]>
Commit | Line | Data |
---|---|---|
c942fddf | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
8cdea7c0 BS |
2 | /* memcontrol.c - Memory Controller |
3 | * | |
4 | * Copyright IBM Corporation, 2007 | |
5 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
6 | * | |
78fb7466 PE |
7 | * Copyright 2007 OpenVZ SWsoft Inc |
8 | * Author: Pavel Emelianov <xemul@openvz.org> | |
9 | * | |
2e72b634 KS |
10 | * Memory thresholds |
11 | * Copyright (C) 2009 Nokia Corporation | |
12 | * Author: Kirill A. Shutemov | |
13 | * | |
7ae1e1d0 GC |
14 | * Kernel Memory Controller |
15 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
16 | * Authors: Glauber Costa and Suleiman Souhlal | |
17 | * | |
1575e68b JW |
18 | * Native page reclaim |
19 | * Charge lifetime sanitation | |
20 | * Lockless page tracking & accounting | |
21 | * Unified hierarchy configuration model | |
22 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
8cdea7c0 BS |
23 | */ |
24 | ||
3e32cb2e | 25 | #include <linux/page_counter.h> |
8cdea7c0 BS |
26 | #include <linux/memcontrol.h> |
27 | #include <linux/cgroup.h> | |
78fb7466 | 28 | #include <linux/mm.h> |
6e84f315 | 29 | #include <linux/sched/mm.h> |
3a4f8a0b | 30 | #include <linux/shmem_fs.h> |
4ffef5fe | 31 | #include <linux/hugetlb.h> |
d13d1443 | 32 | #include <linux/pagemap.h> |
1ff9e6e1 | 33 | #include <linux/vm_event_item.h> |
d52aa412 | 34 | #include <linux/smp.h> |
8a9f3ccd | 35 | #include <linux/page-flags.h> |
66e1707b | 36 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
37 | #include <linux/bit_spinlock.h> |
38 | #include <linux/rcupdate.h> | |
e222432b | 39 | #include <linux/limits.h> |
b9e15baf | 40 | #include <linux/export.h> |
8c7c6e34 | 41 | #include <linux/mutex.h> |
bb4cc1a8 | 42 | #include <linux/rbtree.h> |
b6ac57d5 | 43 | #include <linux/slab.h> |
66e1707b | 44 | #include <linux/swap.h> |
02491447 | 45 | #include <linux/swapops.h> |
66e1707b | 46 | #include <linux/spinlock.h> |
2e72b634 | 47 | #include <linux/eventfd.h> |
79bd9814 | 48 | #include <linux/poll.h> |
2e72b634 | 49 | #include <linux/sort.h> |
66e1707b | 50 | #include <linux/fs.h> |
d2ceb9b7 | 51 | #include <linux/seq_file.h> |
70ddf637 | 52 | #include <linux/vmpressure.h> |
b69408e8 | 53 | #include <linux/mm_inline.h> |
5d1ea48b | 54 | #include <linux/swap_cgroup.h> |
cdec2e42 | 55 | #include <linux/cpu.h> |
158e0a2d | 56 | #include <linux/oom.h> |
0056f4e6 | 57 | #include <linux/lockdep.h> |
79bd9814 | 58 | #include <linux/file.h> |
b23afb93 | 59 | #include <linux/tracehook.h> |
c8713d0b | 60 | #include <linux/seq_buf.h> |
08e552c6 | 61 | #include "internal.h" |
d1a4c0b3 | 62 | #include <net/sock.h> |
4bd2c1ee | 63 | #include <net/ip.h> |
f35c3a8e | 64 | #include "slab.h" |
8cdea7c0 | 65 | |
7c0f6ba6 | 66 | #include <linux/uaccess.h> |
8697d331 | 67 | |
cc8e970c KM |
68 | #include <trace/events/vmscan.h> |
69 | ||
073219e9 TH |
70 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
71 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 72 | |
7d828602 JW |
73 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
74 | ||
a181b0e8 | 75 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 76 | |
f7e1cb6e JW |
77 | /* Socket memory accounting disabled? */ |
78 | static bool cgroup_memory_nosocket; | |
79 | ||
04823c83 VD |
80 | /* Kernel memory accounting disabled? */ |
81 | static bool cgroup_memory_nokmem; | |
82 | ||
21afa38e | 83 | /* Whether the swap controller is active */ |
c255a458 | 84 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 85 | int do_swap_account __read_mostly; |
c077719b | 86 | #else |
a0db00fc | 87 | #define do_swap_account 0 |
c077719b KH |
88 | #endif |
89 | ||
7941d214 JW |
90 | /* Whether legacy memory+swap accounting is active */ |
91 | static bool do_memsw_account(void) | |
92 | { | |
93 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; | |
94 | } | |
95 | ||
71cd3113 | 96 | static const char *const mem_cgroup_lru_names[] = { |
58cf188e SZ |
97 | "inactive_anon", |
98 | "active_anon", | |
99 | "inactive_file", | |
100 | "active_file", | |
101 | "unevictable", | |
102 | }; | |
103 | ||
a0db00fc KS |
104 | #define THRESHOLDS_EVENTS_TARGET 128 |
105 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
106 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 107 | |
bb4cc1a8 AM |
108 | /* |
109 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
110 | * their hierarchy representation | |
111 | */ | |
112 | ||
ef8f2327 | 113 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 114 | struct rb_root rb_root; |
fa90b2fd | 115 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
116 | spinlock_t lock; |
117 | }; | |
118 | ||
bb4cc1a8 AM |
119 | struct mem_cgroup_tree { |
120 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
121 | }; | |
122 | ||
123 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
124 | ||
9490ff27 KH |
125 | /* for OOM */ |
126 | struct mem_cgroup_eventfd_list { | |
127 | struct list_head list; | |
128 | struct eventfd_ctx *eventfd; | |
129 | }; | |
2e72b634 | 130 | |
79bd9814 TH |
131 | /* |
132 | * cgroup_event represents events which userspace want to receive. | |
133 | */ | |
3bc942f3 | 134 | struct mem_cgroup_event { |
79bd9814 | 135 | /* |
59b6f873 | 136 | * memcg which the event belongs to. |
79bd9814 | 137 | */ |
59b6f873 | 138 | struct mem_cgroup *memcg; |
79bd9814 TH |
139 | /* |
140 | * eventfd to signal userspace about the event. | |
141 | */ | |
142 | struct eventfd_ctx *eventfd; | |
143 | /* | |
144 | * Each of these stored in a list by the cgroup. | |
145 | */ | |
146 | struct list_head list; | |
fba94807 TH |
147 | /* |
148 | * register_event() callback will be used to add new userspace | |
149 | * waiter for changes related to this event. Use eventfd_signal() | |
150 | * on eventfd to send notification to userspace. | |
151 | */ | |
59b6f873 | 152 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 153 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
154 | /* |
155 | * unregister_event() callback will be called when userspace closes | |
156 | * the eventfd or on cgroup removing. This callback must be set, | |
157 | * if you want provide notification functionality. | |
158 | */ | |
59b6f873 | 159 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 160 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
161 | /* |
162 | * All fields below needed to unregister event when | |
163 | * userspace closes eventfd. | |
164 | */ | |
165 | poll_table pt; | |
166 | wait_queue_head_t *wqh; | |
ac6424b9 | 167 | wait_queue_entry_t wait; |
79bd9814 TH |
168 | struct work_struct remove; |
169 | }; | |
170 | ||
c0ff4b85 R |
171 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
172 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 173 | |
7dc74be0 DN |
174 | /* Stuffs for move charges at task migration. */ |
175 | /* | |
1dfab5ab | 176 | * Types of charges to be moved. |
7dc74be0 | 177 | */ |
1dfab5ab JW |
178 | #define MOVE_ANON 0x1U |
179 | #define MOVE_FILE 0x2U | |
180 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 181 | |
4ffef5fe DN |
182 | /* "mc" and its members are protected by cgroup_mutex */ |
183 | static struct move_charge_struct { | |
b1dd693e | 184 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 185 | struct mm_struct *mm; |
4ffef5fe DN |
186 | struct mem_cgroup *from; |
187 | struct mem_cgroup *to; | |
1dfab5ab | 188 | unsigned long flags; |
4ffef5fe | 189 | unsigned long precharge; |
854ffa8d | 190 | unsigned long moved_charge; |
483c30b5 | 191 | unsigned long moved_swap; |
8033b97c DN |
192 | struct task_struct *moving_task; /* a task moving charges */ |
193 | wait_queue_head_t waitq; /* a waitq for other context */ | |
194 | } mc = { | |
2bd9bb20 | 195 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
196 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
197 | }; | |
4ffef5fe | 198 | |
4e416953 BS |
199 | /* |
200 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
201 | * limit reclaim to prevent infinite loops, if they ever occur. | |
202 | */ | |
a0db00fc | 203 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 204 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 205 | |
217bc319 KH |
206 | enum charge_type { |
207 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 208 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 209 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 210 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
211 | NR_CHARGE_TYPE, |
212 | }; | |
213 | ||
8c7c6e34 | 214 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
215 | enum res_type { |
216 | _MEM, | |
217 | _MEMSWAP, | |
218 | _OOM_TYPE, | |
510fc4e1 | 219 | _KMEM, |
d55f90bf | 220 | _TCP, |
86ae53e1 GC |
221 | }; |
222 | ||
a0db00fc KS |
223 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
224 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 225 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
226 | /* Used for OOM nofiier */ |
227 | #define OOM_CONTROL (0) | |
8c7c6e34 | 228 | |
b05706f1 KT |
229 | /* |
230 | * Iteration constructs for visiting all cgroups (under a tree). If | |
231 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
232 | * be used for reference counting. | |
233 | */ | |
234 | #define for_each_mem_cgroup_tree(iter, root) \ | |
235 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
236 | iter != NULL; \ | |
237 | iter = mem_cgroup_iter(root, iter, NULL)) | |
238 | ||
239 | #define for_each_mem_cgroup(iter) \ | |
240 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
241 | iter != NULL; \ | |
242 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
243 | ||
7775face TH |
244 | static inline bool should_force_charge(void) |
245 | { | |
246 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
247 | (current->flags & PF_EXITING); | |
248 | } | |
249 | ||
70ddf637 AV |
250 | /* Some nice accessors for the vmpressure. */ |
251 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
252 | { | |
253 | if (!memcg) | |
254 | memcg = root_mem_cgroup; | |
255 | return &memcg->vmpressure; | |
256 | } | |
257 | ||
258 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
259 | { | |
260 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
261 | } | |
262 | ||
84c07d11 | 263 | #ifdef CONFIG_MEMCG_KMEM |
55007d84 | 264 | /* |
f7ce3190 | 265 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
266 | * The main reason for not using cgroup id for this: |
267 | * this works better in sparse environments, where we have a lot of memcgs, | |
268 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
269 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
270 | * 200 entry array for that. | |
55007d84 | 271 | * |
dbcf73e2 VD |
272 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
273 | * will double each time we have to increase it. | |
55007d84 | 274 | */ |
dbcf73e2 VD |
275 | static DEFINE_IDA(memcg_cache_ida); |
276 | int memcg_nr_cache_ids; | |
749c5415 | 277 | |
05257a1a VD |
278 | /* Protects memcg_nr_cache_ids */ |
279 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
280 | ||
281 | void memcg_get_cache_ids(void) | |
282 | { | |
283 | down_read(&memcg_cache_ids_sem); | |
284 | } | |
285 | ||
286 | void memcg_put_cache_ids(void) | |
287 | { | |
288 | up_read(&memcg_cache_ids_sem); | |
289 | } | |
290 | ||
55007d84 GC |
291 | /* |
292 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
293 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
294 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
295 | * tunable, but that is strictly not necessary. | |
296 | * | |
b8627835 | 297 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
298 | * this constant directly from cgroup, but it is understandable that this is |
299 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 300 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
301 | * increase ours as well if it increases. |
302 | */ | |
303 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 304 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 305 | |
d7f25f8a GC |
306 | /* |
307 | * A lot of the calls to the cache allocation functions are expected to be | |
308 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
309 | * conditional to this static branch, we'll have to allow modules that does | |
310 | * kmem_cache_alloc and the such to see this symbol as well | |
311 | */ | |
ef12947c | 312 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 313 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 314 | |
17cc4dfe TH |
315 | struct workqueue_struct *memcg_kmem_cache_wq; |
316 | ||
0a4465d3 KT |
317 | static int memcg_shrinker_map_size; |
318 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); | |
319 | ||
320 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) | |
321 | { | |
322 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); | |
323 | } | |
324 | ||
325 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, | |
326 | int size, int old_size) | |
327 | { | |
328 | struct memcg_shrinker_map *new, *old; | |
329 | int nid; | |
330 | ||
331 | lockdep_assert_held(&memcg_shrinker_map_mutex); | |
332 | ||
333 | for_each_node(nid) { | |
334 | old = rcu_dereference_protected( | |
335 | mem_cgroup_nodeinfo(memcg, nid)->shrinker_map, true); | |
336 | /* Not yet online memcg */ | |
337 | if (!old) | |
338 | return 0; | |
339 | ||
340 | new = kvmalloc(sizeof(*new) + size, GFP_KERNEL); | |
341 | if (!new) | |
342 | return -ENOMEM; | |
343 | ||
344 | /* Set all old bits, clear all new bits */ | |
345 | memset(new->map, (int)0xff, old_size); | |
346 | memset((void *)new->map + old_size, 0, size - old_size); | |
347 | ||
348 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, new); | |
349 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); | |
350 | } | |
351 | ||
352 | return 0; | |
353 | } | |
354 | ||
355 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) | |
356 | { | |
357 | struct mem_cgroup_per_node *pn; | |
358 | struct memcg_shrinker_map *map; | |
359 | int nid; | |
360 | ||
361 | if (mem_cgroup_is_root(memcg)) | |
362 | return; | |
363 | ||
364 | for_each_node(nid) { | |
365 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
366 | map = rcu_dereference_protected(pn->shrinker_map, true); | |
367 | if (map) | |
368 | kvfree(map); | |
369 | rcu_assign_pointer(pn->shrinker_map, NULL); | |
370 | } | |
371 | } | |
372 | ||
373 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
374 | { | |
375 | struct memcg_shrinker_map *map; | |
376 | int nid, size, ret = 0; | |
377 | ||
378 | if (mem_cgroup_is_root(memcg)) | |
379 | return 0; | |
380 | ||
381 | mutex_lock(&memcg_shrinker_map_mutex); | |
382 | size = memcg_shrinker_map_size; | |
383 | for_each_node(nid) { | |
384 | map = kvzalloc(sizeof(*map) + size, GFP_KERNEL); | |
385 | if (!map) { | |
386 | memcg_free_shrinker_maps(memcg); | |
387 | ret = -ENOMEM; | |
388 | break; | |
389 | } | |
390 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); | |
391 | } | |
392 | mutex_unlock(&memcg_shrinker_map_mutex); | |
393 | ||
394 | return ret; | |
395 | } | |
396 | ||
397 | int memcg_expand_shrinker_maps(int new_id) | |
398 | { | |
399 | int size, old_size, ret = 0; | |
400 | struct mem_cgroup *memcg; | |
401 | ||
402 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); | |
403 | old_size = memcg_shrinker_map_size; | |
404 | if (size <= old_size) | |
405 | return 0; | |
406 | ||
407 | mutex_lock(&memcg_shrinker_map_mutex); | |
408 | if (!root_mem_cgroup) | |
409 | goto unlock; | |
410 | ||
411 | for_each_mem_cgroup(memcg) { | |
412 | if (mem_cgroup_is_root(memcg)) | |
413 | continue; | |
414 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); | |
415 | if (ret) | |
416 | goto unlock; | |
417 | } | |
418 | unlock: | |
419 | if (!ret) | |
420 | memcg_shrinker_map_size = size; | |
421 | mutex_unlock(&memcg_shrinker_map_mutex); | |
422 | return ret; | |
423 | } | |
fae91d6d KT |
424 | |
425 | void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id) | |
426 | { | |
427 | if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) { | |
428 | struct memcg_shrinker_map *map; | |
429 | ||
430 | rcu_read_lock(); | |
431 | map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map); | |
f90280d6 KT |
432 | /* Pairs with smp mb in shrink_slab() */ |
433 | smp_mb__before_atomic(); | |
fae91d6d KT |
434 | set_bit(shrinker_id, map->map); |
435 | rcu_read_unlock(); | |
436 | } | |
437 | } | |
438 | ||
0a4465d3 KT |
439 | #else /* CONFIG_MEMCG_KMEM */ |
440 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) | |
441 | { | |
442 | return 0; | |
443 | } | |
444 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) { } | |
84c07d11 | 445 | #endif /* CONFIG_MEMCG_KMEM */ |
a8964b9b | 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(); | |
489 | memcg = READ_ONCE(page->mem_cgroup); | |
490 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
491 | memcg = parent_mem_cgroup(memcg); | |
492 | if (memcg) | |
493 | ino = cgroup_ino(memcg->css.cgroup); | |
494 | rcu_read_unlock(); | |
495 | return ino; | |
496 | } | |
497 | ||
ef8f2327 MG |
498 | static struct mem_cgroup_per_node * |
499 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 500 | { |
97a6c37b | 501 | int nid = page_to_nid(page); |
f64c3f54 | 502 | |
ef8f2327 | 503 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
504 | } |
505 | ||
ef8f2327 MG |
506 | static struct mem_cgroup_tree_per_node * |
507 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 508 | { |
ef8f2327 | 509 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
510 | } |
511 | ||
ef8f2327 | 512 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
513 | soft_limit_tree_from_page(struct page *page) |
514 | { | |
515 | int nid = page_to_nid(page); | |
bb4cc1a8 | 516 | |
ef8f2327 | 517 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
518 | } |
519 | ||
ef8f2327 MG |
520 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
521 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 522 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
523 | { |
524 | struct rb_node **p = &mctz->rb_root.rb_node; | |
525 | struct rb_node *parent = NULL; | |
ef8f2327 | 526 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 527 | bool rightmost = true; |
bb4cc1a8 AM |
528 | |
529 | if (mz->on_tree) | |
530 | return; | |
531 | ||
532 | mz->usage_in_excess = new_usage_in_excess; | |
533 | if (!mz->usage_in_excess) | |
534 | return; | |
535 | while (*p) { | |
536 | parent = *p; | |
ef8f2327 | 537 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 538 | tree_node); |
fa90b2fd | 539 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 540 | p = &(*p)->rb_left; |
fa90b2fd DB |
541 | rightmost = false; |
542 | } | |
543 | ||
bb4cc1a8 AM |
544 | /* |
545 | * We can't avoid mem cgroups that are over their soft | |
546 | * limit by the same amount | |
547 | */ | |
548 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
549 | p = &(*p)->rb_right; | |
550 | } | |
fa90b2fd DB |
551 | |
552 | if (rightmost) | |
553 | mctz->rb_rightmost = &mz->tree_node; | |
554 | ||
bb4cc1a8 AM |
555 | rb_link_node(&mz->tree_node, parent, p); |
556 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
557 | mz->on_tree = true; | |
558 | } | |
559 | ||
ef8f2327 MG |
560 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
561 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
562 | { |
563 | if (!mz->on_tree) | |
564 | return; | |
fa90b2fd DB |
565 | |
566 | if (&mz->tree_node == mctz->rb_rightmost) | |
567 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
568 | ||
bb4cc1a8 AM |
569 | rb_erase(&mz->tree_node, &mctz->rb_root); |
570 | mz->on_tree = false; | |
571 | } | |
572 | ||
ef8f2327 MG |
573 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
574 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 575 | { |
0a31bc97 JW |
576 | unsigned long flags; |
577 | ||
578 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 579 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 580 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
581 | } |
582 | ||
3e32cb2e JW |
583 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
584 | { | |
585 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 586 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
587 | unsigned long excess = 0; |
588 | ||
589 | if (nr_pages > soft_limit) | |
590 | excess = nr_pages - soft_limit; | |
591 | ||
592 | return excess; | |
593 | } | |
bb4cc1a8 AM |
594 | |
595 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
596 | { | |
3e32cb2e | 597 | unsigned long excess; |
ef8f2327 MG |
598 | struct mem_cgroup_per_node *mz; |
599 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 600 | |
e231875b | 601 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
602 | if (!mctz) |
603 | return; | |
bb4cc1a8 AM |
604 | /* |
605 | * Necessary to update all ancestors when hierarchy is used. | |
606 | * because their event counter is not touched. | |
607 | */ | |
608 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 609 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 610 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
611 | /* |
612 | * We have to update the tree if mz is on RB-tree or | |
613 | * mem is over its softlimit. | |
614 | */ | |
615 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
616 | unsigned long flags; |
617 | ||
618 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
619 | /* if on-tree, remove it */ |
620 | if (mz->on_tree) | |
cf2c8127 | 621 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
622 | /* |
623 | * Insert again. mz->usage_in_excess will be updated. | |
624 | * If excess is 0, no tree ops. | |
625 | */ | |
cf2c8127 | 626 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 627 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
628 | } |
629 | } | |
630 | } | |
631 | ||
632 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
633 | { | |
ef8f2327 MG |
634 | struct mem_cgroup_tree_per_node *mctz; |
635 | struct mem_cgroup_per_node *mz; | |
636 | int nid; | |
bb4cc1a8 | 637 | |
e231875b | 638 | for_each_node(nid) { |
ef8f2327 MG |
639 | mz = mem_cgroup_nodeinfo(memcg, nid); |
640 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
641 | if (mctz) |
642 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
643 | } |
644 | } | |
645 | ||
ef8f2327 MG |
646 | static struct mem_cgroup_per_node * |
647 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 648 | { |
ef8f2327 | 649 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
650 | |
651 | retry: | |
652 | mz = NULL; | |
fa90b2fd | 653 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
654 | goto done; /* Nothing to reclaim from */ |
655 | ||
fa90b2fd DB |
656 | mz = rb_entry(mctz->rb_rightmost, |
657 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
658 | /* |
659 | * Remove the node now but someone else can add it back, | |
660 | * we will to add it back at the end of reclaim to its correct | |
661 | * position in the tree. | |
662 | */ | |
cf2c8127 | 663 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 664 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 665 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
666 | goto retry; |
667 | done: | |
668 | return mz; | |
669 | } | |
670 | ||
ef8f2327 MG |
671 | static struct mem_cgroup_per_node * |
672 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 673 | { |
ef8f2327 | 674 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 675 | |
0a31bc97 | 676 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 677 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 678 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
679 | return mz; |
680 | } | |
681 | ||
db9adbcb JW |
682 | /** |
683 | * __mod_memcg_state - update cgroup memory statistics | |
684 | * @memcg: the memory cgroup | |
685 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
686 | * @val: delta to add to the counter, can be negative | |
687 | */ | |
688 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
689 | { | |
690 | long x; | |
691 | ||
692 | if (mem_cgroup_disabled()) | |
693 | return; | |
694 | ||
815744d7 JW |
695 | __this_cpu_add(memcg->vmstats_local->stat[idx], val); |
696 | ||
db9adbcb JW |
697 | x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); |
698 | if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
699 | struct mem_cgroup *mi; |
700 | ||
42a30035 JW |
701 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
702 | atomic_long_add(x, &mi->vmstats[idx]); | |
db9adbcb JW |
703 | x = 0; |
704 | } | |
705 | __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); | |
706 | } | |
707 | ||
42a30035 JW |
708 | static struct mem_cgroup_per_node * |
709 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
710 | { | |
711 | struct mem_cgroup *parent; | |
712 | ||
713 | parent = parent_mem_cgroup(pn->memcg); | |
714 | if (!parent) | |
715 | return NULL; | |
716 | return mem_cgroup_nodeinfo(parent, nid); | |
717 | } | |
718 | ||
db9adbcb JW |
719 | /** |
720 | * __mod_lruvec_state - update lruvec memory statistics | |
721 | * @lruvec: the lruvec | |
722 | * @idx: the stat item | |
723 | * @val: delta to add to the counter, can be negative | |
724 | * | |
725 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
726 | * function updates the all three counters that are affected by a | |
727 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
728 | */ | |
729 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
730 | int val) | |
731 | { | |
42a30035 | 732 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
db9adbcb | 733 | struct mem_cgroup_per_node *pn; |
42a30035 | 734 | struct mem_cgroup *memcg; |
db9adbcb JW |
735 | long x; |
736 | ||
737 | /* Update node */ | |
42a30035 | 738 | __mod_node_page_state(pgdat, idx, val); |
db9adbcb JW |
739 | |
740 | if (mem_cgroup_disabled()) | |
741 | return; | |
742 | ||
743 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); | |
42a30035 | 744 | memcg = pn->memcg; |
db9adbcb JW |
745 | |
746 | /* Update memcg */ | |
42a30035 | 747 | __mod_memcg_state(memcg, idx, val); |
db9adbcb JW |
748 | |
749 | /* Update lruvec */ | |
815744d7 JW |
750 | __this_cpu_add(pn->lruvec_stat_local->count[idx], val); |
751 | ||
db9adbcb JW |
752 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); |
753 | if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
754 | struct mem_cgroup_per_node *pi; |
755 | ||
42a30035 JW |
756 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) |
757 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
758 | x = 0; |
759 | } | |
760 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
761 | } | |
762 | ||
763 | /** | |
764 | * __count_memcg_events - account VM events in a cgroup | |
765 | * @memcg: the memory cgroup | |
766 | * @idx: the event item | |
767 | * @count: the number of events that occured | |
768 | */ | |
769 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
770 | unsigned long count) | |
771 | { | |
772 | unsigned long x; | |
773 | ||
774 | if (mem_cgroup_disabled()) | |
775 | return; | |
776 | ||
815744d7 JW |
777 | __this_cpu_add(memcg->vmstats_local->events[idx], count); |
778 | ||
db9adbcb JW |
779 | x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); |
780 | if (unlikely(x > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
781 | struct mem_cgroup *mi; |
782 | ||
42a30035 JW |
783 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
784 | atomic_long_add(x, &mi->vmevents[idx]); | |
db9adbcb JW |
785 | x = 0; |
786 | } | |
787 | __this_cpu_write(memcg->vmstats_percpu->events[idx], x); | |
788 | } | |
789 | ||
42a30035 | 790 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 791 | { |
871789d4 | 792 | return atomic_long_read(&memcg->vmevents[event]); |
e9f8974f JW |
793 | } |
794 | ||
42a30035 JW |
795 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
796 | { | |
815744d7 JW |
797 | long x = 0; |
798 | int cpu; | |
799 | ||
800 | for_each_possible_cpu(cpu) | |
801 | x += per_cpu(memcg->vmstats_local->events[event], cpu); | |
802 | return x; | |
42a30035 JW |
803 | } |
804 | ||
c0ff4b85 | 805 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 806 | struct page *page, |
f627c2f5 | 807 | bool compound, int nr_pages) |
d52aa412 | 808 | { |
b2402857 KH |
809 | /* |
810 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
811 | * counted as CACHE even if it's on ANON LRU. | |
812 | */ | |
0a31bc97 | 813 | if (PageAnon(page)) |
c9019e9b | 814 | __mod_memcg_state(memcg, MEMCG_RSS, nr_pages); |
9a4caf1e | 815 | else { |
c9019e9b | 816 | __mod_memcg_state(memcg, MEMCG_CACHE, nr_pages); |
9a4caf1e | 817 | if (PageSwapBacked(page)) |
c9019e9b | 818 | __mod_memcg_state(memcg, NR_SHMEM, nr_pages); |
9a4caf1e | 819 | } |
55e462b0 | 820 | |
f627c2f5 KS |
821 | if (compound) { |
822 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
c9019e9b | 823 | __mod_memcg_state(memcg, MEMCG_RSS_HUGE, nr_pages); |
f627c2f5 | 824 | } |
b070e65c | 825 | |
e401f176 KH |
826 | /* pagein of a big page is an event. So, ignore page size */ |
827 | if (nr_pages > 0) | |
c9019e9b | 828 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 829 | else { |
c9019e9b | 830 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
831 | nr_pages = -nr_pages; /* for event */ |
832 | } | |
e401f176 | 833 | |
871789d4 | 834 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
835 | } |
836 | ||
f53d7ce3 JW |
837 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
838 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
839 | { |
840 | unsigned long val, next; | |
841 | ||
871789d4 CD |
842 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
843 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 844 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 845 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
846 | switch (target) { |
847 | case MEM_CGROUP_TARGET_THRESH: | |
848 | next = val + THRESHOLDS_EVENTS_TARGET; | |
849 | break; | |
bb4cc1a8 AM |
850 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
851 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
852 | break; | |
f53d7ce3 JW |
853 | case MEM_CGROUP_TARGET_NUMAINFO: |
854 | next = val + NUMAINFO_EVENTS_TARGET; | |
855 | break; | |
856 | default: | |
857 | break; | |
858 | } | |
871789d4 | 859 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 860 | return true; |
7a159cc9 | 861 | } |
f53d7ce3 | 862 | return false; |
d2265e6f KH |
863 | } |
864 | ||
865 | /* | |
866 | * Check events in order. | |
867 | * | |
868 | */ | |
c0ff4b85 | 869 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
870 | { |
871 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
872 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
873 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 874 | bool do_softlimit; |
82b3f2a7 | 875 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 876 | |
bb4cc1a8 AM |
877 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
878 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
879 | #if MAX_NUMNODES > 1 |
880 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
881 | MEM_CGROUP_TARGET_NUMAINFO); | |
882 | #endif | |
c0ff4b85 | 883 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
884 | if (unlikely(do_softlimit)) |
885 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 886 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 887 | if (unlikely(do_numainfo)) |
c0ff4b85 | 888 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 889 | #endif |
0a31bc97 | 890 | } |
d2265e6f KH |
891 | } |
892 | ||
cf475ad2 | 893 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 894 | { |
31a78f23 BS |
895 | /* |
896 | * mm_update_next_owner() may clear mm->owner to NULL | |
897 | * if it races with swapoff, page migration, etc. | |
898 | * So this can be called with p == NULL. | |
899 | */ | |
900 | if (unlikely(!p)) | |
901 | return NULL; | |
902 | ||
073219e9 | 903 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 904 | } |
33398cf2 | 905 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 906 | |
d46eb14b SB |
907 | /** |
908 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
909 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
