem_cgroup, MEMCG_RSS_HUGE, -HPAGE_PMD_NR); |
ca3e0214 | 3134 | } |
12d27107 | 3135 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 3136 | |
c255a458 | 3137 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
3138 | /** |
3139 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3140 | * @entry: swap entry to be moved | |
3141 | * @from: mem_cgroup which the entry is moved from | |
3142 | * @to: mem_cgroup which the entry is moved to | |
3143 | * | |
3144 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3145 | * as the mem_cgroup's id of @from. | |
3146 | * | |
3147 | * Returns 0 on success, -EINVAL on failure. | |
3148 | * | |
3e32cb2e | 3149 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3150 | * both res and memsw, and called css_get(). |
3151 | */ | |
3152 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3153 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3154 | { |
3155 | unsigned short old_id, new_id; | |
3156 | ||
34c00c31 LZ |
3157 | old_id = mem_cgroup_id(from); |
3158 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3159 | |
3160 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3161 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3162 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3163 | return 0; |
3164 | } | |
3165 | return -EINVAL; | |
3166 | } | |
3167 | #else | |
3168 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3169 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3170 | { |
3171 | return -EINVAL; | |
3172 | } | |
8c7c6e34 | 3173 | #endif |
d13d1443 | 3174 | |
bbec2e15 | 3175 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3176 | |
bbec2e15 RG |
3177 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3178 | unsigned long max, bool memsw) | |
628f4235 | 3179 | { |
3e32cb2e | 3180 | bool enlarge = false; |
bb4a7ea2 | 3181 | bool drained = false; |
3e32cb2e | 3182 | int ret; |
c054a78c YZ |
3183 | bool limits_invariant; |
3184 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3185 | |
3e32cb2e | 3186 | do { |
628f4235 KH |
3187 | if (signal_pending(current)) { |
3188 | ret = -EINTR; | |
3189 | break; | |
3190 | } | |
3e32cb2e | 3191 | |
bbec2e15 | 3192 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3193 | /* |
3194 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3195 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3196 | */ |
bbec2e15 RG |
3197 | limits_invariant = memsw ? max >= memcg->memory.max : |
3198 | max <= memcg->memsw.max; | |
c054a78c | 3199 | if (!limits_invariant) { |
bbec2e15 | 3200 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3201 | ret = -EINVAL; |
8c7c6e34 KH |
3202 | break; |
3203 | } | |
bbec2e15 | 3204 | if (max > counter->max) |
3e32cb2e | 3205 | enlarge = true; |
bbec2e15 RG |
3206 | ret = page_counter_set_max(counter, max); |
3207 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3208 | |
3209 | if (!ret) | |
3210 | break; | |
3211 | ||
bb4a7ea2 SB |
3212 | if (!drained) { |
3213 | drain_all_stock(memcg); | |
3214 | drained = true; | |
3215 | continue; | |
3216 | } | |
3217 | ||
1ab5c056 AR |
3218 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
3219 | GFP_KERNEL, !memsw)) { | |
3220 | ret = -EBUSY; | |
3221 | break; | |
3222 | } | |
3223 | } while (true); | |
3e32cb2e | 3224 | |
3c11ecf4 KH |
3225 | if (!ret && enlarge) |
3226 | memcg_oom_recover(memcg); | |
3e32cb2e | 3227 | |
628f4235 KH |
3228 | return ret; |
3229 | } | |
3230 | ||
ef8f2327 | 3231 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3232 | gfp_t gfp_mask, |
3233 | unsigned long *total_scanned) | |
3234 | { | |
3235 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3236 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3237 | unsigned long reclaimed; |
3238 | int loop = 0; | |
ef8f2327 | 3239 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3240 | unsigned long excess; |
0608f43d AM |
3241 | unsigned long nr_scanned; |
3242 | ||
3243 | if (order > 0) | |
3244 | return 0; | |
3245 | ||
ef8f2327 | 3246 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
3247 | |
3248 | /* | |
3249 | * Do not even bother to check the largest node if the root | |
3250 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3251 | * are acceptable as soft limit is best effort anyway. | |
3252 | */ | |
bfc7228b | 3253 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3254 | return 0; |
3255 | ||
0608f43d AM |
3256 | /* |
3257 | * This loop can run a while, specially if mem_cgroup's continuously | |
3258 | * keep exceeding their soft limit and putting the system under | |
3259 | * pressure | |
3260 | */ | |
3261 | do { | |
3262 | if (next_mz) | |
3263 | mz = next_mz; | |
3264 | else | |
3265 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3266 | if (!mz) | |
3267 | break; | |
3268 | ||
3269 | nr_scanned = 0; | |
ef8f2327 | 3270 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
3271 | gfp_mask, &nr_scanned); |
3272 | nr_reclaimed += reclaimed; | |
3273 | *total_scanned += nr_scanned; | |
0a31bc97 | 3274 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3275 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3276 | |
3277 | /* | |
3278 | * If we failed to reclaim anything from this memory cgroup | |
3279 | * it is time to move on to the next cgroup | |
3280 | */ | |
3281 | next_mz = NULL; | |
bc2f2e7f VD |
3282 | if (!reclaimed) |
3283 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3284 | ||
3e32cb2e | 3285 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3286 | /* |
3287 | * One school of thought says that we should not add | |
3288 | * back the node to the tree if reclaim returns 0. | |
3289 | * But our reclaim could return 0, simply because due | |
3290 | * to priority we are exposing a smaller subset of | |
3291 | * memory to reclaim from. Consider this as a longer | |
3292 | * term TODO. | |
3293 | */ | |
3294 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3295 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3296 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3297 | css_put(&mz->memcg->css); |
3298 | loop++; | |
3299 | /* | |
3300 | * Could not reclaim anything and there are no more | |
3301 | * mem cgroups to try or we seem to be looping without | |
3302 | * reclaiming anything. | |
3303 | */ | |
3304 | if (!nr_reclaimed && | |
3305 | (next_mz == NULL || | |
3306 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3307 | break; | |
3308 | } while (!nr_reclaimed); | |
3309 | if (next_mz) | |
3310 | css_put(&next_mz->memcg->css); | |
3311 | return nr_reclaimed; | |
3312 | } | |
3313 | ||
ea280e7b TH |
3314 | /* |
3315 | * Test whether @memcg has children, dead or alive. Note that this | |
3316 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3317 | * returns %true if there are child csses according to the cgroup | |
3318 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3319 | */ | |
b5f99b53 GC |
3320 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3321 | { | |
ea280e7b TH |
3322 | bool ret; |
3323 | ||
ea280e7b TH |
3324 | rcu_read_lock(); |
3325 | ret = css_next_child(NULL, &memcg->css); | |
3326 | rcu_read_unlock(); | |
3327 | return ret; | |
b5f99b53 GC |
3328 | } |
3329 | ||
c26251f9 | 3330 | /* |
51038171 | 3331 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3332 | * |
3333 | * Caller is responsible for holding css reference for memcg. | |
3334 | */ | |
3335 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3336 | { | |
3337 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3338 | |
c1e862c1 KH |
3339 | /* we call try-to-free pages for make this cgroup empty */ |
3340 | lru_add_drain_all(); | |
d12c60f6 JS |
3341 | |
3342 | drain_all_stock(memcg); | |
3343 | ||
f817ed48 | 3344 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3345 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3346 | int progress; |
c1e862c1 | 3347 | |
c26251f9 MH |
3348 | if (signal_pending(current)) |
3349 | return -EINTR; | |
3350 | ||
b70a2a21 JW |
3351 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3352 | GFP_KERNEL, true); | |
c1e862c1 | 3353 | if (!progress) { |
f817ed48 | 3354 | nr_retries--; |
c1e862c1 | 3355 | /* maybe some writeback is necessary */ |
8aa7e847 | 3356 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3357 | } |
f817ed48 KH |
3358 | |
3359 | } | |
ab5196c2 MH |
3360 | |
3361 | return 0; | |
cc847582 KH |
3362 | } |
3363 | ||
6770c64e TH |
3364 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3365 | char *buf, size_t nbytes, | |
3366 | loff_t off) | |
c1e862c1 | 3367 | { |
6770c64e | 3368 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3369 | |
d8423011 MH |
3370 | if (mem_cgroup_is_root(memcg)) |
3371 | return -EINVAL; | |
6770c64e | 3372 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3373 | } |
3374 | ||
182446d0 TH |
3375 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3376 | struct cftype *cft) | |
18f59ea7 | 3377 | { |
182446d0 | 3378 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3379 | } |
3380 | ||
182446d0 TH |
3381 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3382 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3383 | { |
3384 | int retval = 0; | |
182446d0 | 3385 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3386 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3387 | |
567fb435 | 3388 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 3389 | return 0; |
567fb435 | 3390 | |
18f59ea7 | 3391 | /* |
af901ca1 | 3392 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3393 | * in the child subtrees. If it is unset, then the change can |
3394 | * occur, provided the current cgroup has no children. | |
3395 | * | |
3396 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3397 | * set if there are no children. | |
3398 | */ | |
c0ff4b85 | 3399 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3400 | (val == 1 || val == 0)) { |
ea280e7b | 3401 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3402 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3403 | else |
3404 | retval = -EBUSY; | |
3405 | } else | |
3406 | retval = -EINVAL; | |
567fb435 | 3407 | |
18f59ea7 BS |
3408 | return retval; |
3409 | } | |
3410 | ||
6f646156 | 3411 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3412 | { |
42a30035 | 3413 | unsigned long val; |
ce00a967 | 3414 | |
3e32cb2e | 3415 | if (mem_cgroup_is_root(memcg)) { |
42a30035 JW |
3416 | val = memcg_page_state(memcg, MEMCG_CACHE) + |
3417 | memcg_page_state(memcg, MEMCG_RSS); | |
3418 | if (swap) | |
3419 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3420 | } else { |
ce00a967 | 3421 | if (!swap) |
3e32cb2e | 3422 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3423 | else |
3e32cb2e | 3424 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3425 | } |
c12176d3 | 3426 | return val; |
ce00a967 JW |
3427 | } |
3428 | ||
3e32cb2e JW |
3429 | enum { |
3430 | RES_USAGE, | |
3431 | RES_LIMIT, | |
3432 | RES_MAX_USAGE, | |
3433 | RES_FAILCNT, | |
3434 | RES_SOFT_LIMIT, | |
3435 | }; | |
ce00a967 | 3436 | |
791badbd | 3437 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3438 | struct cftype *cft) |
8cdea7c0 | 3439 | { |
182446d0 | 3440 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3441 | struct page_counter *counter; |
af36f906 | 3442 | |
3e32cb2e | 3443 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3444 | case _MEM: |
3e32cb2e JW |
3445 | counter = &memcg->memory; |
3446 | break; | |
8c7c6e34 | 3447 | case _MEMSWAP: |
3e32cb2e JW |
3448 | counter = &memcg->memsw; |
3449 | break; | |
510fc4e1 | 3450 | case _KMEM: |
3e32cb2e | 3451 | counter = &memcg->kmem; |
510fc4e1 | 3452 | break; |
d55f90bf | 3453 | case _TCP: |
0db15298 | 3454 | counter = &memcg->tcpmem; |
d55f90bf | 3455 | break; |
8c7c6e34 KH |
3456 | default: |
3457 | BUG(); | |
8c7c6e34 | 3458 | } |
3e32cb2e JW |
3459 | |
3460 | switch (MEMFILE_ATTR(cft->private)) { | |
3461 | case RES_USAGE: | |
3462 | if (counter == &memcg->memory) | |
c12176d3 | 3463 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3464 | if (counter == &memcg->memsw) |
c12176d3 | 3465 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3466 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3467 | case RES_LIMIT: | |
bbec2e15 | 3468 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3469 | case RES_MAX_USAGE: |
3470 | return (u64)counter->watermark * PAGE_SIZE; | |
3471 | case RES_FAILCNT: | |
3472 | return counter->failcnt; | |
3473 | case RES_SOFT_LIMIT: | |
3474 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3475 | default: | |
3476 | BUG(); | |
3477 | } | |
8cdea7c0 | 3478 | } |
510fc4e1 | 3479 | |
c2f70949 | 3480 | static void memcg_flush_percpu_vmstats(struct mem_cgroup *memcg) |
c350a99e | 3481 | { |
c2f70949 | 3482 | unsigned long stat[MEMCG_NR_STAT] = {0}; |
c350a99e RG |
3483 | struct mem_cgroup *mi; |
3484 | int node, cpu, i; | |
c350a99e RG |
3485 | |
3486 | for_each_online_cpu(cpu) | |
c2f70949 | 3487 | for (i = 0; i < MEMCG_NR_STAT; i++) |
6c1c2808 | 3488 | stat[i] += per_cpu(memcg->vmstats_percpu->stat[i], cpu); |
c350a99e RG |
3489 | |
3490 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
c2f70949 | 3491 | for (i = 0; i < MEMCG_NR_STAT; i++) |
c350a99e RG |
3492 | atomic_long_add(stat[i], &mi->vmstats[i]); |
3493 | ||
3494 | for_each_node(node) { | |
3495 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; | |
3496 | struct mem_cgroup_per_node *pi; | |
3497 | ||
c2f70949 | 3498 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3499 | stat[i] = 0; |
3500 | ||
3501 | for_each_online_cpu(cpu) | |
c2f70949 | 3502 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
6c1c2808 SB |
3503 | stat[i] += per_cpu( |
3504 | pn->lruvec_stat_cpu->count[i], cpu); | |
c350a99e RG |
3505 | |
3506 | for (pi = pn; pi; pi = parent_nodeinfo(pi, node)) | |
c2f70949 | 3507 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
c350a99e RG |
3508 | atomic_long_add(stat[i], &pi->lruvec_stat[i]); |
3509 | } | |
3510 | } | |
3511 | ||
bb65f89b RG |
3512 | static void memcg_flush_percpu_vmevents(struct mem_cgroup *memcg) |
3513 | { | |
3514 | unsigned long events[NR_VM_EVENT_ITEMS]; | |
3515 | struct mem_cgroup *mi; | |
3516 | int cpu, i; | |
3517 | ||
3518 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3519 | events[i] = 0; | |
3520 | ||
3521 | for_each_online_cpu(cpu) | |
3522 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
6c1c2808 SB |
3523 | events[i] += per_cpu(memcg->vmstats_percpu->events[i], |
3524 | cpu); | |
bb65f89b RG |
3525 | |
3526 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) | |
3527 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) | |
3528 | atomic_long_add(events[i], &mi->vmevents[i]); | |
3529 | } | |
3530 | ||
84c07d11 | 3531 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3532 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3533 | { |
d6441637 VD |
3534 | int memcg_id; |
3535 | ||
b313aeee VD |
3536 | if (cgroup_memory_nokmem) |
3537 | return 0; | |
3538 | ||
2a4db7eb | 3539 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3540 | BUG_ON(memcg->kmem_state); |
d6441637 | 3541 | |
f3bb3043 | 3542 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3543 | if (memcg_id < 0) |
3544 | return memcg_id; | |
d6441637 | 3545 | |
ef12947c | 3546 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 3547 | /* |
567e9ab2 | 3548 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 3549 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
3550 | * guarantee no one starts accounting before all call sites are |
3551 | * patched. | |
3552 | */ | |
900a38f0 | 3553 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3554 | memcg->kmem_state = KMEM_ONLINE; |
bc2791f8 | 3555 | INIT_LIST_HEAD(&memcg->kmem_caches); |
0b8f73e1 JW |
3556 | |
3557 | return 0; | |
d6441637 VD |
3558 | } |
3559 | ||
8e0a8912 JW |
3560 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3561 | { | |
3562 | struct cgroup_subsys_state *css; | |
3563 | struct mem_cgroup *parent, *child; | |
3564 | int kmemcg_id; | |
3565 | ||
3566 | if (memcg->kmem_state != KMEM_ONLINE) | |
3567 | return; | |
3568 | /* | |
3569 | * Clear the online state before clearing memcg_caches array | |
3570 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
3571 | * guarantees that no cache will be created for this cgroup | |
3572 | * after we are done (see memcg_create_kmem_cache()). | |
3573 | */ | |
3574 | memcg->kmem_state = KMEM_ALLOCATED; | |
3575 | ||
8e0a8912 JW |
3576 | parent = parent_mem_cgroup(memcg); |
3577 | if (!parent) | |
3578 | parent = root_mem_cgroup; | |
3579 | ||
bee07b33 | 3580 | /* |
c2f70949 | 3581 | * Deactivate and reparent kmem_caches. |
bee07b33 | 3582 | */ |
fb2f2b0a RG |
3583 | memcg_deactivate_kmem_caches(memcg, parent); |
3584 | ||
3585 | kmemcg_id = memcg->kmemcg_id; | |
3586 | BUG_ON(kmemcg_id < 0); | |
3587 | ||
8e0a8912 JW |
3588 | /* |
3589 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3590 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3591 | * to ones of the parent. After we have finished, all list_lrus | |
3592 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3593 | * ordering is imposed by list_lru_node->lock taken by | |
3594 | * memcg_drain_all_list_lrus(). | |
3595 | */ | |
3a06bb78 | 3596 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3597 | css_for_each_descendant_pre(css, &memcg->css) { |
3598 | child = mem_cgroup_from_css(css); | |
3599 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3600 | child->kmemcg_id = parent->kmemcg_id; | |
3601 | if (!memcg->use_hierarchy) | |
3602 | break; | |
3603 | } | |
3a06bb78 TH |
3604 | rcu_read_unlock(); |
3605 | ||
9bec5c35 | 3606 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3607 | |
3608 | memcg_free_cache_id(kmemcg_id); | |
3609 | } | |
3610 | ||
3611 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3612 | { | |
0b8f73e1 JW |
3613 | /* css_alloc() failed, offlining didn't happen */ |
3614 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3615 | memcg_offline_kmem(memcg); | |
3616 | ||
8e0a8912 | 3617 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
f0a3a24b | 3618 | WARN_ON(!list_empty(&memcg->kmem_caches)); |
8e0a8912 | 3619 | static_branch_dec(&memcg_kmem_enabled_key); |
8e0a8912 | 3620 | } |
8e0a8912 | 3621 | } |
d6441637 | 3622 | #else |
0b8f73e1 | 3623 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3624 | { |
3625 | return 0; | |
3626 | } | |
3627 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3628 | { | |
3629 | } | |
3630 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3631 | { | |
3632 | } | |
84c07d11 | 3633 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3634 | |
bbec2e15 RG |
3635 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3636 | unsigned long max) | |
d6441637 | 3637 | { |
b313aeee | 3638 | int ret; |
127424c8 | 3639 | |
bbec2e15 RG |
3640 | mutex_lock(&memcg_max_mutex); |
3641 | ret = page_counter_set_max(&memcg->kmem, max); | |
3642 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3643 | return ret; |
d6441637 | 3644 | } |
510fc4e1 | 3645 | |
bbec2e15 | 3646 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3647 | { |
3648 | int ret; | |
3649 | ||
bbec2e15 | 3650 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3651 | |
bbec2e15 | 3652 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3653 | if (ret) |
3654 | goto out; | |
3655 | ||
0db15298 | 3656 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3657 | /* |
3658 | * The active flag needs to be written after the static_key | |
3659 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3660 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3661 | * for details, and note that we don't mark any socket as | |
