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(); | |
4d96ba35 RG |
489 | if (PageHead(page) && PageSlab(page)) |
490 | memcg = memcg_from_slab_page(page); | |
491 | else | |
492 | memcg = READ_ONCE(page->mem_cgroup); | |
2fc04524 VD |
493 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
494 | memcg = parent_mem_cgroup(memcg); | |
495 | if (memcg) | |
496 | ino = cgroup_ino(memcg->css.cgroup); | |
497 | rcu_read_unlock(); | |
498 | return ino; | |
499 | } | |
500 | ||
ef8f2327 MG |
501 | static struct mem_cgroup_per_node * |
502 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 503 | { |
97a6c37b | 504 | int nid = page_to_nid(page); |
f64c3f54 | 505 | |
ef8f2327 | 506 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
507 | } |
508 | ||
ef8f2327 MG |
509 | static struct mem_cgroup_tree_per_node * |
510 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 511 | { |
ef8f2327 | 512 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
513 | } |
514 | ||
ef8f2327 | 515 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
516 | soft_limit_tree_from_page(struct page *page) |
517 | { | |
518 | int nid = page_to_nid(page); | |
bb4cc1a8 | 519 | |
ef8f2327 | 520 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
521 | } |
522 | ||
ef8f2327 MG |
523 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
524 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 525 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
526 | { |
527 | struct rb_node **p = &mctz->rb_root.rb_node; | |
528 | struct rb_node *parent = NULL; | |
ef8f2327 | 529 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 530 | bool rightmost = true; |
bb4cc1a8 AM |
531 | |
532 | if (mz->on_tree) | |
533 | return; | |
534 | ||
535 | mz->usage_in_excess = new_usage_in_excess; | |
536 | if (!mz->usage_in_excess) | |
537 | return; | |
538 | while (*p) { | |
539 | parent = *p; | |
ef8f2327 | 540 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 541 | tree_node); |
fa90b2fd | 542 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 543 | p = &(*p)->rb_left; |
fa90b2fd DB |
544 | rightmost = false; |
545 | } | |
546 | ||
bb4cc1a8 AM |
547 | /* |
548 | * We can't avoid mem cgroups that are over their soft | |
549 | * limit by the same amount | |
550 | */ | |
551 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
552 | p = &(*p)->rb_right; | |
553 | } | |
fa90b2fd DB |
554 | |
555 | if (rightmost) | |
556 | mctz->rb_rightmost = &mz->tree_node; | |
557 | ||
bb4cc1a8 AM |
558 | rb_link_node(&mz->tree_node, parent, p); |
559 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
560 | mz->on_tree = true; | |
561 | } | |
562 | ||
ef8f2327 MG |
563 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
564 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
565 | { |
566 | if (!mz->on_tree) | |
567 | return; | |
fa90b2fd DB |
568 | |
569 | if (&mz->tree_node == mctz->rb_rightmost) | |
570 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
571 | ||
bb4cc1a8 AM |
572 | rb_erase(&mz->tree_node, &mctz->rb_root); |
573 | mz->on_tree = false; | |
574 | } | |
575 | ||
ef8f2327 MG |
576 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
577 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 578 | { |
0a31bc97 JW |
579 | unsigned long flags; |
580 | ||
581 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 582 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 583 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
584 | } |
585 | ||
3e32cb2e JW |
586 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
587 | { | |
588 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 589 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
590 | unsigned long excess = 0; |
591 | ||
592 | if (nr_pages > soft_limit) | |
593 | excess = nr_pages - soft_limit; | |
594 | ||
595 | return excess; | |
596 | } | |
bb4cc1a8 AM |
597 | |
598 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
599 | { | |
3e32cb2e | 600 | unsigned long excess; |
ef8f2327 MG |
601 | struct mem_cgroup_per_node *mz; |
602 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 603 | |
e231875b | 604 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
605 | if (!mctz) |
606 | return; | |
bb4cc1a8 AM |
607 | /* |
608 | * Necessary to update all ancestors when hierarchy is used. | |
609 | * because their event counter is not touched. | |
610 | */ | |
611 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 612 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 613 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
614 | /* |
615 | * We have to update the tree if mz is on RB-tree or | |
616 | * mem is over its softlimit. | |
617 | */ | |
618 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
619 | unsigned long flags; |
620 | ||
621 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
622 | /* if on-tree, remove it */ |
623 | if (mz->on_tree) | |
cf2c8127 | 624 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