910 | * | |
911 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
912 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
913 | * returned. | |
914 | */ | |
915 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 916 | { |
d46eb14b SB |
917 | struct mem_cgroup *memcg; |
918 | ||
919 | if (mem_cgroup_disabled()) | |
920 | return NULL; | |
0b7f569e | 921 | |
54595fe2 KH |
922 | rcu_read_lock(); |
923 | do { | |
6f6acb00 MH |
924 | /* |
925 | * Page cache insertions can happen withou an | |
926 | * actual mm context, e.g. during disk probing | |
927 | * on boot, loopback IO, acct() writes etc. | |
928 | */ | |
929 | if (unlikely(!mm)) | |
df381975 | 930 | memcg = root_mem_cgroup; |
6f6acb00 MH |
931 | else { |
932 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
933 | if (unlikely(!memcg)) | |
934 | memcg = root_mem_cgroup; | |
935 | } | |
ec903c0c | 936 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 937 | rcu_read_unlock(); |
c0ff4b85 | 938 | return memcg; |
54595fe2 | 939 | } |
d46eb14b SB |
940 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
941 | ||
f745c6f5 SB |
942 | /** |
943 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
944 | * @page: page from which memcg should be extracted. | |
945 | * | |
946 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
947 | * root_mem_cgroup is returned. | |
948 | */ | |
949 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
950 | { | |
951 | struct mem_cgroup *memcg = page->mem_cgroup; | |
952 | ||
953 | if (mem_cgroup_disabled()) | |
954 | return NULL; | |
955 | ||
956 | rcu_read_lock(); | |
957 | if (!memcg || !css_tryget_online(&memcg->css)) | |
958 | memcg = root_mem_cgroup; | |
959 | rcu_read_unlock(); | |
960 | return memcg; | |
961 | } | |
962 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
963 | ||
d46eb14b SB |
964 | /** |
965 | * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg. | |
966 | */ | |
967 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
968 | { | |
969 | if (unlikely(current->active_memcg)) { | |
970 | struct mem_cgroup *memcg = root_mem_cgroup; | |
971 | ||
972 | rcu_read_lock(); | |
973 | if (css_tryget_online(¤t->active_memcg->css)) | |
974 | memcg = current->active_memcg; | |
975 | rcu_read_unlock(); | |
976 | return memcg; | |
977 | } | |
978 | return get_mem_cgroup_from_mm(current->mm); | |
979 | } | |
54595fe2 | 980 | |
5660048c JW |
981 | /** |
982 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
983 | * @root: hierarchy root | |
984 | * @prev: previously returned memcg, NULL on first invocation | |
985 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
986 | * | |
987 | * Returns references to children of the hierarchy below @root, or | |
988 | * @root itself, or %NULL after a full round-trip. | |
989 | * | |
990 | * Caller must pass the return value in @prev on subsequent | |
991 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
992 | * to cancel a hierarchy walk before the round-trip is complete. | |
993 | * | |
b213b54f | 994 | * Reclaimers can specify a node and a priority level in @reclaim to |
5660048c | 995 | * divide up the memcgs in the hierarchy among all concurrent |
b213b54f | 996 | * reclaimers operating on the same node and priority. |
5660048c | 997 | */ |
694fbc0f | 998 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 999 | struct mem_cgroup *prev, |
694fbc0f | 1000 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1001 | { |
33398cf2 | 1002 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 1003 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1004 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1005 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1006 | |
694fbc0f AM |
1007 | if (mem_cgroup_disabled()) |
1008 | return NULL; | |
5660048c | 1009 | |
9f3a0d09 JW |
1010 | if (!root) |
1011 | root = root_mem_cgroup; | |
7d74b06f | 1012 | |
9f3a0d09 | 1013 | if (prev && !reclaim) |
5ac8fb31 | 1014 | pos = prev; |
14067bb3 | 1015 | |
9f3a0d09 JW |
1016 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1017 | if (prev) | |
5ac8fb31 | 1018 | goto out; |
694fbc0f | 1019 | return root; |
9f3a0d09 | 1020 | } |
14067bb3 | 1021 | |
542f85f9 | 1022 | rcu_read_lock(); |
5f578161 | 1023 | |
5ac8fb31 | 1024 | if (reclaim) { |
ef8f2327 | 1025 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1026 | |
ef8f2327 | 1027 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
5ac8fb31 JW |
1028 | iter = &mz->iter[reclaim->priority]; |
1029 | ||
1030 | if (prev && reclaim->generation != iter->generation) | |
1031 | goto out_unlock; | |
1032 | ||
6df38689 | 1033 | while (1) { |
4db0c3c2 | 1034 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1035 | if (!pos || css_tryget(&pos->css)) |
1036 | break; | |
5ac8fb31 | 1037 | /* |
6df38689 VD |
1038 | * css reference reached zero, so iter->position will |
1039 | * be cleared by ->css_released. However, we should not | |
1040 | * rely on this happening soon, because ->css_released | |
1041 | * is called from a work queue, and by busy-waiting we | |
1042 | * might block it. So we clear iter->position right | |
1043 | * away. | |
5ac8fb31 | 1044 | */ |
6df38689 VD |
1045 | (void)cmpxchg(&iter->position, pos, NULL); |
1046 | } | |
5ac8fb31 JW |
1047 | } |
1048 | ||
1049 | if (pos) | |
1050 | css = &pos->css; | |
1051 | ||
1052 | for (;;) { | |
1053 | css = css_next_descendant_pre(css, &root->css); | |
1054 | if (!css) { | |
1055 | /* | |
1056 | * Reclaimers share the hierarchy walk, and a | |
1057 | * new one might jump in right at the end of | |
1058 | * the hierarchy - make sure they see at least | |
1059 | * one group and restart from the beginning. | |
1060 | */ | |
1061 | if (!prev) | |
1062 | continue; | |
1063 | break; | |
527a5ec9 | 1064 | } |
7d74b06f | 1065 | |
5ac8fb31 JW |
1066 | /* |
1067 | * Verify the css and acquire a reference. The root | |
1068 | * is provided by the caller, so we know it's alive | |
1069 | * and kicking, and don't take an extra reference. | |
1070 | */ | |
1071 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1072 | |
5ac8fb31 JW |
1073 | if (css == &root->css) |
1074 | break; | |
14067bb3 | 1075 | |
0b8f73e1 JW |
1076 | if (css_tryget(css)) |
1077 | break; | |
9f3a0d09 | 1078 | |
5ac8fb31 | 1079 | memcg = NULL; |
9f3a0d09 | 1080 | } |
5ac8fb31 JW |
1081 | |
1082 | if (reclaim) { | |
5ac8fb31 | 1083 | /* |
6df38689 VD |
1084 | * The position could have already been updated by a competing |
1085 | * thread, so check that the value hasn't changed since we read | |
1086 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1087 | */ |
6df38689 VD |
1088 | (void)cmpxchg(&iter->position, pos, memcg); |
1089 | ||
5ac8fb31 JW |
1090 | if (pos) |
1091 | css_put(&pos->css); | |
1092 | ||
1093 | if (!memcg) | |
1094 | iter->generation++; | |
1095 | else if (!prev) | |
1096 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1097 | } |
5ac8fb31 | 1098 | |
542f85f9 MH |
1099 | out_unlock: |
1100 | rcu_read_unlock(); | |
5ac8fb31 | 1101 | out: |
c40046f3 MH |
1102 | if (prev && prev != root) |
1103 | css_put(&prev->css); | |
1104 | ||
9f3a0d09 | 1105 | return memcg; |
14067bb3 | 1106 | } |
7d74b06f | 1107 | |
5660048c JW |
1108 | /** |
1109 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1110 | * @root: hierarchy root | |
1111 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1112 | */ | |
1113 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1114 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1115 | { |
1116 | if (!root) | |
1117 | root = root_mem_cgroup; | |
1118 | if (prev && prev != root) | |
1119 | css_put(&prev->css); | |
1120 | } | |
7d74b06f | 1121 | |
6df38689 VD |
1122 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1123 | { | |
1124 | struct mem_cgroup *memcg = dead_memcg; | |
1125 | struct mem_cgroup_reclaim_iter *iter; | |
ef8f2327 MG |
1126 | struct mem_cgroup_per_node *mz; |
1127 | int nid; | |
6df38689 VD |
1128 | int i; |
1129 | ||
9f15bde6 | 1130 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
6df38689 | 1131 | for_each_node(nid) { |
ef8f2327 MG |
1132 | mz = mem_cgroup_nodeinfo(memcg, nid); |
1133 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
1134 | iter = &mz->iter[i]; | |
1135 | cmpxchg(&iter->position, | |
1136 | dead_memcg, NULL); | |
6df38689 VD |
1137 | } |
1138 | } | |
1139 | } | |
1140 | } | |
1141 | ||
7c5f64f8 VD |
1142 | /** |
1143 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1144 | * @memcg: hierarchy root | |
1145 | * @fn: function to call for each task | |
1146 | * @arg: argument passed to @fn | |
1147 | * | |
1148 | * This function iterates over tasks attached to @memcg or to any of its | |
1149 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1150 | * value, the function breaks the iteration loop and returns the value. | |
1151 | * Otherwise, it will iterate over all tasks and return 0. | |
1152 | * | |
1153 | * This function must not be called for the root memory cgroup. | |
1154 | */ | |
1155 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1156 | int (*fn)(struct task_struct *, void *), void *arg) | |
1157 | { | |
1158 | struct mem_cgroup *iter; | |
1159 | int ret = 0; | |
1160 | ||
1161 | BUG_ON(memcg == root_mem_cgroup); | |
1162 | ||
1163 | for_each_mem_cgroup_tree(iter, memcg) { | |
1164 | struct css_task_iter it; | |
1165 | struct task_struct *task; | |
1166 | ||
bc2fb7ed | 1167 | css_task_iter_start(&iter->css, 0, &it); |
7c5f64f8 VD |
1168 | while (!ret && (task = css_task_iter_next(&it))) |
1169 | ret = fn(task, arg); | |
1170 | css_task_iter_end(&it); | |
1171 | if (ret) { | |
1172 | mem_cgroup_iter_break(memcg, iter); | |
1173 | break; | |
1174 | } | |
1175 | } | |
1176 | return ret; | |
1177 | } | |
1178 | ||
925b7673 | 1179 | /** |
dfe0e773 | 1180 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1181 | * @page: the page |
f144c390 | 1182 | * @pgdat: pgdat of the page |
dfe0e773 JW |
1183 | * |
1184 | * This function is only safe when following the LRU page isolation | |
1185 | * and putback protocol: the LRU lock must be held, and the page must | |
1186 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1187 | */ |
599d0c95 | 1188 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1189 | { |
ef8f2327 | 1190 | struct mem_cgroup_per_node *mz; |
925b7673 | 1191 | struct mem_cgroup *memcg; |
bea8c150 | 1192 | struct lruvec *lruvec; |
6d12e2d8 | 1193 | |
bea8c150 | 1194 | if (mem_cgroup_disabled()) { |
599d0c95 | 1195 | lruvec = &pgdat->lruvec; |
bea8c150 HD |
1196 | goto out; |
1197 | } | |
925b7673 | 1198 | |
1306a85a | 1199 | memcg = page->mem_cgroup; |
7512102c | 1200 | /* |
dfe0e773 | 1201 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1202 | * possibly migrated - before they are charged. |
7512102c | 1203 | */ |
29833315 JW |
1204 | if (!memcg) |
1205 | memcg = root_mem_cgroup; | |
7512102c | 1206 | |
ef8f2327 | 1207 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1208 | lruvec = &mz->lruvec; |
1209 | out: | |
1210 | /* | |
1211 | * Since a node can be onlined after the mem_cgroup was created, | |
1212 | * we have to be prepared to initialize lruvec->zone here; | |
1213 | * and if offlined then reonlined, we need to reinitialize it. | |
1214 | */ | |
599d0c95 MG |
1215 | if (unlikely(lruvec->pgdat != pgdat)) |
1216 | lruvec->pgdat = pgdat; | |
bea8c150 | 1217 | return lruvec; |
08e552c6 | 1218 | } |
b69408e8 | 1219 | |
925b7673 | 1220 | /** |
fa9add64 HD |
1221 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1222 | * @lruvec: mem_cgroup per zone lru vector | |
1223 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1224 | * @zid: zone id of the accounted pages |
fa9add64 | 1225 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1226 | * |
ca707239 HD |
1227 | * This function must be called under lru_lock, just before a page is added |
1228 | * to or just after a page is removed from an lru list (that ordering being | |
1229 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1230 | */ |
fa9add64 | 1231 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1232 | int zid, int nr_pages) |
3f58a829 | 1233 | { |
ef8f2327 | 1234 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1235 | unsigned long *lru_size; |
ca707239 | 1236 | long size; |
3f58a829 MK |
1237 | |
1238 | if (mem_cgroup_disabled()) | |
1239 | return; | |
1240 | ||
ef8f2327 | 1241 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1242 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1243 | |
1244 | if (nr_pages < 0) | |
1245 | *lru_size += nr_pages; | |
1246 | ||
1247 | size = *lru_size; | |
b4536f0c MH |
1248 | if (WARN_ONCE(size < 0, |
1249 | "%s(%p, %d, %d): lru_size %ld\n", | |
1250 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1251 | VM_BUG_ON(1); |
1252 | *lru_size = 0; | |
1253 | } | |
1254 | ||
1255 | if (nr_pages > 0) | |
1256 | *lru_size += nr_pages; | |
08e552c6 | 1257 | } |
544122e5 | 1258 | |
2314b42d | 1259 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1260 | { |
2314b42d | 1261 | struct mem_cgroup *task_memcg; |
158e0a2d | 1262 | struct task_struct *p; |
ffbdccf5 | 1263 | bool ret; |
4c4a2214 | 1264 | |
158e0a2d | 1265 | p = find_lock_task_mm(task); |
de077d22 | 1266 | if (p) { |
2314b42d | 1267 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1268 | task_unlock(p); |
1269 | } else { | |
1270 | /* | |
1271 | * All threads may have already detached their mm's, but the oom | |
1272 | * killer still needs to detect if they have already been oom | |
1273 | * killed to prevent needlessly killing additional tasks. | |
1274 | */ | |
ffbdccf5 | 1275 | rcu_read_lock(); |
2314b42d JW |
1276 | task_memcg = mem_cgroup_from_task(task); |
1277 | css_get(&task_memcg->css); | |
ffbdccf5 | 1278 | rcu_read_unlock(); |
de077d22 | 1279 | } |
2314b42d JW |
1280 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1281 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1282 | return ret; |
1283 | } | |
1284 | ||
19942822 | 1285 | /** |
9d11ea9f | 1286 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1287 | * @memcg: the memory cgroup |
19942822 | 1288 | * |
9d11ea9f | 1289 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1290 | * pages. |
19942822 | 1291 | */ |
c0ff4b85 | 1292 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1293 | { |
3e32cb2e JW |
1294 | unsigned long margin = 0; |
1295 | unsigned long count; | |
1296 | unsigned long limit; | |
9d11ea9f | 1297 | |
3e32cb2e | 1298 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1299 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1300 | if (count < limit) |
1301 | margin = limit - count; | |
1302 | ||
7941d214 | 1303 | if (do_memsw_account()) { |
3e32cb2e | 1304 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1305 | limit = READ_ONCE(memcg->memsw.max); |
3e32cb2e JW |
1306 | if (count <= limit) |
1307 | margin = min(margin, limit - count); | |
cbedbac3 LR |
1308 | else |
1309 | margin = 0; | |
3e32cb2e JW |
1310 | } |
1311 | ||
1312 | return margin; | |
19942822 JW |
1313 | } |
1314 | ||
32047e2a | 1315 | /* |
bdcbb659 | 1316 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1317 | * |
bdcbb659 QH |
1318 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1319 | * moving cgroups. This is for waiting at high-memory pressure | |
1320 | * caused by "move". | |
32047e2a | 1321 | */ |
c0ff4b85 | 1322 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1323 | { |
2bd9bb20 KH |
1324 | struct mem_cgroup *from; |
1325 | struct mem_cgroup *to; | |
4b534334 | 1326 | bool ret = false; |
2bd9bb20 KH |
1327 | /* |
1328 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1329 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1330 | */ | |
1331 | spin_lock(&mc.lock); | |
1332 | from = mc.from; | |
1333 | to = mc.to; | |
1334 | if (!from) | |
1335 | goto unlock; | |
3e92041d | 1336 | |
2314b42d JW |
1337 | ret = mem_cgroup_is_descendant(from, memcg) || |
1338 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1339 | unlock: |
1340 | spin_unlock(&mc.lock); | |
4b534334 KH |
1341 | return ret; |
1342 | } | |
1343 | ||
c0ff4b85 | 1344 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1345 | { |
1346 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1347 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1348 | DEFINE_WAIT(wait); |
1349 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1350 | /* moving charge context might have finished. */ | |
1351 | if (mc.moving_task) | |
1352 | schedule(); | |
1353 | finish_wait(&mc.waitq, &wait); | |
1354 | return true; | |
1355 | } | |
1356 | } | |
1357 | return false; | |
1358 | } | |
1359 | ||
c8713d0b JW |
1360 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1361 | { | |
1362 | struct seq_buf s; | |
1363 | int i; | |
71cd3113 | 1364 | |
c8713d0b JW |
1365 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1366 | if (!s.buffer) | |
1367 | return NULL; | |
1368 | ||
1369 | /* | |
1370 | * Provide statistics on the state of the memory subsystem as | |
1371 | * well as cumulative event counters that show past behavior. | |
1372 | * | |
1373 | * This list is ordered following a combination of these gradients: | |
1374 | * 1) generic big picture -> specifics and details | |
1375 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1376 | * | |
1377 | * Current memory state: | |
1378 | */ | |
1379 | ||
1380 | seq_buf_printf(&s, "anon %llu\n", | |
1381 | (u64)memcg_page_state(memcg, MEMCG_RSS) * | |
1382 | PAGE_SIZE); | |
1383 | seq_buf_printf(&s, "file %llu\n", | |
1384 | (u64)memcg_page_state(memcg, MEMCG_CACHE) * | |
1385 | PAGE_SIZE); | |
1386 | seq_buf_printf(&s, "kernel_stack %llu\n", | |
1387 | (u64)memcg_page_state(memcg, MEMCG_KERNEL_STACK_KB) * | |
1388 | 1024); | |
1389 | seq_buf_printf(&s, "slab %llu\n", | |
1390 | (u64)(memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) + | |
1391 | memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE)) * | |
1392 | PAGE_SIZE); | |
1393 | seq_buf_printf(&s, "sock %llu\n", | |
1394 | (u64)memcg_page_state(memcg, MEMCG_SOCK) * | |
1395 | PAGE_SIZE); | |
1396 | ||
1397 | seq_buf_printf(&s, "shmem %llu\n", | |
1398 | (u64)memcg_page_state(memcg, NR_SHMEM) * | |
1399 | PAGE_SIZE); | |
1400 | seq_buf_printf(&s, "file_mapped %llu\n", | |
1401 | (u64)memcg_page_state(memcg, NR_FILE_MAPPED) * | |
1402 | PAGE_SIZE); | |
1403 | seq_buf_printf(&s, "file_dirty %llu\n", | |
1404 | (u64)memcg_page_state(memcg, NR_FILE_DIRTY) * | |
1405 | PAGE_SIZE); | |
1406 | seq_buf_printf(&s, "file_writeback %llu\n", | |
1407 | (u64)memcg_page_state(memcg, NR_WRITEBACK) * | |
1408 | PAGE_SIZE); | |
1409 | ||
1410 | /* | |
1411 | * TODO: We should eventually replace our own MEMCG_RSS_HUGE counter | |
1412 | * with the NR_ANON_THP vm counter, but right now it's a pain in the | |
1413 | * arse because it requires migrating the work out of rmap to a place | |
1414 | * where the page->mem_cgroup is set up and stable. | |
1415 | */ | |
1416 | seq_buf_printf(&s, "anon_thp %llu\n", | |
1417 | (u64)memcg_page_state(memcg, MEMCG_RSS_HUGE) * | |
1418 | PAGE_SIZE); | |
1419 | ||
1420 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1421 | seq_buf_printf(&s, "%s %llu\n", mem_cgroup_lru_names[i], | |
1422 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * | |
1423 | PAGE_SIZE); | |
1424 | ||
1425 | seq_buf_printf(&s, "slab_reclaimable %llu\n", | |
1426 | (u64)memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) * | |
1427 | PAGE_SIZE); | |
1428 | seq_buf_printf(&s, "slab_unreclaimable %llu\n", | |
1429 | (u64)memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE) * | |
1430 | PAGE_SIZE); | |
1431 | ||
1432 | /* Accumulated memory events */ | |
1433 | ||
1434 | seq_buf_printf(&s, "pgfault %lu\n", memcg_events(memcg, PGFAULT)); | |
1435 | seq_buf_printf(&s, "pgmajfault %lu\n", memcg_events(memcg, PGMAJFAULT)); | |
1436 | ||
1437 | seq_buf_printf(&s, "workingset_refault %lu\n", | |
1438 | memcg_page_state(memcg, WORKINGSET_REFAULT)); | |
1439 | seq_buf_printf(&s, "workingset_activate %lu\n", | |
1440 | memcg_page_state(memcg, WORKINGSET_ACTIVATE)); | |
1441 | seq_buf_printf(&s, "workingset_nodereclaim %lu\n", | |
1442 | memcg_page_state(memcg, WORKINGSET_NODERECLAIM)); | |
1443 | ||
1444 | seq_buf_printf(&s, "pgrefill %lu\n", memcg_events(memcg, PGREFILL)); | |
1445 | seq_buf_printf(&s, "pgscan %lu\n", | |
1446 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1447 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1448 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1449 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1450 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
1451 | seq_buf_printf(&s, "pgactivate %lu\n", memcg_events(memcg, PGACTIVATE)); | |
1452 | seq_buf_printf(&s, "pgdeactivate %lu\n", memcg_events(memcg, PGDEACTIVATE)); | |
1453 | seq_buf_printf(&s, "pglazyfree %lu\n", memcg_events(memcg, PGLAZYFREE)); | |
1454 | seq_buf_printf(&s, "pglazyfreed %lu\n", memcg_events(memcg, PGLAZYFREED)); | |
1455 | ||
1456 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1457 | seq_buf_printf(&s, "thp_fault_alloc %lu\n", | |
1458 | memcg_events(memcg, THP_FAULT_ALLOC)); | |
1459 | seq_buf_printf(&s, "thp_collapse_alloc %lu\n", | |
1460 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); | |
1461 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1462 | ||
1463 | /* The above should easily fit into one page */ | |
1464 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1465 | ||
1466 | return s.buffer; | |
1467 | } | |
71cd3113 | 1468 | |
58cf188e | 1469 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1470 | /** |
f0c867d9 | 1471 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1472 | * memory controller. | |
e222432b BS |
1473 | * @memcg: The memory cgroup that went over limit |
1474 | * @p: Task that is going to be killed | |
1475 | * | |
1476 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1477 | * enabled | |
1478 | */ | |
f0c867d9 | 1479 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1480 | { |
e222432b BS |
1481 | rcu_read_lock(); |
1482 | ||
f0c867d9 | 1483 | if (memcg) { |
1484 | pr_cont(",oom_memcg="); | |
1485 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1486 | } else | |
1487 | pr_cont(",global_oom"); | |
2415b9f5 | 1488 | if (p) { |
f0c867d9 | 1489 | pr_cont(",task_memcg="); |
2415b9f5 | 1490 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1491 | } |
e222432b | 1492 | rcu_read_unlock(); |
f0c867d9 | 1493 | } |
1494 | ||
1495 | /** | |
1496 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1497 | * memory controller. | |
1498 | * @memcg: The memory cgroup that went over limit | |
1499 | */ | |
1500 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1501 | { | |
c8713d0b | 1502 | char *buf; |
e222432b | 1503 | |
3e32cb2e JW |
1504 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1505 | K((u64)page_counter_read(&memcg->memory)), | |
bbec2e15 | 1506 | K((u64)memcg->memory.max), memcg->memory.failcnt); |
c8713d0b JW |
1507 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1508 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1509 | K((u64)page_counter_read(&memcg->swap)), | |
1510 | K((u64)memcg->swap.max), memcg->swap.failcnt); | |
1511 | else { | |
1512 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1513 | K((u64)page_counter_read(&memcg->memsw)), | |
1514 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1515 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1516 | K((u64)page_counter_read(&memcg->kmem)), | |
1517 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1518 | } |
c8713d0b JW |
1519 | |
1520 | pr_info("Memory cgroup stats for "); | |
1521 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1522 | pr_cont(":"); | |
1523 | buf = memory_stat_format(memcg); | |
1524 | if (!buf) | |
1525 | return; | |
1526 | pr_info("%s", buf); | |
1527 | kfree(buf); | |
e222432b BS |
1528 | } |
1529 | ||
a63d83f4 DR |
1530 | /* |
1531 | * Return the memory (and swap, if configured) limit for a memcg. | |
1532 | */ | |
bbec2e15 | 1533 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1534 | { |
bbec2e15 | 1535 | unsigned long max; |
f3e8eb70 | 1536 | |
bbec2e15 | 1537 | max = memcg->memory.max; |
9a5a8f19 | 1538 | if (mem_cgroup_swappiness(memcg)) { |
bbec2e15 RG |
1539 | unsigned long memsw_max; |
1540 | unsigned long swap_max; | |
9a5a8f19 | 1541 | |
bbec2e15 RG |
1542 | memsw_max = memcg->memsw.max; |
1543 | swap_max = memcg->swap.max; | |
1544 | swap_max = min(swap_max, (unsigned long)total_swap_pages); | |
1545 | max = min(max + swap_max, memsw_max); | |
9a5a8f19 | 1546 | } |
bbec2e15 | 1547 | return max; |
a63d83f4 DR |
1548 | } |
1549 | ||
b6e6edcf | 1550 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1551 | int order) |
9cbb78bb | 1552 | { |
6e0fc46d DR |
1553 | struct oom_control oc = { |
1554 | .zonelist = NULL, | |
1555 | .nodemask = NULL, | |
2a966b77 | 1556 | .memcg = memcg, |
6e0fc46d DR |
1557 | .gfp_mask = gfp_mask, |
1558 | .order = order, | |
6e0fc46d | 1559 | }; |
7c5f64f8 | 1560 | bool ret; |
9cbb78bb | 1561 | |
7775face TH |
1562 | if (mutex_lock_killable(&oom_lock)) |
1563 | return true; | |
1564 | /* | |
1565 | * A few threads which were not waiting at mutex_lock_killable() can | |
1566 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1567 | */ | |
1568 | ret = should_force_charge() || out_of_memory(&oc); | |
dc56401f | 1569 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1570 | return ret; |
9cbb78bb DR |
1571 | } |
1572 | ||
ae6e71d3 MC |
1573 | #if MAX_NUMNODES > 1 |
1574 | ||
4d0c066d KH |
1575 | /** |
1576 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1577 | * @memcg: the target memcg |
4d0c066d KH |
1578 | * @nid: the node ID to be checked. |
1579 | * @noswap : specify true here if the user wants flle only information. | |
1580 | * | |
1581 | * This function returns whether the specified memcg contains any | |
1582 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1583 | * pages in the node. | |
1584 | */ | |
c0ff4b85 | 1585 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1586 | int nid, bool noswap) |
1587 | { | |
2b487e59 JW |
1588 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); |
1589 | ||
def0fdae JW |
1590 | if (lruvec_page_state(lruvec, NR_INACTIVE_FILE) || |
1591 | lruvec_page_state(lruvec, NR_ACTIVE_FILE)) | |
4d0c066d KH |
1592 | return true; |
1593 | if (noswap || !total_swap_pages) | |
1594 | return false; | |
def0fdae JW |
1595 | if (lruvec_page_state(lruvec, NR_INACTIVE_ANON) || |
1596 | lruvec_page_state(lruvec, NR_ACTIVE_ANON)) | |
4d0c066d KH |
1597 | return true; |
1598 | return false; | |
1599 | ||
1600 | } | |
889976db YH |
1601 | |
1602 | /* | |
1603 | * Always updating the nodemask is not very good - even if we have an empty | |
1604 | * list or the wrong list here, we can start from some node and traverse all | |
1605 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1606 | * | |
1607 | */ | |
c0ff4b85 | 1608 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1609 | { |
1610 | int nid; | |
453a9bf3 KH |
1611 | /* |
1612 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1613 | * pagein/pageout changes since the last update. | |
1614 | */ | |
c0ff4b85 | 1615 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1616 | return; |
c0ff4b85 | 1617 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1618 | return; |
1619 | ||
889976db | 1620 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1621 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1622 | |
31aaea4a | 1623 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1624 | |
c0ff4b85 R |
1625 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1626 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1627 | } |
453a9bf3 | 1628 | |
c0ff4b85 R |
1629 | atomic_set(&memcg->numainfo_events, 0); |
1630 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1631 | } |
1632 | ||
1633 | /* | |
1634 | * Selecting a node where we start reclaim from. Because what we need is just | |
1635 | * reducing usage counter, start from anywhere is O,K. Considering | |
1636 | * memory reclaim from current node, there are pros. and cons. | |
1637 | * | |
1638 | * Freeing memory from current node means freeing memory from a node which | |
1639 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1640 | * hit limits, it will see a contention on a node. But freeing from remote | |
1641 | * node means more costs for memory reclaim because of memory latency. | |
1642 | * | |
1643 | * Now, we use round-robin. Better algorithm is welcomed. | |
1644 | */ | |
c0ff4b85 | 1645 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1646 | { |
1647 | int node; | |
1648 | ||
c0ff4b85 R |
1649 | mem_cgroup_may_update_nodemask(memcg); |
1650 | node = memcg->last_scanned_node; | |
889976db | 1651 | |
0edaf86c | 1652 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1653 | /* |
fda3d69b MH |
1654 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1655 | * last time it really checked all the LRUs due to rate limiting. | |
1656 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1657 | */ |
1658 | if (unlikely(node == MAX_NUMNODES)) | |
1659 | node = numa_node_id(); | |
1660 | ||
c0ff4b85 | 1661 | memcg->last_scanned_node = node; |
889976db YH |
1662 | return node; |
1663 | } | |
889976db | 1664 | #else |
c0ff4b85 | 1665 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1666 | { |
1667 | return 0; | |
1668 | } | |
1669 | #endif | |
1670 | ||
0608f43d | 1671 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1672 | pg_data_t *pgdat, |
0608f43d AM |
1673 | gfp_t gfp_mask, |
1674 | unsigned long *total_scanned) | |
1675 | { | |
1676 | struct mem_cgroup *victim = NULL; | |
1677 | int total = 0; | |
1678 | int loop = 0; | |
1679 | unsigned long excess; | |
1680 | unsigned long nr_scanned; | |
1681 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1682 | .pgdat = pgdat, |
0608f43d AM |
1683 | .priority = 0, |
1684 | }; | |
1685 | ||
3e32cb2e | 1686 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1687 | |
1688 | while (1) { | |
1689 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1690 | if (!victim) { | |
1691 | loop++; | |
1692 | if (loop >= 2) { | |
1693 | /* | |
1694 | * If we have not been able to reclaim | |
1695 | * anything, it might because there are | |
1696 | * no reclaimable pages under this hierarchy | |
1697 | */ | |
1698 | if (!total) | |
1699 | break; | |
1700 | /* | |
1701 | * We want to do more targeted reclaim. | |
1702 | * excess >> 2 is not to excessive so as to | |
1703 | * reclaim too much, nor too less that we keep | |
1704 | * coming back to reclaim from this cgroup | |
1705 | */ | |
1706 | if (total >= (excess >> 2) || | |
1707 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1708 | break; | |
1709 | } | |
1710 | continue; | |
1711 | } | |
a9dd0a83 | 1712 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1713 | pgdat, &nr_scanned); |
0608f43d | 1714 | *total_scanned += nr_scanned; |
3e32cb2e | 1715 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1716 | break; |
6d61ef40 | 1717 | } |
0608f43d AM |
1718 | mem_cgroup_iter_break(root_memcg, victim); |
1719 | return total; | |
6d61ef40 BS |
1720 | } |
1721 | ||
0056f4e6 JW |
1722 | #ifdef CONFIG_LOCKDEP |
1723 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1724 | .name = "memcg_oom_lock", | |
1725 | }; | |
1726 | #endif | |
1727 | ||
fb2a6fc5 JW |
1728 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1729 | ||
867578cb KH |
1730 | /* |
1731 | * Check OOM-Killer is already running under our hierarchy. | |
1732 | * If someone is running, return false. | |
1733 | */ | |
fb2a6fc5 | 1734 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1735 | { |
79dfdacc | 1736 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1737 | |
fb2a6fc5 JW |
1738 | spin_lock(&memcg_oom_lock); |
1739 | ||
9f3a0d09 | 1740 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1741 | if (iter->oom_lock) { |
79dfdacc MH |
1742 | /* |
1743 | * this subtree of our hierarchy is already locked | |
1744 | * so we cannot give a lock. | |
1745 | */ | |
79dfdacc | 1746 | failed = iter; |
9f3a0d09 JW |
1747 | mem_cgroup_iter_break(memcg, iter); |