3662 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3663 | * |
3664 | * We need to do this, because static_keys will span multiple | |
3665 | * sites, but we can't control their order. If we mark a socket | |
3666 | * as accounted, but the accounting functions are not patched in | |
3667 | * yet, we'll lose accounting. | |
3668 | * | |
2d758073 | 3669 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3670 | * because when this value change, the code to process it is not |
3671 | * patched in yet. | |
3672 | */ | |
3673 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3674 | memcg->tcpmem_active = true; |
d55f90bf VD |
3675 | } |
3676 | out: | |
bbec2e15 | 3677 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3678 | return ret; |
3679 | } | |
d55f90bf | 3680 | |
628f4235 KH |
3681 | /* |
3682 | * The user of this function is... | |
3683 | * RES_LIMIT. | |
3684 | */ | |
451af504 TH |
3685 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3686 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3687 | { |
451af504 | 3688 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3689 | unsigned long nr_pages; |
628f4235 KH |
3690 | int ret; |
3691 | ||
451af504 | 3692 | buf = strstrip(buf); |
650c5e56 | 3693 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3694 | if (ret) |
3695 | return ret; | |
af36f906 | 3696 | |
3e32cb2e | 3697 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3698 | case RES_LIMIT: |
4b3bde4c BS |
3699 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3700 | ret = -EINVAL; | |
3701 | break; | |
3702 | } | |
3e32cb2e JW |
3703 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3704 | case _MEM: | |
bbec2e15 | 3705 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3706 | break; |
3e32cb2e | 3707 | case _MEMSWAP: |
bbec2e15 | 3708 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3709 | break; |
3e32cb2e | 3710 | case _KMEM: |
0158115f MH |
3711 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " |
3712 | "Please report your usecase to linux-mm@kvack.org if you " | |
3713 | "depend on this functionality.\n"); | |
bbec2e15 | 3714 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3715 | break; |
d55f90bf | 3716 | case _TCP: |
bbec2e15 | 3717 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3718 | break; |
3e32cb2e | 3719 | } |
296c81d8 | 3720 | break; |
3e32cb2e JW |
3721 | case RES_SOFT_LIMIT: |
3722 | memcg->soft_limit = nr_pages; | |
3723 | ret = 0; | |
628f4235 KH |
3724 | break; |
3725 | } | |
451af504 | 3726 | return ret ?: nbytes; |
8cdea7c0 BS |
3727 | } |
3728 | ||
6770c64e TH |
3729 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3730 | size_t nbytes, loff_t off) | |
c84872e1 | 3731 | { |
6770c64e | 3732 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3733 | struct page_counter *counter; |
c84872e1 | 3734 | |
3e32cb2e JW |
3735 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3736 | case _MEM: | |
3737 | counter = &memcg->memory; | |
3738 | break; | |
3739 | case _MEMSWAP: | |
3740 | counter = &memcg->memsw; | |
3741 | break; | |
3742 | case _KMEM: | |
3743 | counter = &memcg->kmem; | |
3744 | break; | |
d55f90bf | 3745 | case _TCP: |
0db15298 | 3746 | counter = &memcg->tcpmem; |
d55f90bf | 3747 | break; |
3e32cb2e JW |
3748 | default: |
3749 | BUG(); | |
3750 | } | |
af36f906 | 3751 | |
3e32cb2e | 3752 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3753 | case RES_MAX_USAGE: |
3e32cb2e | 3754 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3755 | break; |
3756 | case RES_FAILCNT: | |
3e32cb2e | 3757 | counter->failcnt = 0; |
29f2a4da | 3758 | break; |
3e32cb2e JW |
3759 | default: |
3760 | BUG(); | |
29f2a4da | 3761 | } |
f64c3f54 | 3762 | |
6770c64e | 3763 | return nbytes; |
c84872e1 PE |
3764 | } |
3765 | ||
182446d0 | 3766 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3767 | struct cftype *cft) |
3768 | { | |
182446d0 | 3769 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3770 | } |
3771 | ||
02491447 | 3772 | #ifdef CONFIG_MMU |
182446d0 | 3773 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3774 | struct cftype *cft, u64 val) |
3775 | { | |
182446d0 | 3776 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3777 | |
1dfab5ab | 3778 | if (val & ~MOVE_MASK) |
7dc74be0 | 3779 | return -EINVAL; |
ee5e8472 | 3780 | |
7dc74be0 | 3781 | /* |
ee5e8472 GC |
3782 | * No kind of locking is needed in here, because ->can_attach() will |
3783 | * check this value once in the beginning of the process, and then carry | |
3784 | * on with stale data. This means that changes to this value will only | |
3785 | * affect task migrations starting after the change. | |
7dc74be0 | 3786 | */ |
c0ff4b85 | 3787 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3788 | return 0; |
3789 | } | |
02491447 | 3790 | #else |
182446d0 | 3791 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3792 | struct cftype *cft, u64 val) |
3793 | { | |
3794 | return -ENOSYS; | |
3795 | } | |
3796 | #endif | |
7dc74be0 | 3797 | |
406eb0c9 | 3798 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3799 | |
3800 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3801 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3802 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3803 | ||
3804 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
3805 | int nid, unsigned int lru_mask) | |
3806 | { | |
3807 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); | |
3808 | unsigned long nr = 0; | |
3809 | enum lru_list lru; | |
3810 | ||
3811 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3812 | ||
3813 | for_each_lru(lru) { | |
3814 | if (!(BIT(lru) & lru_mask)) | |
3815 | continue; | |
205b20cc | 3816 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); |
113b7dfd JW |
3817 | } |
3818 | return nr; | |
3819 | } | |
3820 | ||
3821 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
3822 | unsigned int lru_mask) | |
3823 | { | |
3824 | unsigned long nr = 0; | |
3825 | enum lru_list lru; | |
3826 | ||
3827 | for_each_lru(lru) { | |
3828 | if (!(BIT(lru) & lru_mask)) | |
3829 | continue; | |
205b20cc | 3830 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); |
113b7dfd JW |
3831 | } |
3832 | return nr; | |
3833 | } | |
3834 | ||
2da8ca82 | 3835 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3836 | { |
25485de6 GT |
3837 | struct numa_stat { |
3838 | const char *name; | |
3839 | unsigned int lru_mask; | |
3840 | }; | |
3841 | ||
3842 | static const struct numa_stat stats[] = { | |
3843 | { "total", LRU_ALL }, | |
3844 | { "file", LRU_ALL_FILE }, | |
3845 | { "anon", LRU_ALL_ANON }, | |
3846 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3847 | }; | |
3848 | const struct numa_stat *stat; | |
406eb0c9 | 3849 | int nid; |
25485de6 | 3850 | unsigned long nr; |
aa9694bb | 3851 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3852 | |
25485de6 GT |
3853 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3854 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3855 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3856 | for_each_node_state(nid, N_MEMORY) { | |
3857 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3858 | stat->lru_mask); | |
3859 | seq_printf(m, " N%d=%lu", nid, nr); | |
3860 | } | |
3861 | seq_putc(m, '\n'); | |
406eb0c9 | 3862 | } |
406eb0c9 | 3863 | |
071aee13 YH |
3864 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3865 | struct mem_cgroup *iter; | |
3866 | ||
3867 | nr = 0; | |
3868 | for_each_mem_cgroup_tree(iter, memcg) | |
3869 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3870 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3871 | for_each_node_state(nid, N_MEMORY) { | |
3872 | nr = 0; | |
3873 | for_each_mem_cgroup_tree(iter, memcg) | |
3874 | nr += mem_cgroup_node_nr_lru_pages( | |
3875 | iter, nid, stat->lru_mask); | |
3876 | seq_printf(m, " N%d=%lu", nid, nr); | |
3877 | } | |
3878 | seq_putc(m, '\n'); | |
406eb0c9 | 3879 | } |
406eb0c9 | 3880 | |
406eb0c9 YH |
3881 | return 0; |
3882 | } | |
3883 | #endif /* CONFIG_NUMA */ | |
3884 | ||
c8713d0b JW |
3885 | static const unsigned int memcg1_stats[] = { |
3886 | MEMCG_CACHE, | |
3887 | MEMCG_RSS, | |
3888 | MEMCG_RSS_HUGE, | |
3889 | NR_SHMEM, | |
3890 | NR_FILE_MAPPED, | |
3891 | NR_FILE_DIRTY, | |
3892 | NR_WRITEBACK, | |
3893 | MEMCG_SWAP, | |
3894 | }; | |
3895 | ||
3896 | static const char *const memcg1_stat_names[] = { | |
3897 | "cache", | |
3898 | "rss", | |
3899 | "rss_huge", | |
3900 | "shmem", | |
3901 | "mapped_file", | |
3902 | "dirty", | |
3903 | "writeback", | |
3904 | "swap", | |
3905 | }; | |
3906 | ||
df0e53d0 | 3907 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3908 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
3909 | PGPGIN, |
3910 | PGPGOUT, | |
3911 | PGFAULT, | |
3912 | PGMAJFAULT, | |
3913 | }; | |
3914 | ||
3915 | static const char *const memcg1_event_names[] = { | |
3916 | "pgpgin", | |
3917 | "pgpgout", | |
3918 | "pgfault", | |
3919 | "pgmajfault", | |
3920 | }; | |
3921 | ||
2da8ca82 | 3922 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3923 | { |
aa9694bb | 3924 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 3925 | unsigned long memory, memsw; |
af7c4b0e JW |
3926 | struct mem_cgroup *mi; |
3927 | unsigned int i; | |
406eb0c9 | 3928 | |
71cd3113 | 3929 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c RS |
3930 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3931 | ||
71cd3113 JW |
3932 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
3933 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) | |
1dd3a273 | 3934 | continue; |
71cd3113 | 3935 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], |
205b20cc | 3936 | memcg_page_state_local(memcg, memcg1_stats[i]) * |
71cd3113 | 3937 | PAGE_SIZE); |
1dd3a273 | 3938 | } |
7b854121 | 3939 | |
df0e53d0 JW |
3940 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3941 | seq_printf(m, "%s %lu\n", memcg1_event_names[i], | |
205b20cc | 3942 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
3943 | |
3944 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3945 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
205b20cc | 3946 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 3947 | PAGE_SIZE); |
af7c4b0e | 3948 | |
14067bb3 | 3949 | /* Hierarchical information */ |
3e32cb2e JW |
3950 | memory = memsw = PAGE_COUNTER_MAX; |
3951 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
bbec2e15 RG |
3952 | memory = min(memory, mi->memory.max); |
3953 | memsw = min(memsw, mi->memsw.max); | |
fee7b548 | 3954 | } |
3e32cb2e JW |
3955 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3956 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3957 | if (do_memsw_account()) |
3e32cb2e JW |
3958 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3959 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3960 | |
8de7ecc6 | 3961 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
71cd3113 | 3962 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3963 | continue; |
8de7ecc6 | 3964 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
dd923990 YS |
3965 | (u64)memcg_page_state(memcg, memcg1_stats[i]) * |
3966 | PAGE_SIZE); | |
af7c4b0e JW |
3967 | } |
3968 | ||
8de7ecc6 SB |
3969 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3970 | seq_printf(m, "total_%s %llu\n", memcg1_event_names[i], | |
dd923990 | 3971 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 3972 | |
8de7ecc6 SB |
3973 | for (i = 0; i < NR_LRU_LISTS; i++) |
3974 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], | |
42a30035 JW |
3975 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
3976 | PAGE_SIZE); | |
14067bb3 | 3977 | |
7f016ee8 | 3978 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 3979 | { |
ef8f2327 MG |
3980 | pg_data_t *pgdat; |
3981 | struct mem_cgroup_per_node *mz; | |
89abfab1 | 3982 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3983 | unsigned long recent_rotated[2] = {0, 0}; |
3984 | unsigned long recent_scanned[2] = {0, 0}; | |
3985 | ||
ef8f2327 MG |
3986 | for_each_online_pgdat(pgdat) { |
3987 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
3988 | rstat = &mz->lruvec.reclaim_stat; | |
7f016ee8 | 3989 | |
ef8f2327 MG |
3990 | recent_rotated[0] += rstat->recent_rotated[0]; |
3991 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3992 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3993 | recent_scanned[1] += rstat->recent_scanned[1]; | |
3994 | } | |
78ccf5b5 JW |
3995 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3996 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3997 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3998 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3999 | } |
4000 | #endif | |
4001 | ||
d2ceb9b7 KH |
4002 | return 0; |
4003 | } | |
4004 | ||
182446d0 TH |
4005 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
4006 | struct cftype *cft) | |
a7885eb8 | 4007 | { |
182446d0 | 4008 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4009 | |
1f4c025b | 4010 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4011 | } |
4012 | ||
182446d0 TH |
4013 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
4014 | struct cftype *cft, u64 val) | |
a7885eb8 | 4015 | { |
182446d0 | 4016 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 4017 | |
3dae7fec | 4018 | if (val > 100) |
a7885eb8 KM |
4019 | return -EINVAL; |
4020 | ||
14208b0e | 4021 | if (css->parent) |
3dae7fec JW |
4022 | memcg->swappiness = val; |
4023 | else | |
4024 | vm_swappiness = val; | |
068b38c1 | 4025 | |
a7885eb8 KM |
4026 | return 0; |
4027 | } | |
4028 | ||
2e72b634 KS |
4029 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4030 | { | |
4031 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 4032 | unsigned long usage; |
2e72b634 KS |
4033 | int i; |
4034 | ||
4035 | rcu_read_lock(); | |
4036 | if (!swap) | |
2c488db2 | 4037 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4038 | else |
2c488db2 | 4039 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4040 | |
4041 | if (!t) | |
4042 | goto unlock; | |
4043 | ||
ce00a967 | 4044 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
4045 | |
4046 | /* | |
748dad36 | 4047 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
4048 | * If it's not true, a threshold was crossed after last |
4049 | * call of __mem_cgroup_threshold(). | |
4050 | */ | |
5407a562 | 4051 | i = t->current_threshold; |
2e72b634 KS |
4052 | |
4053 | /* | |
4054 | * Iterate backward over array of thresholds starting from | |
4055 | * current_threshold and check if a threshold is crossed. | |
4056 | * If none of thresholds below usage is crossed, we read | |
4057 | * only one element of the array here. | |
4058 | */ | |
4059 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4060 | eventfd_signal(t->entries[i].eventfd, 1); | |
4061 | ||
4062 | /* i = current_threshold + 1 */ | |
4063 | i++; | |
4064 | ||
4065 | /* | |
4066 | * Iterate forward over array of thresholds starting from | |
4067 | * current_threshold+1 and check if a threshold is crossed. | |
4068 | * If none of thresholds above usage is crossed, we read | |
4069 | * only one element of the array here. | |
4070 | */ | |
4071 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4072 | eventfd_signal(t->entries[i].eventfd, 1); | |
4073 | ||
4074 | /* Update current_threshold */ | |
5407a562 | 4075 | t->current_threshold = i - 1; |
2e72b634 KS |
4076 | unlock: |
4077 | rcu_read_unlock(); | |
4078 | } | |
4079 | ||
4080 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4081 | { | |
ad4ca5f4 KS |
4082 | while (memcg) { |
4083 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 4084 | if (do_memsw_account()) |
ad4ca5f4 KS |
4085 | __mem_cgroup_threshold(memcg, true); |
4086 | ||
4087 | memcg = parent_mem_cgroup(memcg); | |
4088 | } | |
2e72b634 KS |
4089 | } |
4090 | ||
4091 | static int compare_thresholds(const void *a, const void *b) | |
4092 | { | |
4093 | const struct mem_cgroup_threshold *_a = a; | |
4094 | const struct mem_cgroup_threshold *_b = b; | |
4095 | ||
2bff24a3 GT |
4096 | if (_a->threshold > _b->threshold) |
4097 | return 1; | |
4098 | ||
4099 | if (_a->threshold < _b->threshold) | |
4100 | return -1; | |
4101 | ||
4102 | return 0; | |
2e72b634 KS |
4103 | } |
4104 | ||
c0ff4b85 | 4105 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4106 | { |
4107 | struct mem_cgroup_eventfd_list *ev; | |
4108 | ||
2bcf2e92 MH |
4109 | spin_lock(&memcg_oom_lock); |
4110 | ||
c0ff4b85 | 4111 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4112 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4113 | |
4114 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4115 | return 0; |
4116 | } | |
4117 | ||
c0ff4b85 | 4118 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4119 | { |
7d74b06f KH |
4120 | struct mem_cgroup *iter; |
4121 | ||
c0ff4b85 | 4122 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4123 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4124 | } |
4125 | ||
59b6f873 | 4126 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4127 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4128 | { |
2c488db2 KS |
4129 | struct mem_cgroup_thresholds *thresholds; |
4130 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4131 | unsigned long threshold; |
4132 | unsigned long usage; | |
2c488db2 | 4133 | int i, size, ret; |
2e72b634 | 4134 | |
650c5e56 | 4135 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4136 | if (ret) |
4137 | return ret; | |
4138 | ||
4139 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4140 | |
05b84301 | 4141 | if (type == _MEM) { |
2c488db2 | 4142 | thresholds = &memcg->thresholds; |
ce00a967 | 4143 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4144 | } else if (type == _MEMSWAP) { |
2c488db2 | 4145 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4146 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4147 | } else |
2e72b634 KS |
4148 | BUG(); |
4149 | ||
2e72b634 | 4150 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4151 | if (thresholds->primary) |
2e72b634 KS |
4152 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4153 | ||
2c488db2 | 4154 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4155 | |
4156 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4157 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4158 | if (!new) { |
2e72b634 KS |
4159 | ret = -ENOMEM; |
4160 | goto unlock; | |
4161 | } | |
2c488db2 | 4162 | new->size = size; |
2e72b634 KS |
4163 | |
4164 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4165 | if (thresholds->primary) { |
4166 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4167 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4168 | } |
4169 | ||
2e72b634 | 4170 | /* Add new threshold */ |
2c488db2 KS |
4171 | new->entries[size - 1].eventfd = eventfd; |
4172 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4173 | |
4174 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4175 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4176 | compare_thresholds, NULL); |
4177 | ||
4178 | /* Find current threshold */ | |
2c488db2 | 4179 | new->current_threshold = -1; |
2e72b634 | 4180 | for (i = 0; i < size; i++) { |
748dad36 | 4181 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4182 | /* |
2c488db2 KS |
4183 | * new->current_threshold will not be used until |
4184 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4185 | * it here. |
4186 | */ | |
2c488db2 | 4187 | ++new->current_threshold; |
748dad36 SZ |
4188 | } else |
4189 | break; | |
2e72b634 KS |
4190 | } |
4191 | ||
2c488db2 KS |
4192 | /* Free old spare buffer and save old primary buffer as spare */ |
4193 | kfree(thresholds->spare); | |