625 | /* |
626 | * Insert again. mz->usage_in_excess will be updated. | |
627 | * If excess is 0, no tree ops. | |
628 | */ | |
cf2c8127 | 629 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 630 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
631 | } |
632 | } | |
633 | } | |
634 | ||
635 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
636 | { | |
ef8f2327 MG |
637 | struct mem_cgroup_tree_per_node *mctz; |
638 | struct mem_cgroup_per_node *mz; | |
639 | int nid; | |
bb4cc1a8 | 640 | |
e231875b | 641 | for_each_node(nid) { |
ef8f2327 MG |
642 | mz = mem_cgroup_nodeinfo(memcg, nid); |
643 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
644 | if (mctz) |
645 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
646 | } |
647 | } | |
648 | ||
ef8f2327 MG |
649 | static struct mem_cgroup_per_node * |
650 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 651 | { |
ef8f2327 | 652 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
653 | |
654 | retry: | |
655 | mz = NULL; | |
fa90b2fd | 656 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
657 | goto done; /* Nothing to reclaim from */ |
658 | ||
fa90b2fd DB |
659 | mz = rb_entry(mctz->rb_rightmost, |
660 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
661 | /* |
662 | * Remove the node now but someone else can add it back, | |
663 | * we will to add it back at the end of reclaim to its correct | |
664 | * position in the tree. | |
665 | */ | |
cf2c8127 | 666 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 667 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 668 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
669 | goto retry; |
670 | done: | |
671 | return mz; | |
672 | } | |
673 | ||
ef8f2327 MG |
674 | static struct mem_cgroup_per_node * |
675 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 676 | { |
ef8f2327 | 677 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 678 | |
0a31bc97 | 679 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 680 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 681 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
682 | return mz; |
683 | } | |
684 | ||
db9adbcb JW |
685 | /** |
686 | * __mod_memcg_state - update cgroup memory statistics | |
687 | * @memcg: the memory cgroup | |
688 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
689 | * @val: delta to add to the counter, can be negative | |
690 | */ | |
691 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
692 | { | |
693 | long x; | |
694 | ||
695 | if (mem_cgroup_disabled()) | |
696 | return; | |
697 | ||
815744d7 JW |
698 | __this_cpu_add(memcg->vmstats_local->stat[idx], val); |
699 | ||
db9adbcb JW |
700 | x = val + __this_cpu_read(memcg->vmstats_percpu->stat[idx]); |
701 | if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
702 | struct mem_cgroup *mi; |
703 | ||
42a30035 JW |
704 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
705 | atomic_long_add(x, &mi->vmstats[idx]); | |
db9adbcb JW |
706 | x = 0; |
707 | } | |
708 | __this_cpu_write(memcg->vmstats_percpu->stat[idx], x); | |
709 | } | |
710 | ||
42a30035 JW |
711 | static struct mem_cgroup_per_node * |
712 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
713 | { | |
714 | struct mem_cgroup *parent; | |
715 | ||
716 | parent = parent_mem_cgroup(pn->memcg); | |
717 | if (!parent) | |
718 | return NULL; | |
719 | return mem_cgroup_nodeinfo(parent, nid); | |
720 | } | |
721 | ||
db9adbcb JW |
722 | /** |
723 | * __mod_lruvec_state - update lruvec memory statistics | |
724 | * @lruvec: the lruvec | |
725 | * @idx: the stat item | |
726 | * @val: delta to add to the counter, can be negative | |
727 | * | |
728 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
729 | * function updates the all three counters that are affected by a | |
730 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
731 | */ | |
732 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
733 | int val) | |
734 | { | |
42a30035 | 735 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
db9adbcb | 736 | struct mem_cgroup_per_node *pn; |
42a30035 | 737 | struct mem_cgroup *memcg; |
db9adbcb JW |
738 | long x; |
739 | ||
740 | /* Update node */ | |
42a30035 | 741 | __mod_node_page_state(pgdat, idx, val); |
db9adbcb JW |
742 | |
743 | if (mem_cgroup_disabled()) | |
744 | return; | |
745 | ||
746 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); | |
42a30035 | 747 | memcg = pn->memcg; |
db9adbcb JW |
748 | |
749 | /* Update memcg */ | |
42a30035 | 750 | __mod_memcg_state(memcg, idx, val); |
db9adbcb JW |
751 | |
752 | /* Update lruvec */ | |
815744d7 JW |
753 | __this_cpu_add(pn->lruvec_stat_local->count[idx], val); |
754 | ||
db9adbcb JW |
755 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); |
756 | if (unlikely(abs(x) > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
757 | struct mem_cgroup_per_node *pi; |
758 | ||
42a30035 JW |
759 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) |