1748 | break; | |
23751be0 JW |
1749 | } else |
1750 | iter->oom_lock = true; | |
7d74b06f | 1751 | } |
867578cb | 1752 | |
fb2a6fc5 JW |
1753 | if (failed) { |
1754 | /* | |
1755 | * OK, we failed to lock the whole subtree so we have | |
1756 | * to clean up what we set up to the failing subtree | |
1757 | */ | |
1758 | for_each_mem_cgroup_tree(iter, memcg) { | |
1759 | if (iter == failed) { | |
1760 | mem_cgroup_iter_break(memcg, iter); | |
1761 | break; | |
1762 | } | |
1763 | iter->oom_lock = false; | |
79dfdacc | 1764 | } |
0056f4e6 JW |
1765 | } else |
1766 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1767 | |
1768 | spin_unlock(&memcg_oom_lock); | |
1769 | ||
1770 | return !failed; | |
a636b327 | 1771 | } |
0b7f569e | 1772 | |
fb2a6fc5 | 1773 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1774 | { |
7d74b06f KH |
1775 | struct mem_cgroup *iter; |
1776 | ||
fb2a6fc5 | 1777 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1778 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1779 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1780 | iter->oom_lock = false; |
fb2a6fc5 | 1781 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1782 | } |
1783 | ||
c0ff4b85 | 1784 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1785 | { |
1786 | struct mem_cgroup *iter; | |
1787 | ||
c2b42d3c | 1788 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1789 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1790 | iter->under_oom++; |
1791 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1792 | } |
1793 | ||
c0ff4b85 | 1794 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1795 | { |
1796 | struct mem_cgroup *iter; | |
1797 | ||
867578cb KH |
1798 | /* |
1799 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1800 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1801 | */ |
c2b42d3c | 1802 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1803 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1804 | if (iter->under_oom > 0) |
1805 | iter->under_oom--; | |
1806 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1807 | } |
1808 | ||
867578cb KH |
1809 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1810 | ||
dc98df5a | 1811 | struct oom_wait_info { |
d79154bb | 1812 | struct mem_cgroup *memcg; |
ac6424b9 | 1813 | wait_queue_entry_t wait; |
dc98df5a KH |
1814 | }; |
1815 | ||
ac6424b9 | 1816 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1817 | unsigned mode, int sync, void *arg) |
1818 | { | |
d79154bb HD |
1819 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1820 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1821 | struct oom_wait_info *oom_wait_info; |
1822 | ||
1823 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1824 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1825 | |
2314b42d JW |
1826 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1827 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1828 | return 0; |
dc98df5a KH |
1829 | return autoremove_wake_function(wait, mode, sync, arg); |
1830 | } | |
1831 | ||
c0ff4b85 | 1832 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1833 | { |
c2b42d3c TH |
1834 | /* |
1835 | * For the following lockless ->under_oom test, the only required | |
1836 | * guarantee is that it must see the state asserted by an OOM when | |
1837 | * this function is called as a result of userland actions | |
1838 | * triggered by the notification of the OOM. This is trivially | |
1839 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1840 | * triggering notification. | |
1841 | */ | |
1842 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1843 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1844 | } |
1845 | ||
29ef680a MH |
1846 | enum oom_status { |
1847 | OOM_SUCCESS, | |
1848 | OOM_FAILED, | |
1849 | OOM_ASYNC, | |
1850 | OOM_SKIPPED | |
1851 | }; | |
1852 | ||
1853 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1854 | { |
7056d3a3 MH |
1855 | enum oom_status ret; |
1856 | bool locked; | |
1857 | ||
29ef680a MH |
1858 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1859 | return OOM_SKIPPED; | |
1860 | ||
7a1adfdd RG |
1861 | memcg_memory_event(memcg, MEMCG_OOM); |
1862 | ||
867578cb | 1863 | /* |
49426420 JW |
1864 | * We are in the middle of the charge context here, so we |
1865 | * don't want to block when potentially sitting on a callstack | |
1866 | * that holds all kinds of filesystem and mm locks. | |
1867 | * | |
29ef680a MH |
1868 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1869 | * handling until the charge can succeed; remember the context and put | |
1870 | * the task to sleep at the end of the page fault when all locks are | |
1871 | * released. | |
49426420 | 1872 | * |
29ef680a MH |
1873 | * On the other hand, in-kernel OOM killer allows for an async victim |
1874 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1875 | * relying on the oom victim to make a forward progress and we can | |
1876 | * invoke the oom killer here. | |
1877 | * | |
1878 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1879 | * victim and then we have to bail out from the charge path. | |
867578cb | 1880 | */ |
29ef680a MH |
1881 | if (memcg->oom_kill_disable) { |
1882 | if (!current->in_user_fault) | |
1883 | return OOM_SKIPPED; | |
1884 | css_get(&memcg->css); | |
1885 | current->memcg_in_oom = memcg; | |
1886 | current->memcg_oom_gfp_mask = mask; | |
1887 | current->memcg_oom_order = order; | |
1888 | ||
1889 | return OOM_ASYNC; | |
1890 | } | |
1891 | ||
7056d3a3 MH |
1892 | mem_cgroup_mark_under_oom(memcg); |
1893 | ||
1894 | locked = mem_cgroup_oom_trylock(memcg); | |
1895 | ||
1896 | if (locked) | |
1897 | mem_cgroup_oom_notify(memcg); | |
1898 | ||
1899 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1900 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1901 | ret = OOM_SUCCESS; |
1902 | else | |
1903 | ret = OOM_FAILED; | |
1904 | ||
1905 | if (locked) | |
1906 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1907 | |
7056d3a3 | 1908 | return ret; |
3812c8c8 JW |
1909 | } |
1910 | ||
1911 | /** | |
1912 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1913 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1914 | * |
49426420 JW |
1915 | * This has to be called at the end of a page fault if the memcg OOM |
1916 | * handler was enabled. | |
3812c8c8 | 1917 | * |
49426420 | 1918 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1919 | * sleep on a waitqueue until the userspace task resolves the |
1920 | * situation. Sleeping directly in the charge context with all kinds | |
1921 | * of locks held is not a good idea, instead we remember an OOM state | |
1922 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1923 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1924 | * |
1925 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1926 | * completed, %false otherwise. |
3812c8c8 | 1927 | */ |
49426420 | 1928 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1929 | { |
626ebc41 | 1930 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1931 | struct oom_wait_info owait; |
49426420 | 1932 | bool locked; |
3812c8c8 JW |
1933 | |
1934 | /* OOM is global, do not handle */ | |
3812c8c8 | 1935 | if (!memcg) |
49426420 | 1936 | return false; |
3812c8c8 | 1937 | |
7c5f64f8 | 1938 | if (!handle) |
49426420 | 1939 | goto cleanup; |
3812c8c8 JW |
1940 | |
1941 | owait.memcg = memcg; | |
1942 | owait.wait.flags = 0; | |
1943 | owait.wait.func = memcg_oom_wake_function; | |
1944 | owait.wait.private = current; | |
2055da97 | 1945 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1946 | |
3812c8c8 | 1947 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1948 | mem_cgroup_mark_under_oom(memcg); |
1949 | ||
1950 | locked = mem_cgroup_oom_trylock(memcg); | |
1951 | ||
1952 | if (locked) | |
1953 | mem_cgroup_oom_notify(memcg); | |
1954 | ||
1955 | if (locked && !memcg->oom_kill_disable) { | |
1956 | mem_cgroup_unmark_under_oom(memcg); | |
1957 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1958 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1959 | current->memcg_oom_order); | |
49426420 | 1960 | } else { |
3812c8c8 | 1961 | schedule(); |
49426420 JW |
1962 | mem_cgroup_unmark_under_oom(memcg); |
1963 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1964 | } | |
1965 | ||
1966 | if (locked) { | |
fb2a6fc5 JW |
1967 | mem_cgroup_oom_unlock(memcg); |
1968 | /* | |
1969 | * There is no guarantee that an OOM-lock contender | |
1970 | * sees the wakeups triggered by the OOM kill | |
1971 | * uncharges. Wake any sleepers explicitely. | |
1972 | */ | |
1973 | memcg_oom_recover(memcg); | |
1974 | } | |
49426420 | 1975 | cleanup: |
626ebc41 | 1976 | current->memcg_in_oom = NULL; |
3812c8c8 | 1977 | css_put(&memcg->css); |
867578cb | 1978 | return true; |
0b7f569e KH |
1979 | } |
1980 | ||
3d8b38eb RG |
1981 | /** |
1982 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
1983 | * @victim: task to be killed by the OOM killer | |
1984 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
1985 | * | |
1986 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
1987 | * by killing all belonging OOM-killable tasks. | |
1988 | * | |
1989 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
1990 | */ | |
1991 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
1992 | struct mem_cgroup *oom_domain) | |
1993 | { | |
1994 | struct mem_cgroup *oom_group = NULL; | |
1995 | struct mem_cgroup *memcg; | |
1996 | ||
1997 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
1998 | return NULL; | |
1999 | ||
2000 | if (!oom_domain) | |
2001 | oom_domain = root_mem_cgroup; | |
2002 | ||
2003 | rcu_read_lock(); | |
2004 | ||
2005 | memcg = mem_cgroup_from_task(victim); | |
2006 | if (memcg == root_mem_cgroup) | |
2007 | goto out; | |
2008 | ||
2009 | /* | |
2010 | * Traverse the memory cgroup hierarchy from the victim task's | |
2011 | * cgroup up to the OOMing cgroup (or root) to find the | |
2012 | * highest-level memory cgroup with oom.group set. | |
2013 | */ | |
2014 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
2015 | if (memcg->oom_group) | |
2016 | oom_group = memcg; | |
2017 | ||
2018 | if (memcg == oom_domain) | |
2019 | break; | |
2020 | } | |
2021 | ||
2022 | if (oom_group) | |
2023 | css_get(&oom_group->css); | |
2024 | out: | |
2025 | rcu_read_unlock(); | |
2026 | ||
2027 | return oom_group; | |
2028 | } | |
2029 | ||
2030 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2031 | { | |
2032 | pr_info("Tasks in "); | |
2033 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2034 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2035 | } | |
2036 | ||
d7365e78 | 2037 | /** |
81f8c3a4 JW |
2038 | * lock_page_memcg - lock a page->mem_cgroup binding |
2039 | * @page: the page | |
32047e2a | 2040 | * |
81f8c3a4 | 2041 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
2042 | * another cgroup. |
2043 | * | |
2044 | * It ensures lifetime of the returned memcg. Caller is responsible | |
2045 | * for the lifetime of the page; __unlock_page_memcg() is available | |
2046 | * when @page might get freed inside the locked section. | |
d69b042f | 2047 | */ |
739f79fc | 2048 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 KH |
2049 | { |
2050 | struct mem_cgroup *memcg; | |
6de22619 | 2051 | unsigned long flags; |
89c06bd5 | 2052 | |
6de22619 JW |
2053 | /* |
2054 | * The RCU lock is held throughout the transaction. The fast | |
2055 | * path can get away without acquiring the memcg->move_lock | |
2056 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
2057 | * |
2058 | * The RCU lock also protects the memcg from being freed when | |
2059 | * the page state that is going to change is the only thing | |
2060 | * preventing the page itself from being freed. E.g. writeback | |
2061 | * doesn't hold a page reference and relies on PG_writeback to | |
2062 | * keep off truncation, migration and so forth. | |
2063 | */ | |
d7365e78 JW |
2064 | rcu_read_lock(); |
2065 | ||
2066 | if (mem_cgroup_disabled()) | |
739f79fc | 2067 | return NULL; |
89c06bd5 | 2068 | again: |
1306a85a | 2069 | memcg = page->mem_cgroup; |
29833315 | 2070 | if (unlikely(!memcg)) |
739f79fc | 2071 | return NULL; |
d7365e78 | 2072 | |
bdcbb659 | 2073 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 2074 | return memcg; |
89c06bd5 | 2075 | |
6de22619 | 2076 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 2077 | if (memcg != page->mem_cgroup) { |
6de22619 | 2078 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2079 | goto again; |
2080 | } | |
6de22619 JW |
2081 | |
2082 | /* | |
2083 | * When charge migration first begins, we can have locked and | |
2084 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 2085 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
2086 | */ |
2087 | memcg->move_lock_task = current; | |
2088 | memcg->move_lock_flags = flags; | |
d7365e78 | 2089 | |
739f79fc | 2090 | return memcg; |
89c06bd5 | 2091 | } |
81f8c3a4 | 2092 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2093 | |
d7365e78 | 2094 | /** |
739f79fc JW |
2095 | * __unlock_page_memcg - unlock and unpin a memcg |
2096 | * @memcg: the memcg | |
2097 | * | |
2098 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2099 | */ |
739f79fc | 2100 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2101 | { |
6de22619 JW |
2102 | if (memcg && memcg->move_lock_task == current) { |
2103 | unsigned long flags = memcg->move_lock_flags; | |
2104 | ||
2105 | memcg->move_lock_task = NULL; | |
2106 | memcg->move_lock_flags = 0; | |
2107 | ||
2108 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2109 | } | |
89c06bd5 | 2110 | |
d7365e78 | 2111 | rcu_read_unlock(); |
89c06bd5 | 2112 | } |
739f79fc JW |
2113 | |
2114 | /** | |
2115 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
2116 | * @page: the page | |
2117 | */ | |
2118 | void unlock_page_memcg(struct page *page) | |
2119 | { | |
2120 | __unlock_page_memcg(page->mem_cgroup); | |
2121 | } | |
81f8c3a4 | 2122 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2123 | |
cdec2e42 KH |
2124 | struct memcg_stock_pcp { |
2125 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2126 | unsigned int nr_pages; |
cdec2e42 | 2127 | struct work_struct work; |
26fe6168 | 2128 | unsigned long flags; |
a0db00fc | 2129 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2130 | }; |
2131 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2132 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2133 | |
a0956d54 SS |
2134 | /** |
2135 | * consume_stock: Try to consume stocked charge on this cpu. | |
2136 | * @memcg: memcg to consume from. | |
2137 | * @nr_pages: how many pages to charge. | |
2138 | * | |
2139 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2140 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2141 | * service an allocation will refill the stock. | |
2142 | * | |
2143 | * returns true if successful, false otherwise. | |
cdec2e42 | 2144 | */ |
a0956d54 | 2145 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2146 | { |
2147 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2148 | unsigned long flags; |
3e32cb2e | 2149 | bool ret = false; |
cdec2e42 | 2150 | |
a983b5eb | 2151 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2152 | return ret; |
a0956d54 | 2153 | |
db2ba40c JW |
2154 | local_irq_save(flags); |
2155 | ||
2156 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2157 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2158 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2159 | ret = true; |
2160 | } | |
db2ba40c JW |
2161 | |
2162 | local_irq_restore(flags); | |
2163 | ||
cdec2e42 KH |
2164 | return ret; |
2165 | } | |
2166 | ||
2167 | /* | |
3e32cb2e | 2168 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2169 | */ |
2170 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2171 | { | |
2172 | struct mem_cgroup *old = stock->cached; | |
2173 | ||
11c9ea4e | 2174 | if (stock->nr_pages) { |
3e32cb2e | 2175 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2176 | if (do_memsw_account()) |
3e32cb2e | 2177 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2178 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2179 | stock->nr_pages = 0; |
cdec2e42 KH |
2180 | } |
2181 | stock->cached = NULL; | |
cdec2e42 KH |
2182 | } |
2183 | ||
cdec2e42 KH |
2184 | static void drain_local_stock(struct work_struct *dummy) |
2185 | { | |
db2ba40c JW |
2186 | struct memcg_stock_pcp *stock; |
2187 | unsigned long flags; | |
2188 | ||
72f0184c MH |
2189 | /* |
2190 | * The only protection from memory hotplug vs. drain_stock races is | |
2191 | * that we always operate on local CPU stock here with IRQ disabled | |
2192 | */ | |
db2ba40c JW |
2193 | local_irq_save(flags); |
2194 | ||
2195 | stock = this_cpu_ptr(&memcg_stock); | |
cdec2e42 | 2196 | drain_stock(stock); |
26fe6168 | 2197 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2198 | |
2199 | local_irq_restore(flags); | |
cdec2e42 KH |
2200 | } |
2201 | ||
2202 | /* | |
3e32cb2e | 2203 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2204 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2205 | */ |
c0ff4b85 | 2206 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2207 | { |
db2ba40c JW |
2208 | struct memcg_stock_pcp *stock; |
2209 | unsigned long flags; | |
2210 | ||
2211 | local_irq_save(flags); | |
cdec2e42 | 2212 | |
db2ba40c | 2213 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2214 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2215 | drain_stock(stock); |
c0ff4b85 | 2216 | stock->cached = memcg; |
cdec2e42 | 2217 | } |
11c9ea4e | 2218 | stock->nr_pages += nr_pages; |
db2ba40c | 2219 | |
a983b5eb | 2220 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2221 | drain_stock(stock); |
2222 | ||
db2ba40c | 2223 | local_irq_restore(flags); |
cdec2e42 KH |
2224 | } |
2225 | ||
2226 | /* | |
c0ff4b85 | 2227 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2228 | * of the hierarchy under it. |
cdec2e42 | 2229 | */ |
6d3d6aa2 | 2230 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2231 | { |
26fe6168 | 2232 | int cpu, curcpu; |
d38144b7 | 2233 | |
6d3d6aa2 JW |
2234 | /* If someone's already draining, avoid adding running more workers. */ |
2235 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2236 | return; | |
72f0184c MH |
2237 | /* |
2238 | * Notify other cpus that system-wide "drain" is running | |
2239 | * We do not care about races with the cpu hotplug because cpu down | |
2240 | * as well as workers from this path always operate on the local | |
2241 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2242 | */ | |
5af12d0e | 2243 | curcpu = get_cpu(); |
cdec2e42 KH |
2244 | for_each_online_cpu(cpu) { |
2245 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2246 | struct mem_cgroup *memcg; |
26fe6168 | 2247 | |
c0ff4b85 | 2248 | memcg = stock->cached; |
72f0184c | 2249 | if (!memcg || !stock->nr_pages || !css_tryget(&memcg->css)) |
26fe6168 | 2250 | continue; |
72f0184c MH |
2251 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) { |
2252 | css_put(&memcg->css); | |
3e92041d | 2253 | continue; |
72f0184c | 2254 | } |
d1a05b69 MH |
2255 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2256 | if (cpu == curcpu) | |
2257 | drain_local_stock(&stock->work); | |
2258 | else | |
2259 | schedule_work_on(cpu, &stock->work); | |
2260 | } | |
72f0184c | 2261 | css_put(&memcg->css); |
cdec2e42 | 2262 | } |
5af12d0e | 2263 | put_cpu(); |
9f50fad6 | 2264 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2265 | } |
2266 | ||
308167fc | 2267 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2268 | { |
cdec2e42 | 2269 | struct memcg_stock_pcp *stock; |
42a30035 | 2270 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2271 | |
cdec2e42 KH |
2272 | stock = &per_cpu(memcg_stock, cpu); |
2273 | drain_stock(stock); | |
a983b5eb JW |
2274 | |
2275 | for_each_mem_cgroup(memcg) { | |
2276 | int i; | |
2277 | ||
2278 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2279 | int nid; | |
2280 | long x; | |
2281 | ||
871789d4 | 2282 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
815744d7 | 2283 | if (x) |
42a30035 JW |
2284 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2285 | atomic_long_add(x, &memcg->vmstats[i]); | |
a983b5eb JW |
2286 | |
2287 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2288 | continue; | |
2289 | ||
2290 | for_each_node(nid) { | |
2291 | struct mem_cgroup_per_node *pn; | |
2292 | ||
2293 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2294 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
815744d7 | 2295 | if (x) |
42a30035 JW |
2296 | do { |
2297 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2298 | } while ((pn = parent_nodeinfo(pn, nid))); | |
a983b5eb JW |
2299 | } |
2300 | } | |
2301 | ||
e27be240 | 2302 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2303 | long x; |
2304 | ||
871789d4 | 2305 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
815744d7 | 2306 | if (x) |
42a30035 JW |
2307 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2308 | atomic_long_add(x, &memcg->vmevents[i]); | |
a983b5eb JW |
2309 | } |
2310 | } | |
2311 | ||
308167fc | 2312 | return 0; |
cdec2e42 KH |
2313 | } |
2314 | ||
f7e1cb6e JW |
2315 | static void reclaim_high(struct mem_cgroup *memcg, |
2316 | unsigned int nr_pages, | |
2317 | gfp_t gfp_mask) | |
2318 | { | |
2319 | do { | |
2320 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
2321 | continue; | |
e27be240 | 2322 | memcg_memory_event(memcg, MEMCG_HIGH); |
f7e1cb6e JW |
2323 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
2324 | } while ((memcg = parent_mem_cgroup(memcg))); | |
2325 | } | |
2326 | ||
2327 | static void high_work_func(struct work_struct *work) | |
2328 | { | |
2329 | struct mem_cgroup *memcg; | |
2330 | ||
2331 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2332 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2333 | } |
2334 | ||
b23afb93 TH |
2335 | /* |
2336 | * Scheduled by try_charge() to be executed from the userland return path | |
2337 | * and reclaims memory over the high limit. | |
2338 | */ | |
2339 | void mem_cgroup_handle_over_high(void) | |
2340 | { | |
2341 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 2342 | struct mem_cgroup *memcg; |
b23afb93 TH |
2343 | |
2344 | if (likely(!nr_pages)) | |
2345 | return; | |
2346 | ||
f7e1cb6e JW |
2347 | memcg = get_mem_cgroup_from_mm(current->mm); |
2348 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
2349 | css_put(&memcg->css); |
2350 | current->memcg_nr_pages_over_high = 0; | |
2351 | } | |
2352 | ||
00501b53 JW |
2353 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2354 | unsigned int nr_pages) | |
8a9f3ccd | 2355 | { |
a983b5eb | 2356 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
9b130619 | 2357 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2358 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2359 | struct page_counter *counter; |
6539cc05 | 2360 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2361 | bool may_swap = true; |
2362 | bool drained = false; | |
29ef680a | 2363 | enum oom_status oom_status; |
a636b327 | 2364 | |
ce00a967 | 2365 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2366 | return 0; |
6539cc05 | 2367 | retry: |
b6b6cc72 | 2368 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2369 | return 0; |
8a9f3ccd | 2370 | |
7941d214 | 2371 | if (!do_memsw_account() || |
6071ca52 JW |
2372 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2373 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2374 | goto done_restock; |
7941d214 | 2375 | if (do_memsw_account()) |
3e32cb2e JW |
2376 | page_counter_uncharge(&memcg->memsw, batch); |
2377 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2378 | } else { |
3e32cb2e | 2379 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2380 | may_swap = false; |
3fbe7244 | 2381 | } |
7a81b88c | 2382 | |
6539cc05 JW |
2383 | if (batch > nr_pages) { |
2384 | batch = nr_pages; | |
2385 | goto retry; | |
2386 | } | |
6d61ef40 | 2387 | |
06b078fc JW |
2388 | /* |
2389 | * Unlike in global OOM situations, memcg is not in a physical | |
2390 | * memory shortage. Allow dying and OOM-killed tasks to | |
2391 | * bypass the last charges so that they can exit quickly and | |
2392 | * free their memory. | |
2393 | */ | |
7775face | 2394 | if (unlikely(should_force_charge())) |
10d53c74 | 2395 | goto force; |
06b078fc | 2396 | |
89a28483 JW |
2397 | /* |
2398 | * Prevent unbounded recursion when reclaim operations need to | |
2399 | * allocate memory. This might exceed the limits temporarily, | |
2400 | * but we prefer facilitating memory reclaim and getting back | |
2401 | * under the limit over triggering OOM kills in these cases. | |
2402 | */ | |
2403 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2404 | goto force; | |
2405 | ||
06b078fc JW |
2406 | if (unlikely(task_in_memcg_oom(current))) |
2407 | goto nomem; | |
2408 | ||
d0164adc | 2409 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2410 | goto nomem; |
4b534334 | 2411 | |
e27be240 | 2412 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2413 | |
b70a2a21 JW |
2414 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2415 | gfp_mask, may_swap); | |
6539cc05 | 2416 | |
61e02c74 | 2417 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2418 | goto retry; |
28c34c29 | 2419 | |
b70a2a21 | 2420 | if (!drained) { |
6d3d6aa2 | 2421 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2422 | drained = true; |
2423 | goto retry; | |
2424 | } | |
2425 | ||
28c34c29 JW |
2426 | if (gfp_mask & __GFP_NORETRY) |
2427 | goto nomem; | |
6539cc05 JW |
2428 | /* |
2429 | * Even though the limit is exceeded at this point, reclaim | |
2430 | * may have been able to free some pages. Retry the charge | |
2431 | * before killing the task. | |
2432 | * | |
2433 | * Only for regular pages, though: huge pages are rather | |
2434 | * unlikely to succeed so close to the limit, and we fall back | |
2435 | * to regular pages anyway in case of failure. | |
2436 | */ | |
61e02c74 | 2437 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2438 | goto retry; |
2439 | /* | |
2440 | * At task move, charge accounts can be doubly counted. So, it's | |
2441 | * better to wait until the end of task_move if something is going on. | |
2442 | */ | |
2443 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2444 | goto retry; | |
2445 | ||
9b130619 JW |
2446 | if (nr_retries--) |
2447 | goto retry; | |
2448 | ||
38d38493 | 2449 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2450 | goto nomem; |
2451 | ||
06b078fc | 2452 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2453 | goto force; |
06b078fc | 2454 | |
6539cc05 | 2455 | if (fatal_signal_pending(current)) |
10d53c74 | 2456 | goto force; |
6539cc05 | 2457 | |
29ef680a MH |
2458 | /* |
2459 | * keep retrying as long as the memcg oom killer is able to make | |
2460 | * a forward progress or bypass the charge if the oom killer | |
2461 | * couldn't make any progress. | |
2462 | */ | |
2463 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2464 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2465 | switch (oom_status) { |
2466 | case OOM_SUCCESS: | |
2467 | nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
29ef680a MH |
2468 | goto retry; |
2469 | case OOM_FAILED: | |
2470 | goto force; | |
2471 | default: | |
2472 | goto nomem; | |
2473 | } | |
7a81b88c | 2474 | nomem: |
6d1fdc48 | 2475 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2476 | return -ENOMEM; |
10d53c74 TH |
2477 | force: |
2478 | /* | |
2479 | * The allocation either can't fail or will lead to more memory | |
2480 | * being freed very soon. Allow memory usage go over the limit | |
2481 | * temporarily by force charging it. | |
2482 | */ | |
2483 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2484 | if (do_memsw_account()) |
10d53c74 TH |
2485 | page_counter_charge(&memcg->memsw, nr_pages); |
2486 | css_get_many(&memcg->css, nr_pages); | |
2487 | ||
2488 | return 0; | |
6539cc05 JW |
2489 | |
2490 | done_restock: | |
e8ea14cc | 2491 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2492 | if (batch > nr_pages) |
2493 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2494 | |
241994ed | 2495 | /* |
b23afb93 TH |
2496 | * If the hierarchy is above the normal consumption range, schedule |
2497 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2498 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2499 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2500 | * not recorded as it most likely matches current's and won't | |
2501 | * change in the meantime. As high limit is checked again before | |
2502 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2503 | */ |
2504 | do { | |
b23afb93 | 2505 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2506 | /* Don't bother a random interrupted task */ |
2507 | if (in_interrupt()) { | |
2508 | schedule_work(&memcg->high_work); | |
2509 | break; | |
2510 | } | |
9516a18a | 2511 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2512 | set_notify_resume(current); |
2513 | break; | |
2514 | } | |
241994ed | 2515 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2516 | |
2517 | return 0; | |
7a81b88c | 2518 | } |
8a9f3ccd | 2519 | |
00501b53 | 2520 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2521 | { |
ce00a967 JW |
2522 | if (mem_cgroup_is_root(memcg)) |
2523 | return; | |
2524 | ||
3e32cb2e | 2525 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2526 | if (do_memsw_account()) |
3e32cb2e | 2527 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2528 | |
e8ea14cc | 2529 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2530 | } |
2531 | ||
0a31bc97 JW |
2532 | static void lock_page_lru(struct page *page, int *isolated) |
2533 | { | |
f4b7e272 | 2534 | pg_data_t *pgdat = page_pgdat(page); |
0a31bc97 | 2535 | |
f4b7e272 | 2536 | spin_lock_irq(&pgdat->lru_lock); |
0a31bc97 JW |
2537 | if (PageLRU(page)) { |
2538 | struct lruvec *lruvec; | |
2539 | ||
f4b7e272 | 2540 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
0a31bc97 JW |
2541 | ClearPageLRU(page); |
2542 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2543 | *isolated = 1; | |
2544 | } else | |
2545 | *isolated = 0; | |
2546 | } | |
2547 | ||
2548 | static void unlock_page_lru(struct page *page, int isolated) | |
2549 | { | |
f4b7e272 | 2550 | pg_data_t *pgdat = page_pgdat(page); |
0a31bc97 JW |
2551 | |
2552 | if (isolated) { | |
2553 | struct lruvec *lruvec; | |
2554 | ||
f4b7e272 | 2555 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
0a31bc97 JW |
2556 | VM_BUG_ON_PAGE(PageLRU(page), page); |
2557 | SetPageLRU(page); | |
2558 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2559 | } | |
f4b7e272 | 2560 | spin_unlock_irq(&pgdat->lru_lock); |
0a31bc97 JW |
2561 | } |
2562 | ||
00501b53 | 2563 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2564 | bool lrucare) |
7a81b88c | 2565 | { |
0a31bc97 | 2566 | int isolated; |
9ce70c02 | 2567 | |
1306a85a | 2568 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2569 | |
2570 | /* | |
2571 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2572 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2573 | */ | |
0a31bc97 JW |
2574 | if (lrucare) |
2575 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2576 | |
0a31bc97 JW |
2577 | /* |
2578 | * Nobody should be changing or seriously looking at | |
1306a85a | 2579 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2580 | * |
2581 | * - the page is uncharged | |
2582 | * | |
2583 | * - the page is off-LRU | |
2584 | * | |
2585 | * - an anonymous fault has exclusive page access, except for | |
2586 | * a locked page table | |
2587 | * | |
2588 | * - a page cache insertion, a swapin fault, or a migration | |
2589 | * have the page locked | |
2590 | */ | |
1306a85a | 2591 | page->mem_cgroup = memcg; |
9ce70c02 | 2592 | |
0a31bc97 JW |
2593 | if (lrucare) |
2594 | unlock_page_lru(page, isolated); | |
7a81b88c | 2595 | } |
66e1707b | 2596 | |
84c07d11 | 2597 | #ifdef CONFIG_MEMCG_KMEM |
f3bb3043 | 2598 | static int memcg_alloc_cache_id(void) |
55007d84 | 2599 | { |