4194 | thresholds->spare = thresholds->primary; | |
4195 | ||
4196 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4197 | |
907860ed | 4198 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4199 | synchronize_rcu(); |
4200 | ||
2e72b634 KS |
4201 | unlock: |
4202 | mutex_unlock(&memcg->thresholds_lock); | |
4203 | ||
4204 | return ret; | |
4205 | } | |
4206 | ||
59b6f873 | 4207 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4208 | struct eventfd_ctx *eventfd, const char *args) |
4209 | { | |
59b6f873 | 4210 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4211 | } |
4212 | ||
59b6f873 | 4213 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4214 | struct eventfd_ctx *eventfd, const char *args) |
4215 | { | |
59b6f873 | 4216 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4217 | } |
4218 | ||
59b6f873 | 4219 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4220 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4221 | { |
2c488db2 KS |
4222 | struct mem_cgroup_thresholds *thresholds; |
4223 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4224 | unsigned long usage; |
18314d72 | 4225 | int i, j, size, entries; |
2e72b634 KS |
4226 | |
4227 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4228 | |
4229 | if (type == _MEM) { | |
2c488db2 | 4230 | thresholds = &memcg->thresholds; |
ce00a967 | 4231 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4232 | } else if (type == _MEMSWAP) { |
2c488db2 | 4233 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4234 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4235 | } else |
2e72b634 KS |
4236 | BUG(); |
4237 | ||
371528ca AV |
4238 | if (!thresholds->primary) |
4239 | goto unlock; | |
4240 | ||
2e72b634 KS |
4241 | /* Check if a threshold crossed before removing */ |
4242 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4243 | ||
4244 | /* Calculate new number of threshold */ | |
18314d72 | 4245 | size = entries = 0; |
2c488db2 KS |
4246 | for (i = 0; i < thresholds->primary->size; i++) { |
4247 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4248 | size++; |
18314d72 CX |
4249 | else |
4250 | entries++; | |
2e72b634 KS |
4251 | } |
4252 | ||
2c488db2 | 4253 | new = thresholds->spare; |
907860ed | 4254 | |
18314d72 CX |
4255 | /* If no items related to eventfd have been cleared, nothing to do */ |
4256 | if (!entries) | |
4257 | goto unlock; | |
4258 | ||
2e72b634 KS |
4259 | /* Set thresholds array to NULL if we don't have thresholds */ |
4260 | if (!size) { | |
2c488db2 KS |
4261 | kfree(new); |
4262 | new = NULL; | |
907860ed | 4263 | goto swap_buffers; |
2e72b634 KS |
4264 | } |
4265 | ||
2c488db2 | 4266 | new->size = size; |
2e72b634 KS |
4267 | |
4268 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4269 | new->current_threshold = -1; |
4270 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4271 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4272 | continue; |
4273 | ||
2c488db2 | 4274 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4275 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4276 | /* |
2c488db2 | 4277 | * new->current_threshold will not be used |
2e72b634 KS |
4278 | * until rcu_assign_pointer(), so it's safe to increment |
4279 | * it here. | |
4280 | */ | |
2c488db2 | 4281 | ++new->current_threshold; |
2e72b634 KS |
4282 | } |
4283 | j++; | |
4284 | } | |
4285 | ||
907860ed | 4286 | swap_buffers: |
2c488db2 KS |
4287 | /* Swap primary and spare array */ |
4288 | thresholds->spare = thresholds->primary; | |
8c757763 | 4289 | |
2c488db2 | 4290 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4291 | |
907860ed | 4292 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4293 | synchronize_rcu(); |
6611d8d7 MC |
4294 | |
4295 | /* If all events are unregistered, free the spare array */ | |
4296 | if (!new) { | |
4297 | kfree(thresholds->spare); | |
4298 | thresholds->spare = NULL; | |
4299 | } | |
371528ca | 4300 | unlock: |
2e72b634 | 4301 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4302 | } |
c1e862c1 | 4303 | |
59b6f873 | 4304 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4305 | struct eventfd_ctx *eventfd) |
4306 | { | |
59b6f873 | 4307 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4308 | } |
4309 | ||
59b6f873 | 4310 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4311 | struct eventfd_ctx *eventfd) |
4312 | { | |
59b6f873 | 4313 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4314 | } |
4315 | ||
59b6f873 | 4316 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4317 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4318 | { |
9490ff27 | 4319 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4320 | |
9490ff27 KH |
4321 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4322 | if (!event) | |
4323 | return -ENOMEM; | |
4324 | ||
1af8efe9 | 4325 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4326 | |
4327 | event->eventfd = eventfd; | |
4328 | list_add(&event->list, &memcg->oom_notify); | |
4329 | ||
4330 | /* already in OOM ? */ | |
c2b42d3c | 4331 | if (memcg->under_oom) |
9490ff27 | 4332 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4333 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4334 | |
4335 | return 0; | |
4336 | } | |
4337 | ||
59b6f873 | 4338 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4339 | struct eventfd_ctx *eventfd) |
9490ff27 | 4340 | { |
9490ff27 | 4341 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4342 | |
1af8efe9 | 4343 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4344 | |
c0ff4b85 | 4345 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4346 | if (ev->eventfd == eventfd) { |
4347 | list_del(&ev->list); | |
4348 | kfree(ev); | |
4349 | } | |
4350 | } | |
4351 | ||
1af8efe9 | 4352 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4353 | } |
4354 | ||
2da8ca82 | 4355 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4356 | { |
aa9694bb | 4357 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4358 | |
791badbd | 4359 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4360 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4361 | seq_printf(sf, "oom_kill %lu\n", |
4362 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4363 | return 0; |
4364 | } | |
4365 | ||
182446d0 | 4366 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4367 | struct cftype *cft, u64 val) |
4368 | { | |
182446d0 | 4369 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4370 | |
4371 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4372 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4373 | return -EINVAL; |
4374 | ||
c0ff4b85 | 4375 | memcg->oom_kill_disable = val; |
4d845ebf | 4376 | if (!val) |
c0ff4b85 | 4377 | memcg_oom_recover(memcg); |
3dae7fec | 4378 | |
3c11ecf4 KH |
4379 | return 0; |
4380 | } | |
4381 | ||
52ebea74 TH |
4382 | #ifdef CONFIG_CGROUP_WRITEBACK |
4383 | ||
3a8e9ac8 TH |
4384 | #include <trace/events/writeback.h> |
4385 | ||
841710aa TH |
4386 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4387 | { | |
4388 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4389 | } | |
4390 | ||
4391 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4392 | { | |
4393 | wb_domain_exit(&memcg->cgwb_domain); | |
4394 | } | |
4395 | ||
2529bb3a TH |
4396 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4397 | { | |
4398 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4399 | } | |
4400 | ||
841710aa TH |
4401 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4402 | { | |
4403 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4404 | ||
4405 | if (!memcg->css.parent) | |
4406 | return NULL; | |
4407 | ||
4408 | return &memcg->cgwb_domain; | |
4409 | } | |
4410 | ||
0b3d6e6f GT |
4411 | /* |
4412 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
4413 | * Keep in sync with memcg_exact_page(). | |
4414 | */ | |
4415 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
4416 | { | |
871789d4 | 4417 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
4418 | int cpu; |
4419 | ||
4420 | for_each_online_cpu(cpu) | |
871789d4 | 4421 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
4422 | if (x < 0) |
4423 | x = 0; | |
4424 | return x; | |
4425 | } | |
4426 | ||
c2aa723a TH |
4427 | /** |
4428 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4429 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4430 | * @pfilepages: out parameter for number of file pages |
4431 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4432 | * @pdirty: out parameter for number of dirty pages |
4433 | * @pwriteback: out parameter for number of pages under writeback | |
4434 | * | |
c5edf9cd TH |
4435 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4436 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4437 | * is a bit more involved. | |
c2aa723a | 4438 | * |
c5edf9cd TH |
4439 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4440 | * headroom is calculated as the lowest headroom of itself and the | |
4441 | * ancestors. Note that this doesn't consider the actual amount of | |
4442 | * available memory in the system. The caller should further cap | |
4443 | * *@pheadroom accordingly. | |
c2aa723a | 4444 | */ |
c5edf9cd TH |
4445 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4446 | unsigned long *pheadroom, unsigned long *pdirty, | |
4447 | unsigned long *pwriteback) | |
c2aa723a TH |
4448 | { |
4449 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4450 | struct mem_cgroup *parent; | |
c2aa723a | 4451 | |
0b3d6e6f | 4452 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a TH |
4453 | |
4454 | /* this should eventually include NR_UNSTABLE_NFS */ | |
0b3d6e6f | 4455 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4456 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4457 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4458 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4459 | |
c2aa723a | 4460 | while ((parent = parent_mem_cgroup(memcg))) { |
bbec2e15 | 4461 | unsigned long ceiling = min(memcg->memory.max, memcg->high); |
c2aa723a TH |
4462 | unsigned long used = page_counter_read(&memcg->memory); |
4463 | ||
c5edf9cd | 4464 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4465 | memcg = parent; |
4466 | } | |
c2aa723a TH |
4467 | } |
4468 | ||
97b27821 TH |
4469 | /* |
4470 | * Foreign dirty flushing | |
4471 | * | |
4472 | * There's an inherent mismatch between memcg and writeback. The former | |
4473 | * trackes ownership per-page while the latter per-inode. This was a | |
4474 | * deliberate design decision because honoring per-page ownership in the | |
4475 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4476 | * and deemed unnecessary given that write-sharing an inode across | |
4477 | * different cgroups isn't a common use-case. | |
4478 | * | |
4479 | * Combined with inode majority-writer ownership switching, this works well | |
4480 | * enough in most cases but there are some pathological cases. For | |
4481 | * example, let's say there are two cgroups A and B which keep writing to | |
4482 | * different but confined parts of the same inode. B owns the inode and | |
4483 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4484 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4485 | * triggering background writeback. A will be slowed down without a way to | |
4486 | * make writeback of the dirty pages happen. | |
4487 | * | |
4488 | * Conditions like the above can lead to a cgroup getting repatedly and | |
4489 | * severely throttled after making some progress after each | |
4490 | * dirty_expire_interval while the underyling IO device is almost | |
4491 | * completely idle. | |
4492 | * | |
4493 | * Solving this problem completely requires matching the ownership tracking | |
4494 | * granularities between memcg and writeback in either direction. However, | |
4495 | * the more egregious behaviors can be avoided by simply remembering the | |
4496 | * most recent foreign dirtying events and initiating remote flushes on | |
4497 | * them when local writeback isn't enough to keep the memory clean enough. | |
4498 | * | |
4499 | * The following two functions implement such mechanism. When a foreign | |
4500 | * page - a page whose memcg and writeback ownerships don't match - is | |
4501 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4502 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4503 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4504 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4505 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4506 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4507 | * limited to MEMCG_CGWB_FRN_CNT. | |
4508 | * | |
4509 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4510 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4511 | * records are lockless and racy. | |
4512 | */ | |
4513 | void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, | |
4514 | struct bdi_writeback *wb) | |
4515 | { | |
4516 | struct mem_cgroup *memcg = page->mem_cgroup; | |
4517 | struct memcg_cgwb_frn *frn; | |
4518 | u64 now = get_jiffies_64(); | |
4519 | u64 oldest_at = now; | |
4520 | int oldest = -1; | |
4521 | int i; | |
4522 | ||
3a8e9ac8 TH |
4523 | trace_track_foreign_dirty(page, wb); |
4524 | ||
97b27821 TH |
4525 | /* |
4526 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4527 | * using it. If not replace the oldest one which isn't being | |
4528 | * written out. | |
4529 | */ | |
4530 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4531 | frn = &memcg->cgwb_frn[i]; | |
4532 | if (frn->bdi_id == wb->bdi->id && | |
4533 | frn->memcg_id == wb->memcg_css->id) | |
4534 | break; | |
4535 | if (time_before64(frn->at, oldest_at) && | |
4536 | atomic_read(&frn->done.cnt) == 1) { | |
4537 | oldest = i; | |
4538 | oldest_at = frn->at; | |
4539 | } | |
4540 | } | |
4541 | ||
4542 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4543 | /* | |
4544 | * Re-using an existing one. Update timestamp lazily to | |
4545 | * avoid making the cacheline hot. We want them to be | |
4546 | * reasonably up-to-date and significantly shorter than | |
4547 | * dirty_expire_interval as that's what expires the record. | |
4548 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4549 | */ | |
4550 | unsigned long update_intv = | |
4551 | min_t(unsigned long, HZ, | |
4552 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4553 | ||
4554 | if (time_before64(frn->at, now - update_intv)) | |
4555 | frn->at = now; | |
4556 | } else if (oldest >= 0) { | |
4557 | /* replace the oldest free one */ | |
4558 | frn = &memcg->cgwb_frn[oldest]; | |
4559 | frn->bdi_id = wb->bdi->id; | |
4560 | frn->memcg_id = wb->memcg_css->id; | |
4561 | frn->at = now; | |
4562 | } | |
4563 | } | |
4564 | ||
4565 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4566 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4567 | { | |
4568 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4569 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4570 | u64 now = jiffies_64; | |
4571 | int i; | |
4572 | ||
4573 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4574 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4575 | ||
4576 | /* | |
4577 | * If the record is older than dirty_expire_interval, | |
4578 | * writeback on it has already started. No need to kick it | |
4579 | * off again. Also, don't start a new one if there's | |
4580 | * already one in flight. | |
4581 | */ | |
4582 | if (time_after64(frn->at, now - intv) && | |
4583 | atomic_read(&frn->done.cnt) == 1) { | |
4584 | frn->at = 0; | |
3a8e9ac8 | 4585 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
97b27821 TH |
4586 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0, |
4587 | WB_REASON_FOREIGN_FLUSH, | |
4588 | &frn->done); | |
4589 | } | |
4590 | } | |
4591 | } | |
4592 | ||
841710aa TH |
4593 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4594 | ||
4595 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4596 | { | |
4597 | return 0; | |
4598 | } | |
4599 | ||
4600 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4601 | { | |
4602 | } | |
4603 | ||
2529bb3a TH |
4604 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4605 | { | |
4606 | } | |
4607 | ||
52ebea74 TH |
4608 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4609 | ||
3bc942f3 TH |
4610 | /* |
4611 | * DO NOT USE IN NEW FILES. | |
4612 | * | |
4613 | * "cgroup.event_control" implementation. | |
4614 | * | |
4615 | * This is way over-engineered. It tries to support fully configurable | |
4616 | * events for each user. Such level of flexibility is completely | |
4617 | * unnecessary especially in the light of the planned unified hierarchy. | |
4618 | * | |
4619 | * Please deprecate this and replace with something simpler if at all | |
4620 | * possible. | |
4621 | */ | |
4622 | ||
79bd9814 TH |
4623 | /* |
4624 | * Unregister event and free resources. | |
4625 | * | |
4626 | * Gets called from workqueue. | |
4627 | */ | |
3bc942f3 | 4628 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4629 | { |
3bc942f3 TH |
4630 | struct mem_cgroup_event *event = |
4631 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4632 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4633 | |
4634 | remove_wait_queue(event->wqh, &event->wait); | |
4635 | ||
59b6f873 | 4636 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4637 | |
4638 | /* Notify userspace the event is going away. */ | |
4639 | eventfd_signal(event->eventfd, 1); | |
4640 | ||
4641 | eventfd_ctx_put(event->eventfd); | |
4642 | kfree(event); | |
59b6f873 | 4643 | css_put(&memcg->css); |
79bd9814 TH |
4644 | } |
4645 | ||
4646 | /* | |
a9a08845 | 4647 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4648 | * |
4649 | * Called with wqh->lock held and interrupts disabled. | |
4650 | */ | |
ac6424b9 | 4651 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4652 | int sync, void *key) |
79bd9814 | 4653 | { |
3bc942f3 TH |
4654 | struct mem_cgroup_event *event = |
4655 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4656 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4657 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4658 | |
a9a08845 | 4659 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4660 | /* |
4661 | * If the event has been detached at cgroup removal, we | |
4662 | * can simply return knowing the other side will cleanup | |
4663 | * for us. | |
4664 | * | |
4665 | * We can't race against event freeing since the other | |
4666 | * side will require wqh->lock via remove_wait_queue(), | |
4667 | * which we hold. | |
4668 | */ | |
fba94807 | 4669 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4670 | if (!list_empty(&event->list)) { |
4671 | list_del_init(&event->list); | |
4672 | /* | |
4673 | * We are in atomic context, but cgroup_event_remove() | |
4674 | * may sleep, so we have to call it in workqueue. | |
4675 | */ | |
4676 | schedule_work(&event->remove); | |
4677 | } | |
fba94807 | 4678 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4679 | } |
4680 | ||
4681 | return 0; | |
4682 | } | |
4683 | ||
3bc942f3 | 4684 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4685 | wait_queue_head_t *wqh, poll_table *pt) |
4686 | { | |
3bc942f3 TH |
4687 | struct mem_cgroup_event *event = |
4688 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4689 | |
4690 | event->wqh = wqh; | |
4691 | add_wait_queue(wqh, &event->wait); | |
4692 | } | |
4693 | ||
4694 | /* | |
3bc942f3 TH |
4695 | * DO NOT USE IN NEW FILES. |
4696 | * | |
79bd9814 TH |
4697 | * Parse input and register new cgroup event handler. |
4698 | * | |
4699 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4700 | * Interpretation of args is defined by control file implementation. | |