760 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
761 | x = 0; |
762 | } | |
763 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
764 | } | |
765 | ||
766 | /** | |
767 | * __count_memcg_events - account VM events in a cgroup | |
768 | * @memcg: the memory cgroup | |
769 | * @idx: the event item | |
770 | * @count: the number of events that occured | |
771 | */ | |
772 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
773 | unsigned long count) | |
774 | { | |
775 | unsigned long x; | |
776 | ||
777 | if (mem_cgroup_disabled()) | |
778 | return; | |
779 | ||
815744d7 JW |
780 | __this_cpu_add(memcg->vmstats_local->events[idx], count); |
781 | ||
db9adbcb JW |
782 | x = count + __this_cpu_read(memcg->vmstats_percpu->events[idx]); |
783 | if (unlikely(x > MEMCG_CHARGE_BATCH)) { | |
42a30035 JW |
784 | struct mem_cgroup *mi; |
785 | ||
42a30035 JW |
786 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
787 | atomic_long_add(x, &mi->vmevents[idx]); | |
db9adbcb JW |
788 | x = 0; |
789 | } | |
790 | __this_cpu_write(memcg->vmstats_percpu->events[idx], x); | |
791 | } | |
792 | ||
42a30035 | 793 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 794 | { |
871789d4 | 795 | return atomic_long_read(&memcg->vmevents[event]); |
e9f8974f JW |
796 | } |
797 | ||
42a30035 JW |
798 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
799 | { | |
815744d7 JW |
800 | long x = 0; |
801 | int cpu; | |
802 | ||
803 | for_each_possible_cpu(cpu) | |
804 | x += per_cpu(memcg->vmstats_local->events[event], cpu); | |
805 | return x; | |
42a30035 JW |
806 | } |
807 | ||
c0ff4b85 | 808 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 809 | struct page *page, |
f627c2f5 | 810 | bool compound, int nr_pages) |
d52aa412 | 811 | { |
b2402857 KH |
812 | /* |
813 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
814 | * counted as CACHE even if it's on ANON LRU. | |
815 | */ | |
0a31bc97 | 816 | if (PageAnon(page)) |
c9019e9b | 817 | __mod_memcg_state(memcg, MEMCG_RSS, nr_pages); |
9a4caf1e | 818 | else { |
c9019e9b | 819 | __mod_memcg_state(memcg, MEMCG_CACHE, nr_pages); |
9a4caf1e | 820 | if (PageSwapBacked(page)) |
c9019e9b | 821 | __mod_memcg_state(memcg, NR_SHMEM, nr_pages); |
9a4caf1e | 822 | } |
55e462b0 | 823 | |
f627c2f5 KS |
824 | if (compound) { |
825 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
c9019e9b | 826 | __mod_memcg_state(memcg, MEMCG_RSS_HUGE, nr_pages); |
f627c2f5 | 827 | } |
b070e65c | 828 | |
e401f176 KH |
829 | /* pagein of a big page is an event. So, ignore page size */ |
830 | if (nr_pages > 0) | |
c9019e9b | 831 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 832 | else { |
c9019e9b | 833 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
834 | nr_pages = -nr_pages; /* for event */ |
835 | } | |
e401f176 | 836 | |
871789d4 | 837 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
838 | } |
839 | ||
f53d7ce3 JW |
840 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
841 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
842 | { |
843 | unsigned long val, next; | |
844 | ||
871789d4 CD |
845 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
846 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 847 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 848 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
849 | switch (target) { |
850 | case MEM_CGROUP_TARGET_THRESH: | |
851 | next = val + THRESHOLDS_EVENTS_TARGET; | |
852 | break; | |
bb4cc1a8 AM |
853 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
854 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
855 | break; | |
f53d7ce3 JW |
856 | case MEM_CGROUP_TARGET_NUMAINFO: |
857 | next = val + NUMAINFO_EVENTS_TARGET; | |
858 | break; | |
859 | default: | |
860 | break; | |
861 | } | |
871789d4 | 862 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 863 | return true; |
7a159cc9 | 864 | } |
f53d7ce3 | 865 | return false; |
d2265e6f KH |
866 | } |
867 | ||
868 | /* | |
869 | * Check events in order. | |
870 | * | |
871 | */ | |
c0ff4b85 | 872 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
873 | { |
874 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
875 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
876 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 877 | bool do_softlimit; |
82b3f2a7 | 878 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 879 | |
bb4cc1a8 AM |
880 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
881 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
882 | #if MAX_NUMNODES > 1 |
883 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
884 | MEM_CGROUP_TARGET_NUMAINFO); | |
885 | #endif | |
c0ff4b85 | 886 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
887 | if (unlikely(do_softlimit)) |
888 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 889 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 890 | if (unlikely(do_numainfo)) |
c0ff4b85 | 891 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 892 | #endif |