f3bb3043 VD |
2600 | int id, size; |
2601 | int err; | |
2602 | ||
dbcf73e2 | 2603 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2604 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2605 | if (id < 0) | |
2606 | return id; | |
55007d84 | 2607 | |
dbcf73e2 | 2608 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2609 | return id; |
2610 | ||
2611 | /* | |
2612 | * There's no space for the new id in memcg_caches arrays, | |
2613 | * so we have to grow them. | |
2614 | */ | |
05257a1a | 2615 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2616 | |
2617 | size = 2 * (id + 1); | |
55007d84 GC |
2618 | if (size < MEMCG_CACHES_MIN_SIZE) |
2619 | size = MEMCG_CACHES_MIN_SIZE; | |
2620 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2621 | size = MEMCG_CACHES_MAX_SIZE; | |
2622 | ||
f3bb3043 | 2623 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2624 | if (!err) |
2625 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2626 | if (!err) |
2627 | memcg_nr_cache_ids = size; | |
2628 | ||
2629 | up_write(&memcg_cache_ids_sem); | |
2630 | ||
f3bb3043 | 2631 | if (err) { |
dbcf73e2 | 2632 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2633 | return err; |
2634 | } | |
2635 | return id; | |
2636 | } | |
2637 | ||
2638 | static void memcg_free_cache_id(int id) | |
2639 | { | |
dbcf73e2 | 2640 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2641 | } |
2642 | ||
d5b3cf71 | 2643 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2644 | struct mem_cgroup *memcg; |
2645 | struct kmem_cache *cachep; | |
2646 | struct work_struct work; | |
2647 | }; | |
2648 | ||
d5b3cf71 | 2649 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2650 | { |
d5b3cf71 VD |
2651 | struct memcg_kmem_cache_create_work *cw = |
2652 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2653 | struct mem_cgroup *memcg = cw->memcg; |
2654 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2655 | |
d5b3cf71 | 2656 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2657 | |
5722d094 | 2658 | css_put(&memcg->css); |
d7f25f8a GC |
2659 | kfree(cw); |
2660 | } | |
2661 | ||
2662 | /* | |
2663 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2664 | */ |
85cfb245 | 2665 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
d5b3cf71 | 2666 | struct kmem_cache *cachep) |
d7f25f8a | 2667 | { |
d5b3cf71 | 2668 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2669 | |
f0a3a24b RG |
2670 | if (!css_tryget_online(&memcg->css)) |
2671 | return; | |
2672 | ||
c892fd82 | 2673 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT | __GFP_NOWARN); |
8135be5a | 2674 | if (!cw) |
d7f25f8a | 2675 | return; |
8135be5a | 2676 | |
d7f25f8a GC |
2677 | cw->memcg = memcg; |
2678 | cw->cachep = cachep; | |
d5b3cf71 | 2679 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2680 | |
17cc4dfe | 2681 | queue_work(memcg_kmem_cache_wq, &cw->work); |
d7f25f8a GC |
2682 | } |
2683 | ||
45264778 VD |
2684 | static inline bool memcg_kmem_bypass(void) |
2685 | { | |
2686 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
2687 | return true; | |
2688 | return false; | |
2689 | } | |
2690 | ||
2691 | /** | |
2692 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation | |
2693 | * @cachep: the original global kmem cache | |
2694 | * | |
d7f25f8a GC |
2695 | * Return the kmem_cache we're supposed to use for a slab allocation. |
2696 | * We try to use the current memcg's version of the cache. | |
2697 | * | |
45264778 VD |
2698 | * If the cache does not exist yet, if we are the first user of it, we |
2699 | * create it asynchronously in a workqueue and let the current allocation | |
2700 | * go through with the original cache. | |
d7f25f8a | 2701 | * |
45264778 VD |
2702 | * This function takes a reference to the cache it returns to assure it |
2703 | * won't get destroyed while we are working with it. Once the caller is | |
2704 | * done with it, memcg_kmem_put_cache() must be called to release the | |
2705 | * reference. | |
d7f25f8a | 2706 | */ |
45264778 | 2707 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2708 | { |
2709 | struct mem_cgroup *memcg; | |
959c8963 | 2710 | struct kmem_cache *memcg_cachep; |
f0a3a24b | 2711 | struct memcg_cache_array *arr; |
2a4db7eb | 2712 | int kmemcg_id; |
d7f25f8a | 2713 | |
f7ce3190 | 2714 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2715 | |
45264778 | 2716 | if (memcg_kmem_bypass()) |
230e9fc2 VD |
2717 | return cachep; |
2718 | ||
f0a3a24b RG |
2719 | rcu_read_lock(); |
2720 | ||
2721 | if (unlikely(current->active_memcg)) | |
2722 | memcg = current->active_memcg; | |
2723 | else | |
2724 | memcg = mem_cgroup_from_task(current); | |
2725 | ||
2726 | if (!memcg || memcg == root_mem_cgroup) | |
2727 | goto out_unlock; | |
2728 | ||
4db0c3c2 | 2729 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2730 | if (kmemcg_id < 0) |
f0a3a24b | 2731 | goto out_unlock; |
d7f25f8a | 2732 | |
f0a3a24b RG |
2733 | arr = rcu_dereference(cachep->memcg_params.memcg_caches); |
2734 | ||
2735 | /* | |
2736 | * Make sure we will access the up-to-date value. The code updating | |
2737 | * memcg_caches issues a write barrier to match the data dependency | |
2738 | * barrier inside READ_ONCE() (see memcg_create_kmem_cache()). | |
2739 | */ | |
2740 | memcg_cachep = READ_ONCE(arr->entries[kmemcg_id]); | |
ca0dde97 LZ |
2741 | |
2742 | /* | |
2743 | * If we are in a safe context (can wait, and not in interrupt | |
2744 | * context), we could be be predictable and return right away. | |
2745 | * This would guarantee that the allocation being performed | |
2746 | * already belongs in the new cache. | |
2747 | * | |
2748 | * However, there are some clashes that can arrive from locking. | |
2749 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2750 | * memcg_create_kmem_cache, this means no further allocation |
2751 | * could happen with the slab_mutex held. So it's better to | |
2752 | * defer everything. | |
f0a3a24b RG |
2753 | * |
2754 | * If the memcg is dying or memcg_cache is about to be released, | |
2755 | * don't bother creating new kmem_caches. Because memcg_cachep | |
2756 | * is ZEROed as the fist step of kmem offlining, we don't need | |
2757 | * percpu_ref_tryget_live() here. css_tryget_online() check in | |
2758 | * memcg_schedule_kmem_cache_create() will prevent us from | |
2759 | * creation of a new kmem_cache. | |
ca0dde97 | 2760 | */ |
f0a3a24b RG |
2761 | if (unlikely(!memcg_cachep)) |
2762 | memcg_schedule_kmem_cache_create(memcg, cachep); | |
2763 | else if (percpu_ref_tryget(&memcg_cachep->memcg_params.refcnt)) | |
2764 | cachep = memcg_cachep; | |
2765 | out_unlock: | |
2766 | rcu_read_unlock(); | |
ca0dde97 | 2767 | return cachep; |
d7f25f8a | 2768 | } |
d7f25f8a | 2769 | |
45264778 VD |
2770 | /** |
2771 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache | |
2772 | * @cachep: the cache returned by memcg_kmem_get_cache | |
2773 | */ | |
2774 | void memcg_kmem_put_cache(struct kmem_cache *cachep) | |
8135be5a VD |
2775 | { |
2776 | if (!is_root_cache(cachep)) | |
f0a3a24b | 2777 | percpu_ref_put(&cachep->memcg_params.refcnt); |
8135be5a VD |
2778 | } |
2779 | ||
45264778 | 2780 | /** |
60cd4bcd | 2781 | * __memcg_kmem_charge_memcg: charge a kmem page |
45264778 VD |
2782 | * @page: page to charge |
2783 | * @gfp: reclaim mode | |
2784 | * @order: allocation order | |
2785 | * @memcg: memory cgroup to charge | |
2786 | * | |
2787 | * Returns 0 on success, an error code on failure. | |
2788 | */ | |
60cd4bcd | 2789 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
45264778 | 2790 | struct mem_cgroup *memcg) |
7ae1e1d0 | 2791 | { |
f3ccb2c4 VD |
2792 | unsigned int nr_pages = 1 << order; |
2793 | struct page_counter *counter; | |
7ae1e1d0 GC |
2794 | int ret; |
2795 | ||
f3ccb2c4 | 2796 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2797 | if (ret) |
f3ccb2c4 | 2798 | return ret; |
52c29b04 JW |
2799 | |
2800 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2801 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2802 | cancel_charge(memcg, nr_pages); | |
2803 | return -ENOMEM; | |
7ae1e1d0 GC |
2804 | } |
2805 | ||
f3ccb2c4 | 2806 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2807 | |
f3ccb2c4 | 2808 | return 0; |
7ae1e1d0 GC |
2809 | } |
2810 | ||
45264778 | 2811 | /** |
60cd4bcd | 2812 | * __memcg_kmem_charge: charge a kmem page to the current memory cgroup |
45264778 VD |
2813 | * @page: page to charge |
2814 | * @gfp: reclaim mode | |
2815 | * @order: allocation order | |
2816 | * | |
2817 | * Returns 0 on success, an error code on failure. | |
2818 | */ | |
60cd4bcd | 2819 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2820 | { |
f3ccb2c4 | 2821 | struct mem_cgroup *memcg; |
fcff7d7e | 2822 | int ret = 0; |
7ae1e1d0 | 2823 | |
60cd4bcd | 2824 | if (memcg_kmem_bypass()) |
45264778 VD |
2825 | return 0; |
2826 | ||
d46eb14b | 2827 | memcg = get_mem_cgroup_from_current(); |
c4159a75 | 2828 | if (!mem_cgroup_is_root(memcg)) { |
60cd4bcd | 2829 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); |
c4159a75 VD |
2830 | if (!ret) |
2831 | __SetPageKmemcg(page); | |
2832 | } | |
7ae1e1d0 | 2833 | css_put(&memcg->css); |
d05e83a6 | 2834 | return ret; |
7ae1e1d0 | 2835 | } |
49a18eae RG |
2836 | |
2837 | /** | |
2838 | * __memcg_kmem_uncharge_memcg: uncharge a kmem page | |
2839 | * @memcg: memcg to uncharge | |
2840 | * @nr_pages: number of pages to uncharge | |
2841 | */ | |
2842 | void __memcg_kmem_uncharge_memcg(struct mem_cgroup *memcg, | |
2843 | unsigned int nr_pages) | |
2844 | { | |
2845 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2846 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2847 | ||
2848 | page_counter_uncharge(&memcg->memory, nr_pages); | |
2849 | if (do_memsw_account()) | |
2850 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
2851 | } | |
45264778 | 2852 | /** |
60cd4bcd | 2853 | * __memcg_kmem_uncharge: uncharge a kmem page |
45264778 VD |
2854 | * @page: page to uncharge |
2855 | * @order: allocation order | |
2856 | */ | |
60cd4bcd | 2857 | void __memcg_kmem_uncharge(struct page *page, int order) |
7ae1e1d0 | 2858 | { |
1306a85a | 2859 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2860 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2861 | |
7ae1e1d0 GC |
2862 | if (!memcg) |
2863 | return; | |
2864 | ||
309381fe | 2865 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
49a18eae | 2866 | __memcg_kmem_uncharge_memcg(memcg, nr_pages); |
1306a85a | 2867 | page->mem_cgroup = NULL; |
c4159a75 VD |
2868 | |
2869 | /* slab pages do not have PageKmemcg flag set */ | |
2870 | if (PageKmemcg(page)) | |
2871 | __ClearPageKmemcg(page); | |
2872 | ||
f3ccb2c4 | 2873 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2874 | } |
84c07d11 | 2875 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 2876 | |
ca3e0214 KH |
2877 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2878 | ||
ca3e0214 KH |
2879 | /* |
2880 | * Because tail pages are not marked as "used", set it. We're under | |
f4b7e272 | 2881 | * pgdat->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2882 | */ |
e94c8a9c | 2883 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2884 | { |
e94c8a9c | 2885 | int i; |
ca3e0214 | 2886 | |
3d37c4a9 KH |
2887 | if (mem_cgroup_disabled()) |
2888 | return; | |
b070e65c | 2889 | |
29833315 | 2890 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2891 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2892 | |
c9019e9b | 2893 | __mod_memcg_state(head->mem_cgroup, MEMCG_RSS_HUGE, -HPAGE_PMD_NR); |
ca3e0214 | 2894 | } |
12d27107 | 2895 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2896 | |
c255a458 | 2897 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
2898 | /** |
2899 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2900 | * @entry: swap entry to be moved | |
2901 | * @from: mem_cgroup which the entry is moved from | |
2902 | * @to: mem_cgroup which the entry is moved to | |
2903 | * | |
2904 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2905 | * as the mem_cgroup's id of @from. | |
2906 | * | |
2907 | * Returns 0 on success, -EINVAL on failure. | |
2908 | * | |
3e32cb2e | 2909 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2910 | * both res and memsw, and called css_get(). |
2911 | */ | |
2912 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2913 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2914 | { |
2915 | unsigned short old_id, new_id; | |
2916 | ||
34c00c31 LZ |
2917 | old_id = mem_cgroup_id(from); |
2918 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2919 | |
2920 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
2921 | mod_memcg_state(from, MEMCG_SWAP, -1); |
2922 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
2923 | return 0; |
2924 | } | |
2925 | return -EINVAL; | |
2926 | } | |
2927 | #else | |
2928 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2929 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2930 | { |
2931 | return -EINVAL; | |
2932 | } | |
8c7c6e34 | 2933 | #endif |
d13d1443 | 2934 | |
bbec2e15 | 2935 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 2936 | |
bbec2e15 RG |
2937 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
2938 | unsigned long max, bool memsw) | |
628f4235 | 2939 | { |
3e32cb2e | 2940 | bool enlarge = false; |
bb4a7ea2 | 2941 | bool drained = false; |
3e32cb2e | 2942 | int ret; |
c054a78c YZ |
2943 | bool limits_invariant; |
2944 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 2945 | |
3e32cb2e | 2946 | do { |
628f4235 KH |
2947 | if (signal_pending(current)) { |
2948 | ret = -EINTR; | |
2949 | break; | |
2950 | } | |
3e32cb2e | 2951 | |
bbec2e15 | 2952 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
2953 | /* |
2954 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 2955 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 2956 | */ |
bbec2e15 RG |
2957 | limits_invariant = memsw ? max >= memcg->memory.max : |
2958 | max <= memcg->memsw.max; | |
c054a78c | 2959 | if (!limits_invariant) { |
bbec2e15 | 2960 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 2961 | ret = -EINVAL; |
8c7c6e34 KH |
2962 | break; |
2963 | } | |
bbec2e15 | 2964 | if (max > counter->max) |
3e32cb2e | 2965 | enlarge = true; |
bbec2e15 RG |
2966 | ret = page_counter_set_max(counter, max); |
2967 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
2968 | |
2969 | if (!ret) | |
2970 | break; | |
2971 | ||
bb4a7ea2 SB |
2972 | if (!drained) { |
2973 | drain_all_stock(memcg); | |
2974 | drained = true; | |
2975 | continue; | |
2976 | } | |
2977 | ||
1ab5c056 AR |
2978 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
2979 | GFP_KERNEL, !memsw)) { | |
2980 | ret = -EBUSY; | |
2981 | break; | |
2982 | } | |
2983 | } while (true); | |
3e32cb2e | 2984 | |
3c11ecf4 KH |
2985 | if (!ret && enlarge) |
2986 | memcg_oom_recover(memcg); | |
3e32cb2e | 2987 | |
628f4235 KH |
2988 | return ret; |
2989 | } | |
2990 | ||
ef8f2327 | 2991 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
2992 | gfp_t gfp_mask, |
2993 | unsigned long *total_scanned) | |
2994 | { | |
2995 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 2996 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
2997 | unsigned long reclaimed; |
2998 | int loop = 0; | |
ef8f2327 | 2999 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3000 | unsigned long excess; |
0608f43d AM |
3001 | unsigned long nr_scanned; |
3002 | ||
3003 | if (order > 0) | |
3004 | return 0; | |
3005 | ||
ef8f2327 | 3006 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
3007 | |
3008 | /* | |
3009 | * Do not even bother to check the largest node if the root | |
3010 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3011 | * are acceptable as soft limit is best effort anyway. | |
3012 | */ | |
bfc7228b | 3013 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3014 | return 0; |
3015 | ||
0608f43d AM |
3016 | /* |
3017 | * This loop can run a while, specially if mem_cgroup's continuously | |
3018 | * keep exceeding their soft limit and putting the system under | |
3019 | * pressure | |
3020 | */ | |
3021 | do { | |
3022 | if (next_mz) | |
3023 | mz = next_mz; | |
3024 | else | |
3025 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3026 | if (!mz) | |
3027 | break; | |
3028 | ||
3029 | nr_scanned = 0; | |
ef8f2327 | 3030 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
3031 | gfp_mask, &nr_scanned); |
3032 | nr_reclaimed += reclaimed; | |
3033 | *total_scanned += nr_scanned; | |
0a31bc97 | 3034 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3035 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3036 | |
3037 | /* | |
3038 | * If we failed to reclaim anything from this memory cgroup | |
3039 | * it is time to move on to the next cgroup | |
3040 | */ | |
3041 | next_mz = NULL; | |
bc2f2e7f VD |
3042 | if (!reclaimed) |
3043 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3044 | ||
3e32cb2e | 3045 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3046 | /* |
3047 | * One school of thought says that we should not add | |
3048 | * back the node to the tree if reclaim returns 0. | |
3049 | * But our reclaim could return 0, simply because due | |
3050 | * to priority we are exposing a smaller subset of | |
3051 | * memory to reclaim from. Consider this as a longer | |
3052 | * term TODO. | |
3053 | */ | |
3054 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3055 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3056 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3057 | css_put(&mz->memcg->css); |
3058 | loop++; | |
3059 | /* | |
3060 | * Could not reclaim anything and there are no more | |
3061 | * mem cgroups to try or we seem to be looping without | |
3062 | * reclaiming anything. | |
3063 | */ | |
3064 | if (!nr_reclaimed && | |
3065 | (next_mz == NULL || | |
3066 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3067 | break; | |
3068 | } while (!nr_reclaimed); | |
3069 | if (next_mz) | |
3070 | css_put(&next_mz->memcg->css); | |
3071 | return nr_reclaimed; | |
3072 | } | |
3073 | ||
ea280e7b TH |
3074 | /* |
3075 | * Test whether @memcg has children, dead or alive. Note that this | |
3076 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3077 | * returns %true if there are child csses according to the cgroup | |
3078 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3079 | */ | |
b5f99b53 GC |
3080 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3081 | { | |
ea280e7b TH |
3082 | bool ret; |
3083 | ||
ea280e7b TH |
3084 | rcu_read_lock(); |
3085 | ret = css_next_child(NULL, &memcg->css); | |
3086 | rcu_read_unlock(); | |
3087 | return ret; | |
b5f99b53 GC |
3088 | } |
3089 | ||
c26251f9 | 3090 | /* |
51038171 | 3091 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3092 | * |
3093 | * Caller is responsible for holding css reference for memcg. | |
3094 | */ | |
3095 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3096 | { | |
3097 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3098 | |
c1e862c1 KH |
3099 | /* we call try-to-free pages for make this cgroup empty */ |
3100 | lru_add_drain_all(); | |
d12c60f6 JS |
3101 | |
3102 | drain_all_stock(memcg); | |
3103 | ||
f817ed48 | 3104 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3105 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3106 | int progress; |
c1e862c1 | 3107 | |
c26251f9 MH |
3108 | if (signal_pending(current)) |
3109 | return -EINTR; | |
3110 | ||
b70a2a21 JW |
3111 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3112 | GFP_KERNEL, true); | |
c1e862c1 | 3113 | if (!progress) { |
f817ed48 | 3114 | nr_retries--; |
c1e862c1 | 3115 | /* maybe some writeback is necessary */ |
8aa7e847 | 3116 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3117 | } |
f817ed48 KH |
3118 | |
3119 | } | |
ab5196c2 MH |
3120 | |
3121 | return 0; | |
cc847582 KH |
3122 | } |
3123 | ||
6770c64e TH |
3124 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3125 | char *buf, size_t nbytes, | |
3126 | loff_t off) | |
c1e862c1 | 3127 | { |
6770c64e | 3128 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3129 | |
d8423011 MH |
3130 | if (mem_cgroup_is_root(memcg)) |
3131 | return -EINVAL; | |
6770c64e | 3132 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3133 | } |
3134 | ||
182446d0 TH |
3135 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3136 | struct cftype *cft) | |
18f59ea7 | 3137 | { |
182446d0 | 3138 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3139 | } |
3140 | ||
182446d0 TH |
3141 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3142 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3143 | { |
3144 | int retval = 0; | |
182446d0 | 3145 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3146 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3147 | |
567fb435 | 3148 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 3149 | return 0; |
567fb435 | 3150 | |
18f59ea7 | 3151 | /* |
af901ca1 | 3152 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3153 | * in the child subtrees. If it is unset, then the change can |
3154 | * occur, provided the current cgroup has no children. | |
3155 | * | |
3156 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3157 | * set if there are no children. | |
3158 | */ | |
c0ff4b85 | 3159 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3160 | (val == 1 || val == 0)) { |
ea280e7b | 3161 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3162 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3163 | else |
3164 | retval = -EBUSY; | |
3165 | } else | |
3166 | retval = -EINVAL; | |
567fb435 | 3167 | |
18f59ea7 BS |
3168 | return retval; |
3169 | } | |
3170 | ||
6f646156 | 3171 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3172 | { |
42a30035 | 3173 | unsigned long val; |
ce00a967 | 3174 | |
3e32cb2e | 3175 | if (mem_cgroup_is_root(memcg)) { |
42a30035 JW |
3176 | val = memcg_page_state(memcg, MEMCG_CACHE) + |
3177 | memcg_page_state(memcg, MEMCG_RSS); | |
3178 | if (swap) | |
3179 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3180 | } else { |
ce00a967 | 3181 | if (!swap) |
3e32cb2e | 3182 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3183 | else |
3e32cb2e | 3184 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3185 | } |
c12176d3 | 3186 | return val; |
ce00a967 JW |
3187 | } |
3188 | ||
3e32cb2e JW |
3189 | enum { |
3190 | RES_USAGE, | |
3191 | RES_LIMIT, | |
3192 | RES_MAX_USAGE, | |
3193 | RES_FAILCNT, | |
3194 | RES_SOFT_LIMIT, | |
3195 | }; | |
ce00a967 | 3196 | |
791badbd | 3197 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3198 | struct cftype *cft) |
8cdea7c0 | 3199 | { |
182446d0 | 3200 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3201 | struct page_counter *counter; |
af36f906 | 3202 | |
3e32cb2e | 3203 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3204 | case _MEM: |
3e32cb2e JW |
3205 | counter = &memcg->memory; |
3206 | break; | |
8c7c6e34 | 3207 | case _MEMSWAP: |
3e32cb2e JW |
3208 | counter = &memcg->memsw; |
3209 | break; | |
510fc4e1 | 3210 | case _KMEM: |
3e32cb2e | 3211 | counter = &memcg->kmem; |
510fc4e1 | 3212 | break; |
d55f90bf | 3213 | case _TCP: |
0db15298 | 3214 | counter = &memcg->tcpmem; |
d55f90bf | 3215 | break; |
8c7c6e34 KH |
3216 | default: |
3217 | BUG(); | |
8c7c6e34 | 3218 | } |
3e32cb2e JW |
3219 | |
3220 | switch (MEMFILE_ATTR(cft->private)) { | |
3221 | case RES_USAGE: | |
3222 | if (counter == &memcg->memory) | |
c12176d3 | 3223 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3224 | if (counter == &memcg->memsw) |
c12176d3 | 3225 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3226 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3227 | case RES_LIMIT: | |
bbec2e15 | 3228 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3229 | case RES_MAX_USAGE: |
3230 | return (u64)counter->watermark * PAGE_SIZE; | |
3231 | case RES_FAILCNT: | |
3232 | return counter->failcnt; | |
3233 | case RES_SOFT_LIMIT: | |
3234 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3235 | default: | |
3236 | BUG(); | |
3237 | } | |
8cdea7c0 | 3238 | } |
510fc4e1 | 3239 | |
84c07d11 | 3240 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3241 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3242 | { |
d6441637 VD |
3243 | int memcg_id; |
3244 | ||
b313aeee VD |
3245 | if (cgroup_memory_nokmem) |
3246 | return 0; | |
3247 | ||
2a4db7eb | 3248 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3249 | BUG_ON(memcg->kmem_state); |
d6441637 | 3250 | |
f3bb3043 | 3251 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3252 | if (memcg_id < 0) |
3253 | return memcg_id; | |
d6441637 | 3254 | |
ef12947c | 3255 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 3256 | /* |
567e9ab2 | 3257 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 3258 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
3259 | * guarantee no one starts accounting before all call sites are |
3260 | * patched. | |
3261 | */ | |
900a38f0 | 3262 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3263 | memcg->kmem_state = KMEM_ONLINE; |
bc2791f8 | 3264 | INIT_LIST_HEAD(&memcg->kmem_caches); |
0b8f73e1 JW |
3265 | |
3266 | return 0; | |
d6441637 VD |
3267 | } |
3268 | ||
8e0a8912 JW |
3269 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3270 | { | |
3271 | struct cgroup_subsys_state *css; | |
3272 | struct mem_cgroup *parent, *child; | |
3273 | int kmemcg_id; | |
3274 | ||
3275 | if (memcg->kmem_state != KMEM_ONLINE) | |
3276 | return; | |
3277 | /* | |
3278 | * Clear the online state before clearing memcg_caches array | |
3279 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
3280 | * guarantees that no cache will be created for this cgroup | |
3281 | * after we are done (see memcg_create_kmem_cache()). | |
3282 | */ | |
3283 | memcg->kmem_state = KMEM_ALLOCATED; | |
3284 | ||
3285 | memcg_deactivate_kmem_caches(memcg); | |
3286 | ||
3287 | kmemcg_id = memcg->kmemcg_id; | |
3288 | BUG_ON(kmemcg_id < 0); | |
3289 | ||
3290 | parent = parent_mem_cgroup(memcg); | |
3291 | if (!parent) | |
3292 | parent = root_mem_cgroup; | |
3293 | ||
3294 | /* | |
3295 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3296 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3297 | * to ones of the parent. After we have finished, all list_lrus | |
3298 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3299 | * ordering is imposed by list_lru_node->lock taken by | |
3300 | * memcg_drain_all_list_lrus(). | |
3301 | */ | |
3a06bb78 | 3302 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3303 | css_for_each_descendant_pre(css, &memcg->css) { |
3304 | child = mem_cgroup_from_css(css); | |
3305 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3306 | child->kmemcg_id = parent->kmemcg_id; | |
3307 | if (!memcg->use_hierarchy) | |
3308 | break; | |
3309 | } | |
3a06bb78 TH |
3310 | rcu_read_unlock(); |
3311 | ||
9bec5c35 | 3312 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3313 | |
3314 | memcg_free_cache_id(kmemcg_id); | |
3315 | } | |
3316 | ||
3317 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3318 | { | |
0b8f73e1 JW |
3319 | /* css_alloc() failed, offlining didn't happen */ |
3320 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3321 | memcg_offline_kmem(memcg); | |
3322 | ||
8e0a8912 | 3323 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
f0a3a24b | 3324 | WARN_ON(!list_empty(&memcg->kmem_caches)); |
8e0a8912 JW |
3325 | static_branch_dec(&memcg_kmem_enabled_key); |
3326 | WARN_ON(page_counter_read(&memcg->kmem)); | |
3327 | } | |
8e0a8912 | 3328 | } |
d6441637 | 3329 | #else |
0b8f73e1 | 3330 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3331 | { |
3332 | return 0; | |
3333 | } | |
3334 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3335 | { | |
3336 | } | |
3337 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3338 | { | |
3339 | } | |
84c07d11 | 3340 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3341 | |
bbec2e15 RG |
3342 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3343 | unsigned long max) | |
d6441637 | 3344 | { |
b313aeee | 3345 | int ret; |
127424c8 | 3346 | |
bbec2e15 RG |
3347 | mutex_lock(&memcg_max_mutex); |
3348 | ret = page_counter_set_max(&memcg->kmem, max); | |
3349 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3350 | return ret; |
d6441637 | 3351 | } |
510fc4e1 | 3352 | |
bbec2e15 | 3353 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3354 | { |
3355 | int ret; | |
3356 | ||
bbec2e15 | 3357 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3358 | |
bbec2e15 | 3359 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3360 | if (ret) |
3361 | goto out; | |
3362 | ||
0db15298 | 3363 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3364 | /* |
3365 | * The active flag needs to be written after the static_key | |
3366 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3367 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3368 | * for details, and note that we don't mark any socket as | |
3369 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3370 | * |
3371 | * We need to do this, because static_keys will span multiple | |
3372 | * sites, but we can't control their order. If we mark a socket | |
3373 | * as accounted, but the accounting functions are not patched in | |
3374 | * yet, we'll lose accounting. | |
3375 | * | |
2d758073 | 3376 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3377 | * because when this value change, the code to process it is not |
3378 | * patched in yet. | |
3379 | */ | |
3380 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3381 | memcg->tcpmem_active = true; |
d55f90bf VD |
3382 | } |
3383 | out: | |
bbec2e15 | 3384 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3385 | return ret; |
3386 | } | |
d55f90bf | 3387 | |
628f4235 KH |
3388 | /* |
3389 | * The user of this function is... | |
3390 | * RES_LIMIT. | |
3391 | */ | |
451af504 TH |
3392 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3393 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3394 | { |
451af504 | 3395 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3396 | unsigned long nr_pages; |
628f4235 KH |
3397 | int ret; |
3398 | ||
451af504 | 3399 | buf = strstrip(buf); |
650c5e56 | 3400 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3401 | if (ret) |
3402 | return ret; | |
af36f906 | 3403 | |
3e32cb2e | 3404 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3405 | case RES_LIMIT: |
4b3bde4c BS |
3406 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3407 | ret = -EINVAL; | |
3408 | break; | |
3409 | } | |
3e32cb2e JW |
3410 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3411 | case _MEM: | |
bbec2e15 | 3412 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3413 | break; |
3e32cb2e | 3414 | case _MEMSWAP: |
bbec2e15 | 3415 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3416 | break; |
3e32cb2e | 3417 | case _KMEM: |
bbec2e15 | 3418 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3419 | break; |
d55f90bf | 3420 | case _TCP: |
bbec2e15 | 3421 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3422 | break; |
3e32cb2e | 3423 | } |
296c81d8 | 3424 | break; |
3e32cb2e JW |
3425 | case RES_SOFT_LIMIT: |
3426 | memcg->soft_limit = nr_pages; | |
3427 | ret = 0; | |
628f4235 KH |
3428 | break; |
3429 | } | |
451af504 | 3430 | return ret ?: nbytes; |
8cdea7c0 BS |
3431 | } |
3432 | ||
6770c64e TH |
3433 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3434 | size_t nbytes, loff_t off) | |
c84872e1 | 3435 | { |
6770c64e | 3436 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3437 | struct page_counter *counter; |
c84872e1 | 3438 | |
3e32cb2e JW |
3439 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3440 | case _MEM: | |
3441 | counter = &memcg->memory; | |
3442 | break; | |
3443 | case _MEMSWAP: | |
3444 | counter = &memcg->memsw; | |
3445 | break; | |
3446 | case _KMEM: | |
3447 | counter = &memcg->kmem; | |
3448 | break; | |
d55f90bf | 3449 | case _TCP: |
0db15298 | 3450 | counter = &memcg->tcpmem; |
d55f90bf | 3451 | break; |
3e32cb2e JW |
3452 | default: |
3453 | BUG(); | |
3454 | } | |
af36f906 | 3455 | |