4701 | */ | |
451af504 TH |
4702 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4703 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4704 | { |
451af504 | 4705 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4706 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4707 | struct mem_cgroup_event *event; |
79bd9814 TH |
4708 | struct cgroup_subsys_state *cfile_css; |
4709 | unsigned int efd, cfd; | |
4710 | struct fd efile; | |
4711 | struct fd cfile; | |
fba94807 | 4712 | const char *name; |
79bd9814 TH |
4713 | char *endp; |
4714 | int ret; | |
4715 | ||
451af504 TH |
4716 | buf = strstrip(buf); |
4717 | ||
4718 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4719 | if (*endp != ' ') |
4720 | return -EINVAL; | |
451af504 | 4721 | buf = endp + 1; |
79bd9814 | 4722 | |
451af504 | 4723 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4724 | if ((*endp != ' ') && (*endp != '\0')) |
4725 | return -EINVAL; | |
451af504 | 4726 | buf = endp + 1; |
79bd9814 TH |
4727 | |
4728 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4729 | if (!event) | |
4730 | return -ENOMEM; | |
4731 | ||
59b6f873 | 4732 | event->memcg = memcg; |
79bd9814 | 4733 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4734 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4735 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4736 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4737 | |
4738 | efile = fdget(efd); | |
4739 | if (!efile.file) { | |
4740 | ret = -EBADF; | |
4741 | goto out_kfree; | |
4742 | } | |
4743 | ||
4744 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4745 | if (IS_ERR(event->eventfd)) { | |
4746 | ret = PTR_ERR(event->eventfd); | |
4747 | goto out_put_efile; | |
4748 | } | |
4749 | ||
4750 | cfile = fdget(cfd); | |
4751 | if (!cfile.file) { | |
4752 | ret = -EBADF; | |
4753 | goto out_put_eventfd; | |
4754 | } | |
4755 | ||
4756 | /* the process need read permission on control file */ | |
4757 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4758 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4759 | if (ret < 0) | |
4760 | goto out_put_cfile; | |
4761 | ||
fba94807 TH |
4762 | /* |
4763 | * Determine the event callbacks and set them in @event. This used | |
4764 | * to be done via struct cftype but cgroup core no longer knows | |
4765 | * about these events. The following is crude but the whole thing | |
4766 | * is for compatibility anyway. | |
3bc942f3 TH |
4767 | * |
4768 | * DO NOT ADD NEW FILES. | |
fba94807 | 4769 | */ |
b583043e | 4770 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4771 | |
4772 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4773 | event->register_event = mem_cgroup_usage_register_event; | |
4774 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4775 | } else if (!strcmp(name, "memory.oom_control")) { | |
4776 | event->register_event = mem_cgroup_oom_register_event; | |
4777 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4778 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4779 | event->register_event = vmpressure_register_event; | |
4780 | event->unregister_event = vmpressure_unregister_event; | |
4781 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4782 | event->register_event = memsw_cgroup_usage_register_event; |
4783 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4784 | } else { |
4785 | ret = -EINVAL; | |
4786 | goto out_put_cfile; | |
4787 | } | |
4788 | ||
79bd9814 | 4789 | /* |
b5557c4c TH |
4790 | * Verify @cfile should belong to @css. Also, remaining events are |
4791 | * automatically removed on cgroup destruction but the removal is | |
4792 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4793 | */ |
b583043e | 4794 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4795 | &memory_cgrp_subsys); |
79bd9814 | 4796 | ret = -EINVAL; |
5a17f543 | 4797 | if (IS_ERR(cfile_css)) |
79bd9814 | 4798 | goto out_put_cfile; |
5a17f543 TH |
4799 | if (cfile_css != css) { |
4800 | css_put(cfile_css); | |
79bd9814 | 4801 | goto out_put_cfile; |
5a17f543 | 4802 | } |
79bd9814 | 4803 | |
451af504 | 4804 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4805 | if (ret) |
4806 | goto out_put_css; | |
4807 | ||
9965ed17 | 4808 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4809 | |
fba94807 TH |
4810 | spin_lock(&memcg->event_list_lock); |
4811 | list_add(&event->list, &memcg->event_list); | |
4812 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4813 | |
4814 | fdput(cfile); | |
4815 | fdput(efile); | |
4816 | ||
451af504 | 4817 | return nbytes; |
79bd9814 TH |
4818 | |
4819 | out_put_css: | |
b5557c4c | 4820 | css_put(css); |
79bd9814 TH |
4821 | out_put_cfile: |
4822 | fdput(cfile); | |
4823 | out_put_eventfd: | |
4824 | eventfd_ctx_put(event->eventfd); | |
4825 | out_put_efile: | |
4826 | fdput(efile); | |
4827 | out_kfree: | |
4828 | kfree(event); | |
4829 | ||
4830 | return ret; | |
4831 | } | |
4832 | ||
241994ed | 4833 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4834 | { |
0eea1030 | 4835 | .name = "usage_in_bytes", |
8c7c6e34 | 4836 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4837 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4838 | }, |
c84872e1 PE |
4839 | { |
4840 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4841 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4842 | .write = mem_cgroup_reset, |
791badbd | 4843 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4844 | }, |
8cdea7c0 | 4845 | { |
0eea1030 | 4846 | .name = "limit_in_bytes", |
8c7c6e34 | 4847 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4848 | .write = mem_cgroup_write, |
791badbd | 4849 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4850 | }, |
296c81d8 BS |
4851 | { |
4852 | .name = "soft_limit_in_bytes", | |
4853 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4854 | .write = mem_cgroup_write, |
791badbd | 4855 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4856 | }, |
8cdea7c0 BS |
4857 | { |
4858 | .name = "failcnt", | |
8c7c6e34 | 4859 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4860 | .write = mem_cgroup_reset, |
791badbd | 4861 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4862 | }, |
d2ceb9b7 KH |
4863 | { |
4864 | .name = "stat", | |
2da8ca82 | 4865 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4866 | }, |
c1e862c1 KH |
4867 | { |
4868 | .name = "force_empty", | |
6770c64e | 4869 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4870 | }, |
18f59ea7 BS |
4871 | { |
4872 | .name = "use_hierarchy", | |
4873 | .write_u64 = mem_cgroup_hierarchy_write, | |
4874 | .read_u64 = mem_cgroup_hierarchy_read, | |
4875 | }, | |
79bd9814 | 4876 | { |
3bc942f3 | 4877 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4878 | .write = memcg_write_event_control, |
7dbdb199 | 4879 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4880 | }, |
a7885eb8 KM |
4881 | { |
4882 | .name = "swappiness", | |
4883 | .read_u64 = mem_cgroup_swappiness_read, | |
4884 | .write_u64 = mem_cgroup_swappiness_write, | |
4885 | }, | |
7dc74be0 DN |
4886 | { |
4887 | .name = "move_charge_at_immigrate", | |
4888 | .read_u64 = mem_cgroup_move_charge_read, | |
4889 | .write_u64 = mem_cgroup_move_charge_write, | |
4890 | }, | |
9490ff27 KH |
4891 | { |
4892 | .name = "oom_control", | |
2da8ca82 | 4893 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4894 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4895 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4896 | }, | |
70ddf637 AV |
4897 | { |
4898 | .name = "pressure_level", | |
70ddf637 | 4899 | }, |
406eb0c9 YH |
4900 | #ifdef CONFIG_NUMA |
4901 | { | |
4902 | .name = "numa_stat", | |
2da8ca82 | 4903 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4904 | }, |
4905 | #endif | |
510fc4e1 GC |
4906 | { |
4907 | .name = "kmem.limit_in_bytes", | |
4908 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4909 | .write = mem_cgroup_write, |
791badbd | 4910 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4911 | }, |
4912 | { | |
4913 | .name = "kmem.usage_in_bytes", | |
4914 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4915 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4916 | }, |
4917 | { | |
4918 | .name = "kmem.failcnt", | |
4919 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4920 | .write = mem_cgroup_reset, |
791badbd | 4921 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4922 | }, |
4923 | { | |
4924 | .name = "kmem.max_usage_in_bytes", | |
4925 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4926 | .write = mem_cgroup_reset, |
791badbd | 4927 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4928 | }, |
5b365771 | 4929 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
749c5415 GC |
4930 | { |
4931 | .name = "kmem.slabinfo", | |
bc2791f8 TH |
4932 | .seq_start = memcg_slab_start, |
4933 | .seq_next = memcg_slab_next, | |
4934 | .seq_stop = memcg_slab_stop, | |
b047501c | 4935 | .seq_show = memcg_slab_show, |
749c5415 GC |
4936 | }, |
4937 | #endif | |
d55f90bf VD |
4938 | { |
4939 | .name = "kmem.tcp.limit_in_bytes", | |
4940 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4941 | .write = mem_cgroup_write, | |
4942 | .read_u64 = mem_cgroup_read_u64, | |
4943 | }, | |
4944 | { | |
4945 | .name = "kmem.tcp.usage_in_bytes", | |
4946 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4947 | .read_u64 = mem_cgroup_read_u64, | |
4948 | }, | |
4949 | { | |
4950 | .name = "kmem.tcp.failcnt", | |
4951 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4952 | .write = mem_cgroup_reset, | |
4953 | .read_u64 = mem_cgroup_read_u64, | |
4954 | }, | |
4955 | { | |
4956 | .name = "kmem.tcp.max_usage_in_bytes", | |
4957 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4958 | .write = mem_cgroup_reset, | |
4959 | .read_u64 = mem_cgroup_read_u64, | |
4960 | }, | |
6bc10349 | 4961 | { }, /* terminate */ |
af36f906 | 4962 | }; |
8c7c6e34 | 4963 | |
73f576c0 JW |
4964 | /* |
4965 | * Private memory cgroup IDR | |
4966 | * | |
4967 | * Swap-out records and page cache shadow entries need to store memcg | |
4968 | * references in constrained space, so we maintain an ID space that is | |
4969 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4970 | * memory-controlled cgroups to 64k. | |
4971 | * | |
4972 | * However, there usually are many references to the oflline CSS after | |
4973 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4974 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4975 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4976 | * relatively small ID space and prevent the creation of new cgroups | |
4977 | * even when there are much fewer than 64k cgroups - possibly none. | |
4978 | * | |
4979 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4980 | * be freed and recycled when it's no longer needed, which is usually | |
4981 | * when the CSS is offlined. | |
4982 | * | |
4983 | * The only exception to that are records of swapped out tmpfs/shmem | |
4984 | * pages that need to be attributed to live ancestors on swapin. But | |
4985 | * those references are manageable from userspace. | |
4986 | */ | |
4987 | ||
4988 | static DEFINE_IDR(mem_cgroup_idr); | |
4989 | ||
7e97de0b KT |
4990 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
4991 | { | |
4992 | if (memcg->id.id > 0) { | |
4993 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
4994 | memcg->id.id = 0; | |
4995 | } | |
4996 | } | |
4997 | ||
615d66c3 | 4998 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4999 | { |
1c2d479a | 5000 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
5001 | } |
5002 | ||
615d66c3 | 5003 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 5004 | { |
1c2d479a | 5005 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 5006 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
5007 | |
5008 | /* Memcg ID pins CSS */ | |
5009 | css_put(&memcg->css); | |
5010 | } | |
5011 | } | |
5012 | ||
615d66c3 VD |
5013 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
5014 | { | |
5015 | mem_cgroup_id_put_many(memcg, 1); | |
5016 | } | |
5017 | ||
73f576c0 JW |
5018 | /** |
5019 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
5020 | * @id: the memcg id to look up | |
5021 | * | |
5022 | * Caller must hold rcu_read_lock(). | |
5023 | */ | |
5024 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
5025 | { | |
5026 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5027 | return idr_find(&mem_cgroup_idr, id); | |
5028 | } | |
5029 | ||
ef8f2327 | 5030 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
5031 | { |
5032 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 5033 | int tmp = node; |
1ecaab2b KH |
5034 | /* |
5035 | * This routine is called against possible nodes. | |
5036 | * But it's BUG to call kmalloc() against offline node. | |
5037 | * | |
5038 | * TODO: this routine can waste much memory for nodes which will | |
5039 | * never be onlined. It's better to use memory hotplug callback | |
5040 | * function. | |
5041 | */ | |
41e3355d KH |
5042 | if (!node_state(node, N_NORMAL_MEMORY)) |
5043 | tmp = -1; | |
17295c88 | 5044 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
5045 | if (!pn) |
5046 | return 1; | |
1ecaab2b | 5047 | |
815744d7 JW |
5048 | pn->lruvec_stat_local = alloc_percpu(struct lruvec_stat); |
5049 | if (!pn->lruvec_stat_local) { | |
5050 | kfree(pn); | |
5051 | return 1; | |
5052 | } | |
5053 | ||
a983b5eb JW |
5054 | pn->lruvec_stat_cpu = alloc_percpu(struct lruvec_stat); |
5055 | if (!pn->lruvec_stat_cpu) { | |
815744d7 | 5056 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c JW |
5057 | kfree(pn); |
5058 | return 1; | |
5059 | } | |
5060 | ||
ef8f2327 MG |
5061 | lruvec_init(&pn->lruvec); |
5062 | pn->usage_in_excess = 0; | |
5063 | pn->on_tree = false; | |
5064 | pn->memcg = memcg; | |
5065 | ||
54f72fe0 | 5066 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
5067 | return 0; |
5068 | } | |
5069 | ||
ef8f2327 | 5070 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 5071 | { |
00f3ca2c JW |
5072 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
5073 | ||
4eaf431f MH |
5074 | if (!pn) |
5075 | return; | |
5076 | ||
a983b5eb | 5077 | free_percpu(pn->lruvec_stat_cpu); |
815744d7 | 5078 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c | 5079 | kfree(pn); |
1ecaab2b KH |
5080 | } |
5081 | ||
40e952f9 | 5082 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 5083 | { |
c8b2a36f | 5084 | int node; |
59927fb9 | 5085 | |
c8b2a36f | 5086 | for_each_node(node) |
ef8f2327 | 5087 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 5088 | free_percpu(memcg->vmstats_percpu); |
815744d7 | 5089 | free_percpu(memcg->vmstats_local); |
8ff69e2c | 5090 | kfree(memcg); |
59927fb9 | 5091 | } |
3afe36b1 | 5092 | |
40e952f9 TE |
5093 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
5094 | { | |
5095 | memcg_wb_domain_exit(memcg); | |
7961eee3 SB |
5096 | /* |
5097 | * Flush percpu vmstats and vmevents to guarantee the value correctness | |
5098 | * on parent's and all ancestor levels. | |
5099 | */ | |
c2f70949 | 5100 | memcg_flush_percpu_vmstats(memcg); |
7961eee3 | 5101 | memcg_flush_percpu_vmevents(memcg); |
40e952f9 TE |
5102 | __mem_cgroup_free(memcg); |
5103 | } | |
5104 | ||
0b8f73e1 | 5105 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 5106 | { |
d142e3e6 | 5107 | struct mem_cgroup *memcg; |
b9726c26 | 5108 | unsigned int size; |
6d12e2d8 | 5109 | int node; |
97b27821 | 5110 | int __maybe_unused i; |
b9f426a6 | 5111 | long error = -ENOMEM; |
8cdea7c0 | 5112 | |
0b8f73e1 JW |
5113 | size = sizeof(struct mem_cgroup); |
5114 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
5115 | ||
5116 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 5117 | if (!memcg) |
b9f426a6 | 5118 | return ERR_PTR(error); |
0b8f73e1 | 5119 | |
73f576c0 JW |
5120 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
5121 | 1, MEM_CGROUP_ID_MAX, | |
5122 | GFP_KERNEL); | |
b9f426a6 YS |
5123 | if (memcg->id.id < 0) { |
5124 | error = memcg->id.id; | |
73f576c0 | 5125 | goto fail; |
b9f426a6 | 5126 | } |
73f576c0 | 5127 | |
815744d7 JW |
5128 | memcg->vmstats_local = alloc_percpu(struct memcg_vmstats_percpu); |
5129 | if (!memcg->vmstats_local) | |
5130 | goto fail; | |
5131 | ||
871789d4 CD |
5132 | memcg->vmstats_percpu = alloc_percpu(struct memcg_vmstats_percpu); |
5133 | if (!memcg->vmstats_percpu) | |
0b8f73e1 | 5134 | goto fail; |
78fb7466 | 5135 | |
3ed28fa1 | 5136 | for_each_node(node) |
ef8f2327 | 5137 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5138 | goto fail; |
f64c3f54 | 5139 | |
0b8f73e1 JW |
5140 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5141 | goto fail; | |
28dbc4b6 | 5142 | |
f7e1cb6e | 5143 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
5144 | memcg->last_scanned_node = MAX_NUMNODES; |
5145 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
5146 | mutex_init(&memcg->thresholds_lock); |
5147 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5148 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5149 | INIT_LIST_HEAD(&memcg->event_list); |
5150 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5151 | memcg->socket_pressure = jiffies; |
84c07d11 | 5152 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5153 | memcg->kmemcg_id = -1; |
900a38f0 | 5154 | #endif |
52ebea74 TH |
5155 | #ifdef CONFIG_CGROUP_WRITEBACK |
5156 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5157 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5158 | memcg->cgwb_frn[i].done = | |
5159 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5160 | #endif |
5161 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5162 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5163 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5164 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5165 | #endif |
73f576c0 | 5166 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
5167 | return memcg; |
5168 | fail: | |
7e97de0b | 5169 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5170 | __mem_cgroup_free(memcg); |
b9f426a6 | 5171 | return ERR_PTR(error); |
d142e3e6 GC |
5172 | } |
5173 | ||
0b8f73e1 JW |
5174 | static struct cgroup_subsys_state * __ref |
5175 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5176 | { |
0b8f73e1 JW |
5177 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
5178 | struct mem_cgroup *memcg; | |
5179 | long error = -ENOMEM; | |
d142e3e6 | 5180 | |
0b8f73e1 | 5181 | memcg = mem_cgroup_alloc(); |
b9f426a6 YS |
5182 | if (IS_ERR(memcg)) |
5183 | return ERR_CAST(memcg); | |
d142e3e6 | 5184 | |
0b8f73e1 JW |
5185 | memcg->high = PAGE_COUNTER_MAX; |
5186 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
5187 | if (parent) { | |
5188 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5189 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
5190 | } | |
5191 | if (parent && parent->use_hierarchy) { | |
5192 | memcg->use_hierarchy = true; | |
3e32cb2e | 5193 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5194 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
5195 | page_counter_init(&memcg->memsw, &parent->memsw); |
5196 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 5197 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5198 | } else { |
3e32cb2e | 5199 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 5200 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
5201 | page_counter_init(&memcg->memsw, NULL); |
5202 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 5203 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
5204 | /* |
5205 | * Deeper hierachy with use_hierarchy == false doesn't make | |
5206 | * much sense so let cgroup subsystem know about this | |