0a31bc97 | 893 | } |
d2265e6f KH |
894 | } |
895 | ||
cf475ad2 | 896 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 897 | { |
31a78f23 BS |
898 | /* |
899 | * mm_update_next_owner() may clear mm->owner to NULL | |
900 | * if it races with swapoff, page migration, etc. | |
901 | * So this can be called with p == NULL. | |
902 | */ | |
903 | if (unlikely(!p)) | |
904 | return NULL; | |
905 | ||
073219e9 | 906 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 907 | } |
33398cf2 | 908 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 909 | |
d46eb14b SB |
910 | /** |
911 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
912 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
913 | * | |
914 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
915 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
916 | * returned. | |
917 | */ | |
918 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 919 | { |
d46eb14b SB |
920 | struct mem_cgroup *memcg; |
921 | ||
922 | if (mem_cgroup_disabled()) | |
923 | return NULL; | |
0b7f569e | 924 | |
54595fe2 KH |
925 | rcu_read_lock(); |
926 | do { | |
6f6acb00 MH |
927 | /* |
928 | * Page cache insertions can happen withou an | |
929 | * actual mm context, e.g. during disk probing | |
930 | * on boot, loopback IO, acct() writes etc. | |
931 | */ | |
932 | if (unlikely(!mm)) | |
df381975 | 933 | memcg = root_mem_cgroup; |
6f6acb00 MH |
934 | else { |
935 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
936 | if (unlikely(!memcg)) | |
937 | memcg = root_mem_cgroup; | |
938 | } | |
ec903c0c | 939 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 940 | rcu_read_unlock(); |
c0ff4b85 | 941 | return memcg; |
54595fe2 | 942 | } |
d46eb14b SB |
943 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
944 | ||
f745c6f5 SB |
945 | /** |
946 | * get_mem_cgroup_from_page: Obtain a reference on given page's memcg. | |
947 | * @page: page from which memcg should be extracted. | |
948 | * | |
949 | * Obtain a reference on page->memcg and returns it if successful. Otherwise | |
950 | * root_mem_cgroup is returned. | |
951 | */ | |
952 | struct mem_cgroup *get_mem_cgroup_from_page(struct page *page) | |
953 | { | |
954 | struct mem_cgroup *memcg = page->mem_cgroup; | |
955 | ||
956 | if (mem_cgroup_disabled()) | |
957 | return NULL; | |
958 | ||
959 | rcu_read_lock(); | |
960 | if (!memcg || !css_tryget_online(&memcg->css)) | |
961 | memcg = root_mem_cgroup; | |
962 | rcu_read_unlock(); | |
963 | return memcg; | |
964 | } | |
965 | EXPORT_SYMBOL(get_mem_cgroup_from_page); | |
966 | ||
d46eb14b SB |
967 | /** |
968 | * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg. | |
969 | */ | |
970 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) | |
971 | { | |
972 | if (unlikely(current->active_memcg)) { | |
973 | struct mem_cgroup *memcg = root_mem_cgroup; | |
974 | ||
975 | rcu_read_lock(); | |
976 | if (css_tryget_online(¤t->active_memcg->css)) | |
977 | memcg = current->active_memcg; | |
978 | rcu_read_unlock(); | |
979 | return memcg; | |
980 | } | |
981 | return get_mem_cgroup_from_mm(current->mm); | |
982 | } | |
54595fe2 | 983 | |
5660048c JW |
984 | /** |
985 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
986 | * @root: hierarchy root | |
987 | * @prev: previously returned memcg, NULL on first invocation | |
988 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
989 | * | |
990 | * Returns references to children of the hierarchy below @root, or | |
991 | * @root itself, or %NULL after a full round-trip. | |
992 | * | |
993 | * Caller must pass the return value in @prev on subsequent | |
994 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
995 | * to cancel a hierarchy walk before the round-trip is complete. | |
996 | * | |
b213b54f | 997 | * Reclaimers can specify a node and a priority level in @reclaim to |
5660048c | 998 | * divide up the memcgs in the hierarchy among all concurrent |
b213b54f | 999 | * reclaimers operating on the same node and priority. |
5660048c | 1000 | */ |
694fbc0f | 1001 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 1002 | struct mem_cgroup *prev, |
694fbc0f | 1003 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 1004 | { |
33398cf2 | 1005 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 1006 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 1007 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 1008 | struct mem_cgroup *pos = NULL; |
711d3d2c | 1009 | |
694fbc0f AM |
1010 | if (mem_cgroup_disabled()) |
1011 | return NULL; | |
5660048c | 1012 | |
9f3a0d09 JW |
1013 | if (!root) |
1014 | root = root_mem_cgroup; | |
7d74b06f | 1015 | |
9f3a0d09 | 1016 | if (prev && !reclaim) |
5ac8fb31 | 1017 | pos = prev; |
14067bb3 | 1018 | |
9f3a0d09 JW |
1019 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
1020 | if (prev) | |
5ac8fb31 | 1021 | goto out; |
694fbc0f | 1022 | return root; |
9f3a0d09 | 1023 | } |
14067bb3 | 1024 | |