3e32cb2e | 3456 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3457 | case RES_MAX_USAGE: |
3e32cb2e | 3458 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3459 | break; |
3460 | case RES_FAILCNT: | |
3e32cb2e | 3461 | counter->failcnt = 0; |
29f2a4da | 3462 | break; |
3e32cb2e JW |
3463 | default: |
3464 | BUG(); | |
29f2a4da | 3465 | } |
f64c3f54 | 3466 | |
6770c64e | 3467 | return nbytes; |
c84872e1 PE |
3468 | } |
3469 | ||
182446d0 | 3470 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3471 | struct cftype *cft) |
3472 | { | |
182446d0 | 3473 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3474 | } |
3475 | ||
02491447 | 3476 | #ifdef CONFIG_MMU |
182446d0 | 3477 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3478 | struct cftype *cft, u64 val) |
3479 | { | |
182446d0 | 3480 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3481 | |
1dfab5ab | 3482 | if (val & ~MOVE_MASK) |
7dc74be0 | 3483 | return -EINVAL; |
ee5e8472 | 3484 | |
7dc74be0 | 3485 | /* |
ee5e8472 GC |
3486 | * No kind of locking is needed in here, because ->can_attach() will |
3487 | * check this value once in the beginning of the process, and then carry | |
3488 | * on with stale data. This means that changes to this value will only | |
3489 | * affect task migrations starting after the change. | |
7dc74be0 | 3490 | */ |
c0ff4b85 | 3491 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3492 | return 0; |
3493 | } | |
02491447 | 3494 | #else |
182446d0 | 3495 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3496 | struct cftype *cft, u64 val) |
3497 | { | |
3498 | return -ENOSYS; | |
3499 | } | |
3500 | #endif | |
7dc74be0 | 3501 | |
406eb0c9 | 3502 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3503 | |
3504 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3505 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3506 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3507 | ||
3508 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
3509 | int nid, unsigned int lru_mask) | |
3510 | { | |
3511 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); | |
3512 | unsigned long nr = 0; | |
3513 | enum lru_list lru; | |
3514 | ||
3515 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3516 | ||
3517 | for_each_lru(lru) { | |
3518 | if (!(BIT(lru) & lru_mask)) | |
3519 | continue; | |
205b20cc | 3520 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); |
113b7dfd JW |
3521 | } |
3522 | return nr; | |
3523 | } | |
3524 | ||
3525 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
3526 | unsigned int lru_mask) | |
3527 | { | |
3528 | unsigned long nr = 0; | |
3529 | enum lru_list lru; | |
3530 | ||
3531 | for_each_lru(lru) { | |
3532 | if (!(BIT(lru) & lru_mask)) | |
3533 | continue; | |
205b20cc | 3534 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); |
113b7dfd JW |
3535 | } |
3536 | return nr; | |
3537 | } | |
3538 | ||
2da8ca82 | 3539 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3540 | { |
25485de6 GT |
3541 | struct numa_stat { |
3542 | const char *name; | |
3543 | unsigned int lru_mask; | |
3544 | }; | |
3545 | ||
3546 | static const struct numa_stat stats[] = { | |
3547 | { "total", LRU_ALL }, | |
3548 | { "file", LRU_ALL_FILE }, | |
3549 | { "anon", LRU_ALL_ANON }, | |
3550 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3551 | }; | |
3552 | const struct numa_stat *stat; | |
406eb0c9 | 3553 | int nid; |
25485de6 | 3554 | unsigned long nr; |
aa9694bb | 3555 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3556 | |
25485de6 GT |
3557 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3558 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3559 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3560 | for_each_node_state(nid, N_MEMORY) { | |
3561 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3562 | stat->lru_mask); | |
3563 | seq_printf(m, " N%d=%lu", nid, nr); | |
3564 | } | |
3565 | seq_putc(m, '\n'); | |
406eb0c9 | 3566 | } |
406eb0c9 | 3567 | |
071aee13 YH |
3568 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3569 | struct mem_cgroup *iter; | |
3570 | ||
3571 | nr = 0; | |
3572 | for_each_mem_cgroup_tree(iter, memcg) | |
3573 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3574 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3575 | for_each_node_state(nid, N_MEMORY) { | |
3576 | nr = 0; | |
3577 | for_each_mem_cgroup_tree(iter, memcg) | |
3578 | nr += mem_cgroup_node_nr_lru_pages( | |
3579 | iter, nid, stat->lru_mask); | |
3580 | seq_printf(m, " N%d=%lu", nid, nr); | |
3581 | } | |
3582 | seq_putc(m, '\n'); | |
406eb0c9 | 3583 | } |
406eb0c9 | 3584 | |
406eb0c9 YH |
3585 | return 0; |
3586 | } | |
3587 | #endif /* CONFIG_NUMA */ | |
3588 | ||
c8713d0b JW |
3589 | static const unsigned int memcg1_stats[] = { |
3590 | MEMCG_CACHE, | |
3591 | MEMCG_RSS, | |
3592 | MEMCG_RSS_HUGE, | |
3593 | NR_SHMEM, | |
3594 | NR_FILE_MAPPED, | |
3595 | NR_FILE_DIRTY, | |
3596 | NR_WRITEBACK, | |
3597 | MEMCG_SWAP, | |
3598 | }; | |
3599 | ||
3600 | static const char *const memcg1_stat_names[] = { | |
3601 | "cache", | |
3602 | "rss", | |
3603 | "rss_huge", | |
3604 | "shmem", | |
3605 | "mapped_file", | |
3606 | "dirty", | |
3607 | "writeback", | |
3608 | "swap", | |
3609 | }; | |
3610 | ||
df0e53d0 | 3611 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3612 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
3613 | PGPGIN, |
3614 | PGPGOUT, | |
3615 | PGFAULT, | |
3616 | PGMAJFAULT, | |
3617 | }; | |
3618 | ||
3619 | static const char *const memcg1_event_names[] = { | |
3620 | "pgpgin", | |
3621 | "pgpgout", | |
3622 | "pgfault", | |
3623 | "pgmajfault", | |
3624 | }; | |
3625 | ||
2da8ca82 | 3626 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3627 | { |
aa9694bb | 3628 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 3629 | unsigned long memory, memsw; |
af7c4b0e JW |
3630 | struct mem_cgroup *mi; |
3631 | unsigned int i; | |
406eb0c9 | 3632 | |
71cd3113 | 3633 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c RS |
3634 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3635 | ||
71cd3113 JW |
3636 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
3637 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) | |
1dd3a273 | 3638 | continue; |
71cd3113 | 3639 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], |
205b20cc | 3640 | memcg_page_state_local(memcg, memcg1_stats[i]) * |
71cd3113 | 3641 | PAGE_SIZE); |
1dd3a273 | 3642 | } |
7b854121 | 3643 | |
df0e53d0 JW |
3644 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3645 | seq_printf(m, "%s %lu\n", memcg1_event_names[i], | |
205b20cc | 3646 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
3647 | |
3648 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3649 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
205b20cc | 3650 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 3651 | PAGE_SIZE); |
af7c4b0e | 3652 | |
14067bb3 | 3653 | /* Hierarchical information */ |
3e32cb2e JW |
3654 | memory = memsw = PAGE_COUNTER_MAX; |
3655 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
bbec2e15 RG |
3656 | memory = min(memory, mi->memory.max); |
3657 | memsw = min(memsw, mi->memsw.max); | |
fee7b548 | 3658 | } |
3e32cb2e JW |
3659 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3660 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3661 | if (do_memsw_account()) |
3e32cb2e JW |
3662 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3663 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3664 | |
8de7ecc6 | 3665 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
71cd3113 | 3666 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3667 | continue; |
8de7ecc6 | 3668 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
dd923990 YS |
3669 | (u64)memcg_page_state(memcg, memcg1_stats[i]) * |
3670 | PAGE_SIZE); | |
af7c4b0e JW |
3671 | } |
3672 | ||
8de7ecc6 SB |
3673 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3674 | seq_printf(m, "total_%s %llu\n", memcg1_event_names[i], | |
dd923990 | 3675 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 3676 | |
8de7ecc6 SB |
3677 | for (i = 0; i < NR_LRU_LISTS; i++) |
3678 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], | |
42a30035 JW |
3679 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
3680 | PAGE_SIZE); | |
14067bb3 | 3681 | |
7f016ee8 | 3682 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 3683 | { |
ef8f2327 MG |
3684 | pg_data_t *pgdat; |
3685 | struct mem_cgroup_per_node *mz; | |
89abfab1 | 3686 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3687 | unsigned long recent_rotated[2] = {0, 0}; |
3688 | unsigned long recent_scanned[2] = {0, 0}; | |
3689 | ||
ef8f2327 MG |
3690 | for_each_online_pgdat(pgdat) { |
3691 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
3692 | rstat = &mz->lruvec.reclaim_stat; | |
7f016ee8 | 3693 | |
ef8f2327 MG |
3694 | recent_rotated[0] += rstat->recent_rotated[0]; |
3695 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3696 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3697 | recent_scanned[1] += rstat->recent_scanned[1]; | |
3698 | } | |
78ccf5b5 JW |
3699 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3700 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3701 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3702 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3703 | } |
3704 | #endif | |
3705 | ||
d2ceb9b7 KH |
3706 | return 0; |
3707 | } | |
3708 | ||
182446d0 TH |
3709 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3710 | struct cftype *cft) | |
a7885eb8 | 3711 | { |
182446d0 | 3712 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3713 | |
1f4c025b | 3714 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3715 | } |
3716 | ||
182446d0 TH |
3717 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3718 | struct cftype *cft, u64 val) | |
a7885eb8 | 3719 | { |
182446d0 | 3720 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3721 | |
3dae7fec | 3722 | if (val > 100) |
a7885eb8 KM |
3723 | return -EINVAL; |
3724 | ||
14208b0e | 3725 | if (css->parent) |
3dae7fec JW |
3726 | memcg->swappiness = val; |
3727 | else | |
3728 | vm_swappiness = val; | |
068b38c1 | 3729 | |
a7885eb8 KM |
3730 | return 0; |
3731 | } | |
3732 | ||
2e72b634 KS |
3733 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3734 | { | |
3735 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3736 | unsigned long usage; |
2e72b634 KS |
3737 | int i; |
3738 | ||
3739 | rcu_read_lock(); | |
3740 | if (!swap) | |
2c488db2 | 3741 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3742 | else |
2c488db2 | 3743 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3744 | |
3745 | if (!t) | |
3746 | goto unlock; | |
3747 | ||
ce00a967 | 3748 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3749 | |
3750 | /* | |
748dad36 | 3751 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3752 | * If it's not true, a threshold was crossed after last |
3753 | * call of __mem_cgroup_threshold(). | |
3754 | */ | |
5407a562 | 3755 | i = t->current_threshold; |
2e72b634 KS |
3756 | |
3757 | /* | |
3758 | * Iterate backward over array of thresholds starting from | |
3759 | * current_threshold and check if a threshold is crossed. | |
3760 | * If none of thresholds below usage is crossed, we read | |
3761 | * only one element of the array here. | |
3762 | */ | |
3763 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3764 | eventfd_signal(t->entries[i].eventfd, 1); | |
3765 | ||
3766 | /* i = current_threshold + 1 */ | |
3767 | i++; | |
3768 | ||
3769 | /* | |
3770 | * Iterate forward over array of thresholds starting from | |
3771 | * current_threshold+1 and check if a threshold is crossed. | |
3772 | * If none of thresholds above usage is crossed, we read | |
3773 | * only one element of the array here. | |
3774 | */ | |
3775 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3776 | eventfd_signal(t->entries[i].eventfd, 1); | |
3777 | ||
3778 | /* Update current_threshold */ | |
5407a562 | 3779 | t->current_threshold = i - 1; |
2e72b634 KS |
3780 | unlock: |
3781 | rcu_read_unlock(); | |
3782 | } | |
3783 | ||
3784 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3785 | { | |
ad4ca5f4 KS |
3786 | while (memcg) { |
3787 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3788 | if (do_memsw_account()) |
ad4ca5f4 KS |
3789 | __mem_cgroup_threshold(memcg, true); |
3790 | ||
3791 | memcg = parent_mem_cgroup(memcg); | |
3792 | } | |
2e72b634 KS |
3793 | } |
3794 | ||
3795 | static int compare_thresholds(const void *a, const void *b) | |
3796 | { | |
3797 | const struct mem_cgroup_threshold *_a = a; | |
3798 | const struct mem_cgroup_threshold *_b = b; | |
3799 | ||
2bff24a3 GT |
3800 | if (_a->threshold > _b->threshold) |
3801 | return 1; | |
3802 | ||
3803 | if (_a->threshold < _b->threshold) | |
3804 | return -1; | |
3805 | ||
3806 | return 0; | |
2e72b634 KS |
3807 | } |
3808 | ||
c0ff4b85 | 3809 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3810 | { |
3811 | struct mem_cgroup_eventfd_list *ev; | |
3812 | ||
2bcf2e92 MH |
3813 | spin_lock(&memcg_oom_lock); |
3814 | ||
c0ff4b85 | 3815 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3816 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3817 | |
3818 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3819 | return 0; |
3820 | } | |
3821 | ||
c0ff4b85 | 3822 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3823 | { |
7d74b06f KH |
3824 | struct mem_cgroup *iter; |
3825 | ||
c0ff4b85 | 3826 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3827 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3828 | } |
3829 | ||
59b6f873 | 3830 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3831 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3832 | { |
2c488db2 KS |
3833 | struct mem_cgroup_thresholds *thresholds; |
3834 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3835 | unsigned long threshold; |
3836 | unsigned long usage; | |
2c488db2 | 3837 | int i, size, ret; |
2e72b634 | 3838 | |
650c5e56 | 3839 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3840 | if (ret) |
3841 | return ret; | |
3842 | ||
3843 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3844 | |
05b84301 | 3845 | if (type == _MEM) { |
2c488db2 | 3846 | thresholds = &memcg->thresholds; |
ce00a967 | 3847 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3848 | } else if (type == _MEMSWAP) { |
2c488db2 | 3849 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3850 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3851 | } else |
2e72b634 KS |
3852 | BUG(); |
3853 | ||
2e72b634 | 3854 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3855 | if (thresholds->primary) |
2e72b634 KS |
3856 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3857 | ||
2c488db2 | 3858 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3859 | |
3860 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 3861 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 3862 | if (!new) { |
2e72b634 KS |
3863 | ret = -ENOMEM; |
3864 | goto unlock; | |
3865 | } | |
2c488db2 | 3866 | new->size = size; |
2e72b634 KS |
3867 | |
3868 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3869 | if (thresholds->primary) { |
3870 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3871 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3872 | } |
3873 | ||
2e72b634 | 3874 | /* Add new threshold */ |
2c488db2 KS |
3875 | new->entries[size - 1].eventfd = eventfd; |
3876 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3877 | |
3878 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3879 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3880 | compare_thresholds, NULL); |
3881 | ||
3882 | /* Find current threshold */ | |
2c488db2 | 3883 | new->current_threshold = -1; |
2e72b634 | 3884 | for (i = 0; i < size; i++) { |
748dad36 | 3885 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3886 | /* |
2c488db2 KS |
3887 | * new->current_threshold will not be used until |
3888 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3889 | * it here. |
3890 | */ | |
2c488db2 | 3891 | ++new->current_threshold; |
748dad36 SZ |
3892 | } else |
3893 | break; | |
2e72b634 KS |
3894 | } |
3895 | ||
2c488db2 KS |
3896 | /* Free old spare buffer and save old primary buffer as spare */ |
3897 | kfree(thresholds->spare); | |
3898 | thresholds->spare = thresholds->primary; | |
3899 | ||
3900 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3901 | |
907860ed | 3902 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3903 | synchronize_rcu(); |
3904 | ||
2e72b634 KS |
3905 | unlock: |
3906 | mutex_unlock(&memcg->thresholds_lock); | |
3907 | ||
3908 | return ret; | |
3909 | } | |
3910 | ||
59b6f873 | 3911 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3912 | struct eventfd_ctx *eventfd, const char *args) |
3913 | { | |
59b6f873 | 3914 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3915 | } |
3916 | ||
59b6f873 | 3917 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3918 | struct eventfd_ctx *eventfd, const char *args) |
3919 | { | |
59b6f873 | 3920 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3921 | } |
3922 | ||
59b6f873 | 3923 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3924 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3925 | { |
2c488db2 KS |
3926 | struct mem_cgroup_thresholds *thresholds; |
3927 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3928 | unsigned long usage; |
2c488db2 | 3929 | int i, j, size; |
2e72b634 KS |
3930 | |
3931 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3932 | |
3933 | if (type == _MEM) { | |
2c488db2 | 3934 | thresholds = &memcg->thresholds; |
ce00a967 | 3935 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3936 | } else if (type == _MEMSWAP) { |
2c488db2 | 3937 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3938 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3939 | } else |
2e72b634 KS |
3940 | BUG(); |
3941 | ||
371528ca AV |
3942 | if (!thresholds->primary) |
3943 | goto unlock; | |
3944 | ||
2e72b634 KS |
3945 | /* Check if a threshold crossed before removing */ |
3946 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3947 | ||
3948 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3949 | size = 0; |
3950 | for (i = 0; i < thresholds->primary->size; i++) { | |
3951 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3952 | size++; |
3953 | } | |
3954 | ||
2c488db2 | 3955 | new = thresholds->spare; |
907860ed | 3956 | |
2e72b634 KS |
3957 | /* Set thresholds array to NULL if we don't have thresholds */ |
3958 | if (!size) { | |
2c488db2 KS |
3959 | kfree(new); |
3960 | new = NULL; | |
907860ed | 3961 | goto swap_buffers; |
2e72b634 KS |
3962 | } |
3963 | ||
2c488db2 | 3964 | new->size = size; |
2e72b634 KS |
3965 | |
3966 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3967 | new->current_threshold = -1; |
3968 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3969 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3970 | continue; |
3971 | ||
2c488db2 | 3972 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3973 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3974 | /* |
2c488db2 | 3975 | * new->current_threshold will not be used |
2e72b634 KS |
3976 | * until rcu_assign_pointer(), so it's safe to increment |
3977 | * it here. | |
3978 | */ | |
2c488db2 | 3979 | ++new->current_threshold; |
2e72b634 KS |
3980 | } |
3981 | j++; | |
3982 | } | |
3983 | ||
907860ed | 3984 | swap_buffers: |
2c488db2 KS |
3985 | /* Swap primary and spare array */ |
3986 | thresholds->spare = thresholds->primary; | |
8c757763 | 3987 | |
2c488db2 | 3988 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3989 | |
907860ed | 3990 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3991 | synchronize_rcu(); |
6611d8d7 MC |
3992 | |
3993 | /* If all events are unregistered, free the spare array */ | |
3994 | if (!new) { | |
3995 | kfree(thresholds->spare); | |
3996 | thresholds->spare = NULL; | |
3997 | } | |
371528ca | 3998 | unlock: |
2e72b634 | 3999 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4000 | } |
c1e862c1 | 4001 | |
59b6f873 | 4002 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4003 | struct eventfd_ctx *eventfd) |
4004 | { | |
59b6f873 | 4005 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4006 | } |
4007 | ||
59b6f873 | 4008 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4009 | struct eventfd_ctx *eventfd) |
4010 | { | |
59b6f873 | 4011 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4012 | } |
4013 | ||
59b6f873 | 4014 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4015 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4016 | { |
9490ff27 | 4017 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4018 | |
9490ff27 KH |
4019 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4020 | if (!event) | |
4021 | return -ENOMEM; | |
4022 | ||
1af8efe9 | 4023 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4024 | |
4025 | event->eventfd = eventfd; | |
4026 | list_add(&event->list, &memcg->oom_notify); | |
4027 | ||
4028 | /* already in OOM ? */ | |
c2b42d3c | 4029 | if (memcg->under_oom) |
9490ff27 | 4030 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4031 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4032 | |
4033 | return 0; | |
4034 | } | |
4035 | ||
59b6f873 | 4036 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4037 | struct eventfd_ctx *eventfd) |
9490ff27 | 4038 | { |
9490ff27 | 4039 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4040 | |
1af8efe9 | 4041 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4042 | |
c0ff4b85 | 4043 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4044 | if (ev->eventfd == eventfd) { |
4045 | list_del(&ev->list); | |
4046 | kfree(ev); | |
4047 | } | |
4048 | } | |
4049 | ||
1af8efe9 | 4050 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4051 | } |
4052 | ||
2da8ca82 | 4053 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4054 | { |
aa9694bb | 4055 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4056 | |
791badbd | 4057 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4058 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4059 | seq_printf(sf, "oom_kill %lu\n", |
4060 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4061 | return 0; |
4062 | } | |
4063 | ||
182446d0 | 4064 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4065 | struct cftype *cft, u64 val) |
4066 | { | |
182446d0 | 4067 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4068 | |
4069 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4070 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4071 | return -EINVAL; |
4072 | ||
c0ff4b85 | 4073 | memcg->oom_kill_disable = val; |
4d845ebf | 4074 | if (!val) |
c0ff4b85 | 4075 | memcg_oom_recover(memcg); |
3dae7fec | 4076 | |
3c11ecf4 KH |
4077 | return 0; |
4078 | } | |
4079 | ||
52ebea74 TH |
4080 | #ifdef CONFIG_CGROUP_WRITEBACK |
4081 | ||
841710aa TH |
4082 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4083 | { | |
4084 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4085 | } | |
4086 | ||
4087 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4088 | { | |
4089 | wb_domain_exit(&memcg->cgwb_domain); | |
4090 | } | |
4091 | ||
2529bb3a TH |
4092 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4093 | { | |
4094 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4095 | } | |
4096 | ||
841710aa TH |
4097 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4098 | { | |
4099 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4100 | ||
4101 | if (!memcg->css.parent) | |
4102 | return NULL; | |
4103 | ||
4104 | return &memcg->cgwb_domain; | |
4105 | } | |
4106 | ||
0b3d6e6f GT |
4107 | /* |
4108 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
4109 | * Keep in sync with memcg_exact_page(). | |
4110 | */ | |
4111 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
4112 | { | |
871789d4 | 4113 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
4114 | int cpu; |
4115 | ||
4116 | for_each_online_cpu(cpu) | |
871789d4 | 4117 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
4118 | if (x < 0) |
4119 | x = 0; | |
4120 | return x; | |
4121 | } | |
4122 | ||
c2aa723a TH |
4123 | /** |
4124 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4125 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4126 | * @pfilepages: out parameter for number of file pages |
4127 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4128 | * @pdirty: out parameter for number of dirty pages |
4129 | * @pwriteback: out parameter for number of pages under writeback | |
4130 | * | |
c5edf9cd TH |
4131 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4132 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4133 | * is a bit more involved. | |
c2aa723a | 4134 | * |
c5edf9cd TH |
4135 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4136 | * headroom is calculated as the lowest headroom of itself and the | |
4137 | * ancestors. Note that this doesn't consider the actual amount of | |
4138 | * available memory in the system. The caller should further cap | |
4139 | * *@pheadroom accordingly. | |
c2aa723a | 4140 | */ |
c5edf9cd TH |
4141 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4142 | unsigned long *pheadroom, unsigned long *pdirty, | |
4143 | unsigned long *pwriteback) | |
c2aa723a TH |
4144 | { |
4145 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4146 | struct mem_cgroup *parent; | |
c2aa723a | 4147 | |
0b3d6e6f | 4148 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a TH |
4149 | |
4150 | /* this should eventually include NR_UNSTABLE_NFS */ | |
0b3d6e6f | 4151 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4152 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4153 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4154 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4155 | |
c2aa723a | 4156 | while ((parent = parent_mem_cgroup(memcg))) { |
bbec2e15 | 4157 | unsigned long ceiling = min(memcg->memory.max, memcg->high); |
c2aa723a TH |
4158 | unsigned long used = page_counter_read(&memcg->memory); |
4159 | ||
c5edf9cd | 4160 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4161 | memcg = parent; |
4162 | } | |
c2aa723a TH |
4163 | } |
4164 | ||
841710aa TH |
4165 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4166 | ||
4167 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4168 | { | |
4169 | return 0; | |
4170 | } | |
4171 | ||
4172 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4173 | { | |
4174 | } | |
4175 | ||
2529bb3a TH |
4176 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4177 | { | |
4178 | } | |
4179 | ||
52ebea74 TH |
4180 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4181 | ||
3bc942f3 TH |
4182 | /* |
4183 | * DO NOT USE IN NEW FILES. | |
4184 | * | |
4185 | * "cgroup.event_control" implementation. | |
4186 | * | |
4187 | * This is way over-engineered. It tries to support fully configurable | |
4188 | * events for each user. Such level of flexibility is completely | |
4189 | * unnecessary especially in the light of the planned unified hierarchy. | |
4190 | * | |
4191 | * Please deprecate this and replace with something simpler if at all | |
4192 | * possible. | |
4193 | */ | |
4194 | ||
79bd9814 TH |
4195 | /* |
4196 | * Unregister event and free resources. | |
4197 | * | |
4198 | * Gets called from workqueue. | |
4199 | */ | |
3bc942f3 | 4200 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4201 | { |
3bc942f3 TH |
4202 | struct mem_cgroup_event *event = |
4203 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4204 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4205 | |
4206 | remove_wait_queue(event->wqh, &event->wait); | |
4207 | ||
59b6f873 | 4208 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4209 | |
4210 | /* Notify userspace the event is going away. */ | |
4211 | eventfd_signal(event->eventfd, 1); | |
4212 | ||
4213 | eventfd_ctx_put(event->eventfd); | |
4214 | kfree(event); | |
59b6f873 | 4215 | css_put(&memcg->css); |
79bd9814 TH |
4216 | } |
4217 | ||
4218 | /* | |
a9a08845 | 4219 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4220 | * |
4221 | * Called with wqh->lock held and interrupts disabled. | |
4222 | */ | |
ac6424b9 | 4223 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4224 | int sync, void *key) |
79bd9814 | 4225 | { |
3bc942f3 TH |
4226 | struct mem_cgroup_event *event = |
4227 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4228 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4229 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4230 | |
a9a08845 | 4231 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4232 | /* |
4233 | * If the event has been detached at cgroup removal, we | |
4234 | * can simply return knowing the other side will cleanup | |
4235 | * for us. | |
4236 | * | |
4237 | * We can't race against event freeing since the other | |
4238 | * side will require wqh->lock via remove_wait_queue(), | |
4239 | * which we hold. | |
4240 | */ | |
fba94807 | 4241 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4242 | if (!list_empty(&event->list)) { |
4243 | list_del_init(&event->list); | |
4244 | /* | |
4245 | * We are in atomic context, but cgroup_event_remove() | |
4246 | * may sleep, so we have to call it in workqueue. | |
4247 | */ | |
4248 | schedule_work(&event->remove); | |
4249 | } | |
fba94807 | 4250 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4251 | } |
4252 | ||
4253 | return 0; | |
4254 | } | |
4255 | ||
3bc942f3 | 4256 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4257 | wait_queue_head_t *wqh, poll_table *pt) |
4258 | { | |
3bc942f3 TH |
4259 | struct mem_cgroup_event *event = |
4260 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4261 | |
4262 | event->wqh = wqh; | |
4263 | add_wait_queue(wqh, &event->wait); | |
4264 | } | |
4265 | ||
4266 | /* | |
3bc942f3 TH |
4267 | * DO NOT USE IN NEW FILES. |
4268 | * | |
79bd9814 TH |
4269 | * Parse input and register new cgroup event handler. |
4270 | * | |
4271 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4272 | * Interpretation of args is defined by control file implementation. | |
4273 | */ | |
451af504 TH |
4274 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4275 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4276 | { |
451af504 | 4277 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4278 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4279 | struct mem_cgroup_event *event; |
79bd9814 TH |
4280 | struct cgroup_subsys_state *cfile_css; |
4281 | unsigned int efd, cfd; | |
4282 | struct fd efile; | |
4283 | struct fd cfile; | |
fba94807 | 4284 | const char *name; |
79bd9814 TH |
4285 | char *endp; |
4286 | int ret; | |
4287 | ||
451af504 TH |
4288 | buf = strstrip(buf); |
4289 | ||
4290 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4291 | if (*endp != ' ') |
4292 | return -EINVAL; | |
451af504 | 4293 | buf = endp + 1; |
79bd9814 | 4294 | |
451af504 | 4295 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4296 | if ((*endp != ' ') && (*endp != '\0')) |
4297 | return -EINVAL; | |
451af504 | 4298 | buf = endp + 1; |
79bd9814 TH |
4299 | |
4300 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4301 | if (!event) | |
4302 | return -ENOMEM; | |
4303 | ||
59b6f873 | 4304 | event->memcg = memcg; |
79bd9814 | 4305 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4306 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4307 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4308 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4309 | |
4310 | efile = fdget(efd); | |
4311 | if (!efile.file) { | |
4312 | ret = -EBADF; | |
4313 | goto out_kfree; | |
4314 | } | |
4315 | ||
4316 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4317 | if (IS_ERR(event->eventfd)) { | |
4318 | ret = PTR_ERR(event->eventfd); | |
4319 | goto out_put_efile; | |
4320 | } | |
4321 | ||
4322 | cfile = fdget(cfd); | |
4323 | if (!cfile.file) { | |
4324 | ret = -EBADF; | |
4325 | goto out_put_eventfd; | |
4326 | } | |
4327 | ||
4328 | /* the process need read permission on control file */ | |