5207 | * unfortunate state in our controller. | |
5208 | */ | |
d142e3e6 | 5209 | if (parent != root_mem_cgroup) |
073219e9 | 5210 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 5211 | } |
d6441637 | 5212 | |
0b8f73e1 JW |
5213 | /* The following stuff does not apply to the root */ |
5214 | if (!parent) { | |
fb2f2b0a RG |
5215 | #ifdef CONFIG_MEMCG_KMEM |
5216 | INIT_LIST_HEAD(&memcg->kmem_caches); | |
5217 | #endif | |
0b8f73e1 JW |
5218 | root_mem_cgroup = memcg; |
5219 | return &memcg->css; | |
5220 | } | |
5221 | ||
b313aeee | 5222 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
5223 | if (error) |
5224 | goto fail; | |
127424c8 | 5225 | |
f7e1cb6e | 5226 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5227 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5228 | |
0b8f73e1 JW |
5229 | return &memcg->css; |
5230 | fail: | |
7e97de0b | 5231 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 5232 | mem_cgroup_free(memcg); |
b9f426a6 | 5233 | return ERR_PTR(error); |
0b8f73e1 JW |
5234 | } |
5235 | ||
73f576c0 | 5236 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5237 | { |
58fa2a55 VD |
5238 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5239 | ||
0a4465d3 KT |
5240 | /* |
5241 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
5242 | * by the time the maps are allocated. So, we allocate maps | |
5243 | * here, when for_each_mem_cgroup() can't skip it. | |
5244 | */ | |
5245 | if (memcg_alloc_shrinker_maps(memcg)) { | |
5246 | mem_cgroup_id_remove(memcg); | |
5247 | return -ENOMEM; | |
5248 | } | |
5249 | ||
73f576c0 | 5250 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5251 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5252 | css_get(css); |
2f7dd7a4 | 5253 | return 0; |
8cdea7c0 BS |
5254 | } |
5255 | ||
eb95419b | 5256 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5257 | { |
eb95419b | 5258 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5259 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5260 | |
5261 | /* | |
5262 | * Unregister events and notify userspace. | |
5263 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5264 | * directory to avoid race between userspace and kernelspace. | |
5265 | */ | |
fba94807 TH |
5266 | spin_lock(&memcg->event_list_lock); |
5267 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5268 | list_del_init(&event->list); |
5269 | schedule_work(&event->remove); | |
5270 | } | |
fba94807 | 5271 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5272 | |
bf8d5d52 | 5273 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5274 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5275 | |
567e9ab2 | 5276 | memcg_offline_kmem(memcg); |
52ebea74 | 5277 | wb_memcg_offline(memcg); |
73f576c0 | 5278 | |
591edfb1 RG |
5279 | drain_all_stock(memcg); |
5280 | ||
73f576c0 | 5281 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5282 | } |
5283 | ||
6df38689 VD |
5284 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5285 | { | |
5286 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5287 | ||
5288 | invalidate_reclaim_iterators(memcg); | |
5289 | } | |
5290 | ||
eb95419b | 5291 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5292 | { |
eb95419b | 5293 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5294 | int __maybe_unused i; |
c268e994 | 5295 | |
97b27821 TH |
5296 | #ifdef CONFIG_CGROUP_WRITEBACK |
5297 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5298 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5299 | #endif | |
f7e1cb6e | 5300 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5301 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5302 | |
0db15298 | 5303 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5304 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5305 | |
0b8f73e1 JW |
5306 | vmpressure_cleanup(&memcg->vmpressure); |
5307 | cancel_work_sync(&memcg->high_work); | |
5308 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 5309 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 5310 | memcg_free_kmem(memcg); |
0b8f73e1 | 5311 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5312 | } |
5313 | ||
1ced953b TH |
5314 | /** |
5315 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5316 | * @css: the target css | |
5317 | * | |
5318 | * Reset the states of the mem_cgroup associated with @css. This is | |
5319 | * invoked when the userland requests disabling on the default hierarchy | |
5320 | * but the memcg is pinned through dependency. The memcg should stop | |
5321 | * applying policies and should revert to the vanilla state as it may be | |
5322 | * made visible again. | |
5323 | * | |
5324 | * The current implementation only resets the essential configurations. | |
5325 | * This needs to be expanded to cover all the visible parts. | |
5326 | */ | |
5327 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5328 | { | |
5329 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5330 | ||
bbec2e15 RG |
5331 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5332 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
5333 | page_counter_set_max(&memcg->memsw, PAGE_COUNTER_MAX); | |
5334 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); | |
5335 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5336 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5337 | page_counter_set_low(&memcg->memory, 0); |
241994ed | 5338 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 5339 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 5340 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5341 | } |
5342 | ||
02491447 | 5343 | #ifdef CONFIG_MMU |
7dc74be0 | 5344 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5345 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5346 | { |
05b84301 | 5347 | int ret; |
9476db97 | 5348 | |
d0164adc MG |
5349 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5350 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5351 | if (!ret) { |
854ffa8d | 5352 | mc.precharge += count; |
854ffa8d DN |
5353 | return ret; |
5354 | } | |
9476db97 | 5355 | |
3674534b | 5356 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5357 | while (count--) { |
3674534b | 5358 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5359 | if (ret) |
38c5d72f | 5360 | return ret; |
854ffa8d | 5361 | mc.precharge++; |
9476db97 | 5362 | cond_resched(); |
854ffa8d | 5363 | } |
9476db97 | 5364 | return 0; |
4ffef5fe DN |
5365 | } |
5366 | ||
4ffef5fe DN |
5367 | union mc_target { |
5368 | struct page *page; | |
02491447 | 5369 | swp_entry_t ent; |
4ffef5fe DN |
5370 | }; |
5371 | ||
4ffef5fe | 5372 | enum mc_target_type { |
8d32ff84 | 5373 | MC_TARGET_NONE = 0, |
4ffef5fe | 5374 | MC_TARGET_PAGE, |
02491447 | 5375 | MC_TARGET_SWAP, |
c733a828 | 5376 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5377 | }; |
5378 | ||
90254a65 DN |
5379 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5380 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5381 | { |
25b2995a | 5382 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5383 | |
90254a65 DN |
5384 | if (!page || !page_mapped(page)) |
5385 | return NULL; | |
5386 | if (PageAnon(page)) { | |
1dfab5ab | 5387 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5388 | return NULL; |
1dfab5ab JW |
5389 | } else { |
5390 | if (!(mc.flags & MOVE_FILE)) | |
5391 | return NULL; | |
5392 | } | |
90254a65 DN |
5393 | if (!get_page_unless_zero(page)) |
5394 | return NULL; | |
5395 | ||
5396 | return page; | |
5397 | } | |
5398 | ||
c733a828 | 5399 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5400 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5401 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5402 | { |
90254a65 DN |
5403 | struct page *page = NULL; |
5404 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5405 | ||
569434e2 | 5406 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5407 | return NULL; |
c733a828 JG |
5408 | |
5409 | /* | |
5410 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
5411 | * a device and because they are not accessible by CPU they are store | |
5412 | * as special swap entry in the CPU page table. | |
5413 | */ | |
5414 | if (is_device_private_entry(ent)) { | |
5415 | page = device_private_entry_to_page(ent); | |
5416 | /* | |
5417 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
5418 | * a refcount of 1 when free (unlike normal page) | |
5419 | */ | |
5420 | if (!page_ref_add_unless(page, 1, 1)) | |
5421 | return NULL; | |
5422 | return page; | |
5423 | } | |
5424 | ||
569434e2 RC |
5425 | if (non_swap_entry(ent)) |
5426 | return NULL; | |
5427 | ||
4b91355e KH |
5428 | /* |
5429 | * Because lookup_swap_cache() updates some statistics counter, | |
5430 | * we call find_get_page() with swapper_space directly. | |
5431 | */ | |
f6ab1f7f | 5432 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
7941d214 | 5433 | if (do_memsw_account()) |
90254a65 DN |
5434 | entry->val = ent.val; |
5435 | ||
5436 | return page; | |
5437 | } | |
4b91355e KH |
5438 | #else |
5439 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5440 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5441 | { |
5442 | return NULL; | |
5443 | } | |
5444 | #endif | |
90254a65 | 5445 | |
87946a72 DN |
5446 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5447 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5448 | { | |
5449 | struct page *page = NULL; | |
87946a72 DN |
5450 | struct address_space *mapping; |
5451 | pgoff_t pgoff; | |
5452 | ||
5453 | if (!vma->vm_file) /* anonymous vma */ | |
5454 | return NULL; | |
1dfab5ab | 5455 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5456 | return NULL; |
5457 | ||
87946a72 | 5458 | mapping = vma->vm_file->f_mapping; |
0661a336 | 5459 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
5460 | |
5461 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5462 | #ifdef CONFIG_SWAP |
5463 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
5464 | if (shmem_mapping(mapping)) { |
5465 | page = find_get_entry(mapping, pgoff); | |
3159f943 | 5466 | if (xa_is_value(page)) { |
139b6a6f | 5467 | swp_entry_t swp = radix_to_swp_entry(page); |
7941d214 | 5468 | if (do_memsw_account()) |
139b6a6f | 5469 | *entry = swp; |
f6ab1f7f HY |
5470 | page = find_get_page(swap_address_space(swp), |
5471 | swp_offset(swp)); | |
139b6a6f JW |
5472 | } |
5473 | } else | |
5474 | page = find_get_page(mapping, pgoff); | |
5475 | #else | |
5476 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 5477 | #endif |
87946a72 DN |
5478 | return page; |
5479 | } | |
5480 | ||
b1b0deab CG |
5481 | /** |
5482 | * mem_cgroup_move_account - move account of the page | |
5483 | * @page: the page | |
25843c2b | 5484 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5485 | * @from: mem_cgroup which the page is moved from. |
5486 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5487 | * | |
3ac808fd | 5488 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5489 | * |
5490 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5491 | * from old cgroup. | |
5492 | */ | |
5493 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5494 | bool compound, |
b1b0deab CG |
5495 | struct mem_cgroup *from, |
5496 | struct mem_cgroup *to) | |
5497 | { | |
ae8af438 KK |
5498 | struct lruvec *from_vec, *to_vec; |
5499 | struct pglist_data *pgdat; | |
f627c2f5 | 5500 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 5501 | int ret; |
c4843a75 | 5502 | bool anon; |
b1b0deab CG |
5503 | |
5504 | VM_BUG_ON(from == to); | |
5505 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5506 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5507 | |
5508 | /* | |
6a93ca8f | 5509 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 5510 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 5511 | */ |
f627c2f5 | 5512 | ret = -EBUSY; |
b1b0deab CG |
5513 | if (!trylock_page(page)) |
5514 | goto out; | |
5515 | ||
5516 | ret = -EINVAL; | |
5517 | if (page->mem_cgroup != from) | |
5518 | goto out_unlock; | |
5519 | ||
c4843a75 GT |
5520 | anon = PageAnon(page); |
5521 | ||
ae8af438 KK |
5522 | pgdat = page_pgdat(page); |
5523 | from_vec = mem_cgroup_lruvec(pgdat, from); | |
5524 | to_vec = mem_cgroup_lruvec(pgdat, to); | |
5525 | ||
a4e242b1 | 5526 | lock_page_memcg(page); |
b1b0deab | 5527 | |
c4843a75 | 5528 | if (!anon && page_mapped(page)) { |
ae8af438 KK |
5529 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); |
5530 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
b1b0deab CG |
5531 | } |
5532 | ||
c4843a75 GT |
5533 | if (!anon && PageDirty(page)) { |
5534 | struct address_space *mapping = page_mapping(page); | |
5535 | ||
5536 | if (mapping_cap_account_dirty(mapping)) { | |
ae8af438 KK |
5537 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, -nr_pages); |
5538 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, nr_pages); | |
c4843a75 GT |
5539 | } |
5540 | } | |
5541 | ||
b1b0deab | 5542 | if (PageWriteback(page)) { |
ae8af438 KK |
5543 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5544 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5545 | } |
5546 | ||
5547 | /* | |
a4e242b1 JW |
5548 | * All state has been migrated, let's switch to the new memcg. |
5549 | * | |
b1b0deab | 5550 | * It is safe to change page->mem_cgroup here because the page |
a4e242b1 JW |
5551 | * is referenced, charged, isolated, and locked: we can't race |
5552 | * with (un)charging, migration, LRU putback, or anything else | |
5553 | * that would rely on a stable page->mem_cgroup. | |
5554 | * | |
5555 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
5556 | * to save space. As soon as we switch page->mem_cgroup to a | |
5557 | * new memcg that isn't locked, the above state can change | |
5558 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5559 | */ |
a4e242b1 | 5560 | smp_mb(); |
b1b0deab | 5561 | |
a4e242b1 | 5562 | page->mem_cgroup = to; /* caller should have done css_get */ |
87eaceb3 | 5563 | |
a4e242b1 | 5564 | __unlock_page_memcg(from); |
b1b0deab CG |
5565 | |
5566 | ret = 0; | |
5567 | ||
5568 | local_irq_disable(); | |
f627c2f5 | 5569 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 5570 | memcg_check_events(to, page); |
f627c2f5 | 5571 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
5572 | memcg_check_events(from, page); |
5573 | local_irq_enable(); | |
5574 | out_unlock: | |
5575 | unlock_page(page); | |
5576 | out: | |
5577 | return ret; | |
5578 | } | |
5579 | ||
7cf7806c LR |
5580 | /** |
5581 | * get_mctgt_type - get target type of moving charge | |
5582 | * @vma: the vma the pte to be checked belongs | |
5583 | * @addr: the address corresponding to the pte to be checked | |
5584 | * @ptent: the pte to be checked | |
5585 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5586 | * | |
5587 | * Returns | |
5588 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5589 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5590 | * move charge. if @target is not NULL, the page is stored in target->page | |
5591 | * with extra refcnt got(Callers should handle it). | |
5592 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5593 | * target for charge migration. if @target is not NULL, the entry is stored | |
5594 | * in target->ent. | |
25b2995a CH |
5595 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
5596 | * (so ZONE_DEVICE page and thus not on the lru). | |
df6ad698 JG |
5597 | * For now we such page is charge like a regular page would be as for all |
5598 | * intent and purposes it is just special memory taking the place of a | |
5599 | * regular page. | |
c733a828 JG |
5600 | * |
5601 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5602 | * |
5603 | * Called with pte lock held. | |
5604 | */ | |
5605 | ||
8d32ff84 | 5606 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5607 | unsigned long addr, pte_t ptent, union mc_target *target) |
5608 | { | |
5609 | struct page *page = NULL; | |
8d32ff84 | 5610 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5611 | swp_entry_t ent = { .val = 0 }; |
5612 | ||
5613 | if (pte_present(ptent)) | |
5614 | page = mc_handle_present_pte(vma, addr, ptent); | |
5615 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5616 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5617 | else if (pte_none(ptent)) |
87946a72 | 5618 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5619 | |
5620 | if (!page && !ent.val) | |
8d32ff84 | 5621 | return ret; |
02491447 | 5622 | if (page) { |
02491447 | 5623 | /* |
0a31bc97 | 5624 | * Do only loose check w/o serialization. |
1306a85a | 5625 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5626 | * not under LRU exclusion. |
02491447 | 5627 | */ |
1306a85a | 5628 | if (page->mem_cgroup == mc.from) { |
02491447 | 5629 | ret = MC_TARGET_PAGE; |
25b2995a | 5630 | if (is_device_private_page(page)) |
c733a828 | 5631 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5632 | if (target) |
5633 | target->page = page; | |
5634 | } | |
5635 | if (!ret || !target) | |
5636 | put_page(page); | |
5637 | } | |
3e14a57b HY |
5638 | /* |
5639 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5640 | * But we cannot move a tail-page in a THP. | |
5641 | */ | |
5642 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5643 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5644 | ret = MC_TARGET_SWAP; |
5645 | if (target) | |
5646 | target->ent = ent; | |
4ffef5fe | 5647 | } |
4ffef5fe DN |
5648 | return ret; |
5649 | } | |
5650 | ||
12724850 NH |
5651 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5652 | /* | |
d6810d73 HY |
5653 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5654 | * not support them for now. | |
12724850 NH |
5655 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5656 | */ | |
5657 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5658 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5659 | { | |
5660 | struct page *page = NULL; | |
12724850 NH |
5661 | enum mc_target_type ret = MC_TARGET_NONE; |
5662 | ||
84c3fc4e ZY |
5663 | if (unlikely(is_swap_pmd(pmd))) { |
5664 | VM_BUG_ON(thp_migration_supported() && | |
5665 | !is_pmd_migration_entry(pmd)); | |
5666 | return ret; | |
5667 | } | |
12724850 | 5668 | page = pmd_page(pmd); |
309381fe | 5669 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5670 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5671 | return ret; |
1306a85a | 5672 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
5673 | ret = MC_TARGET_PAGE; |
5674 | if (target) { | |
5675 | get_page(page); | |
5676 | target->page = page; | |
5677 | } | |
5678 | } | |
5679 | return ret; | |
5680 | } | |
5681 | #else | |
5682 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5683 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5684 | { | |
5685 | return MC_TARGET_NONE; | |
5686 | } | |
5687 | #endif | |
5688 | ||
4ffef5fe DN |
5689 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5690 | unsigned long addr, unsigned long end, | |
5691 | struct mm_walk *walk) | |
5692 | { | |
26bcd64a | 5693 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5694 | pte_t *pte; |
5695 | spinlock_t *ptl; | |
5696 | ||
b6ec57f4 KS |
5697 | ptl = pmd_trans_huge_lock(pmd, vma); |
5698 | if (ptl) { | |
c733a828 JG |
5699 | /* |
5700 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5701 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5702 | * this might change. | |
c733a828 | 5703 | */ |
12724850 NH |
5704 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5705 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5706 | spin_unlock(ptl); |
1a5a9906 | 5707 | return 0; |
12724850 | 5708 | } |
03319327 | 5709 | |
45f83cef AA |