542f85f9 | 1025 | rcu_read_lock(); |
5f578161 | 1026 | |
5ac8fb31 | 1027 | if (reclaim) { |
ef8f2327 | 1028 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 1029 | |
ef8f2327 | 1030 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
5ac8fb31 JW |
1031 | iter = &mz->iter[reclaim->priority]; |
1032 | ||
1033 | if (prev && reclaim->generation != iter->generation) | |
1034 | goto out_unlock; | |
1035 | ||
6df38689 | 1036 | while (1) { |
4db0c3c2 | 1037 | pos = READ_ONCE(iter->position); |
6df38689 VD |
1038 | if (!pos || css_tryget(&pos->css)) |
1039 | break; | |
5ac8fb31 | 1040 | /* |
6df38689 VD |
1041 | * css reference reached zero, so iter->position will |
1042 | * be cleared by ->css_released. However, we should not | |
1043 | * rely on this happening soon, because ->css_released | |
1044 | * is called from a work queue, and by busy-waiting we | |
1045 | * might block it. So we clear iter->position right | |
1046 | * away. | |
5ac8fb31 | 1047 | */ |
6df38689 VD |
1048 | (void)cmpxchg(&iter->position, pos, NULL); |
1049 | } | |
5ac8fb31 JW |
1050 | } |
1051 | ||
1052 | if (pos) | |
1053 | css = &pos->css; | |
1054 | ||
1055 | for (;;) { | |
1056 | css = css_next_descendant_pre(css, &root->css); | |
1057 | if (!css) { | |
1058 | /* | |
1059 | * Reclaimers share the hierarchy walk, and a | |
1060 | * new one might jump in right at the end of | |
1061 | * the hierarchy - make sure they see at least | |
1062 | * one group and restart from the beginning. | |
1063 | */ | |
1064 | if (!prev) | |
1065 | continue; | |
1066 | break; | |
527a5ec9 | 1067 | } |
7d74b06f | 1068 | |
5ac8fb31 JW |
1069 | /* |
1070 | * Verify the css and acquire a reference. The root | |
1071 | * is provided by the caller, so we know it's alive | |
1072 | * and kicking, and don't take an extra reference. | |
1073 | */ | |
1074 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1075 | |
5ac8fb31 JW |
1076 | if (css == &root->css) |
1077 | break; | |
14067bb3 | 1078 | |
0b8f73e1 JW |
1079 | if (css_tryget(css)) |
1080 | break; | |
9f3a0d09 | 1081 | |
5ac8fb31 | 1082 | memcg = NULL; |
9f3a0d09 | 1083 | } |
5ac8fb31 JW |
1084 | |
1085 | if (reclaim) { | |
5ac8fb31 | 1086 | /* |
6df38689 VD |
1087 | * The position could have already been updated by a competing |
1088 | * thread, so check that the value hasn't changed since we read | |
1089 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1090 | */ |
6df38689 VD |
1091 | (void)cmpxchg(&iter->position, pos, memcg); |
1092 | ||
5ac8fb31 JW |
1093 | if (pos) |
1094 | css_put(&pos->css); | |
1095 | ||
1096 | if (!memcg) | |
1097 | iter->generation++; | |
1098 | else if (!prev) | |
1099 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1100 | } |
5ac8fb31 | 1101 | |
542f85f9 MH |
1102 | out_unlock: |
1103 | rcu_read_unlock(); | |
5ac8fb31 | 1104 | out: |
c40046f3 MH |
1105 | if (prev && prev != root) |
1106 | css_put(&prev->css); | |
1107 | ||
9f3a0d09 | 1108 | return memcg; |
14067bb3 | 1109 | } |
7d74b06f | 1110 | |
5660048c JW |
1111 | /** |
1112 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1113 | * @root: hierarchy root | |
1114 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1115 | */ | |
1116 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1117 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1118 | { |
1119 | if (!root) | |
1120 | root = root_mem_cgroup; | |
1121 | if (prev && prev != root) | |
1122 | css_put(&prev->css); | |
1123 | } | |
7d74b06f | 1124 | |
6df38689 VD |
1125 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1126 | { | |
1127 | struct mem_cgroup *memcg = dead_memcg; | |
1128 | struct mem_cgroup_reclaim_iter *iter; | |
ef8f2327 MG |
1129 | struct mem_cgroup_per_node *mz; |
1130 | int nid; | |
6df38689 VD |
1131 | int i; |
1132 | ||
9f15bde6 | 1133 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
6df38689 | 1134 | for_each_node(nid) { |
ef8f2327 MG |
1135 | mz = mem_cgroup_nodeinfo(memcg, nid); |
1136 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
1137 | iter = &mz->iter[i]; | |
1138 | cmpxchg(&iter->position, | |
1139 | dead_memcg, NULL); | |
6df38689 VD |
1140 | } |
1141 | } | |
1142 | } | |
1143 | } | |
1144 | ||
7c5f64f8 VD |
1145 | /** |
1146 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1147 | * @memcg: hierarchy root | |
1148 | * @fn: function to call for each task | |
1149 | * @arg: argument passed to @fn | |
1150 | * | |
1151 | * This function iterates over tasks attached to @memcg or to any of its | |
1152 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1153 | * value, the function breaks the iteration loop and returns the value. | |
1154 | * Otherwise, it will iterate over all tasks and return 0. | |
1155 | * | |
1156 | * This function must not be called for the root memory cgroup. | |
1157 | */ | |
1158 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1159 | int (*fn)(struct task_struct *, void *), void *arg) | |
1160 | { | |
1161 | struct mem_cgroup *iter; | |