4329 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4330 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4331 | if (ret < 0) | |
4332 | goto out_put_cfile; | |
4333 | ||
fba94807 TH |
4334 | /* |
4335 | * Determine the event callbacks and set them in @event. This used | |
4336 | * to be done via struct cftype but cgroup core no longer knows | |
4337 | * about these events. The following is crude but the whole thing | |
4338 | * is for compatibility anyway. | |
3bc942f3 TH |
4339 | * |
4340 | * DO NOT ADD NEW FILES. | |
fba94807 | 4341 | */ |
b583043e | 4342 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4343 | |
4344 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4345 | event->register_event = mem_cgroup_usage_register_event; | |
4346 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4347 | } else if (!strcmp(name, "memory.oom_control")) { | |
4348 | event->register_event = mem_cgroup_oom_register_event; | |
4349 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4350 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4351 | event->register_event = vmpressure_register_event; | |
4352 | event->unregister_event = vmpressure_unregister_event; | |
4353 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4354 | event->register_event = memsw_cgroup_usage_register_event; |
4355 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4356 | } else { |
4357 | ret = -EINVAL; | |
4358 | goto out_put_cfile; | |
4359 | } | |
4360 | ||
79bd9814 | 4361 | /* |
b5557c4c TH |
4362 | * Verify @cfile should belong to @css. Also, remaining events are |
4363 | * automatically removed on cgroup destruction but the removal is | |
4364 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4365 | */ |
b583043e | 4366 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4367 | &memory_cgrp_subsys); |
79bd9814 | 4368 | ret = -EINVAL; |
5a17f543 | 4369 | if (IS_ERR(cfile_css)) |
79bd9814 | 4370 | goto out_put_cfile; |
5a17f543 TH |
4371 | if (cfile_css != css) { |
4372 | css_put(cfile_css); | |
79bd9814 | 4373 | goto out_put_cfile; |
5a17f543 | 4374 | } |
79bd9814 | 4375 | |
451af504 | 4376 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4377 | if (ret) |
4378 | goto out_put_css; | |
4379 | ||
9965ed17 | 4380 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4381 | |
fba94807 TH |
4382 | spin_lock(&memcg->event_list_lock); |
4383 | list_add(&event->list, &memcg->event_list); | |
4384 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4385 | |
4386 | fdput(cfile); | |
4387 | fdput(efile); | |
4388 | ||
451af504 | 4389 | return nbytes; |
79bd9814 TH |
4390 | |
4391 | out_put_css: | |
b5557c4c | 4392 | css_put(css); |
79bd9814 TH |
4393 | out_put_cfile: |
4394 | fdput(cfile); | |
4395 | out_put_eventfd: | |
4396 | eventfd_ctx_put(event->eventfd); | |
4397 | out_put_efile: | |
4398 | fdput(efile); | |
4399 | out_kfree: | |
4400 | kfree(event); | |
4401 | ||
4402 | return ret; | |
4403 | } | |
4404 | ||
241994ed | 4405 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4406 | { |
0eea1030 | 4407 | .name = "usage_in_bytes", |
8c7c6e34 | 4408 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4409 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4410 | }, |
c84872e1 PE |
4411 | { |
4412 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4413 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4414 | .write = mem_cgroup_reset, |
791badbd | 4415 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4416 | }, |
8cdea7c0 | 4417 | { |
0eea1030 | 4418 | .name = "limit_in_bytes", |
8c7c6e34 | 4419 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4420 | .write = mem_cgroup_write, |
791badbd | 4421 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4422 | }, |
296c81d8 BS |
4423 | { |
4424 | .name = "soft_limit_in_bytes", | |
4425 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4426 | .write = mem_cgroup_write, |
791badbd | 4427 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4428 | }, |
8cdea7c0 BS |
4429 | { |
4430 | .name = "failcnt", | |
8c7c6e34 | 4431 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4432 | .write = mem_cgroup_reset, |
791badbd | 4433 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4434 | }, |
d2ceb9b7 KH |
4435 | { |
4436 | .name = "stat", | |
2da8ca82 | 4437 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4438 | }, |
c1e862c1 KH |
4439 | { |
4440 | .name = "force_empty", | |
6770c64e | 4441 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4442 | }, |
18f59ea7 BS |
4443 | { |
4444 | .name = "use_hierarchy", | |
4445 | .write_u64 = mem_cgroup_hierarchy_write, | |
4446 | .read_u64 = mem_cgroup_hierarchy_read, | |
4447 | }, | |
79bd9814 | 4448 | { |
3bc942f3 | 4449 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4450 | .write = memcg_write_event_control, |
7dbdb199 | 4451 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4452 | }, |
a7885eb8 KM |
4453 | { |
4454 | .name = "swappiness", | |
4455 | .read_u64 = mem_cgroup_swappiness_read, | |
4456 | .write_u64 = mem_cgroup_swappiness_write, | |
4457 | }, | |
7dc74be0 DN |
4458 | { |
4459 | .name = "move_charge_at_immigrate", | |
4460 | .read_u64 = mem_cgroup_move_charge_read, | |
4461 | .write_u64 = mem_cgroup_move_charge_write, | |
4462 | }, | |
9490ff27 KH |
4463 | { |
4464 | .name = "oom_control", | |
2da8ca82 | 4465 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4466 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4467 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4468 | }, | |
70ddf637 AV |
4469 | { |
4470 | .name = "pressure_level", | |
70ddf637 | 4471 | }, |
406eb0c9 YH |
4472 | #ifdef CONFIG_NUMA |
4473 | { | |
4474 | .name = "numa_stat", | |
2da8ca82 | 4475 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4476 | }, |
4477 | #endif | |
510fc4e1 GC |
4478 | { |
4479 | .name = "kmem.limit_in_bytes", | |
4480 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4481 | .write = mem_cgroup_write, |
791badbd | 4482 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4483 | }, |
4484 | { | |
4485 | .name = "kmem.usage_in_bytes", | |
4486 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4487 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4488 | }, |
4489 | { | |
4490 | .name = "kmem.failcnt", | |
4491 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4492 | .write = mem_cgroup_reset, |
791badbd | 4493 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4494 | }, |
4495 | { | |
4496 | .name = "kmem.max_usage_in_bytes", | |
4497 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4498 | .write = mem_cgroup_reset, |
791badbd | 4499 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4500 | }, |
5b365771 | 4501 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
749c5415 GC |
4502 | { |
4503 | .name = "kmem.slabinfo", | |
bc2791f8 TH |
4504 | .seq_start = memcg_slab_start, |
4505 | .seq_next = memcg_slab_next, | |
4506 | .seq_stop = memcg_slab_stop, | |
b047501c | 4507 | .seq_show = memcg_slab_show, |
749c5415 GC |
4508 | }, |
4509 | #endif | |
d55f90bf VD |
4510 | { |
4511 | .name = "kmem.tcp.limit_in_bytes", | |
4512 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4513 | .write = mem_cgroup_write, | |
4514 | .read_u64 = mem_cgroup_read_u64, | |
4515 | }, | |
4516 | { | |
4517 | .name = "kmem.tcp.usage_in_bytes", | |
4518 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4519 | .read_u64 = mem_cgroup_read_u64, | |
4520 | }, | |
4521 | { | |
4522 | .name = "kmem.tcp.failcnt", | |
4523 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4524 | .write = mem_cgroup_reset, | |
4525 | .read_u64 = mem_cgroup_read_u64, | |
4526 | }, | |
4527 | { | |
4528 | .name = "kmem.tcp.max_usage_in_bytes", | |
4529 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4530 | .write = mem_cgroup_reset, | |
4531 | .read_u64 = mem_cgroup_read_u64, | |
4532 | }, | |
6bc10349 | 4533 | { }, /* terminate */ |
af36f906 | 4534 | }; |
8c7c6e34 | 4535 | |
73f576c0 JW |
4536 | /* |
4537 | * Private memory cgroup IDR | |
4538 | * | |
4539 | * Swap-out records and page cache shadow entries need to store memcg | |
4540 | * references in constrained space, so we maintain an ID space that is | |
4541 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4542 | * memory-controlled cgroups to 64k. | |
4543 | * | |
4544 | * However, there usually are many references to the oflline CSS after | |
4545 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4546 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4547 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4548 | * relatively small ID space and prevent the creation of new cgroups | |
4549 | * even when there are much fewer than 64k cgroups - possibly none. | |
4550 | * | |
4551 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4552 | * be freed and recycled when it's no longer needed, which is usually | |
4553 | * when the CSS is offlined. | |
4554 | * | |
4555 | * The only exception to that are records of swapped out tmpfs/shmem | |
4556 | * pages that need to be attributed to live ancestors on swapin. But | |
4557 | * those references are manageable from userspace. | |
4558 | */ | |
4559 | ||
4560 | static DEFINE_IDR(mem_cgroup_idr); | |
4561 | ||
7e97de0b KT |
4562 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
4563 | { | |
4564 | if (memcg->id.id > 0) { | |
4565 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
4566 | memcg->id.id = 0; | |
4567 | } | |
4568 | } | |
4569 | ||
615d66c3 | 4570 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4571 | { |
1c2d479a | 4572 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
4573 | } |
4574 | ||
615d66c3 | 4575 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4576 | { |
1c2d479a | 4577 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 4578 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
4579 | |
4580 | /* Memcg ID pins CSS */ | |
4581 | css_put(&memcg->css); | |
4582 | } | |
4583 | } | |
4584 | ||
615d66c3 VD |
4585 | static inline void mem_cgroup_id_get(struct mem_cgroup *memcg) |
4586 | { | |
4587 | mem_cgroup_id_get_many(memcg, 1); | |
4588 | } | |
4589 | ||
4590 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) | |
4591 | { | |
4592 | mem_cgroup_id_put_many(memcg, 1); | |
4593 | } | |
4594 | ||
73f576c0 JW |
4595 | /** |
4596 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4597 | * @id: the memcg id to look up | |
4598 | * | |
4599 | * Caller must hold rcu_read_lock(). | |
4600 | */ | |
4601 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4602 | { | |
4603 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4604 | return idr_find(&mem_cgroup_idr, id); | |
4605 | } | |
4606 | ||
ef8f2327 | 4607 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4608 | { |
4609 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 4610 | int tmp = node; |
1ecaab2b KH |
4611 | /* |
4612 | * This routine is called against possible nodes. | |
4613 | * But it's BUG to call kmalloc() against offline node. | |
4614 | * | |
4615 | * TODO: this routine can waste much memory for nodes which will | |
4616 | * never be onlined. It's better to use memory hotplug callback | |
4617 | * function. | |
4618 | */ | |
41e3355d KH |
4619 | if (!node_state(node, N_NORMAL_MEMORY)) |
4620 | tmp = -1; | |
17295c88 | 4621 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4622 | if (!pn) |
4623 | return 1; | |
1ecaab2b | 4624 | |
815744d7 JW |
4625 | pn->lruvec_stat_local = alloc_percpu(struct lruvec_stat); |
4626 | if (!pn->lruvec_stat_local) { | |
4627 | kfree(pn); | |
4628 | return 1; | |
4629 | } | |
4630 | ||
a983b5eb JW |
4631 | pn->lruvec_stat_cpu = alloc_percpu(struct lruvec_stat); |
4632 | if (!pn->lruvec_stat_cpu) { | |
815744d7 | 4633 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c JW |
4634 | kfree(pn); |
4635 | return 1; | |
4636 | } | |
4637 | ||
ef8f2327 MG |
4638 | lruvec_init(&pn->lruvec); |
4639 | pn->usage_in_excess = 0; | |
4640 | pn->on_tree = false; | |
4641 | pn->memcg = memcg; | |
4642 | ||
54f72fe0 | 4643 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4644 | return 0; |
4645 | } | |
4646 | ||
ef8f2327 | 4647 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4648 | { |
00f3ca2c JW |
4649 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
4650 | ||
4eaf431f MH |
4651 | if (!pn) |
4652 | return; | |
4653 | ||
a983b5eb | 4654 | free_percpu(pn->lruvec_stat_cpu); |
815744d7 | 4655 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c | 4656 | kfree(pn); |
1ecaab2b KH |
4657 | } |
4658 | ||
40e952f9 | 4659 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4660 | { |
c8b2a36f | 4661 | int node; |
59927fb9 | 4662 | |
c8b2a36f | 4663 | for_each_node(node) |
ef8f2327 | 4664 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 4665 | free_percpu(memcg->vmstats_percpu); |
815744d7 | 4666 | free_percpu(memcg->vmstats_local); |
8ff69e2c | 4667 | kfree(memcg); |
59927fb9 | 4668 | } |
3afe36b1 | 4669 | |
40e952f9 TE |
4670 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
4671 | { | |
4672 | memcg_wb_domain_exit(memcg); | |
4673 | __mem_cgroup_free(memcg); | |
4674 | } | |
4675 | ||
0b8f73e1 | 4676 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4677 | { |
d142e3e6 | 4678 | struct mem_cgroup *memcg; |
b9726c26 | 4679 | unsigned int size; |
6d12e2d8 | 4680 | int node; |
8cdea7c0 | 4681 | |
0b8f73e1 JW |
4682 | size = sizeof(struct mem_cgroup); |
4683 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4684 | ||
4685 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4686 | if (!memcg) |
0b8f73e1 JW |
4687 | return NULL; |
4688 | ||
73f576c0 JW |
4689 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
4690 | 1, MEM_CGROUP_ID_MAX, | |
4691 | GFP_KERNEL); | |
4692 | if (memcg->id.id < 0) | |
4693 | goto fail; | |
4694 | ||
815744d7 JW |
4695 | memcg->vmstats_local = alloc_percpu(struct memcg_vmstats_percpu); |
4696 | if (!memcg->vmstats_local) | |
4697 | goto fail; | |
4698 | ||
871789d4 CD |
4699 | memcg->vmstats_percpu = alloc_percpu(struct memcg_vmstats_percpu); |
4700 | if (!memcg->vmstats_percpu) | |
0b8f73e1 | 4701 | goto fail; |
78fb7466 | 4702 | |
3ed28fa1 | 4703 | for_each_node(node) |
ef8f2327 | 4704 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 4705 | goto fail; |
f64c3f54 | 4706 | |
0b8f73e1 JW |
4707 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4708 | goto fail; | |
28dbc4b6 | 4709 | |
f7e1cb6e | 4710 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4711 | memcg->last_scanned_node = MAX_NUMNODES; |
4712 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4713 | mutex_init(&memcg->thresholds_lock); |
4714 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4715 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4716 | INIT_LIST_HEAD(&memcg->event_list); |
4717 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4718 | memcg->socket_pressure = jiffies; |
84c07d11 | 4719 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 4720 | memcg->kmemcg_id = -1; |
900a38f0 | 4721 | #endif |
52ebea74 TH |
4722 | #ifdef CONFIG_CGROUP_WRITEBACK |
4723 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4724 | #endif | |
73f576c0 | 4725 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
4726 | return memcg; |
4727 | fail: | |
7e97de0b | 4728 | mem_cgroup_id_remove(memcg); |
40e952f9 | 4729 | __mem_cgroup_free(memcg); |
0b8f73e1 | 4730 | return NULL; |
d142e3e6 GC |
4731 | } |
4732 | ||
0b8f73e1 JW |
4733 | static struct cgroup_subsys_state * __ref |
4734 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4735 | { |
0b8f73e1 JW |
4736 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4737 | struct mem_cgroup *memcg; | |
4738 | long error = -ENOMEM; | |
d142e3e6 | 4739 | |
0b8f73e1 JW |
4740 | memcg = mem_cgroup_alloc(); |
4741 | if (!memcg) | |
4742 | return ERR_PTR(error); | |
d142e3e6 | 4743 | |
0b8f73e1 JW |
4744 | memcg->high = PAGE_COUNTER_MAX; |
4745 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4746 | if (parent) { | |
4747 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4748 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4749 | } | |
4750 | if (parent && parent->use_hierarchy) { | |
4751 | memcg->use_hierarchy = true; | |
3e32cb2e | 4752 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4753 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4754 | page_counter_init(&memcg->memsw, &parent->memsw); |
4755 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4756 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4757 | } else { |
3e32cb2e | 4758 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4759 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4760 | page_counter_init(&memcg->memsw, NULL); |
4761 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4762 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4763 | /* |
4764 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4765 | * much sense so let cgroup subsystem know about this | |
4766 | * unfortunate state in our controller. | |
4767 | */ | |
d142e3e6 | 4768 | if (parent != root_mem_cgroup) |
073219e9 | 4769 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4770 | } |
d6441637 | 4771 | |
0b8f73e1 JW |
4772 | /* The following stuff does not apply to the root */ |
4773 | if (!parent) { | |
4774 | root_mem_cgroup = memcg; | |
4775 | return &memcg->css; | |
4776 | } | |
4777 | ||
b313aeee | 4778 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
4779 | if (error) |
4780 | goto fail; | |
127424c8 | 4781 | |
f7e1cb6e | 4782 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4783 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4784 | |
0b8f73e1 JW |
4785 | return &memcg->css; |
4786 | fail: | |
7e97de0b | 4787 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 4788 | mem_cgroup_free(memcg); |
ea3a9645 | 4789 | return ERR_PTR(-ENOMEM); |
0b8f73e1 JW |
4790 | } |
4791 | ||
73f576c0 | 4792 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 4793 | { |
58fa2a55 VD |
4794 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
4795 | ||
0a4465d3 KT |
4796 | /* |
4797 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
4798 | * by the time the maps are allocated. So, we allocate maps | |
4799 | * here, when for_each_mem_cgroup() can't skip it. | |
4800 | */ | |
4801 | if (memcg_alloc_shrinker_maps(memcg)) { | |
4802 | mem_cgroup_id_remove(memcg); | |
4803 | return -ENOMEM; | |
4804 | } | |
4805 | ||
73f576c0 | 4806 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 4807 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 4808 | css_get(css); |
2f7dd7a4 | 4809 | return 0; |
8cdea7c0 BS |
4810 | } |
4811 | ||
eb95419b | 4812 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4813 | { |
eb95419b | 4814 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4815 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4816 | |
4817 | /* | |
4818 | * Unregister events and notify userspace. | |
4819 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4820 | * directory to avoid race between userspace and kernelspace. | |
4821 | */ | |
fba94807 TH |
4822 | spin_lock(&memcg->event_list_lock); |
4823 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4824 | list_del_init(&event->list); |
4825 | schedule_work(&event->remove); | |
4826 | } | |
fba94807 | 4827 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4828 | |
bf8d5d52 | 4829 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4830 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 4831 | |
567e9ab2 | 4832 | memcg_offline_kmem(memcg); |
52ebea74 | 4833 | wb_memcg_offline(memcg); |
73f576c0 | 4834 | |
591edfb1 RG |
4835 | drain_all_stock(memcg); |
4836 | ||
73f576c0 | 4837 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
4838 | } |
4839 | ||
6df38689 VD |
4840 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4841 | { | |
4842 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4843 | ||
4844 | invalidate_reclaim_iterators(memcg); | |
4845 | } | |
4846 | ||
eb95419b | 4847 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4848 | { |
eb95419b | 4849 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4850 | |
f7e1cb6e | 4851 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4852 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4853 | |
0db15298 | 4854 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4855 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4856 | |
0b8f73e1 JW |
4857 | vmpressure_cleanup(&memcg->vmpressure); |
4858 | cancel_work_sync(&memcg->high_work); | |
4859 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 4860 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 4861 | memcg_free_kmem(memcg); |
0b8f73e1 | 4862 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4863 | } |
4864 | ||
1ced953b TH |
4865 | /** |
4866 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4867 | * @css: the target css | |
4868 | * | |
4869 | * Reset the states of the mem_cgroup associated with @css. This is | |
4870 | * invoked when the userland requests disabling on the default hierarchy | |
4871 | * but the memcg is pinned through dependency. The memcg should stop | |
4872 | * applying policies and should revert to the vanilla state as it may be | |
4873 | * made visible again. | |
4874 | * | |
4875 | * The current implementation only resets the essential configurations. | |
4876 | * This needs to be expanded to cover all the visible parts. | |
4877 | */ | |
4878 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4879 | { | |
4880 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4881 | ||
bbec2e15 RG |
4882 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
4883 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
4884 | page_counter_set_max(&memcg->memsw, PAGE_COUNTER_MAX); | |
4885 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); | |
4886 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 4887 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4888 | page_counter_set_low(&memcg->memory, 0); |
241994ed | 4889 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4890 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4891 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4892 | } |
4893 | ||
02491447 | 4894 | #ifdef CONFIG_MMU |
7dc74be0 | 4895 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4896 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4897 | { |
05b84301 | 4898 | int ret; |
9476db97 | 4899 | |
d0164adc MG |
4900 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4901 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4902 | if (!ret) { |
854ffa8d | 4903 | mc.precharge += count; |
854ffa8d DN |
4904 | return ret; |
4905 | } | |
9476db97 | 4906 | |
3674534b | 4907 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 4908 | while (count--) { |
3674534b | 4909 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 4910 | if (ret) |
38c5d72f | 4911 | return ret; |
854ffa8d | 4912 | mc.precharge++; |
9476db97 | 4913 | cond_resched(); |
854ffa8d | 4914 | } |
9476db97 | 4915 | return 0; |
4ffef5fe DN |
4916 | } |
4917 | ||
4ffef5fe DN |
4918 | union mc_target { |
4919 | struct page *page; | |
02491447 | 4920 | swp_entry_t ent; |
4ffef5fe DN |
4921 | }; |
4922 | ||
4ffef5fe | 4923 | enum mc_target_type { |
8d32ff84 | 4924 | MC_TARGET_NONE = 0, |
4ffef5fe | 4925 | MC_TARGET_PAGE, |
02491447 | 4926 | MC_TARGET_SWAP, |
c733a828 | 4927 | MC_TARGET_DEVICE, |
4ffef5fe DN |
4928 | }; |
4929 | ||
90254a65 DN |
4930 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4931 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4932 | { |
c733a828 | 4933 | struct page *page = _vm_normal_page(vma, addr, ptent, true); |
4ffef5fe | 4934 | |
90254a65 DN |
4935 | if (!page || !page_mapped(page)) |
4936 | return NULL; | |
4937 | if (PageAnon(page)) { | |
1dfab5ab | 4938 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4939 | return NULL; |
1dfab5ab JW |
4940 | } else { |
4941 | if (!(mc.flags & MOVE_FILE)) | |
4942 | return NULL; | |
4943 | } | |
90254a65 DN |
4944 | if (!get_page_unless_zero(page)) |
4945 | return NULL; | |
4946 | ||
4947 | return page; | |
4948 | } | |
4949 | ||
c733a828 | 4950 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 4951 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 4952 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 4953 | { |
90254a65 DN |
4954 | struct page *page = NULL; |
4955 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4956 | ||
1dfab5ab | 4957 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4958 | return NULL; |
c733a828 JG |
4959 | |
4960 | /* | |
4961 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
4962 | * a device and because they are not accessible by CPU they are store | |
4963 | * as special swap entry in the CPU page table. | |
4964 | */ | |
4965 | if (is_device_private_entry(ent)) { | |
4966 | page = device_private_entry_to_page(ent); | |
4967 | /* | |
4968 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
4969 | * a refcount of 1 when free (unlike normal page) | |
4970 | */ | |
4971 | if (!page_ref_add_unless(page, 1, 1)) | |
4972 | return NULL; | |
4973 | return page; | |
4974 | } | |
4975 | ||
4b91355e KH |
4976 | /* |
4977 | * Because lookup_swap_cache() updates some statistics counter, | |
4978 | * we call find_get_page() with swapper_space directly. | |
4979 | */ | |
f6ab1f7f | 4980 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
7941d214 | 4981 | if (do_memsw_account()) |
90254a65 DN |
4982 | entry->val = ent.val; |
4983 | ||
4984 | return page; | |
4985 | } | |
4b91355e KH |
4986 | #else |
4987 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 4988 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
4989 | { |
4990 | return NULL; | |
4991 | } | |
4992 | #endif | |
90254a65 | 4993 | |
87946a72 DN |
4994 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4995 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4996 | { | |
4997 | struct page *page = NULL; | |
87946a72 DN |
4998 | struct address_space *mapping; |
4999 | pgoff_t pgoff; | |
5000 | ||
5001 | if (!vma->vm_file) /* anonymous vma */ | |
5002 | return NULL; | |
1dfab5ab | 5003 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5004 | return NULL; |
5005 | ||
87946a72 | 5006 | mapping = vma->vm_file->f_mapping; |
0661a336 | 5007 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
5008 | |
5009 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5010 | #ifdef CONFIG_SWAP |
5011 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
5012 | if (shmem_mapping(mapping)) { |
5013 | page = find_get_entry(mapping, pgoff); | |
3159f943 | 5014 | if (xa_is_value(page)) { |
139b6a6f | 5015 | swp_entry_t swp = radix_to_swp_entry(page); |
7941d214 | 5016 | if (do_memsw_account()) |
139b6a6f | 5017 | *entry = swp; |
f6ab1f7f HY |
5018 | page = find_get_page(swap_address_space(swp), |
5019 | swp_offset(swp)); | |
139b6a6f JW |
5020 | } |
5021 | } else | |
5022 | page = find_get_page(mapping, pgoff); | |
5023 | #else | |
5024 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 5025 | #endif |
87946a72 DN |
5026 | return page; |
5027 | } | |
5028 | ||
b1b0deab CG |
5029 | /** |
5030 | * mem_cgroup_move_account - move account of the page | |
5031 | * @page: the page | |
25843c2b | 5032 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5033 | * @from: mem_cgroup which the page is moved from. |
5034 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5035 | * | |
3ac808fd | 5036 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5037 | * |
5038 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5039 | * from old cgroup. | |
5040 | */ | |
5041 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5042 | bool compound, |
b1b0deab CG |
5043 | struct mem_cgroup *from, |
5044 | struct mem_cgroup *to) | |
5045 | { | |
5046 | unsigned long flags; | |
f627c2f5 | 5047 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 5048 | int ret; |
c4843a75 | 5049 | bool anon; |
b1b0deab CG |
5050 | |
5051 | VM_BUG_ON(from == to); | |
5052 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5053 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5054 | |
5055 | /* | |
6a93ca8f | 5056 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 5057 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 5058 | */ |
f627c2f5 | 5059 | ret = -EBUSY; |
b1b0deab CG |
5060 | if (!trylock_page(page)) |
5061 | goto out; | |
5062 | ||
5063 | ret = -EINVAL; | |
5064 | if (page->mem_cgroup != from) | |
5065 | goto out_unlock; | |
5066 | ||
c4843a75 GT |
5067 | anon = PageAnon(page); |
5068 | ||
b1b0deab CG |
5069 | spin_lock_irqsave(&from->move_lock, flags); |
5070 | ||
c4843a75 | 5071 | if (!anon && page_mapped(page)) { |
c9019e9b JW |
5072 | __mod_memcg_state(from, NR_FILE_MAPPED, -nr_pages); |
5073 | __mod_memcg_state(to, NR_FILE_MAPPED, nr_pages); | |
b1b0deab CG |
5074 | } |
5075 | ||
c4843a75 GT |
5076 | /* |
5077 | * move_lock grabbed above and caller set from->moving_account, so | |
ccda7f43 | 5078 | * mod_memcg_page_state will serialize updates to PageDirty. |
c4843a75 GT |
5079 | * So mapping should be stable for dirty pages. |
5080 | */ | |
5081 | if (!anon && PageDirty(page)) { | |
5082 | struct address_space *mapping = page_mapping(page); | |
5083 | ||
5084 | if (mapping_cap_account_dirty(mapping)) { | |
c9019e9b JW |
5085 | __mod_memcg_state(from, NR_FILE_DIRTY, -nr_pages); |
5086 | __mod_memcg_state(to, NR_FILE_DIRTY, nr_pages); | |
c4843a75 GT |
5087 | } |
5088 | } | |
5089 | ||
b1b0deab | 5090 | if (PageWriteback(page)) { |
c9019e9b JW |
5091 | __mod_memcg_state(from, NR_WRITEBACK, -nr_pages); |
5092 | __mod_memcg_state(to, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5093 | } |
5094 | ||
5095 | /* | |
5096 | * It is safe to change page->mem_cgroup here because the page | |
5097 | * is referenced, charged, and isolated - we can't race with | |
5098 | * uncharging, charging, migration, or LRU putback. | |
5099 | */ | |
5100 | ||
5101 | /* caller should have done css_get */ | |
5102 | page->mem_cgroup = to; | |
5103 | spin_unlock_irqrestore(&from->move_lock, flags); | |
5104 | ||
5105 | ret = 0; | |
5106 | ||
5107 | local_irq_disable(); | |
f627c2f5 | 5108 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 5109 | memcg_check_events(to, page); |
f627c2f5 | 5110 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
5111 | memcg_check_events(from, page); |
5112 | local_irq_enable(); | |
5113 | out_unlock: | |
5114 | unlock_page(page); | |
5115 | out: | |
5116 | return ret; | |
5117 | } | |
5118 | ||
7cf7806c LR |
5119 | /** |
5120 | * get_mctgt_type - get target type of moving charge | |
5121 | * @vma: the vma the pte to be checked belongs | |
5122 | * @addr: the address corresponding to the pte to be checked | |
5123 | * @ptent: the pte to be checked | |
5124 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5125 | * | |
5126 | * Returns | |
5127 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5128 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5129 | * move charge. if @target is not NULL, the page is stored in target->page | |
5130 | * with extra refcnt got(Callers should handle it). | |
5131 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5132 | * target for charge migration. if @target is not NULL, the entry is stored | |
5133 | * in target->ent. | |
df6ad698 JG |
5134 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PUBLIC |
5135 | * or MEMORY_DEVICE_PRIVATE (so ZONE_DEVICE page and thus not on the lru). | |
5136 | * For now we such page is charge like a regular page would be as for all | |
5137 | * intent and purposes it is just special memory taking the place of a | |
5138 | * regular page. | |
c733a828 JG |
5139 | * |
5140 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5141 | * |
5142 | * Called with pte lock held. | |
5143 | */ | |
5144 | ||
8d32ff84 | 5145 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5146 | unsigned long addr, pte_t ptent, union mc_target *target) |
5147 | { | |
5148 | struct page *page = NULL; | |
8d32ff84 | 5149 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5150 | swp_entry_t ent = { .val = 0 }; |
5151 | ||
5152 | if (pte_present(ptent)) | |
5153 | page = mc_handle_present_pte(vma, addr, ptent); | |