5710 | if (pmd_trans_unstable(pmd)) |
5711 | return 0; | |
4ffef5fe DN |
5712 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5713 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5714 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5715 | mc.precharge++; /* increment precharge temporarily */ |
5716 | pte_unmap_unlock(pte - 1, ptl); | |
5717 | cond_resched(); | |
5718 | ||
7dc74be0 DN |
5719 | return 0; |
5720 | } | |
5721 | ||
7b86ac33 CH |
5722 | static const struct mm_walk_ops precharge_walk_ops = { |
5723 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5724 | }; | |
5725 | ||
4ffef5fe DN |
5726 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5727 | { | |
5728 | unsigned long precharge; | |
4ffef5fe | 5729 | |
dfe076b0 | 5730 | down_read(&mm->mmap_sem); |
7b86ac33 | 5731 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
dfe076b0 | 5732 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5733 | |
5734 | precharge = mc.precharge; | |
5735 | mc.precharge = 0; | |
5736 | ||
5737 | return precharge; | |
5738 | } | |
5739 | ||
4ffef5fe DN |
5740 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5741 | { | |
dfe076b0 DN |
5742 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5743 | ||
5744 | VM_BUG_ON(mc.moving_task); | |
5745 | mc.moving_task = current; | |
5746 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5747 | } |
5748 | ||
dfe076b0 DN |
5749 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5750 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5751 | { |
2bd9bb20 KH |
5752 | struct mem_cgroup *from = mc.from; |
5753 | struct mem_cgroup *to = mc.to; | |
5754 | ||
4ffef5fe | 5755 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5756 | if (mc.precharge) { |
00501b53 | 5757 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5758 | mc.precharge = 0; |
5759 | } | |
5760 | /* | |
5761 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5762 | * we must uncharge here. | |
5763 | */ | |
5764 | if (mc.moved_charge) { | |
00501b53 | 5765 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5766 | mc.moved_charge = 0; |
4ffef5fe | 5767 | } |
483c30b5 DN |
5768 | /* we must fixup refcnts and charges */ |
5769 | if (mc.moved_swap) { | |
483c30b5 | 5770 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5771 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5772 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5773 | |
615d66c3 VD |
5774 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5775 | ||
05b84301 | 5776 | /* |
3e32cb2e JW |
5777 | * we charged both to->memory and to->memsw, so we |
5778 | * should uncharge to->memory. | |
05b84301 | 5779 | */ |
ce00a967 | 5780 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5781 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5782 | ||
615d66c3 | 5783 | css_put_many(&mc.to->css, mc.moved_swap); |
3e32cb2e | 5784 | |
483c30b5 DN |
5785 | mc.moved_swap = 0; |
5786 | } | |
dfe076b0 DN |
5787 | memcg_oom_recover(from); |
5788 | memcg_oom_recover(to); | |
5789 | wake_up_all(&mc.waitq); | |
5790 | } | |
5791 | ||
5792 | static void mem_cgroup_clear_mc(void) | |
5793 | { | |
264a0ae1 TH |
5794 | struct mm_struct *mm = mc.mm; |
5795 | ||
dfe076b0 DN |
5796 | /* |
5797 | * we must clear moving_task before waking up waiters at the end of | |
5798 | * task migration. | |
5799 | */ | |
5800 | mc.moving_task = NULL; | |
5801 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5802 | spin_lock(&mc.lock); |
4ffef5fe DN |
5803 | mc.from = NULL; |
5804 | mc.to = NULL; | |
264a0ae1 | 5805 | mc.mm = NULL; |
2bd9bb20 | 5806 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5807 | |
5808 | mmput(mm); | |
4ffef5fe DN |
5809 | } |
5810 | ||
1f7dd3e5 | 5811 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5812 | { |
1f7dd3e5 | 5813 | struct cgroup_subsys_state *css; |
eed67d75 | 5814 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5815 | struct mem_cgroup *from; |
4530eddb | 5816 | struct task_struct *leader, *p; |
9f2115f9 | 5817 | struct mm_struct *mm; |
1dfab5ab | 5818 | unsigned long move_flags; |
9f2115f9 | 5819 | int ret = 0; |
7dc74be0 | 5820 | |
1f7dd3e5 TH |
5821 | /* charge immigration isn't supported on the default hierarchy */ |
5822 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5823 | return 0; |
5824 | ||
4530eddb TH |
5825 | /* |
5826 | * Multi-process migrations only happen on the default hierarchy | |
5827 | * where charge immigration is not used. Perform charge | |
5828 | * immigration if @tset contains a leader and whine if there are | |
5829 | * multiple. | |
5830 | */ | |
5831 | p = NULL; | |
1f7dd3e5 | 5832 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5833 | WARN_ON_ONCE(p); |
5834 | p = leader; | |
1f7dd3e5 | 5835 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5836 | } |
5837 | if (!p) | |
5838 | return 0; | |
5839 | ||
1f7dd3e5 TH |
5840 | /* |
5841 | * We are now commited to this value whatever it is. Changes in this | |
5842 | * tunable will only affect upcoming migrations, not the current one. | |
5843 | * So we need to save it, and keep it going. | |
5844 | */ | |
5845 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5846 | if (!move_flags) | |
5847 | return 0; | |
5848 | ||
9f2115f9 TH |
5849 | from = mem_cgroup_from_task(p); |
5850 | ||
5851 | VM_BUG_ON(from == memcg); | |
5852 | ||
5853 | mm = get_task_mm(p); | |
5854 | if (!mm) | |
5855 | return 0; | |
5856 | /* We move charges only when we move a owner of the mm */ | |
5857 | if (mm->owner == p) { | |
5858 | VM_BUG_ON(mc.from); | |
5859 | VM_BUG_ON(mc.to); | |
5860 | VM_BUG_ON(mc.precharge); | |
5861 | VM_BUG_ON(mc.moved_charge); | |
5862 | VM_BUG_ON(mc.moved_swap); | |
5863 | ||
5864 | spin_lock(&mc.lock); | |
264a0ae1 | 5865 | mc.mm = mm; |
9f2115f9 TH |
5866 | mc.from = from; |
5867 | mc.to = memcg; | |
5868 | mc.flags = move_flags; | |
5869 | spin_unlock(&mc.lock); | |
5870 | /* We set mc.moving_task later */ | |
5871 | ||
5872 | ret = mem_cgroup_precharge_mc(mm); | |
5873 | if (ret) | |
5874 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5875 | } else { |
5876 | mmput(mm); | |
7dc74be0 DN |
5877 | } |
5878 | return ret; | |
5879 | } | |
5880 | ||
1f7dd3e5 | 5881 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5882 | { |
4e2f245d JW |
5883 | if (mc.to) |
5884 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5885 | } |
5886 | ||
4ffef5fe DN |
5887 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5888 | unsigned long addr, unsigned long end, | |
5889 | struct mm_walk *walk) | |
7dc74be0 | 5890 | { |
4ffef5fe | 5891 | int ret = 0; |
26bcd64a | 5892 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5893 | pte_t *pte; |
5894 | spinlock_t *ptl; | |
12724850 NH |
5895 | enum mc_target_type target_type; |
5896 | union mc_target target; | |
5897 | struct page *page; | |
4ffef5fe | 5898 | |
b6ec57f4 KS |
5899 | ptl = pmd_trans_huge_lock(pmd, vma); |
5900 | if (ptl) { | |
62ade86a | 5901 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5902 | spin_unlock(ptl); |
12724850 NH |
5903 | return 0; |
5904 | } | |
5905 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5906 | if (target_type == MC_TARGET_PAGE) { | |
5907 | page = target.page; | |
5908 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5909 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5910 | mc.from, mc.to)) { |
12724850 NH |
5911 | mc.precharge -= HPAGE_PMD_NR; |
5912 | mc.moved_charge += HPAGE_PMD_NR; | |
5913 | } | |
5914 | putback_lru_page(page); | |
5915 | } | |
5916 | put_page(page); | |
c733a828 JG |
5917 | } else if (target_type == MC_TARGET_DEVICE) { |
5918 | page = target.page; | |
5919 | if (!mem_cgroup_move_account(page, true, | |
5920 | mc.from, mc.to)) { | |
5921 | mc.precharge -= HPAGE_PMD_NR; | |
5922 | mc.moved_charge += HPAGE_PMD_NR; | |
5923 | } | |
5924 | put_page(page); | |
12724850 | 5925 | } |
bf929152 | 5926 | spin_unlock(ptl); |
1a5a9906 | 5927 | return 0; |
12724850 NH |
5928 | } |
5929 | ||
45f83cef AA |
5930 | if (pmd_trans_unstable(pmd)) |
5931 | return 0; | |
4ffef5fe DN |
5932 | retry: |
5933 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5934 | for (; addr != end; addr += PAGE_SIZE) { | |
5935 | pte_t ptent = *(pte++); | |
c733a828 | 5936 | bool device = false; |
02491447 | 5937 | swp_entry_t ent; |
4ffef5fe DN |
5938 | |
5939 | if (!mc.precharge) | |
5940 | break; | |
5941 | ||
8d32ff84 | 5942 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
5943 | case MC_TARGET_DEVICE: |
5944 | device = true; | |
5945 | /* fall through */ | |
4ffef5fe DN |
5946 | case MC_TARGET_PAGE: |
5947 | page = target.page; | |
53f9263b KS |
5948 | /* |
5949 | * We can have a part of the split pmd here. Moving it | |
5950 | * can be done but it would be too convoluted so simply | |
5951 | * ignore such a partial THP and keep it in original | |
5952 | * memcg. There should be somebody mapping the head. | |
5953 | */ | |
5954 | if (PageTransCompound(page)) | |
5955 | goto put; | |
c733a828 | 5956 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 5957 | goto put; |
f627c2f5 KS |
5958 | if (!mem_cgroup_move_account(page, false, |
5959 | mc.from, mc.to)) { | |
4ffef5fe | 5960 | mc.precharge--; |
854ffa8d DN |
5961 | /* we uncharge from mc.from later. */ |
5962 | mc.moved_charge++; | |
4ffef5fe | 5963 | } |
c733a828 JG |
5964 | if (!device) |
5965 | putback_lru_page(page); | |
8d32ff84 | 5966 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5967 | put_page(page); |
5968 | break; | |
02491447 DN |
5969 | case MC_TARGET_SWAP: |
5970 | ent = target.ent; | |
e91cbb42 | 5971 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5972 | mc.precharge--; |
da63188b HD |
5973 | mem_cgroup_id_get_many(mc.to, 1); |
5974 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
5975 | mc.moved_swap++; |
5976 | } | |
02491447 | 5977 | break; |
4ffef5fe DN |
5978 | default: |
5979 | break; | |
5980 | } | |
5981 | } | |
5982 | pte_unmap_unlock(pte - 1, ptl); | |
5983 | cond_resched(); | |
5984 | ||
5985 | if (addr != end) { | |
5986 | /* | |
5987 | * We have consumed all precharges we got in can_attach(). | |
5988 | * We try charge one by one, but don't do any additional | |
5989 | * charges to mc.to if we have failed in charge once in attach() | |
5990 | * phase. | |
5991 | */ | |
854ffa8d | 5992 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5993 | if (!ret) |
5994 | goto retry; | |
5995 | } | |
5996 | ||
5997 | return ret; | |
5998 | } | |
5999 | ||
7b86ac33 CH |
6000 | static const struct mm_walk_ops charge_walk_ops = { |
6001 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
6002 | }; | |
6003 | ||
264a0ae1 | 6004 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 6005 | { |
4ffef5fe | 6006 | lru_add_drain_all(); |
312722cb | 6007 | /* |
81f8c3a4 JW |
6008 | * Signal lock_page_memcg() to take the memcg's move_lock |
6009 | * while we're moving its pages to another memcg. Then wait | |
6010 | * for already started RCU-only updates to finish. | |
312722cb JW |
6011 | */ |
6012 | atomic_inc(&mc.from->moving_account); | |
6013 | synchronize_rcu(); | |
dfe076b0 | 6014 | retry: |
264a0ae1 | 6015 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
6016 | /* |
6017 | * Someone who are holding the mmap_sem might be waiting in | |
6018 | * waitq. So we cancel all extra charges, wake up all waiters, | |
6019 | * and retry. Because we cancel precharges, we might not be able | |
6020 | * to move enough charges, but moving charge is a best-effort | |
6021 | * feature anyway, so it wouldn't be a big problem. | |
6022 | */ | |
6023 | __mem_cgroup_clear_mc(); | |
6024 | cond_resched(); | |
6025 | goto retry; | |
6026 | } | |
26bcd64a NH |
6027 | /* |
6028 | * When we have consumed all precharges and failed in doing | |
6029 | * additional charge, the page walk just aborts. | |
6030 | */ | |
7b86ac33 CH |
6031 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
6032 | NULL); | |
0247f3f4 | 6033 | |
264a0ae1 | 6034 | up_read(&mc.mm->mmap_sem); |
312722cb | 6035 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
6036 | } |
6037 | ||
264a0ae1 | 6038 | static void mem_cgroup_move_task(void) |
67e465a7 | 6039 | { |
264a0ae1 TH |
6040 | if (mc.to) { |
6041 | mem_cgroup_move_charge(); | |
a433658c | 6042 | mem_cgroup_clear_mc(); |
264a0ae1 | 6043 | } |
67e465a7 | 6044 | } |
5cfb80a7 | 6045 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 6046 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6047 | { |
6048 | return 0; | |
6049 | } | |
1f7dd3e5 | 6050 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
6051 | { |
6052 | } | |
264a0ae1 | 6053 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
6054 | { |
6055 | } | |
6056 | #endif | |
67e465a7 | 6057 | |
f00baae7 TH |
6058 | /* |
6059 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
6060 | * to verify whether we're attached to the default hierarchy on each mount |
6061 | * attempt. | |
f00baae7 | 6062 | */ |
eb95419b | 6063 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
6064 | { |
6065 | /* | |
aa6ec29b | 6066 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
6067 | * guarantees that @root doesn't have any children, so turning it |
6068 | * on for the root memcg is enough. | |
6069 | */ | |
9e10a130 | 6070 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
6071 | root_mem_cgroup->use_hierarchy = true; |
6072 | else | |
6073 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
6074 | } |
6075 | ||
677dc973 CD |
6076 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
6077 | { | |
6078 | if (value == PAGE_COUNTER_MAX) | |
6079 | seq_puts(m, "max\n"); | |
6080 | else | |
6081 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6082 | ||
6083 | return 0; | |
6084 | } | |
6085 | ||
241994ed JW |
6086 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6087 | struct cftype *cft) | |
6088 | { | |
f5fc3c5d JW |
6089 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6090 | ||
6091 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6092 | } |
6093 | ||
bf8d5d52 RG |
6094 | static int memory_min_show(struct seq_file *m, void *v) |
6095 | { | |
677dc973 CD |
6096 | return seq_puts_memcg_tunable(m, |
6097 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6098 | } |
6099 | ||
6100 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6101 | char *buf, size_t nbytes, loff_t off) | |
6102 | { | |
6103 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6104 | unsigned long min; | |
6105 | int err; | |
6106 | ||
6107 | buf = strstrip(buf); | |
6108 | err = page_counter_memparse(buf, "max", &min); | |
6109 | if (err) | |
6110 | return err; | |
6111 | ||
6112 | page_counter_set_min(&memcg->memory, min); | |
6113 | ||
6114 | return nbytes; | |
6115 | } | |
6116 | ||
241994ed JW |
6117 | static int memory_low_show(struct seq_file *m, void *v) |
6118 | { | |
677dc973 CD |
6119 | return seq_puts_memcg_tunable(m, |
6120 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6121 | } |
6122 | ||
6123 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6124 | char *buf, size_t nbytes, loff_t off) | |
6125 | { | |
6126 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6127 | unsigned long low; | |
6128 | int err; | |
6129 | ||
6130 | buf = strstrip(buf); | |
d2973697 | 6131 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6132 | if (err) |
6133 | return err; | |
6134 | ||
23067153 | 6135 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6136 | |
6137 | return nbytes; | |
6138 | } | |
6139 | ||
6140 | static int memory_high_show(struct seq_file *m, void *v) | |
6141 | { | |
677dc973 | 6142 | return seq_puts_memcg_tunable(m, READ_ONCE(mem_cgroup_from_seq(m)->high)); |
241994ed JW |
6143 | } |
6144 | ||
6145 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6146 | char *buf, size_t nbytes, loff_t off) | |
6147 | { | |
6148 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 6149 | unsigned long nr_pages; |
241994ed JW |
6150 | unsigned long high; |
6151 | int err; | |
6152 | ||
6153 | buf = strstrip(buf); | |
d2973697 | 6154 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6155 | if (err) |
6156 | return err; | |
6157 | ||
6158 | memcg->high = high; | |
6159 | ||
588083bb JW |
6160 | nr_pages = page_counter_read(&memcg->memory); |
6161 | if (nr_pages > high) | |
6162 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
6163 | GFP_KERNEL, true); | |
6164 | ||
2529bb3a | 6165 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6166 | return nbytes; |
6167 | } | |
6168 | ||
6169 | static int memory_max_show(struct seq_file *m, void *v) | |
6170 | { | |
677dc973 CD |
6171 | return seq_puts_memcg_tunable(m, |
6172 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6173 | } |
6174 | ||
6175 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6176 | char *buf, size_t nbytes, loff_t off) | |
6177 | { | |
6178 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
6179 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
6180 | bool drained = false; | |
241994ed JW |
6181 | unsigned long max; |
6182 | int err; | |
6183 | ||
6184 | buf = strstrip(buf); | |
d2973697 | 6185 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6186 | if (err) |
6187 | return err; | |
6188 | ||
bbec2e15 | 6189 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6190 | |
6191 | for (;;) { | |
6192 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6193 | ||
6194 | if (nr_pages <= max) | |
6195 | break; | |
6196 | ||
6197 | if (signal_pending(current)) { | |
6198 | err = -EINTR; | |
6199 | break; | |
6200 | } | |
6201 | ||
6202 | if (!drained) { | |
6203 | drain_all_stock(memcg); | |
6204 | drained = true; | |
6205 | continue; | |
6206 | } | |
6207 | ||
6208 | if (nr_reclaims) { | |
6209 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
6210 | GFP_KERNEL, true)) | |
6211 | nr_reclaims--; | |
6212 | continue; | |
6213 | } | |
6214 | ||
e27be240 | 6215 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6216 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6217 | break; | |
6218 | } | |
241994ed | 6219 | |
2529bb3a | 6220 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6221 | return nbytes; |
6222 | } | |
6223 | ||
1e577f97 SB |
6224 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6225 | { | |
6226 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6227 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6228 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6229 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6230 | seq_printf(m, "oom_kill %lu\n", | |
6231 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
6232 | } | |
6233 | ||
241994ed JW |
6234 | static int memory_events_show(struct seq_file *m, void *v) |
6235 | { | |
aa9694bb | 6236 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6237 | |
1e577f97 SB |
6238 | __memory_events_show(m, memcg->memory_events); |
6239 | return 0; | |
6240 | } | |
6241 | ||
6242 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6243 | { | |
6244 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6245 | |
1e577f97 | 6246 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6247 | return 0; |
6248 | } | |
6249 | ||
587d9f72 JW |
6250 | static int memory_stat_show(struct seq_file *m, void *v) |
6251 | { | |
aa9694bb | 6252 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 6253 | char *buf; |
1ff9e6e1 | 6254 | |
c8713d0b JW |
6255 | buf = memory_stat_format(memcg); |
6256 | if (!buf) | |
6257 | return -ENOMEM; | |
6258 | seq_puts(m, buf); | |
6259 | kfree(buf); | |
587d9f72 JW |
6260 | return 0; |
6261 | } | |
6262 | ||
3d8b38eb RG |
6263 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6264 | { | |
aa9694bb | 6265 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6266 | |