1162 | int ret = 0; | |
1163 | ||
1164 | BUG_ON(memcg == root_mem_cgroup); | |
1165 | ||
1166 | for_each_mem_cgroup_tree(iter, memcg) { | |
1167 | struct css_task_iter it; | |
1168 | struct task_struct *task; | |
1169 | ||
f168a9a5 | 1170 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1171 | while (!ret && (task = css_task_iter_next(&it))) |
1172 | ret = fn(task, arg); | |
1173 | css_task_iter_end(&it); | |
1174 | if (ret) { | |
1175 | mem_cgroup_iter_break(memcg, iter); | |
1176 | break; | |
1177 | } | |
1178 | } | |
1179 | return ret; | |
1180 | } | |
1181 | ||
925b7673 | 1182 | /** |
dfe0e773 | 1183 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 1184 | * @page: the page |
f144c390 | 1185 | * @pgdat: pgdat of the page |
dfe0e773 JW |
1186 | * |
1187 | * This function is only safe when following the LRU page isolation | |
1188 | * and putback protocol: the LRU lock must be held, and the page must | |
1189 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 1190 | */ |
599d0c95 | 1191 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 1192 | { |
ef8f2327 | 1193 | struct mem_cgroup_per_node *mz; |
925b7673 | 1194 | struct mem_cgroup *memcg; |
bea8c150 | 1195 | struct lruvec *lruvec; |
6d12e2d8 | 1196 | |
bea8c150 | 1197 | if (mem_cgroup_disabled()) { |
599d0c95 | 1198 | lruvec = &pgdat->lruvec; |
bea8c150 HD |
1199 | goto out; |
1200 | } | |
925b7673 | 1201 | |
1306a85a | 1202 | memcg = page->mem_cgroup; |
7512102c | 1203 | /* |
dfe0e773 | 1204 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 1205 | * possibly migrated - before they are charged. |
7512102c | 1206 | */ |
29833315 JW |
1207 | if (!memcg) |
1208 | memcg = root_mem_cgroup; | |
7512102c | 1209 | |
ef8f2327 | 1210 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1211 | lruvec = &mz->lruvec; |
1212 | out: | |
1213 | /* | |
1214 | * Since a node can be onlined after the mem_cgroup was created, | |
1215 | * we have to be prepared to initialize lruvec->zone here; | |
1216 | * and if offlined then reonlined, we need to reinitialize it. | |
1217 | */ | |
599d0c95 MG |
1218 | if (unlikely(lruvec->pgdat != pgdat)) |
1219 | lruvec->pgdat = pgdat; | |
bea8c150 | 1220 | return lruvec; |
08e552c6 | 1221 | } |
b69408e8 | 1222 | |
925b7673 | 1223 | /** |
fa9add64 HD |
1224 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1225 | * @lruvec: mem_cgroup per zone lru vector | |
1226 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1227 | * @zid: zone id of the accounted pages |
fa9add64 | 1228 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1229 | * |
ca707239 HD |
1230 | * This function must be called under lru_lock, just before a page is added |
1231 | * to or just after a page is removed from an lru list (that ordering being | |
1232 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1233 | */ |
fa9add64 | 1234 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1235 | int zid, int nr_pages) |
3f58a829 | 1236 | { |
ef8f2327 | 1237 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1238 | unsigned long *lru_size; |
ca707239 | 1239 | long size; |
3f58a829 MK |
1240 | |
1241 | if (mem_cgroup_disabled()) | |
1242 | return; | |
1243 | ||
ef8f2327 | 1244 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1245 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1246 | |
1247 | if (nr_pages < 0) | |
1248 | *lru_size += nr_pages; | |
1249 | ||
1250 | size = *lru_size; | |
b4536f0c MH |
1251 | if (WARN_ONCE(size < 0, |
1252 | "%s(%p, %d, %d): lru_size %ld\n", | |
1253 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1254 | VM_BUG_ON(1); |
1255 | *lru_size = 0; | |
1256 | } | |
1257 | ||
1258 | if (nr_pages > 0) | |
1259 | *lru_size += nr_pages; | |
08e552c6 | 1260 | } |
544122e5 | 1261 | |
19942822 | 1262 | /** |
9d11ea9f | 1263 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1264 | * @memcg: the memory cgroup |
19942822 | 1265 | * |
9d11ea9f | 1266 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1267 | * pages. |
19942822 | 1268 | */ |
c0ff4b85 | 1269 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1270 | { |
3e32cb2e JW |
1271 | unsigned long margin = 0; |
1272 | unsigned long count; | |
1273 | unsigned long limit; | |
9d11ea9f | 1274 | |
3e32cb2e | 1275 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1276 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1277 | if (count < limit) |
1278 | margin = limit - count; | |
1279 | ||
7941d214 | 1280 | if (do_memsw_account()) { |
3e32cb2e | 1281 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1282 | limit = READ_ONCE(memcg->memsw.max); |
3e32cb2e JW |
1283 | if (count <= limit) |
1284 | margin = min(margin, limit - count); | |
cbedbac3 LR |
1285 | else |
1286 | margin = 0; | |
3e32cb2e JW |
1287 | } |
1288 | ||
1289 | return margin; | |
19942822 JW |
1290 | } |
1291 | ||
32047e2a | 1292 | /* |
bdcbb659 | 1293 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1294 | * |
bdcbb659 QH |