5154 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5155 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5156 | else if (pte_none(ptent)) |
87946a72 | 5157 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5158 | |
5159 | if (!page && !ent.val) | |
8d32ff84 | 5160 | return ret; |
02491447 | 5161 | if (page) { |
02491447 | 5162 | /* |
0a31bc97 | 5163 | * Do only loose check w/o serialization. |
1306a85a | 5164 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5165 | * not under LRU exclusion. |
02491447 | 5166 | */ |
1306a85a | 5167 | if (page->mem_cgroup == mc.from) { |
02491447 | 5168 | ret = MC_TARGET_PAGE; |
df6ad698 JG |
5169 | if (is_device_private_page(page) || |
5170 | is_device_public_page(page)) | |
c733a828 | 5171 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5172 | if (target) |
5173 | target->page = page; | |
5174 | } | |
5175 | if (!ret || !target) | |
5176 | put_page(page); | |
5177 | } | |
3e14a57b HY |
5178 | /* |
5179 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5180 | * But we cannot move a tail-page in a THP. | |
5181 | */ | |
5182 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5183 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5184 | ret = MC_TARGET_SWAP; |
5185 | if (target) | |
5186 | target->ent = ent; | |
4ffef5fe | 5187 | } |
4ffef5fe DN |
5188 | return ret; |
5189 | } | |
5190 | ||
12724850 NH |
5191 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5192 | /* | |
d6810d73 HY |
5193 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5194 | * not support them for now. | |
12724850 NH |
5195 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5196 | */ | |
5197 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5198 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5199 | { | |
5200 | struct page *page = NULL; | |
12724850 NH |
5201 | enum mc_target_type ret = MC_TARGET_NONE; |
5202 | ||
84c3fc4e ZY |
5203 | if (unlikely(is_swap_pmd(pmd))) { |
5204 | VM_BUG_ON(thp_migration_supported() && | |
5205 | !is_pmd_migration_entry(pmd)); | |
5206 | return ret; | |
5207 | } | |
12724850 | 5208 | page = pmd_page(pmd); |
309381fe | 5209 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5210 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5211 | return ret; |
1306a85a | 5212 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
5213 | ret = MC_TARGET_PAGE; |
5214 | if (target) { | |
5215 | get_page(page); | |
5216 | target->page = page; | |
5217 | } | |
5218 | } | |
5219 | return ret; | |
5220 | } | |
5221 | #else | |
5222 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5223 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5224 | { | |
5225 | return MC_TARGET_NONE; | |
5226 | } | |
5227 | #endif | |
5228 | ||
4ffef5fe DN |
5229 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5230 | unsigned long addr, unsigned long end, | |
5231 | struct mm_walk *walk) | |
5232 | { | |
26bcd64a | 5233 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5234 | pte_t *pte; |
5235 | spinlock_t *ptl; | |
5236 | ||
b6ec57f4 KS |
5237 | ptl = pmd_trans_huge_lock(pmd, vma); |
5238 | if (ptl) { | |
c733a828 JG |
5239 | /* |
5240 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
5241 | * support transparent huge page with MEMORY_DEVICE_PUBLIC or | |
5242 | * MEMORY_DEVICE_PRIVATE but this might change. | |
5243 | */ | |
12724850 NH |
5244 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5245 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5246 | spin_unlock(ptl); |
1a5a9906 | 5247 | return 0; |
12724850 | 5248 | } |
03319327 | 5249 | |
45f83cef AA |
5250 | if (pmd_trans_unstable(pmd)) |
5251 | return 0; | |
4ffef5fe DN |
5252 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5253 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5254 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5255 | mc.precharge++; /* increment precharge temporarily */ |
5256 | pte_unmap_unlock(pte - 1, ptl); | |
5257 | cond_resched(); | |
5258 | ||
7dc74be0 DN |
5259 | return 0; |
5260 | } | |
5261 | ||
4ffef5fe DN |
5262 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5263 | { | |
5264 | unsigned long precharge; | |
4ffef5fe | 5265 | |
26bcd64a NH |
5266 | struct mm_walk mem_cgroup_count_precharge_walk = { |
5267 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5268 | .mm = mm, | |
5269 | }; | |
dfe076b0 | 5270 | down_read(&mm->mmap_sem); |
0247f3f4 JM |
5271 | walk_page_range(0, mm->highest_vm_end, |
5272 | &mem_cgroup_count_precharge_walk); | |
dfe076b0 | 5273 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5274 | |
5275 | precharge = mc.precharge; | |
5276 | mc.precharge = 0; | |
5277 | ||
5278 | return precharge; | |
5279 | } | |
5280 | ||
4ffef5fe DN |
5281 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5282 | { | |
dfe076b0 DN |
5283 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5284 | ||
5285 | VM_BUG_ON(mc.moving_task); | |
5286 | mc.moving_task = current; | |
5287 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5288 | } |
5289 | ||
dfe076b0 DN |
5290 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5291 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5292 | { |
2bd9bb20 KH |
5293 | struct mem_cgroup *from = mc.from; |
5294 | struct mem_cgroup *to = mc.to; | |
5295 | ||
4ffef5fe | 5296 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5297 | if (mc.precharge) { |
00501b53 | 5298 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5299 | mc.precharge = 0; |
5300 | } | |
5301 | /* | |
5302 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5303 | * we must uncharge here. | |
5304 | */ | |
5305 | if (mc.moved_charge) { | |
00501b53 | 5306 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5307 | mc.moved_charge = 0; |
4ffef5fe | 5308 | } |
483c30b5 DN |
5309 | /* we must fixup refcnts and charges */ |
5310 | if (mc.moved_swap) { | |
483c30b5 | 5311 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5312 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5313 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5314 | |
615d66c3 VD |
5315 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5316 | ||
05b84301 | 5317 | /* |
3e32cb2e JW |
5318 | * we charged both to->memory and to->memsw, so we |
5319 | * should uncharge to->memory. | |
05b84301 | 5320 | */ |
ce00a967 | 5321 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5322 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5323 | ||
615d66c3 VD |
5324 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
5325 | css_put_many(&mc.to->css, mc.moved_swap); | |
3e32cb2e | 5326 | |
483c30b5 DN |
5327 | mc.moved_swap = 0; |
5328 | } | |
dfe076b0 DN |
5329 | memcg_oom_recover(from); |
5330 | memcg_oom_recover(to); | |
5331 | wake_up_all(&mc.waitq); | |
5332 | } | |
5333 | ||
5334 | static void mem_cgroup_clear_mc(void) | |
5335 | { | |
264a0ae1 TH |
5336 | struct mm_struct *mm = mc.mm; |
5337 | ||
dfe076b0 DN |
5338 | /* |
5339 | * we must clear moving_task before waking up waiters at the end of | |
5340 | * task migration. | |
5341 | */ | |
5342 | mc.moving_task = NULL; | |
5343 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5344 | spin_lock(&mc.lock); |
4ffef5fe DN |
5345 | mc.from = NULL; |
5346 | mc.to = NULL; | |
264a0ae1 | 5347 | mc.mm = NULL; |
2bd9bb20 | 5348 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5349 | |
5350 | mmput(mm); | |
4ffef5fe DN |
5351 | } |
5352 | ||
1f7dd3e5 | 5353 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5354 | { |
1f7dd3e5 | 5355 | struct cgroup_subsys_state *css; |
eed67d75 | 5356 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5357 | struct mem_cgroup *from; |
4530eddb | 5358 | struct task_struct *leader, *p; |
9f2115f9 | 5359 | struct mm_struct *mm; |
1dfab5ab | 5360 | unsigned long move_flags; |
9f2115f9 | 5361 | int ret = 0; |
7dc74be0 | 5362 | |
1f7dd3e5 TH |
5363 | /* charge immigration isn't supported on the default hierarchy */ |
5364 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5365 | return 0; |
5366 | ||
4530eddb TH |
5367 | /* |
5368 | * Multi-process migrations only happen on the default hierarchy | |
5369 | * where charge immigration is not used. Perform charge | |
5370 | * immigration if @tset contains a leader and whine if there are | |
5371 | * multiple. | |
5372 | */ | |
5373 | p = NULL; | |
1f7dd3e5 | 5374 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5375 | WARN_ON_ONCE(p); |
5376 | p = leader; | |
1f7dd3e5 | 5377 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5378 | } |
5379 | if (!p) | |
5380 | return 0; | |
5381 | ||
1f7dd3e5 TH |
5382 | /* |
5383 | * We are now commited to this value whatever it is. Changes in this | |
5384 | * tunable will only affect upcoming migrations, not the current one. | |
5385 | * So we need to save it, and keep it going. | |
5386 | */ | |
5387 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5388 | if (!move_flags) | |
5389 | return 0; | |
5390 | ||
9f2115f9 TH |
5391 | from = mem_cgroup_from_task(p); |
5392 | ||
5393 | VM_BUG_ON(from == memcg); | |
5394 | ||
5395 | mm = get_task_mm(p); | |
5396 | if (!mm) | |
5397 | return 0; | |
5398 | /* We move charges only when we move a owner of the mm */ | |
5399 | if (mm->owner == p) { | |
5400 | VM_BUG_ON(mc.from); | |
5401 | VM_BUG_ON(mc.to); | |
5402 | VM_BUG_ON(mc.precharge); | |
5403 | VM_BUG_ON(mc.moved_charge); | |
5404 | VM_BUG_ON(mc.moved_swap); | |
5405 | ||
5406 | spin_lock(&mc.lock); | |
264a0ae1 | 5407 | mc.mm = mm; |
9f2115f9 TH |
5408 | mc.from = from; |
5409 | mc.to = memcg; | |
5410 | mc.flags = move_flags; | |
5411 | spin_unlock(&mc.lock); | |
5412 | /* We set mc.moving_task later */ | |
5413 | ||
5414 | ret = mem_cgroup_precharge_mc(mm); | |
5415 | if (ret) | |
5416 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5417 | } else { |
5418 | mmput(mm); | |
7dc74be0 DN |
5419 | } |
5420 | return ret; | |
5421 | } | |
5422 | ||
1f7dd3e5 | 5423 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5424 | { |
4e2f245d JW |
5425 | if (mc.to) |
5426 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5427 | } |
5428 | ||
4ffef5fe DN |
5429 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5430 | unsigned long addr, unsigned long end, | |
5431 | struct mm_walk *walk) | |
7dc74be0 | 5432 | { |
4ffef5fe | 5433 | int ret = 0; |
26bcd64a | 5434 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5435 | pte_t *pte; |
5436 | spinlock_t *ptl; | |
12724850 NH |
5437 | enum mc_target_type target_type; |
5438 | union mc_target target; | |
5439 | struct page *page; | |
4ffef5fe | 5440 | |
b6ec57f4 KS |
5441 | ptl = pmd_trans_huge_lock(pmd, vma); |
5442 | if (ptl) { | |
62ade86a | 5443 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5444 | spin_unlock(ptl); |
12724850 NH |
5445 | return 0; |
5446 | } | |
5447 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5448 | if (target_type == MC_TARGET_PAGE) { | |
5449 | page = target.page; | |
5450 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5451 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5452 | mc.from, mc.to)) { |
12724850 NH |
5453 | mc.precharge -= HPAGE_PMD_NR; |
5454 | mc.moved_charge += HPAGE_PMD_NR; | |
5455 | } | |
5456 | putback_lru_page(page); | |
5457 | } | |
5458 | put_page(page); | |
c733a828 JG |
5459 | } else if (target_type == MC_TARGET_DEVICE) { |
5460 | page = target.page; | |
5461 | if (!mem_cgroup_move_account(page, true, | |
5462 | mc.from, mc.to)) { | |
5463 | mc.precharge -= HPAGE_PMD_NR; | |
5464 | mc.moved_charge += HPAGE_PMD_NR; | |
5465 | } | |
5466 | put_page(page); | |
12724850 | 5467 | } |
bf929152 | 5468 | spin_unlock(ptl); |
1a5a9906 | 5469 | return 0; |
12724850 NH |
5470 | } |
5471 | ||
45f83cef AA |
5472 | if (pmd_trans_unstable(pmd)) |
5473 | return 0; | |
4ffef5fe DN |
5474 | retry: |
5475 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5476 | for (; addr != end; addr += PAGE_SIZE) { | |
5477 | pte_t ptent = *(pte++); | |
c733a828 | 5478 | bool device = false; |
02491447 | 5479 | swp_entry_t ent; |
4ffef5fe DN |
5480 | |
5481 | if (!mc.precharge) | |
5482 | break; | |
5483 | ||
8d32ff84 | 5484 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
5485 | case MC_TARGET_DEVICE: |
5486 | device = true; | |
5487 | /* fall through */ | |
4ffef5fe DN |
5488 | case MC_TARGET_PAGE: |
5489 | page = target.page; | |
53f9263b KS |
5490 | /* |
5491 | * We can have a part of the split pmd here. Moving it | |
5492 | * can be done but it would be too convoluted so simply | |
5493 | * ignore such a partial THP and keep it in original | |
5494 | * memcg. There should be somebody mapping the head. | |
5495 | */ | |
5496 | if (PageTransCompound(page)) | |
5497 | goto put; | |
c733a828 | 5498 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 5499 | goto put; |
f627c2f5 KS |
5500 | if (!mem_cgroup_move_account(page, false, |
5501 | mc.from, mc.to)) { | |
4ffef5fe | 5502 | mc.precharge--; |
854ffa8d DN |
5503 | /* we uncharge from mc.from later. */ |
5504 | mc.moved_charge++; | |
4ffef5fe | 5505 | } |
c733a828 JG |
5506 | if (!device) |
5507 | putback_lru_page(page); | |
8d32ff84 | 5508 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5509 | put_page(page); |
5510 | break; | |
02491447 DN |
5511 | case MC_TARGET_SWAP: |
5512 | ent = target.ent; | |
e91cbb42 | 5513 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5514 | mc.precharge--; |
483c30b5 DN |
5515 | /* we fixup refcnts and charges later. */ |
5516 | mc.moved_swap++; | |
5517 | } | |
02491447 | 5518 | break; |
4ffef5fe DN |
5519 | default: |
5520 | break; | |
5521 | } | |
5522 | } | |
5523 | pte_unmap_unlock(pte - 1, ptl); | |
5524 | cond_resched(); | |
5525 | ||
5526 | if (addr != end) { | |
5527 | /* | |
5528 | * We have consumed all precharges we got in can_attach(). | |
5529 | * We try charge one by one, but don't do any additional | |
5530 | * charges to mc.to if we have failed in charge once in attach() | |
5531 | * phase. | |
5532 | */ | |
854ffa8d | 5533 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5534 | if (!ret) |
5535 | goto retry; | |
5536 | } | |
5537 | ||
5538 | return ret; | |
5539 | } | |
5540 | ||
264a0ae1 | 5541 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 5542 | { |
26bcd64a NH |
5543 | struct mm_walk mem_cgroup_move_charge_walk = { |
5544 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
264a0ae1 | 5545 | .mm = mc.mm, |
26bcd64a | 5546 | }; |
4ffef5fe DN |
5547 | |
5548 | lru_add_drain_all(); | |
312722cb | 5549 | /* |
81f8c3a4 JW |
5550 | * Signal lock_page_memcg() to take the memcg's move_lock |
5551 | * while we're moving its pages to another memcg. Then wait | |
5552 | * for already started RCU-only updates to finish. | |
312722cb JW |
5553 | */ |
5554 | atomic_inc(&mc.from->moving_account); | |
5555 | synchronize_rcu(); | |
dfe076b0 | 5556 | retry: |
264a0ae1 | 5557 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
5558 | /* |
5559 | * Someone who are holding the mmap_sem might be waiting in | |
5560 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5561 | * and retry. Because we cancel precharges, we might not be able | |
5562 | * to move enough charges, but moving charge is a best-effort | |
5563 | * feature anyway, so it wouldn't be a big problem. | |
5564 | */ | |
5565 | __mem_cgroup_clear_mc(); | |
5566 | cond_resched(); | |
5567 | goto retry; | |
5568 | } | |
26bcd64a NH |
5569 | /* |
5570 | * When we have consumed all precharges and failed in doing | |
5571 | * additional charge, the page walk just aborts. | |
5572 | */ | |
0247f3f4 JM |
5573 | walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk); |
5574 | ||
264a0ae1 | 5575 | up_read(&mc.mm->mmap_sem); |
312722cb | 5576 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5577 | } |
5578 | ||
264a0ae1 | 5579 | static void mem_cgroup_move_task(void) |
67e465a7 | 5580 | { |
264a0ae1 TH |
5581 | if (mc.to) { |
5582 | mem_cgroup_move_charge(); | |
a433658c | 5583 | mem_cgroup_clear_mc(); |
264a0ae1 | 5584 | } |
67e465a7 | 5585 | } |
5cfb80a7 | 5586 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5587 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5588 | { |
5589 | return 0; | |
5590 | } | |
1f7dd3e5 | 5591 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5592 | { |
5593 | } | |
264a0ae1 | 5594 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
5595 | { |
5596 | } | |
5597 | #endif | |
67e465a7 | 5598 | |
f00baae7 TH |
5599 | /* |
5600 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5601 | * to verify whether we're attached to the default hierarchy on each mount |
5602 | * attempt. | |
f00baae7 | 5603 | */ |
eb95419b | 5604 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5605 | { |
5606 | /* | |
aa6ec29b | 5607 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5608 | * guarantees that @root doesn't have any children, so turning it |
5609 | * on for the root memcg is enough. | |
5610 | */ | |
9e10a130 | 5611 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5612 | root_mem_cgroup->use_hierarchy = true; |
5613 | else | |
5614 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5615 | } |
5616 | ||
677dc973 CD |
5617 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
5618 | { | |
5619 | if (value == PAGE_COUNTER_MAX) | |
5620 | seq_puts(m, "max\n"); | |
5621 | else | |
5622 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
5623 | ||
5624 | return 0; | |
5625 | } | |
5626 | ||
241994ed JW |
5627 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5628 | struct cftype *cft) | |
5629 | { | |
f5fc3c5d JW |
5630 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5631 | ||
5632 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5633 | } |
5634 | ||
bf8d5d52 RG |
5635 | static int memory_min_show(struct seq_file *m, void *v) |
5636 | { | |
677dc973 CD |
5637 | return seq_puts_memcg_tunable(m, |
5638 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
5639 | } |
5640 | ||
5641 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
5642 | char *buf, size_t nbytes, loff_t off) | |
5643 | { | |
5644 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5645 | unsigned long min; | |
5646 | int err; | |
5647 | ||
5648 | buf = strstrip(buf); | |
5649 | err = page_counter_memparse(buf, "max", &min); | |
5650 | if (err) | |
5651 | return err; | |
5652 | ||
5653 | page_counter_set_min(&memcg->memory, min); | |
5654 | ||
5655 | return nbytes; | |
5656 | } | |
5657 | ||
241994ed JW |
5658 | static int memory_low_show(struct seq_file *m, void *v) |
5659 | { | |
677dc973 CD |
5660 | return seq_puts_memcg_tunable(m, |
5661 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
5662 | } |
5663 | ||
5664 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5665 | char *buf, size_t nbytes, loff_t off) | |
5666 | { | |
5667 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5668 | unsigned long low; | |
5669 | int err; | |
5670 | ||
5671 | buf = strstrip(buf); | |
d2973697 | 5672 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5673 | if (err) |
5674 | return err; | |
5675 | ||
23067153 | 5676 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
5677 | |
5678 | return nbytes; | |
5679 | } | |
5680 | ||
5681 | static int memory_high_show(struct seq_file *m, void *v) | |
5682 | { | |
677dc973 | 5683 | return seq_puts_memcg_tunable(m, READ_ONCE(mem_cgroup_from_seq(m)->high)); |
241994ed JW |
5684 | } |
5685 | ||
5686 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5687 | char *buf, size_t nbytes, loff_t off) | |
5688 | { | |
5689 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 5690 | unsigned long nr_pages; |
241994ed JW |
5691 | unsigned long high; |
5692 | int err; | |
5693 | ||
5694 | buf = strstrip(buf); | |
d2973697 | 5695 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5696 | if (err) |
5697 | return err; | |
5698 | ||
5699 | memcg->high = high; | |
5700 | ||
588083bb JW |
5701 | nr_pages = page_counter_read(&memcg->memory); |
5702 | if (nr_pages > high) | |
5703 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5704 | GFP_KERNEL, true); | |
5705 | ||
2529bb3a | 5706 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5707 | return nbytes; |
5708 | } | |
5709 | ||
5710 | static int memory_max_show(struct seq_file *m, void *v) | |
5711 | { | |
677dc973 CD |
5712 | return seq_puts_memcg_tunable(m, |
5713 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
5714 | } |
5715 | ||
5716 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5717 | char *buf, size_t nbytes, loff_t off) | |
5718 | { | |
5719 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
5720 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5721 | bool drained = false; | |
241994ed JW |
5722 | unsigned long max; |
5723 | int err; | |
5724 | ||
5725 | buf = strstrip(buf); | |
d2973697 | 5726 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5727 | if (err) |
5728 | return err; | |
5729 | ||
bbec2e15 | 5730 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
5731 | |
5732 | for (;;) { | |
5733 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5734 | ||
5735 | if (nr_pages <= max) | |
5736 | break; | |
5737 | ||
5738 | if (signal_pending(current)) { | |
5739 | err = -EINTR; | |
5740 | break; | |
5741 | } | |
5742 | ||
5743 | if (!drained) { | |
5744 | drain_all_stock(memcg); | |
5745 | drained = true; | |
5746 | continue; | |
5747 | } | |
5748 | ||
5749 | if (nr_reclaims) { | |
5750 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5751 | GFP_KERNEL, true)) | |
5752 | nr_reclaims--; | |
5753 | continue; | |
5754 | } | |
5755 | ||
e27be240 | 5756 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
5757 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
5758 | break; | |
5759 | } | |
241994ed | 5760 | |
2529bb3a | 5761 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5762 | return nbytes; |
5763 | } | |
5764 | ||
1e577f97 SB |
5765 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
5766 | { | |
5767 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
5768 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
5769 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
5770 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
5771 | seq_printf(m, "oom_kill %lu\n", | |
5772 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
5773 | } | |
5774 | ||
241994ed JW |
5775 | static int memory_events_show(struct seq_file *m, void *v) |
5776 | { | |
aa9694bb | 5777 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 5778 | |
1e577f97 SB |
5779 | __memory_events_show(m, memcg->memory_events); |
5780 | return 0; | |
5781 | } | |
5782 | ||
5783 | static int memory_events_local_show(struct seq_file *m, void *v) | |
5784 | { | |
5785 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 5786 | |
1e577f97 | 5787 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
5788 | return 0; |
5789 | } | |
5790 | ||
587d9f72 JW |
5791 | static int memory_stat_show(struct seq_file *m, void *v) |
5792 | { | |
aa9694bb | 5793 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 5794 | char *buf; |
1ff9e6e1 | 5795 | |
c8713d0b JW |
5796 | buf = memory_stat_format(memcg); |
5797 | if (!buf) | |
5798 | return -ENOMEM; | |
5799 | seq_puts(m, buf); | |
5800 | kfree(buf); | |
587d9f72 JW |
5801 | return 0; |
5802 | } | |
5803 | ||
3d8b38eb RG |
5804 | static int memory_oom_group_show(struct seq_file *m, void *v) |
5805 | { | |
aa9694bb | 5806 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
5807 | |
5808 | seq_printf(m, "%d\n", memcg->oom_group); | |
5809 | ||
5810 | return 0; | |
5811 | } | |
5812 | ||
5813 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
5814 | char *buf, size_t nbytes, loff_t off) | |
5815 | { | |
5816 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5817 | int ret, oom_group; | |
5818 | ||
5819 | buf = strstrip(buf); | |
5820 | if (!buf) | |
5821 | return -EINVAL; | |
5822 | ||
5823 | ret = kstrtoint(buf, 0, &oom_group); | |
5824 | if (ret) | |
5825 | return ret; | |
5826 | ||
5827 | if (oom_group != 0 && oom_group != 1) | |
5828 | return -EINVAL; | |
5829 | ||
5830 | memcg->oom_group = oom_group; | |
5831 | ||
5832 | return nbytes; | |
5833 | } | |
5834 | ||
241994ed JW |
5835 | static struct cftype memory_files[] = { |
5836 | { | |
5837 | .name = "current", | |
f5fc3c5d | 5838 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5839 | .read_u64 = memory_current_read, |
5840 | }, | |
bf8d5d52 RG |
5841 | { |
5842 | .name = "min", | |
5843 | .flags = CFTYPE_NOT_ON_ROOT, | |
5844 | .seq_show = memory_min_show, | |
5845 | .write = memory_min_write, | |
5846 | }, | |
241994ed JW |
5847 | { |
5848 | .name = "low", | |
5849 | .flags = CFTYPE_NOT_ON_ROOT, | |
5850 | .seq_show = memory_low_show, | |
5851 | .write = memory_low_write, | |
5852 | }, | |
5853 | { | |
5854 | .name = "high", | |
5855 | .flags = CFTYPE_NOT_ON_ROOT, | |
5856 | .seq_show = memory_high_show, | |
5857 | .write = memory_high_write, | |
5858 | }, | |
5859 | { | |
5860 | .name = "max", | |
5861 | .flags = CFTYPE_NOT_ON_ROOT, | |
5862 | .seq_show = memory_max_show, | |
5863 | .write = memory_max_write, | |
5864 | }, | |
5865 | { | |
5866 | .name = "events", | |
5867 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5868 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5869 | .seq_show = memory_events_show, |
5870 | }, | |
1e577f97 SB |
5871 | { |
5872 | .name = "events.local", | |
5873 | .flags = CFTYPE_NOT_ON_ROOT, | |
5874 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
5875 | .seq_show = memory_events_local_show, | |
5876 | }, | |
587d9f72 JW |
5877 | { |
5878 | .name = "stat", | |
5879 | .flags = CFTYPE_NOT_ON_ROOT, | |
5880 | .seq_show = memory_stat_show, | |
5881 | }, | |
3d8b38eb RG |
5882 | { |
5883 | .name = "oom.group", | |
5884 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
5885 | .seq_show = memory_oom_group_show, | |
5886 | .write = memory_oom_group_write, | |
5887 | }, | |
241994ed JW |
5888 | { } /* terminate */ |
5889 | }; | |
5890 | ||
073219e9 | 5891 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5892 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5893 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5894 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5895 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5896 | .css_free = mem_cgroup_css_free, |
1ced953b | 5897 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5898 | .can_attach = mem_cgroup_can_attach, |
5899 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 5900 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5901 | .bind = mem_cgroup_bind, |
241994ed JW |
5902 | .dfl_cftypes = memory_files, |
5903 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5904 | .early_init = 0, |
8cdea7c0 | 5905 | }; |
c077719b | 5906 | |
241994ed | 5907 | /** |
bf8d5d52 | 5908 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 5909 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
5910 | * @memcg: the memory cgroup to check |
5911 | * | |
23067153 RG |
5912 | * WARNING: This function is not stateless! It can only be used as part |
5913 | * of a top-down tree iteration, not for isolated queries. | |
34c81057 | 5914 | * |
bf8d5d52 RG |
5915 | * Returns one of the following: |
5916 | * MEMCG_PROT_NONE: cgroup memory is not protected | |
5917 | * MEMCG_PROT_LOW: cgroup memory is protected as long there is | |
5918 | * an unprotected supply of reclaimable memory from other cgroups. | |
5919 | * MEMCG_PROT_MIN: cgroup memory is protected | |
34c81057 | 5920 | * |
bf8d5d52 | 5921 | * @root is exclusive; it is never protected when looked at directly |
34c81057 | 5922 | * |
bf8d5d52 RG |
5923 | * To provide a proper hierarchical behavior, effective memory.min/low values |
5924 | * are used. Below is the description of how effective memory.low is calculated. | |
5925 | * Effective memory.min values is calculated in the same way. | |
34c81057 | 5926 | * |
23067153 RG |
5927 | * Effective memory.low is always equal or less than the original memory.low. |
5928 | * If there is no memory.low overcommittment (which is always true for | |
5929 | * top-level memory cgroups), these two values are equal. | |
5930 | * Otherwise, it's a part of parent's effective memory.low, | |
5931 | * calculated as a cgroup's memory.low usage divided by sum of sibling's | |
5932 | * memory.low usages, where memory.low usage is the size of actually | |
5933 | * protected memory. | |
34c81057 | 5934 | * |
23067153 RG |
5935 | * low_usage |
5936 | * elow = min( memory.low, parent->elow * ------------------ ), | |
5937 | * siblings_low_usage | |
34c81057 | 5938 | * |
23067153 RG |
5939 | * | memory.current, if memory.current < memory.low |
5940 | * low_usage = | | |
82ede7ee | 5941 | * | 0, otherwise. |
34c81057 | 5942 | * |
23067153 RG |
5943 | * |
5944 | * Such definition of the effective memory.low provides the expected | |
5945 | * hierarchical behavior: parent's memory.low value is limiting | |
5946 | * children, unprotected memory is reclaimed first and cgroups, | |
5947 | * which are not using their guarantee do not affect actual memory | |
5948 | * distribution. | |
5949 | * | |
5950 | * For example, if there are memcgs A, A/B, A/C, A/D and A/E: | |
5951 | * | |
5952 | * A A/memory.low = 2G, A/memory.current = 6G | |
5953 | * //\\ | |
5954 | * BC DE B/memory.low = 3G B/memory.current = 2G | |
5955 | * C/memory.low = 1G C/memory.current = 2G | |
5956 | * D/memory.low = 0 D/memory.current = 2G | |
5957 | * E/memory.low = 10G E/memory.current = 0 | |
5958 | * | |
5959 | * and the memory pressure is applied, the following memory distribution | |
5960 | * is expected (approximately): | |
5961 | * | |
5962 | * A/memory.current = 2G | |
5963 | * | |
5964 | * B/memory.current = 1.3G | |
5965 | * C/memory.current = 0.6G | |
5966 | * D/memory.current = 0 | |
5967 | * E/memory.current = 0 | |
5968 | * | |
5969 | * These calculations require constant tracking of the actual low usages | |
bf8d5d52 RG |
5970 | * (see propagate_protected_usage()), as well as recursive calculation of |
5971 | * effective memory.low values. But as we do call mem_cgroup_protected() | |
23067153 RG |
5972 | * path for each memory cgroup top-down from the reclaim, |
5973 | * it's possible to optimize this part, and save calculated elow | |
5974 | * for next usage. This part is intentionally racy, but it's ok, | |
5975 | * as memory.low is a best-effort mechanism. | |
241994ed | 5976 | */ |
bf8d5d52 RG |
5977 | enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root, |
5978 | struct mem_cgroup *memcg) | |
241994ed | 5979 | { |
23067153 | 5980 | struct mem_cgroup *parent; |
bf8d5d52 RG |
5981 | unsigned long emin, parent_emin; |
5982 | unsigned long elow, parent_elow; | |
5983 | unsigned long usage; | |
23067153 | 5984 | |
241994ed | 5985 | if (mem_cgroup_disabled()) |
bf8d5d52 | 5986 | return MEMCG_PROT_NONE; |
241994ed | 5987 | |
34c81057 SC |
5988 | if (!root) |
5989 | root = root_mem_cgroup; | |
5990 | if (memcg == root) | |
bf8d5d52 | 5991 | return MEMCG_PROT_NONE; |
241994ed | 5992 | |
23067153 | 5993 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 RG |
5994 | if (!usage) |
5995 | return MEMCG_PROT_NONE; | |
5996 | ||
5997 | emin = memcg->memory.min; | |
5998 | elow = memcg->memory.low; | |
34c81057 | 5999 | |
bf8d5d52 | 6000 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
6001 | /* No parent means a non-hierarchical mode on v1 memcg */ |
6002 | if (!parent) | |
6003 | return MEMCG_PROT_NONE; | |
6004 | ||
23067153 RG |
6005 | if (parent == root) |
6006 | goto exit; | |
6007 | ||
bf8d5d52 RG |
6008 | parent_emin = READ_ONCE(parent->memory.emin); |
6009 | emin = min(emin, parent_emin); | |
6010 | if (emin && parent_emin) { | |
6011 | unsigned long min_usage, siblings_min_usage; | |
6012 | ||
6013 | min_usage = min(usage, memcg->memory.min); | |
6014 | siblings_min_usage = atomic_long_read( | |
6015 | &parent->memory.children_min_usage); | |
6016 | ||
6017 | if (min_usage && siblings_min_usage) | |
6018 | emin = min(emin, parent_emin * min_usage / | |
6019 | siblings_min_usage); | |
6020 | } | |
6021 | ||
23067153 RG |
6022 | parent_elow = READ_ONCE(parent->memory.elow); |
6023 | elow = min(elow, parent_elow); | |
bf8d5d52 RG |