6267 | seq_printf(m, "%d\n", memcg->oom_group); | |
6268 | ||
6269 | return 0; | |
6270 | } | |
6271 | ||
6272 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6273 | char *buf, size_t nbytes, loff_t off) | |
6274 | { | |
6275 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6276 | int ret, oom_group; | |
6277 | ||
6278 | buf = strstrip(buf); | |
6279 | if (!buf) | |
6280 | return -EINVAL; | |
6281 | ||
6282 | ret = kstrtoint(buf, 0, &oom_group); | |
6283 | if (ret) | |
6284 | return ret; | |
6285 | ||
6286 | if (oom_group != 0 && oom_group != 1) | |
6287 | return -EINVAL; | |
6288 | ||
6289 | memcg->oom_group = oom_group; | |
6290 | ||
6291 | return nbytes; | |
6292 | } | |
6293 | ||
241994ed JW |
6294 | static struct cftype memory_files[] = { |
6295 | { | |
6296 | .name = "current", | |
f5fc3c5d | 6297 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6298 | .read_u64 = memory_current_read, |
6299 | }, | |
bf8d5d52 RG |
6300 | { |
6301 | .name = "min", | |
6302 | .flags = CFTYPE_NOT_ON_ROOT, | |
6303 | .seq_show = memory_min_show, | |
6304 | .write = memory_min_write, | |
6305 | }, | |
241994ed JW |
6306 | { |
6307 | .name = "low", | |
6308 | .flags = CFTYPE_NOT_ON_ROOT, | |
6309 | .seq_show = memory_low_show, | |
6310 | .write = memory_low_write, | |
6311 | }, | |
6312 | { | |
6313 | .name = "high", | |
6314 | .flags = CFTYPE_NOT_ON_ROOT, | |
6315 | .seq_show = memory_high_show, | |
6316 | .write = memory_high_write, | |
6317 | }, | |
6318 | { | |
6319 | .name = "max", | |
6320 | .flags = CFTYPE_NOT_ON_ROOT, | |
6321 | .seq_show = memory_max_show, | |
6322 | .write = memory_max_write, | |
6323 | }, | |
6324 | { | |
6325 | .name = "events", | |
6326 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6327 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6328 | .seq_show = memory_events_show, |
6329 | }, | |
1e577f97 SB |
6330 | { |
6331 | .name = "events.local", | |
6332 | .flags = CFTYPE_NOT_ON_ROOT, | |
6333 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6334 | .seq_show = memory_events_local_show, | |
6335 | }, | |
587d9f72 JW |
6336 | { |
6337 | .name = "stat", | |
6338 | .flags = CFTYPE_NOT_ON_ROOT, | |
6339 | .seq_show = memory_stat_show, | |
6340 | }, | |
3d8b38eb RG |
6341 | { |
6342 | .name = "oom.group", | |
6343 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6344 | .seq_show = memory_oom_group_show, | |
6345 | .write = memory_oom_group_write, | |
6346 | }, | |
241994ed JW |
6347 | { } /* terminate */ |
6348 | }; | |
6349 | ||
073219e9 | 6350 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6351 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6352 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6353 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6354 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6355 | .css_free = mem_cgroup_css_free, |
1ced953b | 6356 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
6357 | .can_attach = mem_cgroup_can_attach, |
6358 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 6359 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 6360 | .bind = mem_cgroup_bind, |
241994ed JW |
6361 | .dfl_cftypes = memory_files, |
6362 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6363 | .early_init = 0, |
8cdea7c0 | 6364 | }; |
c077719b | 6365 | |
241994ed | 6366 | /** |
bf8d5d52 | 6367 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 6368 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6369 | * @memcg: the memory cgroup to check |
6370 | * | |
23067153 RG |
6371 | * WARNING: This function is not stateless! It can only be used as part |
6372 | * of a top-down tree iteration, not for isolated queries. | |
34c81057 | 6373 | * |
bf8d5d52 RG |
6374 | * Returns one of the following: |
6375 | * MEMCG_PROT_NONE: cgroup memory is not protected | |
6376 | * MEMCG_PROT_LOW: cgroup memory is protected as long there is | |
6377 | * an unprotected supply of reclaimable memory from other cgroups. | |
6378 | * MEMCG_PROT_MIN: cgroup memory is protected | |
34c81057 | 6379 | * |
bf8d5d52 | 6380 | * @root is exclusive; it is never protected when looked at directly |
34c81057 | 6381 | * |
bf8d5d52 RG |
6382 | * To provide a proper hierarchical behavior, effective memory.min/low values |
6383 | * are used. Below is the description of how effective memory.low is calculated. | |
6384 | * Effective memory.min values is calculated in the same way. | |
34c81057 | 6385 | * |
23067153 RG |
6386 | * Effective memory.low is always equal or less than the original memory.low. |
6387 | * If there is no memory.low overcommittment (which is always true for | |
6388 | * top-level memory cgroups), these two values are equal. | |
6389 | * Otherwise, it's a part of parent's effective memory.low, | |
6390 | * calculated as a cgroup's memory.low usage divided by sum of sibling's | |
6391 | * memory.low usages, where memory.low usage is the size of actually | |
6392 | * protected memory. | |
34c81057 | 6393 | * |
23067153 RG |
6394 | * low_usage |
6395 | * elow = min( memory.low, parent->elow * ------------------ ), | |
6396 | * siblings_low_usage | |
34c81057 | 6397 | * |
23067153 RG |
6398 | * | memory.current, if memory.current < memory.low |
6399 | * low_usage = | | |
82ede7ee | 6400 | * | 0, otherwise. |
34c81057 | 6401 | * |
23067153 RG |
6402 | * |
6403 | * Such definition of the effective memory.low provides the expected | |
6404 | * hierarchical behavior: parent's memory.low value is limiting | |
6405 | * children, unprotected memory is reclaimed first and cgroups, | |
6406 | * which are not using their guarantee do not affect actual memory | |
6407 | * distribution. | |
6408 | * | |
6409 | * For example, if there are memcgs A, A/B, A/C, A/D and A/E: | |
6410 | * | |
6411 | * A A/memory.low = 2G, A/memory.current = 6G | |
6412 | * //\\ | |
6413 | * BC DE B/memory.low = 3G B/memory.current = 2G | |
6414 | * C/memory.low = 1G C/memory.current = 2G | |
6415 | * D/memory.low = 0 D/memory.current = 2G | |
6416 | * E/memory.low = 10G E/memory.current = 0 | |
6417 | * | |
6418 | * and the memory pressure is applied, the following memory distribution | |
6419 | * is expected (approximately): | |
6420 | * | |
6421 | * A/memory.current = 2G | |
6422 | * | |
6423 | * B/memory.current = 1.3G | |
6424 | * C/memory.current = 0.6G | |
6425 | * D/memory.current = 0 | |
6426 | * E/memory.current = 0 | |
6427 | * | |
6428 | * These calculations require constant tracking of the actual low usages | |
bf8d5d52 RG |
6429 | * (see propagate_protected_usage()), as well as recursive calculation of |
6430 | * effective memory.low values. But as we do call mem_cgroup_protected() | |
23067153 RG |
6431 | * path for each memory cgroup top-down from the reclaim, |
6432 | * it's possible to optimize this part, and save calculated elow | |
6433 | * for next usage. This part is intentionally racy, but it's ok, | |
6434 | * as memory.low is a best-effort mechanism. | |
241994ed | 6435 | */ |
bf8d5d52 RG |
6436 | enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root, |
6437 | struct mem_cgroup *memcg) | |
241994ed | 6438 | { |
23067153 | 6439 | struct mem_cgroup *parent; |
bf8d5d52 RG |
6440 | unsigned long emin, parent_emin; |
6441 | unsigned long elow, parent_elow; | |
6442 | unsigned long usage; | |
23067153 | 6443 | |
241994ed | 6444 | if (mem_cgroup_disabled()) |
bf8d5d52 | 6445 | return MEMCG_PROT_NONE; |
241994ed | 6446 | |
34c81057 SC |
6447 | if (!root) |
6448 | root = root_mem_cgroup; | |
5f969cf0 YS |
6449 | |
6450 | /* | |
6451 | * Effective values of the reclaim targets are ignored so they | |
6452 | * can be stale. Have a look at mem_cgroup_protection for more | |
6453 | * details. | |
6454 | * TODO: calculation should be more robust so that we do not need | |
6455 | * that special casing. | |
6456 | */ | |
34c81057 | 6457 | if (memcg == root) |
bf8d5d52 | 6458 | return MEMCG_PROT_NONE; |
241994ed | 6459 | |
23067153 | 6460 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 RG |
6461 | if (!usage) |
6462 | return MEMCG_PROT_NONE; | |
6463 | ||
6464 | emin = memcg->memory.min; | |
6465 | elow = memcg->memory.low; | |
34c81057 | 6466 | |
bf8d5d52 | 6467 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
6468 | /* No parent means a non-hierarchical mode on v1 memcg */ |
6469 | if (!parent) | |
6470 | return MEMCG_PROT_NONE; | |
6471 | ||
23067153 RG |
6472 | if (parent == root) |
6473 | goto exit; | |
6474 | ||
bf8d5d52 RG |
6475 | parent_emin = READ_ONCE(parent->memory.emin); |
6476 | emin = min(emin, parent_emin); | |
6477 | if (emin && parent_emin) { | |
6478 | unsigned long min_usage, siblings_min_usage; | |
6479 | ||
6480 | min_usage = min(usage, memcg->memory.min); | |
6481 | siblings_min_usage = atomic_long_read( | |
6482 | &parent->memory.children_min_usage); | |
6483 | ||
6484 | if (min_usage && siblings_min_usage) | |
6485 | emin = min(emin, parent_emin * min_usage / | |
6486 | siblings_min_usage); | |
6487 | } | |
6488 | ||
23067153 RG |
6489 | parent_elow = READ_ONCE(parent->memory.elow); |
6490 | elow = min(elow, parent_elow); | |
bf8d5d52 RG |
6491 | if (elow && parent_elow) { |
6492 | unsigned long low_usage, siblings_low_usage; | |
23067153 | 6493 | |
bf8d5d52 RG |
6494 | low_usage = min(usage, memcg->memory.low); |
6495 | siblings_low_usage = atomic_long_read( | |
6496 | &parent->memory.children_low_usage); | |
23067153 | 6497 | |
bf8d5d52 RG |
6498 | if (low_usage && siblings_low_usage) |
6499 | elow = min(elow, parent_elow * low_usage / | |
6500 | siblings_low_usage); | |
6501 | } | |
23067153 | 6502 | |
23067153 | 6503 | exit: |
bf8d5d52 | 6504 | memcg->memory.emin = emin; |
23067153 | 6505 | memcg->memory.elow = elow; |
bf8d5d52 RG |
6506 | |
6507 | if (usage <= emin) | |
6508 | return MEMCG_PROT_MIN; | |
6509 | else if (usage <= elow) | |
6510 | return MEMCG_PROT_LOW; | |
6511 | else | |
6512 | return MEMCG_PROT_NONE; | |
241994ed JW |
6513 | } |
6514 | ||
00501b53 JW |
6515 | /** |
6516 | * mem_cgroup_try_charge - try charging a page | |
6517 | * @page: page to charge | |
6518 | * @mm: mm context of the victim | |
6519 | * @gfp_mask: reclaim mode | |
6520 | * @memcgp: charged memcg return | |
25843c2b | 6521 | * @compound: charge the page as compound or small page |
00501b53 JW |
6522 | * |
6523 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6524 | * pages according to @gfp_mask if necessary. | |
6525 | * | |
6526 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
6527 | * Otherwise, an error code is returned. | |
6528 | * | |
6529 | * After page->mapping has been set up, the caller must finalize the | |
6530 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
6531 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
6532 | */ | |
6533 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
6534 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
6535 | bool compound) | |
00501b53 JW |
6536 | { |
6537 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 6538 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6539 | int ret = 0; |
6540 | ||
6541 | if (mem_cgroup_disabled()) | |
6542 | goto out; | |
6543 | ||
6544 | if (PageSwapCache(page)) { | |
00501b53 JW |
6545 | /* |
6546 | * Every swap fault against a single page tries to charge the | |
6547 | * page, bail as early as possible. shmem_unuse() encounters | |
6548 | * already charged pages, too. The USED bit is protected by | |
6549 | * the page lock, which serializes swap cache removal, which | |
6550 | * in turn serializes uncharging. | |
6551 | */ | |
e993d905 | 6552 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 6553 | if (compound_head(page)->mem_cgroup) |
00501b53 | 6554 | goto out; |
e993d905 | 6555 | |
37e84351 | 6556 | if (do_swap_account) { |
e993d905 VD |
6557 | swp_entry_t ent = { .val = page_private(page), }; |
6558 | unsigned short id = lookup_swap_cgroup_id(ent); | |
6559 | ||
6560 | rcu_read_lock(); | |
6561 | memcg = mem_cgroup_from_id(id); | |
6562 | if (memcg && !css_tryget_online(&memcg->css)) | |
6563 | memcg = NULL; | |
6564 | rcu_read_unlock(); | |
6565 | } | |
00501b53 JW |
6566 | } |
6567 | ||
00501b53 JW |
6568 | if (!memcg) |
6569 | memcg = get_mem_cgroup_from_mm(mm); | |
6570 | ||
6571 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
6572 | ||
6573 | css_put(&memcg->css); | |
00501b53 JW |
6574 | out: |
6575 | *memcgp = memcg; | |
6576 | return ret; | |
6577 | } | |
6578 | ||
2cf85583 TH |
6579 | int mem_cgroup_try_charge_delay(struct page *page, struct mm_struct *mm, |
6580 | gfp_t gfp_mask, struct mem_cgroup **memcgp, | |
6581 | bool compound) | |
6582 | { | |
6583 | struct mem_cgroup *memcg; | |
6584 | int ret; | |
6585 | ||
6586 | ret = mem_cgroup_try_charge(page, mm, gfp_mask, memcgp, compound); | |
6587 | memcg = *memcgp; | |
6588 | mem_cgroup_throttle_swaprate(memcg, page_to_nid(page), gfp_mask); | |
6589 | return ret; | |
6590 | } | |
6591 | ||
00501b53 JW |
6592 | /** |
6593 | * mem_cgroup_commit_charge - commit a page charge | |
6594 | * @page: page to charge | |
6595 | * @memcg: memcg to charge the page to | |
6596 | * @lrucare: page might be on LRU already | |
25843c2b | 6597 | * @compound: charge the page as compound or small page |
00501b53 JW |
6598 | * |
6599 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
6600 | * after page->mapping has been set up. This must happen atomically | |
6601 | * as part of the page instantiation, i.e. under the page table lock | |
6602 | * for anonymous pages, under the page lock for page and swap cache. | |
6603 | * | |
6604 | * In addition, the page must not be on the LRU during the commit, to | |
6605 | * prevent racing with task migration. If it might be, use @lrucare. | |
6606 | * | |
6607 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
6608 | */ | |
6609 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 6610 | bool lrucare, bool compound) |
00501b53 | 6611 | { |
f627c2f5 | 6612 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6613 | |
6614 | VM_BUG_ON_PAGE(!page->mapping, page); | |
6615 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
6616 | ||
6617 | if (mem_cgroup_disabled()) | |
6618 | return; | |
6619 | /* | |
6620 | * Swap faults will attempt to charge the same page multiple | |
6621 | * times. But reuse_swap_page() might have removed the page | |
6622 | * from swapcache already, so we can't check PageSwapCache(). | |
6623 | */ | |
6624 | if (!memcg) | |
6625 | return; | |
6626 | ||
6abb5a86 JW |
6627 | commit_charge(page, memcg, lrucare); |
6628 | ||
6abb5a86 | 6629 | local_irq_disable(); |
f627c2f5 | 6630 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
6631 | memcg_check_events(memcg, page); |
6632 | local_irq_enable(); | |
00501b53 | 6633 | |
7941d214 | 6634 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
6635 | swp_entry_t entry = { .val = page_private(page) }; |
6636 | /* | |
6637 | * The swap entry might not get freed for a long time, | |
6638 | * let's not wait for it. The page already received a | |
6639 | * memory+swap charge, drop the swap entry duplicate. | |
6640 | */ | |
38d8b4e6 | 6641 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 JW |
6642 | } |
6643 | } | |
6644 | ||
6645 | /** | |
6646 | * mem_cgroup_cancel_charge - cancel a page charge | |
6647 | * @page: page to charge | |
6648 | * @memcg: memcg to charge the page to | |
25843c2b | 6649 | * @compound: charge the page as compound or small page |
00501b53 JW |
6650 | * |
6651 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
6652 | */ | |
f627c2f5 KS |
6653 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
6654 | bool compound) | |
00501b53 | 6655 | { |
f627c2f5 | 6656 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6657 | |
6658 | if (mem_cgroup_disabled()) | |
6659 | return; | |
6660 | /* | |
6661 | * Swap faults will attempt to charge the same page multiple | |
6662 | * times. But reuse_swap_page() might have removed the page | |
6663 | * from swapcache already, so we can't check PageSwapCache(). | |
6664 | */ | |
6665 | if (!memcg) | |
6666 | return; | |
6667 | ||
00501b53 JW |
6668 | cancel_charge(memcg, nr_pages); |
6669 | } | |
6670 | ||
a9d5adee JG |
6671 | struct uncharge_gather { |
6672 | struct mem_cgroup *memcg; | |
6673 | unsigned long pgpgout; | |
6674 | unsigned long nr_anon; | |
6675 | unsigned long nr_file; | |
6676 | unsigned long nr_kmem; | |
6677 | unsigned long nr_huge; | |
6678 | unsigned long nr_shmem; | |
6679 | struct page *dummy_page; | |
6680 | }; | |
6681 | ||
6682 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6683 | { |
a9d5adee JG |
6684 | memset(ug, 0, sizeof(*ug)); |
6685 | } | |
6686 | ||
6687 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6688 | { | |
6689 | unsigned long nr_pages = ug->nr_anon + ug->nr_file + ug->nr_kmem; | |
747db954 JW |
6690 | unsigned long flags; |
6691 | ||
a9d5adee JG |
6692 | if (!mem_cgroup_is_root(ug->memcg)) { |
6693 | page_counter_uncharge(&ug->memcg->memory, nr_pages); | |
7941d214 | 6694 | if (do_memsw_account()) |
a9d5adee JG |
6695 | page_counter_uncharge(&ug->memcg->memsw, nr_pages); |
6696 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) | |
6697 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6698 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6699 | } |
747db954 JW |
6700 | |
6701 | local_irq_save(flags); | |
c9019e9b JW |
6702 | __mod_memcg_state(ug->memcg, MEMCG_RSS, -ug->nr_anon); |
6703 | __mod_memcg_state(ug->memcg, MEMCG_CACHE, -ug->nr_file); | |
6704 | __mod_memcg_state(ug->memcg, MEMCG_RSS_HUGE, -ug->nr_huge); | |
6705 | __mod_memcg_state(ug->memcg, NR_SHMEM, -ug->nr_shmem); | |
6706 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); | |
871789d4 | 6707 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, nr_pages); |
a9d5adee | 6708 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6709 | local_irq_restore(flags); |
e8ea14cc | 6710 | |
a9d5adee JG |
6711 | if (!mem_cgroup_is_root(ug->memcg)) |
6712 | css_put_many(&ug->memcg->css, nr_pages); | |
6713 | } | |
6714 | ||
6715 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6716 | { | |
6717 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
3f2eb028 JG |
6718 | VM_BUG_ON_PAGE(page_count(page) && !is_zone_device_page(page) && |
6719 | !PageHWPoison(page) , page); | |
a9d5adee JG |
6720 | |
6721 | if (!page->mem_cgroup) | |
6722 | return; | |
6723 | ||
6724 | /* | |
6725 | * Nobody should be changing or seriously looking at | |
6726 | * page->mem_cgroup at this point, we have fully | |
6727 | * exclusive access to the page. | |
6728 | */ | |
6729 | ||
6730 | if (ug->memcg != page->mem_cgroup) { | |
6731 | if (ug->memcg) { | |
6732 | uncharge_batch(ug); | |
6733 | uncharge_gather_clear(ug); | |
6734 | } | |
6735 | ug->memcg = page->mem_cgroup; | |
6736 | } | |
6737 | ||
6738 | if (!PageKmemcg(page)) { | |
6739 | unsigned int nr_pages = 1; | |
6740 | ||
6741 | if (PageTransHuge(page)) { | |
d8c6546b | 6742 | nr_pages = compound_nr(page); |
a9d5adee JG |
6743 | ug->nr_huge += nr_pages; |
6744 | } | |
6745 | if (PageAnon(page)) | |
6746 | ug->nr_anon += nr_pages; | |
6747 | else { | |
6748 | ug->nr_file += nr_pages; | |
6749 | if (PageSwapBacked(page)) | |
6750 | ug->nr_shmem += nr_pages; | |
6751 | } | |
6752 | ug->pgpgout++; | |
6753 | } else { | |
d8c6546b | 6754 | ug->nr_kmem += compound_nr(page); |
a9d5adee JG |
6755 | __ClearPageKmemcg(page); |
6756 | } | |
6757 | ||
6758 | ug->dummy_page = page; | |
6759 | page->mem_cgroup = NULL; | |
747db954 JW |
6760 | } |
6761 | ||
6762 | static void uncharge_list(struct list_head *page_list) | |
6763 | { | |
a9d5adee | 6764 | struct uncharge_gather ug; |
747db954 | 6765 | struct list_head *next; |