1295 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1296 | * moving cgroups. This is for waiting at high-memory pressure | |
1297 | * caused by "move". | |
32047e2a | 1298 | */ |
c0ff4b85 | 1299 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1300 | { |
2bd9bb20 KH |
1301 | struct mem_cgroup *from; |
1302 | struct mem_cgroup *to; | |
4b534334 | 1303 | bool ret = false; |
2bd9bb20 KH |
1304 | /* |
1305 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1306 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1307 | */ | |
1308 | spin_lock(&mc.lock); | |
1309 | from = mc.from; | |
1310 | to = mc.to; | |
1311 | if (!from) | |
1312 | goto unlock; | |
3e92041d | 1313 | |
2314b42d JW |
1314 | ret = mem_cgroup_is_descendant(from, memcg) || |
1315 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1316 | unlock: |
1317 | spin_unlock(&mc.lock); | |
4b534334 KH |
1318 | return ret; |
1319 | } | |
1320 | ||
c0ff4b85 | 1321 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1322 | { |
1323 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1324 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1325 | DEFINE_WAIT(wait); |
1326 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1327 | /* moving charge context might have finished. */ | |
1328 | if (mc.moving_task) | |
1329 | schedule(); | |
1330 | finish_wait(&mc.waitq, &wait); | |
1331 | return true; | |
1332 | } | |
1333 | } | |
1334 | return false; | |
1335 | } | |
1336 | ||
c8713d0b JW |
1337 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1338 | { | |
1339 | struct seq_buf s; | |
1340 | int i; | |
71cd3113 | 1341 | |
c8713d0b JW |
1342 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1343 | if (!s.buffer) | |
1344 | return NULL; | |
1345 | ||
1346 | /* | |
1347 | * Provide statistics on the state of the memory subsystem as | |
1348 | * well as cumulative event counters that show past behavior. | |
1349 | * | |
1350 | * This list is ordered following a combination of these gradients: | |
1351 | * 1) generic big picture -> specifics and details | |
1352 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1353 | * | |
1354 | * Current memory state: | |
1355 | */ | |
1356 | ||
1357 | seq_buf_printf(&s, "anon %llu\n", | |
1358 | (u64)memcg_page_state(memcg, MEMCG_RSS) * | |
1359 | PAGE_SIZE); | |
1360 | seq_buf_printf(&s, "file %llu\n", | |
1361 | (u64)memcg_page_state(memcg, MEMCG_CACHE) * | |
1362 | PAGE_SIZE); | |
1363 | seq_buf_printf(&s, "kernel_stack %llu\n", | |
1364 | (u64)memcg_page_state(memcg, MEMCG_KERNEL_STACK_KB) * | |
1365 | 1024); | |
1366 | seq_buf_printf(&s, "slab %llu\n", | |
1367 | (u64)(memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) + | |
1368 | memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE)) * | |
1369 | PAGE_SIZE); | |
1370 | seq_buf_printf(&s, "sock %llu\n", | |
1371 | (u64)memcg_page_state(memcg, MEMCG_SOCK) * | |
1372 | PAGE_SIZE); | |
1373 | ||
1374 | seq_buf_printf(&s, "shmem %llu\n", | |
1375 | (u64)memcg_page_state(memcg, NR_SHMEM) * | |
1376 | PAGE_SIZE); | |
1377 | seq_buf_printf(&s, "file_mapped %llu\n", | |
1378 | (u64)memcg_page_state(memcg, NR_FILE_MAPPED) * | |
1379 | PAGE_SIZE); | |
1380 | seq_buf_printf(&s, "file_dirty %llu\n", | |
1381 | (u64)memcg_page_state(memcg, NR_FILE_DIRTY) * | |
1382 | PAGE_SIZE); | |
1383 | seq_buf_printf(&s, "file_writeback %llu\n", | |
1384 | (u64)memcg_page_state(memcg, NR_WRITEBACK) * | |
1385 | PAGE_SIZE); | |
1386 | ||
1387 | /* | |
1388 | * TODO: We should eventually replace our own MEMCG_RSS_HUGE counter | |
1389 | * with the NR_ANON_THP vm counter, but right now it's a pain in the | |
1390 | * arse because it requires migrating the work out of rmap to a place | |
1391 | * where the page->mem_cgroup is set up and stable. | |
1392 | */ | |
1393 | seq_buf_printf(&s, "anon_thp %llu\n", | |
1394 | (u64)memcg_page_state(memcg, MEMCG_RSS_HUGE) * | |
1395 | PAGE_SIZE); | |
1396 | ||
1397 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1398 | seq_buf_printf(&s, "%s %llu\n", mem_cgroup_lru_names[i], | |
1399 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * | |
1400 | PAGE_SIZE); | |
1401 | ||
1402 | seq_buf_printf(&s, "slab_reclaimable %llu\n", | |
1403 | (u64)memcg_page_state(memcg, NR_SLAB_RECLAIMABLE) * | |
1404 | PAGE_SIZE); | |
1405 | seq_buf_printf(&s, "slab_unreclaimable %llu\n", | |
1406 | (u64)memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE) * | |
1407 | PAGE_SIZE); | |
1408 | ||
1409 | /* Accumulated memory events */ | |
1410 | ||
1411 | seq_buf_printf(&s, "pgfault %lu\n", memcg_events(memcg, PGFAULT)); | |
1412 | seq_buf_printf(&s, "pgmajfault %lu\n", memcg_events(memcg, PGMAJFAULT)); | |
1413 | ||
1414 | seq_buf_printf(&s, "workingset_refault %lu\n", | |
1415 | memcg_page_state(memcg, WORKINGSET_REFAULT)); | |
1416 | seq_buf_printf(&s, "workingset_activate %lu\n", | |
1417 | memcg_page_state(memcg, WORKINGSET_ACTIVATE)); | |
1418 | seq_buf_printf(&s, "workingset_nodereclaim %lu\n", | |