6024 | if (elow && parent_elow) { |
6025 | unsigned long low_usage, siblings_low_usage; | |
23067153 | 6026 | |
bf8d5d52 RG |
6027 | low_usage = min(usage, memcg->memory.low); |
6028 | siblings_low_usage = atomic_long_read( | |
6029 | &parent->memory.children_low_usage); | |
23067153 | 6030 | |
bf8d5d52 RG |
6031 | if (low_usage && siblings_low_usage) |
6032 | elow = min(elow, parent_elow * low_usage / | |
6033 | siblings_low_usage); | |
6034 | } | |
23067153 | 6035 | |
23067153 | 6036 | exit: |
bf8d5d52 | 6037 | memcg->memory.emin = emin; |
23067153 | 6038 | memcg->memory.elow = elow; |
bf8d5d52 RG |
6039 | |
6040 | if (usage <= emin) | |
6041 | return MEMCG_PROT_MIN; | |
6042 | else if (usage <= elow) | |
6043 | return MEMCG_PROT_LOW; | |
6044 | else | |
6045 | return MEMCG_PROT_NONE; | |
241994ed JW |
6046 | } |
6047 | ||
00501b53 JW |
6048 | /** |
6049 | * mem_cgroup_try_charge - try charging a page | |
6050 | * @page: page to charge | |
6051 | * @mm: mm context of the victim | |
6052 | * @gfp_mask: reclaim mode | |
6053 | * @memcgp: charged memcg return | |
25843c2b | 6054 | * @compound: charge the page as compound or small page |
00501b53 JW |
6055 | * |
6056 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6057 | * pages according to @gfp_mask if necessary. | |
6058 | * | |
6059 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
6060 | * Otherwise, an error code is returned. | |
6061 | * | |
6062 | * After page->mapping has been set up, the caller must finalize the | |
6063 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
6064 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
6065 | */ | |
6066 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
6067 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
6068 | bool compound) | |
00501b53 JW |
6069 | { |
6070 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 6071 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6072 | int ret = 0; |
6073 | ||
6074 | if (mem_cgroup_disabled()) | |
6075 | goto out; | |
6076 | ||
6077 | if (PageSwapCache(page)) { | |
00501b53 JW |
6078 | /* |
6079 | * Every swap fault against a single page tries to charge the | |
6080 | * page, bail as early as possible. shmem_unuse() encounters | |
6081 | * already charged pages, too. The USED bit is protected by | |
6082 | * the page lock, which serializes swap cache removal, which | |
6083 | * in turn serializes uncharging. | |
6084 | */ | |
e993d905 | 6085 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 6086 | if (compound_head(page)->mem_cgroup) |
00501b53 | 6087 | goto out; |
e993d905 | 6088 | |
37e84351 | 6089 | if (do_swap_account) { |
e993d905 VD |
6090 | swp_entry_t ent = { .val = page_private(page), }; |
6091 | unsigned short id = lookup_swap_cgroup_id(ent); | |
6092 | ||
6093 | rcu_read_lock(); | |
6094 | memcg = mem_cgroup_from_id(id); | |
6095 | if (memcg && !css_tryget_online(&memcg->css)) | |
6096 | memcg = NULL; | |
6097 | rcu_read_unlock(); | |
6098 | } | |
00501b53 JW |
6099 | } |
6100 | ||
00501b53 JW |
6101 | if (!memcg) |
6102 | memcg = get_mem_cgroup_from_mm(mm); | |
6103 | ||
6104 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
6105 | ||
6106 | css_put(&memcg->css); | |
00501b53 JW |
6107 | out: |
6108 | *memcgp = memcg; | |
6109 | return ret; | |
6110 | } | |
6111 | ||
2cf85583 TH |
6112 | int mem_cgroup_try_charge_delay(struct page *page, struct mm_struct *mm, |
6113 | gfp_t gfp_mask, struct mem_cgroup **memcgp, | |
6114 | bool compound) | |
6115 | { | |
6116 | struct mem_cgroup *memcg; | |
6117 | int ret; | |
6118 | ||
6119 | ret = mem_cgroup_try_charge(page, mm, gfp_mask, memcgp, compound); | |
6120 | memcg = *memcgp; | |
6121 | mem_cgroup_throttle_swaprate(memcg, page_to_nid(page), gfp_mask); | |
6122 | return ret; | |
6123 | } | |
6124 | ||
00501b53 JW |
6125 | /** |
6126 | * mem_cgroup_commit_charge - commit a page charge | |
6127 | * @page: page to charge | |
6128 | * @memcg: memcg to charge the page to | |
6129 | * @lrucare: page might be on LRU already | |
25843c2b | 6130 | * @compound: charge the page as compound or small page |
00501b53 JW |
6131 | * |
6132 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
6133 | * after page->mapping has been set up. This must happen atomically | |
6134 | * as part of the page instantiation, i.e. under the page table lock | |
6135 | * for anonymous pages, under the page lock for page and swap cache. | |
6136 | * | |
6137 | * In addition, the page must not be on the LRU during the commit, to | |
6138 | * prevent racing with task migration. If it might be, use @lrucare. | |
6139 | * | |
6140 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
6141 | */ | |
6142 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 6143 | bool lrucare, bool compound) |
00501b53 | 6144 | { |
f627c2f5 | 6145 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6146 | |
6147 | VM_BUG_ON_PAGE(!page->mapping, page); | |
6148 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
6149 | ||
6150 | if (mem_cgroup_disabled()) | |
6151 | return; | |
6152 | /* | |
6153 | * Swap faults will attempt to charge the same page multiple | |
6154 | * times. But reuse_swap_page() might have removed the page | |
6155 | * from swapcache already, so we can't check PageSwapCache(). | |
6156 | */ | |
6157 | if (!memcg) | |
6158 | return; | |
6159 | ||
6abb5a86 JW |
6160 | commit_charge(page, memcg, lrucare); |
6161 | ||
6abb5a86 | 6162 | local_irq_disable(); |
f627c2f5 | 6163 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
6164 | memcg_check_events(memcg, page); |
6165 | local_irq_enable(); | |
00501b53 | 6166 | |
7941d214 | 6167 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
6168 | swp_entry_t entry = { .val = page_private(page) }; |
6169 | /* | |
6170 | * The swap entry might not get freed for a long time, | |
6171 | * let's not wait for it. The page already received a | |
6172 | * memory+swap charge, drop the swap entry duplicate. | |
6173 | */ | |
38d8b4e6 | 6174 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 JW |
6175 | } |
6176 | } | |
6177 | ||
6178 | /** | |
6179 | * mem_cgroup_cancel_charge - cancel a page charge | |
6180 | * @page: page to charge | |
6181 | * @memcg: memcg to charge the page to | |
25843c2b | 6182 | * @compound: charge the page as compound or small page |
00501b53 JW |
6183 | * |
6184 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
6185 | */ | |
f627c2f5 KS |
6186 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
6187 | bool compound) | |
00501b53 | 6188 | { |
f627c2f5 | 6189 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6190 | |
6191 | if (mem_cgroup_disabled()) | |
6192 | return; | |
6193 | /* | |
6194 | * Swap faults will attempt to charge the same page multiple | |
6195 | * times. But reuse_swap_page() might have removed the page | |
6196 | * from swapcache already, so we can't check PageSwapCache(). | |
6197 | */ | |
6198 | if (!memcg) | |
6199 | return; | |
6200 | ||
00501b53 JW |
6201 | cancel_charge(memcg, nr_pages); |
6202 | } | |
6203 | ||
a9d5adee JG |
6204 | struct uncharge_gather { |
6205 | struct mem_cgroup *memcg; | |
6206 | unsigned long pgpgout; | |
6207 | unsigned long nr_anon; | |
6208 | unsigned long nr_file; | |
6209 | unsigned long nr_kmem; | |
6210 | unsigned long nr_huge; | |
6211 | unsigned long nr_shmem; | |
6212 | struct page *dummy_page; | |
6213 | }; | |
6214 | ||
6215 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6216 | { |
a9d5adee JG |
6217 | memset(ug, 0, sizeof(*ug)); |
6218 | } | |
6219 | ||
6220 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6221 | { | |
6222 | unsigned long nr_pages = ug->nr_anon + ug->nr_file + ug->nr_kmem; | |
747db954 JW |
6223 | unsigned long flags; |
6224 | ||
a9d5adee JG |
6225 | if (!mem_cgroup_is_root(ug->memcg)) { |
6226 | page_counter_uncharge(&ug->memcg->memory, nr_pages); | |
7941d214 | 6227 | if (do_memsw_account()) |
a9d5adee JG |
6228 | page_counter_uncharge(&ug->memcg->memsw, nr_pages); |
6229 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) | |
6230 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6231 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6232 | } |
747db954 JW |
6233 | |
6234 | local_irq_save(flags); | |
c9019e9b JW |
6235 | __mod_memcg_state(ug->memcg, MEMCG_RSS, -ug->nr_anon); |
6236 | __mod_memcg_state(ug->memcg, MEMCG_CACHE, -ug->nr_file); | |
6237 | __mod_memcg_state(ug->memcg, MEMCG_RSS_HUGE, -ug->nr_huge); | |
6238 | __mod_memcg_state(ug->memcg, NR_SHMEM, -ug->nr_shmem); | |
6239 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); | |
871789d4 | 6240 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, nr_pages); |
a9d5adee | 6241 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6242 | local_irq_restore(flags); |
e8ea14cc | 6243 | |
a9d5adee JG |
6244 | if (!mem_cgroup_is_root(ug->memcg)) |
6245 | css_put_many(&ug->memcg->css, nr_pages); | |
6246 | } | |
6247 | ||
6248 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6249 | { | |
6250 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
3f2eb028 JG |
6251 | VM_BUG_ON_PAGE(page_count(page) && !is_zone_device_page(page) && |
6252 | !PageHWPoison(page) , page); | |
a9d5adee JG |
6253 | |
6254 | if (!page->mem_cgroup) | |
6255 | return; | |
6256 | ||
6257 | /* | |
6258 | * Nobody should be changing or seriously looking at | |
6259 | * page->mem_cgroup at this point, we have fully | |
6260 | * exclusive access to the page. | |
6261 | */ | |
6262 | ||
6263 | if (ug->memcg != page->mem_cgroup) { | |
6264 | if (ug->memcg) { | |
6265 | uncharge_batch(ug); | |
6266 | uncharge_gather_clear(ug); | |
6267 | } | |
6268 | ug->memcg = page->mem_cgroup; | |
6269 | } | |
6270 | ||
6271 | if (!PageKmemcg(page)) { | |
6272 | unsigned int nr_pages = 1; | |
6273 | ||
6274 | if (PageTransHuge(page)) { | |
6275 | nr_pages <<= compound_order(page); | |
6276 | ug->nr_huge += nr_pages; | |
6277 | } | |
6278 | if (PageAnon(page)) | |
6279 | ug->nr_anon += nr_pages; | |
6280 | else { | |
6281 | ug->nr_file += nr_pages; | |
6282 | if (PageSwapBacked(page)) | |
6283 | ug->nr_shmem += nr_pages; | |
6284 | } | |
6285 | ug->pgpgout++; | |
6286 | } else { | |
6287 | ug->nr_kmem += 1 << compound_order(page); | |
6288 | __ClearPageKmemcg(page); | |
6289 | } | |
6290 | ||
6291 | ug->dummy_page = page; | |
6292 | page->mem_cgroup = NULL; | |
747db954 JW |
6293 | } |
6294 | ||
6295 | static void uncharge_list(struct list_head *page_list) | |
6296 | { | |
a9d5adee | 6297 | struct uncharge_gather ug; |
747db954 | 6298 | struct list_head *next; |
a9d5adee JG |
6299 | |
6300 | uncharge_gather_clear(&ug); | |
747db954 | 6301 | |
8b592656 JW |
6302 | /* |
6303 | * Note that the list can be a single page->lru; hence the | |
6304 | * do-while loop instead of a simple list_for_each_entry(). | |
6305 | */ | |
747db954 JW |
6306 | next = page_list->next; |
6307 | do { | |
a9d5adee JG |
6308 | struct page *page; |
6309 | ||
747db954 JW |
6310 | page = list_entry(next, struct page, lru); |
6311 | next = page->lru.next; | |
6312 | ||
a9d5adee | 6313 | uncharge_page(page, &ug); |
747db954 JW |
6314 | } while (next != page_list); |
6315 | ||
a9d5adee JG |
6316 | if (ug.memcg) |
6317 | uncharge_batch(&ug); | |
747db954 JW |
6318 | } |
6319 | ||
0a31bc97 JW |
6320 | /** |
6321 | * mem_cgroup_uncharge - uncharge a page | |
6322 | * @page: page to uncharge | |
6323 | * | |
6324 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
6325 | * mem_cgroup_commit_charge(). | |
6326 | */ | |
6327 | void mem_cgroup_uncharge(struct page *page) | |
6328 | { | |
a9d5adee JG |
6329 | struct uncharge_gather ug; |
6330 | ||
0a31bc97 JW |
6331 | if (mem_cgroup_disabled()) |
6332 | return; | |
6333 | ||
747db954 | 6334 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6335 | if (!page->mem_cgroup) |
0a31bc97 JW |
6336 | return; |
6337 | ||
a9d5adee JG |
6338 | uncharge_gather_clear(&ug); |
6339 | uncharge_page(page, &ug); | |
6340 | uncharge_batch(&ug); | |
747db954 | 6341 | } |
0a31bc97 | 6342 | |
747db954 JW |
6343 | /** |
6344 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6345 | * @page_list: list of pages to uncharge | |
6346 | * | |
6347 | * Uncharge a list of pages previously charged with | |
6348 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
6349 | */ | |
6350 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6351 | { | |
6352 | if (mem_cgroup_disabled()) | |
6353 | return; | |
0a31bc97 | 6354 | |
747db954 JW |
6355 | if (!list_empty(page_list)) |
6356 | uncharge_list(page_list); | |
0a31bc97 JW |
6357 | } |
6358 | ||
6359 | /** | |
6a93ca8f JW |
6360 | * mem_cgroup_migrate - charge a page's replacement |
6361 | * @oldpage: currently circulating page | |
6362 | * @newpage: replacement page | |
0a31bc97 | 6363 | * |
6a93ca8f JW |
6364 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6365 | * be uncharged upon free. | |
0a31bc97 JW |
6366 | * |
6367 | * Both pages must be locked, @newpage->mapping must be set up. | |
6368 | */ | |
6a93ca8f | 6369 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6370 | { |
29833315 | 6371 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
6372 | unsigned int nr_pages; |
6373 | bool compound; | |
d93c4130 | 6374 | unsigned long flags; |
0a31bc97 JW |
6375 | |
6376 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6377 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6378 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6379 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6380 | newpage); | |
0a31bc97 JW |
6381 | |
6382 | if (mem_cgroup_disabled()) | |
6383 | return; | |
6384 | ||
6385 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6386 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6387 | return; |
6388 | ||
45637bab | 6389 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6390 | memcg = oldpage->mem_cgroup; |
29833315 | 6391 | if (!memcg) |
0a31bc97 JW |
6392 | return; |
6393 | ||
44b7a8d3 JW |
6394 | /* Force-charge the new page. The old one will be freed soon */ |
6395 | compound = PageTransHuge(newpage); | |
6396 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
6397 | ||
6398 | page_counter_charge(&memcg->memory, nr_pages); | |
6399 | if (do_memsw_account()) | |
6400 | page_counter_charge(&memcg->memsw, nr_pages); | |
6401 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 6402 | |
9cf7666a | 6403 | commit_charge(newpage, memcg, false); |
44b7a8d3 | 6404 | |
d93c4130 | 6405 | local_irq_save(flags); |
44b7a8d3 JW |
6406 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
6407 | memcg_check_events(memcg, newpage); | |
d93c4130 | 6408 | local_irq_restore(flags); |
0a31bc97 JW |
6409 | } |
6410 | ||
ef12947c | 6411 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6412 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6413 | ||
2d758073 | 6414 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6415 | { |
6416 | struct mem_cgroup *memcg; | |
6417 | ||
2d758073 JW |
6418 | if (!mem_cgroup_sockets_enabled) |
6419 | return; | |
6420 | ||
edbe69ef RG |
6421 | /* |
6422 | * Socket cloning can throw us here with sk_memcg already | |
6423 | * filled. It won't however, necessarily happen from | |
6424 | * process context. So the test for root memcg given | |
6425 | * the current task's memcg won't help us in this case. | |
6426 | * | |
6427 | * Respecting the original socket's memcg is a better | |
6428 | * decision in this case. | |
6429 | */ | |
6430 | if (sk->sk_memcg) { | |
6431 | css_get(&sk->sk_memcg->css); | |
6432 | return; | |
6433 | } | |
6434 | ||
11092087 JW |
6435 | rcu_read_lock(); |
6436 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6437 | if (memcg == root_mem_cgroup) |
6438 | goto out; | |
0db15298 | 6439 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6440 | goto out; |
f7e1cb6e | 6441 | if (css_tryget_online(&memcg->css)) |
11092087 | 6442 | sk->sk_memcg = memcg; |
f7e1cb6e | 6443 | out: |
11092087 JW |
6444 | rcu_read_unlock(); |
6445 | } | |
11092087 | 6446 | |
2d758073 | 6447 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6448 | { |
2d758073 JW |
6449 | if (sk->sk_memcg) |
6450 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6451 | } |
6452 | ||
6453 | /** | |
6454 | * mem_cgroup_charge_skmem - charge socket memory | |
6455 | * @memcg: memcg to charge | |
6456 | * @nr_pages: number of pages to charge | |
6457 | * | |
6458 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6459 | * @memcg's configured limit, %false if the charge had to be forced. | |
6460 | */ | |
6461 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6462 | { | |
f7e1cb6e | 6463 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6464 | |
f7e1cb6e | 6465 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6466 | struct page_counter *fail; |
f7e1cb6e | 6467 | |
0db15298 JW |
6468 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6469 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6470 | return true; |
6471 | } | |
0db15298 JW |
6472 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6473 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6474 | return false; |
11092087 | 6475 | } |
d886f4e4 | 6476 | |
f7e1cb6e JW |
6477 | /* Don't block in the packet receive path */ |
6478 | if (in_softirq()) | |
6479 | gfp_mask = GFP_NOWAIT; | |
6480 | ||
c9019e9b | 6481 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6482 | |
f7e1cb6e JW |
6483 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6484 | return true; | |
6485 | ||
6486 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6487 | return false; |
6488 | } | |
6489 | ||
6490 | /** | |
6491 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
6492 | * @memcg: memcg to uncharge |
6493 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
6494 | */ |
6495 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6496 | { | |
f7e1cb6e | 6497 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6498 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
6499 | return; |
6500 | } | |
d886f4e4 | 6501 | |
c9019e9b | 6502 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 6503 | |
475d0487 | 6504 | refill_stock(memcg, nr_pages); |
11092087 JW |
6505 | } |
6506 | ||
f7e1cb6e JW |
6507 | static int __init cgroup_memory(char *s) |
6508 | { | |
6509 | char *token; | |
6510 | ||
6511 | while ((token = strsep(&s, ",")) != NULL) { | |
6512 | if (!*token) | |
6513 | continue; | |
6514 | if (!strcmp(token, "nosocket")) | |
6515 | cgroup_memory_nosocket = true; | |
04823c83 VD |
6516 | if (!strcmp(token, "nokmem")) |
6517 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
6518 | } |
6519 | return 0; | |
6520 | } | |
6521 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 6522 | |
2d11085e | 6523 | /* |
1081312f MH |
6524 | * subsys_initcall() for memory controller. |
6525 | * | |
308167fc SAS |
6526 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
6527 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
6528 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
6529 | * should be initialized from here. | |
2d11085e MH |
6530 | */ |
6531 | static int __init mem_cgroup_init(void) | |
6532 | { | |
95a045f6 JW |
6533 | int cpu, node; |
6534 | ||
84c07d11 | 6535 | #ifdef CONFIG_MEMCG_KMEM |
13583c3d VD |
6536 | /* |
6537 | * Kmem cache creation is mostly done with the slab_mutex held, | |
17cc4dfe TH |
6538 | * so use a workqueue with limited concurrency to avoid stalling |
6539 | * all worker threads in case lots of cgroups are created and | |
6540 | * destroyed simultaneously. | |
13583c3d | 6541 | */ |
17cc4dfe TH |
6542 | memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1); |
6543 | BUG_ON(!memcg_kmem_cache_wq); | |
13583c3d VD |
6544 | #endif |
6545 | ||
308167fc SAS |
6546 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
6547 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
6548 | |
6549 | for_each_possible_cpu(cpu) | |
6550 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
6551 | drain_local_stock); | |
6552 | ||
6553 | for_each_node(node) { | |
6554 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
6555 | |
6556 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
6557 | node_online(node) ? node : NUMA_NO_NODE); | |
6558 | ||
ef8f2327 | 6559 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 6560 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 6561 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
6562 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
6563 | } | |
6564 | ||
2d11085e MH |
6565 | return 0; |
6566 | } | |
6567 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
6568 | |
6569 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
6570 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
6571 | { | |
1c2d479a | 6572 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
6573 | /* |
6574 | * The root cgroup cannot be destroyed, so it's refcount must | |
6575 | * always be >= 1. | |
6576 | */ | |
6577 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
6578 | VM_BUG_ON(1); | |
6579 | break; | |
6580 | } | |
6581 | memcg = parent_mem_cgroup(memcg); | |
6582 | if (!memcg) | |
6583 | memcg = root_mem_cgroup; | |
6584 | } | |
6585 | return memcg; | |
6586 | } | |
6587 | ||
21afa38e JW |
6588 | /** |
6589 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
6590 | * @page: page whose memsw charge to transfer | |
6591 | * @entry: swap entry to move the charge to | |
6592 | * | |
6593 | * Transfer the memsw charge of @page to @entry. | |
6594 | */ | |
6595 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
6596 | { | |
1f47b61f | 6597 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 6598 | unsigned int nr_entries; |
21afa38e JW |
6599 | unsigned short oldid; |
6600 | ||
6601 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6602 | VM_BUG_ON_PAGE(page_count(page), page); | |
6603 | ||
7941d214 | 6604 | if (!do_memsw_account()) |
21afa38e JW |
6605 | return; |
6606 | ||
6607 | memcg = page->mem_cgroup; | |
6608 | ||
6609 | /* Readahead page, never charged */ | |
6610 | if (!memcg) | |
6611 | return; | |
6612 | ||
1f47b61f VD |
6613 | /* |
6614 | * In case the memcg owning these pages has been offlined and doesn't | |
6615 | * have an ID allocated to it anymore, charge the closest online | |
6616 | * ancestor for the swap instead and transfer the memory+swap charge. | |
6617 | */ | |
6618 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
d6810d73 HY |
6619 | nr_entries = hpage_nr_pages(page); |
6620 | /* Get references for the tail pages, too */ | |
6621 | if (nr_entries > 1) | |
6622 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
6623 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
6624 | nr_entries); | |
21afa38e | 6625 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6626 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
6627 | |
6628 | page->mem_cgroup = NULL; | |
6629 | ||
6630 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 6631 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 6632 | |
1f47b61f VD |
6633 | if (memcg != swap_memcg) { |
6634 | if (!mem_cgroup_is_root(swap_memcg)) | |
d6810d73 HY |
6635 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
6636 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
6637 | } |
6638 | ||
ce9ce665 SAS |
6639 | /* |
6640 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 6641 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 6642 | * important here to have the interrupts disabled because it is the |
b93b0163 | 6643 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
6644 | */ |
6645 | VM_BUG_ON(!irqs_disabled()); | |
d6810d73 HY |
6646 | mem_cgroup_charge_statistics(memcg, page, PageTransHuge(page), |
6647 | -nr_entries); | |
21afa38e | 6648 | memcg_check_events(memcg, page); |
73f576c0 JW |
6649 | |
6650 | if (!mem_cgroup_is_root(memcg)) | |
d08afa14 | 6651 | css_put_many(&memcg->css, nr_entries); |
21afa38e JW |
6652 | } |
6653 | ||
38d8b4e6 HY |
6654 | /** |
6655 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
6656 | * @page: page being added to swap |
6657 | * @entry: swap entry to charge | |
6658 | * | |
38d8b4e6 | 6659 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
6660 | * |
6661 | * Returns 0 on success, -ENOMEM on failure. | |
6662 | */ | |
6663 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
6664 | { | |
38d8b4e6 | 6665 | unsigned int nr_pages = hpage_nr_pages(page); |
37e84351 | 6666 | struct page_counter *counter; |
38d8b4e6 | 6667 | struct mem_cgroup *memcg; |
37e84351 VD |
6668 | unsigned short oldid; |
6669 | ||
6670 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
6671 | return 0; | |
6672 | ||
6673 | memcg = page->mem_cgroup; | |
6674 | ||
6675 | /* Readahead page, never charged */ | |
6676 | if (!memcg) | |
6677 | return 0; | |
6678 | ||
f3a53a3a TH |
6679 | if (!entry.val) { |
6680 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 6681 | return 0; |
f3a53a3a | 6682 | } |
bb98f2c5 | 6683 | |
1f47b61f VD |
6684 | memcg = mem_cgroup_id_get_online(memcg); |
6685 | ||
37e84351 | 6686 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 6687 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
6688 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
6689 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 6690 | mem_cgroup_id_put(memcg); |
37e84351 | 6691 | return -ENOMEM; |
1f47b61f | 6692 | } |
37e84351 | 6693 | |
38d8b4e6 HY |
6694 | /* Get references for the tail pages, too */ |
6695 | if (nr_pages > 1) | |
6696 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
6697 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 6698 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6699 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 6700 | |
37e84351 VD |
6701 | return 0; |
6702 | } | |
6703 | ||
21afa38e | 6704 | /** |
38d8b4e6 | 6705 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 6706 | * @entry: swap entry to uncharge |
38d8b4e6 | 6707 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 6708 | */ |
38d8b4e6 | 6709 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
6710 | { |
6711 | struct mem_cgroup *memcg; | |
6712 | unsigned short id; | |
6713 | ||
37e84351 | 6714 | if (!do_swap_account) |
21afa38e JW |
6715 | return; |
6716 | ||
38d8b4e6 | 6717 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 6718 | rcu_read_lock(); |
adbe427b | 6719 | memcg = mem_cgroup_from_id(id); |
21afa38e | 6720 | if (memcg) { |
37e84351 VD |
6721 | if (!mem_cgroup_is_root(memcg)) { |
6722 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
38d8b4e6 | 6723 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 6724 | else |
38d8b4e6 | 6725 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 6726 | } |
c9019e9b | 6727 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 6728 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
6729 | } |
6730 | rcu_read_unlock(); | |
6731 | } | |
6732 | ||
d8b38438 VD |
6733 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
6734 | { | |
6735 | long nr_swap_pages = get_nr_swap_pages(); | |
6736 | ||
6737 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6738 | return nr_swap_pages; | |
6739 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
6740 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 6741 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
6742 | page_counter_read(&memcg->swap)); |
6743 | return nr_swap_pages; | |
6744 | } | |
6745 | ||
5ccc5aba VD |
6746 | bool mem_cgroup_swap_full(struct page *page) |
6747 | { | |
6748 | struct mem_cgroup *memcg; | |
6749 | ||
6750 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
6751 | ||
6752 | if (vm_swap_full()) | |
6753 | return true; | |
6754 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6755 | return false; | |
6756 | ||
6757 | memcg = page->mem_cgroup; | |
6758 | if (!memcg) | |
6759 | return false; | |
6760 | ||
6761 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
bbec2e15 | 6762 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.max) |
5ccc5aba VD |
6763 | return true; |
6764 | ||
6765 | return false; | |
6766 | } | |
6767 | ||
21afa38e JW |
6768 | /* for remember boot option*/ |
6769 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
6770 | static int really_do_swap_account __initdata = 1; | |
6771 | #else | |
6772 | static int really_do_swap_account __initdata; | |
6773 | #endif | |
6774 | ||
6775 | static int __init enable_swap_account(char *s) | |
6776 | { | |
6777 | if (!strcmp(s, "1")) | |
6778 | really_do_swap_account = 1; | |
6779 | else if (!strcmp(s, "0")) | |
6780 | really_do_swap_account = 0; | |
6781 | return 1; | |
6782 | } | |
6783 | __setup("swapaccount=", enable_swap_account); | |
6784 | ||
37e84351 VD |
6785 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
6786 | struct cftype *cft) | |
6787 | { | |
6788 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6789 | ||
6790 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
6791 | } | |
6792 | ||
6793 | static int swap_max_show(struct seq_file *m, void *v) | |
6794 | { | |
677dc973 CD |
6795 | return seq_puts_memcg_tunable(m, |
6796 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
6797 | } |
6798 | ||
6799 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
6800 | char *buf, size_t nbytes, loff_t off) | |
6801 | { | |
6802 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6803 | unsigned long max; | |
6804 | int err; | |
6805 | ||
6806 | buf = strstrip(buf); | |
6807 | err = page_counter_memparse(buf, "max", &max); | |
6808 | if (err) | |
6809 | return err; | |
6810 | ||
be09102b | 6811 | xchg(&memcg->swap.max, max); |
37e84351 VD |
6812 | |
6813 | return nbytes; | |
6814 | } | |
6815 | ||
f3a53a3a TH |
6816 | static int swap_events_show(struct seq_file *m, void *v) |
6817 | { | |
aa9694bb | 6818 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a TH |
6819 | |
6820 | seq_printf(m, "max %lu\n", | |
6821 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
6822 | seq_printf(m, "fail %lu\n", | |
6823 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
6824 | ||
6825 | return 0; | |
6826 | } | |
6827 | ||
37e84351 VD |
6828 | static struct cftype swap_files[] = { |
6829 | { | |
6830 | .name = "swap.current", | |
6831 | .flags = CFTYPE_NOT_ON_ROOT, | |
6832 | .read_u64 = swap_current_read, | |
6833 | }, | |
6834 | { | |
6835 | .name = "swap.max", | |
6836 | .flags = CFTYPE_NOT_ON_ROOT, | |
6837 | .seq_show = swap_max_show, | |
6838 | .write = swap_max_write, | |
6839 | }, | |
f3a53a3a TH |
6840 | { |
6841 | .name = "swap.events", | |
6842 | .flags = CFTYPE_NOT_ON_ROOT, | |
6843 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
6844 | .seq_show = swap_events_show, | |
6845 | }, | |
37e84351 VD |
6846 | { } /* terminate */ |
6847 | }; | |
6848 | ||
21afa38e JW |
6849 | static struct cftype memsw_cgroup_files[] = { |
6850 | { | |
6851 | .name = "memsw.usage_in_bytes", | |
6852 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6853 | .read_u64 = mem_cgroup_read_u64, | |
6854 | }, | |
6855 | { | |
6856 | .name = "memsw.max_usage_in_bytes", | |
6857 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6858 | .write = mem_cgroup_reset, | |
6859 | .read_u64 = mem_cgroup_read_u64, | |
6860 | }, | |
6861 | { | |
6862 | .name = "memsw.limit_in_bytes", | |
6863 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6864 | .write = mem_cgroup_write, | |
6865 | .read_u64 = mem_cgroup_read_u64, | |
6866 | }, | |
6867 | { | |
6868 | .name = "memsw.failcnt", | |
6869 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6870 | .write = mem_cgroup_reset, | |
6871 | .read_u64 = mem_cgroup_read_u64, | |
6872 | }, | |
6873 | { }, /* terminate */ | |
6874 | }; | |
6875 | ||
6876 | static int __init mem_cgroup_swap_init(void) | |
6877 | { | |
6878 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6879 | do_swap_account = 1; | |
37e84351 VD |
6880 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
6881 | swap_files)); | |
21afa38e JW |
6882 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6883 | memsw_cgroup_files)); | |
6884 | } | |
6885 | return 0; | |
6886 | } | |
6887 | subsys_initcall(mem_cgroup_swap_init); | |
6888 | ||
6889 | #endif /* CONFIG_MEMCG_SWAP */ |