a9d5adee JG |
6766 | |
6767 | uncharge_gather_clear(&ug); | |
747db954 | 6768 | |
8b592656 JW |
6769 | /* |
6770 | * Note that the list can be a single page->lru; hence the | |
6771 | * do-while loop instead of a simple list_for_each_entry(). | |
6772 | */ | |
747db954 JW |
6773 | next = page_list->next; |
6774 | do { | |
a9d5adee JG |
6775 | struct page *page; |
6776 | ||
747db954 JW |
6777 | page = list_entry(next, struct page, lru); |
6778 | next = page->lru.next; | |
6779 | ||
a9d5adee | 6780 | uncharge_page(page, &ug); |
747db954 JW |
6781 | } while (next != page_list); |
6782 | ||
a9d5adee JG |
6783 | if (ug.memcg) |
6784 | uncharge_batch(&ug); | |
747db954 JW |
6785 | } |
6786 | ||
0a31bc97 JW |
6787 | /** |
6788 | * mem_cgroup_uncharge - uncharge a page | |
6789 | * @page: page to uncharge | |
6790 | * | |
6791 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
6792 | * mem_cgroup_commit_charge(). | |
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 | |
6815 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
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 JW |
6839 | unsigned int nr_pages; |
6840 | bool compound; | |
d93c4130 | 6841 | unsigned long flags; |
0a31bc97 JW |
6842 | |
6843 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6844 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6845 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6846 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6847 | newpage); | |
0a31bc97 JW |
6848 | |
6849 | if (mem_cgroup_disabled()) | |
6850 | return; | |
6851 | ||
6852 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6853 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6854 | return; |
6855 | ||
45637bab | 6856 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6857 | memcg = oldpage->mem_cgroup; |
29833315 | 6858 | if (!memcg) |
0a31bc97 JW |
6859 | return; |
6860 | ||
44b7a8d3 JW |
6861 | /* Force-charge the new page. The old one will be freed soon */ |
6862 | compound = PageTransHuge(newpage); | |
6863 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
6864 | ||
6865 | page_counter_charge(&memcg->memory, nr_pages); | |
6866 | if (do_memsw_account()) | |
6867 | page_counter_charge(&memcg->memsw, nr_pages); | |
6868 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 6869 | |
9cf7666a | 6870 | commit_charge(newpage, memcg, false); |
44b7a8d3 | 6871 | |
d93c4130 | 6872 | local_irq_save(flags); |
44b7a8d3 JW |
6873 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
6874 | memcg_check_events(memcg, newpage); | |
d93c4130 | 6875 | local_irq_restore(flags); |
0a31bc97 JW |
6876 | } |
6877 | ||
ef12947c | 6878 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6879 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6880 | ||
2d758073 | 6881 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6882 | { |
6883 | struct mem_cgroup *memcg; | |
6884 | ||
2d758073 JW |
6885 | if (!mem_cgroup_sockets_enabled) |
6886 | return; | |
6887 | ||
ffa9ca5c SB |
6888 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
6889 | if (in_interrupt()) | |
6890 | return; | |
6891 | ||
11092087 JW |
6892 | rcu_read_lock(); |
6893 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6894 | if (memcg == root_mem_cgroup) |
6895 | goto out; | |
0db15298 | 6896 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6897 | goto out; |
f7e1cb6e | 6898 | if (css_tryget_online(&memcg->css)) |
11092087 | 6899 | sk->sk_memcg = memcg; |
f7e1cb6e | 6900 | out: |
11092087 JW |
6901 | rcu_read_unlock(); |
6902 | } | |
11092087 | 6903 | |
2d758073 | 6904 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6905 | { |
2d758073 JW |
6906 | if (sk->sk_memcg) |
6907 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6908 | } |
6909 | ||
6910 | /** | |
6911 | * mem_cgroup_charge_skmem - charge socket memory | |
6912 | * @memcg: memcg to charge | |
6913 | * @nr_pages: number of pages to charge | |
6914 | * | |
6915 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6916 | * @memcg's configured limit, %false if the charge had to be forced. | |
6917 | */ | |
6918 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6919 | { | |
f7e1cb6e | 6920 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6921 | |
f7e1cb6e | 6922 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6923 | struct page_counter *fail; |
f7e1cb6e | 6924 | |
0db15298 JW |
6925 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6926 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6927 | return true; |
6928 | } | |
0db15298 JW |
6929 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6930 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6931 | return false; |
11092087 | 6932 | } |
d886f4e4 | 6933 | |
f7e1cb6e JW |
6934 | /* Don't block in the packet receive path */ |
6935 | if (in_softirq()) | |
6936 | gfp_mask = GFP_NOWAIT; | |
6937 | ||
c9019e9b | 6938 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6939 | |
f7e1cb6e JW |
6940 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6941 | return true; | |
6942 | ||
6943 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6944 | return false; |
6945 | } | |
6946 | ||
6947 | /** | |
6948 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
6949 | * @memcg: memcg to uncharge |
6950 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
6951 | */ |
6952 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6953 | { | |
f7e1cb6e | 6954 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6955 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
6956 | return; |
6957 | } | |
d886f4e4 | 6958 | |
c9019e9b | 6959 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 6960 | |
475d0487 | 6961 | refill_stock(memcg, nr_pages); |
11092087 JW |
6962 | } |
6963 | ||
f7e1cb6e JW |
6964 | static int __init cgroup_memory(char *s) |
6965 | { | |
6966 | char *token; | |
6967 | ||
6968 | while ((token = strsep(&s, ",")) != NULL) { | |
6969 | if (!*token) | |
6970 | continue; | |
6971 | if (!strcmp(token, "nosocket")) | |
6972 | cgroup_memory_nosocket = true; | |
04823c83 VD |
6973 | if (!strcmp(token, "nokmem")) |
6974 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
6975 | } |
6976 | return 0; | |
6977 | } | |
6978 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 6979 | |
2d11085e | 6980 | /* |
1081312f MH |
6981 | * subsys_initcall() for memory controller. |
6982 | * | |
308167fc SAS |
6983 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
6984 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
6985 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
6986 | * should be initialized from here. | |
2d11085e MH |
6987 | */ |
6988 | static int __init mem_cgroup_init(void) | |
6989 | { | |
95a045f6 JW |
6990 | int cpu, node; |
6991 | ||
84c07d11 | 6992 | #ifdef CONFIG_MEMCG_KMEM |
13583c3d VD |
6993 | /* |
6994 | * Kmem cache creation is mostly done with the slab_mutex held, | |
17cc4dfe TH |
6995 | * so use a workqueue with limited concurrency to avoid stalling |
6996 | * all worker threads in case lots of cgroups are created and | |
6997 | * destroyed simultaneously. | |
13583c3d | 6998 | */ |
17cc4dfe TH |
6999 | memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1); |
7000 | BUG_ON(!memcg_kmem_cache_wq); | |
13583c3d VD |
7001 | #endif |
7002 | ||
308167fc SAS |
7003 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
7004 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
7005 | |
7006 | for_each_possible_cpu(cpu) | |
7007 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
7008 | drain_local_stock); | |
7009 | ||
7010 | for_each_node(node) { | |
7011 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
7012 | |
7013 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
7014 | node_online(node) ? node : NUMA_NO_NODE); | |
7015 | ||
ef8f2327 | 7016 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 7017 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 7018 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
7019 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
7020 | } | |
7021 | ||
2d11085e MH |
7022 | return 0; |
7023 | } | |
7024 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
7025 | |
7026 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
7027 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
7028 | { | |
1c2d479a | 7029 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
7030 | /* |
7031 | * The root cgroup cannot be destroyed, so it's refcount must | |
7032 | * always be >= 1. | |
7033 | */ | |
7034 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
7035 | VM_BUG_ON(1); | |
7036 | break; | |
7037 | } | |
7038 | memcg = parent_mem_cgroup(memcg); | |
7039 | if (!memcg) | |
7040 | memcg = root_mem_cgroup; | |
7041 | } | |
7042 | return memcg; | |
7043 | } | |
7044 | ||
21afa38e JW |
7045 | /** |
7046 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
7047 | * @page: page whose memsw charge to transfer | |
7048 | * @entry: swap entry to move the charge to | |
7049 | * | |
7050 | * Transfer the memsw charge of @page to @entry. | |
7051 | */ | |
7052 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
7053 | { | |
1f47b61f | 7054 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 7055 | unsigned int nr_entries; |
21afa38e JW |
7056 | unsigned short oldid; |
7057 | ||
7058 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
7059 | VM_BUG_ON_PAGE(page_count(page), page); | |
7060 | ||
7941d214 | 7061 | if (!do_memsw_account()) |
21afa38e JW |
7062 | return; |
7063 | ||
7064 | memcg = page->mem_cgroup; | |
7065 | ||
7066 | /* Readahead page, never charged */ | |
7067 | if (!memcg) | |
7068 | return; | |
7069 | ||
1f47b61f VD |
7070 | /* |
7071 | * In case the memcg owning these pages has been offlined and doesn't | |
7072 | * have an ID allocated to it anymore, charge the closest online | |
7073 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7074 | */ | |
7075 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
d6810d73 HY |
7076 | nr_entries = hpage_nr_pages(page); |
7077 | /* Get references for the tail pages, too */ | |
7078 | if (nr_entries > 1) | |
7079 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7080 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7081 | nr_entries); | |
21afa38e | 7082 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7083 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
7084 | |
7085 | page->mem_cgroup = NULL; | |
7086 | ||
7087 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7088 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7089 | |
1f47b61f VD |
7090 | if (memcg != swap_memcg) { |
7091 | if (!mem_cgroup_is_root(swap_memcg)) | |
d6810d73 HY |
7092 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7093 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7094 | } |
7095 | ||
ce9ce665 SAS |
7096 | /* |
7097 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7098 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7099 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7100 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
7101 | */ |
7102 | VM_BUG_ON(!irqs_disabled()); | |
d6810d73 HY |
7103 | mem_cgroup_charge_statistics(memcg, page, PageTransHuge(page), |
7104 | -nr_entries); | |
21afa38e | 7105 | memcg_check_events(memcg, page); |
73f576c0 JW |
7106 | |
7107 | if (!mem_cgroup_is_root(memcg)) | |
d08afa14 | 7108 | css_put_many(&memcg->css, nr_entries); |
21afa38e JW |
7109 | } |
7110 | ||
38d8b4e6 HY |
7111 | /** |
7112 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
7113 | * @page: page being added to swap |
7114 | * @entry: swap entry to charge | |
7115 | * | |
38d8b4e6 | 7116 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
7117 | * |
7118 | * Returns 0 on success, -ENOMEM on failure. | |
7119 | */ | |
7120 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
7121 | { | |
38d8b4e6 | 7122 | unsigned int nr_pages = hpage_nr_pages(page); |
37e84351 | 7123 | struct page_counter *counter; |
38d8b4e6 | 7124 | struct mem_cgroup *memcg; |
37e84351 VD |
7125 | unsigned short oldid; |
7126 | ||
7127 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
7128 | return 0; | |
7129 | ||
7130 | memcg = page->mem_cgroup; | |
7131 | ||
7132 | /* Readahead page, never charged */ | |
7133 | if (!memcg) | |
7134 | return 0; | |
7135 | ||
f3a53a3a TH |
7136 | if (!entry.val) { |
7137 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7138 | return 0; |
f3a53a3a | 7139 | } |
bb98f2c5 | 7140 | |
1f47b61f VD |
7141 | memcg = mem_cgroup_id_get_online(memcg); |
7142 | ||
37e84351 | 7143 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7144 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7145 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7146 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7147 | mem_cgroup_id_put(memcg); |
37e84351 | 7148 | return -ENOMEM; |
1f47b61f | 7149 | } |
37e84351 | 7150 | |
38d8b4e6 HY |
7151 | /* Get references for the tail pages, too */ |
7152 | if (nr_pages > 1) | |
7153 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7154 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 7155 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7156 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7157 | |
37e84351 VD |
7158 | return 0; |
7159 | } | |
7160 | ||
21afa38e | 7161 | /** |
38d8b4e6 | 7162 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7163 | * @entry: swap entry to uncharge |
38d8b4e6 | 7164 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7165 | */ |
38d8b4e6 | 7166 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7167 | { |
7168 | struct mem_cgroup *memcg; | |
7169 | unsigned short id; | |
7170 | ||
37e84351 | 7171 | if (!do_swap_account) |
21afa38e JW |
7172 | return; |
7173 | ||
38d8b4e6 | 7174 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7175 | rcu_read_lock(); |
adbe427b | 7176 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7177 | if (memcg) { |
37e84351 VD |
7178 | if (!mem_cgroup_is_root(memcg)) { |
7179 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
38d8b4e6 | 7180 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7181 | else |
38d8b4e6 | 7182 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7183 | } |
c9019e9b | 7184 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7185 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7186 | } |
7187 | rcu_read_unlock(); | |
7188 | } | |
7189 | ||
d8b38438 VD |
7190 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7191 | { | |
7192 | long nr_swap_pages = get_nr_swap_pages(); | |
7193 | ||
7194 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
7195 | return nr_swap_pages; | |
7196 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7197 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7198 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7199 | page_counter_read(&memcg->swap)); |
7200 | return nr_swap_pages; | |
7201 | } | |
7202 | ||
5ccc5aba VD |
7203 | bool mem_cgroup_swap_full(struct page *page) |
7204 | { | |
7205 | struct mem_cgroup *memcg; | |
7206 | ||
7207 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
7208 | ||
7209 | if (vm_swap_full()) | |
7210 | return true; | |
7211 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
7212 | return false; | |
7213 | ||
7214 | memcg = page->mem_cgroup; | |
7215 | if (!memcg) | |
7216 | return false; | |
7217 | ||
7218 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
bbec2e15 | 7219 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.max) |
5ccc5aba VD |
7220 | return true; |
7221 | ||
7222 | return false; | |
7223 | } | |
7224 | ||
21afa38e JW |
7225 | /* for remember boot option*/ |
7226 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
7227 | static int really_do_swap_account __initdata = 1; | |
7228 | #else | |
7229 | static int really_do_swap_account __initdata; | |
7230 | #endif | |
7231 | ||
7232 | static int __init enable_swap_account(char *s) | |
7233 | { | |
7234 | if (!strcmp(s, "1")) | |
7235 | really_do_swap_account = 1; | |
7236 | else if (!strcmp(s, "0")) | |
7237 | really_do_swap_account = 0; | |
7238 | return 1; | |
7239 | } | |
7240 | __setup("swapaccount=", enable_swap_account); | |
7241 | ||
37e84351 VD |
7242 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7243 | struct cftype *cft) | |
7244 | { | |
7245 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7246 | ||
7247 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7248 | } | |
7249 | ||
7250 | static int swap_max_show(struct seq_file *m, void *v) | |
7251 | { | |
677dc973 CD |
7252 | return seq_puts_memcg_tunable(m, |
7253 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7254 | } |
7255 | ||
7256 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7257 | char *buf, size_t nbytes, loff_t off) | |
7258 | { | |
7259 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7260 | unsigned long max; | |
7261 | int err; | |
7262 | ||
7263 | buf = strstrip(buf); | |
7264 | err = page_counter_memparse(buf, "max", &max); | |
7265 | if (err) | |
7266 | return err; | |
7267 | ||
be09102b | 7268 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7269 | |
7270 | return nbytes; | |
7271 | } | |
7272 | ||
f3a53a3a TH |
7273 | static int swap_events_show(struct seq_file *m, void *v) |
7274 | { | |
aa9694bb | 7275 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a TH |
7276 | |
7277 | seq_printf(m, "max %lu\n", | |
7278 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7279 | seq_printf(m, "fail %lu\n", | |
7280 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7281 | ||
7282 | return 0; | |
7283 | } | |
7284 | ||
37e84351 VD |
7285 | static struct cftype swap_files[] = { |
7286 | { | |
7287 | .name = "swap.current", | |
7288 | .flags = CFTYPE_NOT_ON_ROOT, | |
7289 | .read_u64 = swap_current_read, | |
7290 | }, | |
7291 | { | |
7292 | .name = "swap.max", | |
7293 | .flags = CFTYPE_NOT_ON_ROOT, | |
7294 | .seq_show = swap_max_show, | |
7295 | .write = swap_max_write, | |
7296 | }, | |
f3a53a3a TH |
7297 | { |
7298 | .name = "swap.events", | |
7299 | .flags = CFTYPE_NOT_ON_ROOT, | |
7300 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7301 | .seq_show = swap_events_show, | |
7302 | }, | |
37e84351 VD |
7303 | { } /* terminate */ |
7304 | }; | |
7305 | ||
21afa38e JW |
7306 | static struct cftype memsw_cgroup_files[] = { |
7307 | { | |
7308 | .name = "memsw.usage_in_bytes", | |
7309 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7310 | .read_u64 = mem_cgroup_read_u64, | |
7311 | }, | |
7312 | { | |
7313 | .name = "memsw.max_usage_in_bytes", | |
7314 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7315 | .write = mem_cgroup_reset, | |
7316 | .read_u64 = mem_cgroup_read_u64, | |
7317 | }, | |
7318 | { | |
7319 | .name = "memsw.limit_in_bytes", | |
7320 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7321 | .write = mem_cgroup_write, | |
7322 | .read_u64 = mem_cgroup_read_u64, | |
7323 | }, | |
7324 | { | |
7325 | .name = "memsw.failcnt", | |
7326 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7327 | .write = mem_cgroup_reset, | |
7328 | .read_u64 = mem_cgroup_read_u64, | |
7329 | }, | |
7330 | { }, /* terminate */ | |
7331 | }; | |
7332 | ||
7333 | static int __init mem_cgroup_swap_init(void) | |
7334 | { | |
7335 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
7336 | do_swap_account = 1; | |
37e84351 VD |
7337 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
7338 | swap_files)); | |
21afa38e JW |
7339 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
7340 | memsw_cgroup_files)); | |
7341 | } | |
7342 | return 0; | |
7343 | } | |
7344 | subsys_initcall(mem_cgroup_swap_init); | |
7345 | ||
7346 | #endif /* CONFIG_MEMCG_SWAP */ |