1419 | memcg_page_state(memcg, WORKINGSET_NODERECLAIM)); | |
1420 | ||
1421 | seq_buf_printf(&s, "pgrefill %lu\n", memcg_events(memcg, PGREFILL)); | |
1422 | seq_buf_printf(&s, "pgscan %lu\n", | |
1423 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1424 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1425 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1426 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1427 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
1428 | seq_buf_printf(&s, "pgactivate %lu\n", memcg_events(memcg, PGACTIVATE)); | |
1429 | seq_buf_printf(&s, "pgdeactivate %lu\n", memcg_events(memcg, PGDEACTIVATE)); | |
1430 | seq_buf_printf(&s, "pglazyfree %lu\n", memcg_events(memcg, PGLAZYFREE)); | |
1431 | seq_buf_printf(&s, "pglazyfreed %lu\n", memcg_events(memcg, PGLAZYFREED)); | |
1432 | ||
1433 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
1434 | seq_buf_printf(&s, "thp_fault_alloc %lu\n", | |
1435 | memcg_events(memcg, THP_FAULT_ALLOC)); | |
1436 | seq_buf_printf(&s, "thp_collapse_alloc %lu\n", | |
1437 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); | |
1438 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1439 | ||
1440 | /* The above should easily fit into one page */ | |
1441 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1442 | ||
1443 | return s.buffer; | |
1444 | } | |
71cd3113 | 1445 | |
58cf188e | 1446 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1447 | /** |
f0c867d9 | 1448 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1449 | * memory controller. | |
e222432b BS |
1450 | * @memcg: The memory cgroup that went over limit |
1451 | * @p: Task that is going to be killed | |
1452 | * | |
1453 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1454 | * enabled | |
1455 | */ | |
f0c867d9 | 1456 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1457 | { |
e222432b BS |
1458 | rcu_read_lock(); |
1459 | ||
f0c867d9 | 1460 | if (memcg) { |
1461 | pr_cont(",oom_memcg="); | |
1462 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1463 | } else | |
1464 | pr_cont(",global_oom"); | |
2415b9f5 | 1465 | if (p) { |
f0c867d9 | 1466 | pr_cont(",task_memcg="); |
2415b9f5 | 1467 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1468 | } |
e222432b | 1469 | rcu_read_unlock(); |
f0c867d9 | 1470 | } |
1471 | ||
1472 | /** | |
1473 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1474 | * memory controller. | |
1475 | * @memcg: The memory cgroup that went over limit | |
1476 | */ | |
1477 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1478 | { | |
c8713d0b | 1479 | char *buf; |
e222432b | 1480 | |
3e32cb2e JW |
1481 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1482 | K((u64)page_counter_read(&memcg->memory)), | |
bbec2e15 | 1483 | K((u64)memcg->memory.max), memcg->memory.failcnt); |
c8713d0b JW |
1484 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1485 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1486 | K((u64)page_counter_read(&memcg->swap)), | |
1487 | K((u64)memcg->swap.max), memcg->swap.failcnt); | |
1488 | else { | |
1489 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1490 | K((u64)page_counter_read(&memcg->memsw)), | |
1491 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1492 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1493 | K((u64)page_counter_read(&memcg->kmem)), | |
1494 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1495 | } |
c8713d0b JW |
1496 | |
1497 | pr_info("Memory cgroup stats for "); | |
1498 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1499 | pr_cont(":"); | |
1500 | buf = memory_stat_format(memcg); | |
1501 | if (!buf) | |
1502 | return; | |
1503 | pr_info("%s", buf); | |
1504 | kfree(buf); | |
e222432b BS |
1505 | } |
1506 | ||
a63d83f4 DR |
1507 | /* |
1508 | * Return the memory (and swap, if configured) limit for a memcg. | |
1509 | */ | |
bbec2e15 | 1510 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1511 | { |
bbec2e15 | 1512 | unsigned long max; |
f3e8eb70 | 1513 | |
bbec2e15 | 1514 | max = memcg->memory.max; |
9a5a8f19 | 1515 | if (mem_cgroup_swappiness(memcg)) { |
bbec2e15 RG |
1516 | unsigned long memsw_max; |
1517 | unsigned long swap_max; | |
9a5a8f19 | 1518 | |
bbec2e15 RG |
1519 | memsw_max = memcg->memsw.max; |
1520 | swap_max = memcg->swap.max; | |
1521 | swap_max = min(swap_max, (unsigned long)total_swap_pages); | |
1522 | max = min(max + swap_max, memsw_max); | |
9a5a8f19 | 1523 | } |
bbec2e15 | 1524 | return max; |
a63d83f4 DR |
1525 | } |
1526 | ||
b6e6edcf | 1527 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1528 | int order) |
9cbb78bb | 1529 | { |
6e0fc46d DR |
1530 | struct oom_control oc = { |
1531 | .zonelist = NULL, | |
1532 | .nodemask = NULL, | |
2a966b77 | 1533 | .memcg = memcg, |
6e0fc46d DR |
1534 | .gfp_mask = gfp_mask, |
1535 | .order = order, | |
6e0fc46d | 1536 | }; |
7c5f64f8 | 1537 | bool ret; |
9cbb78bb | 1538 | |
7775face TH |
1539 | if (mutex_lock_killable(&oom_lock)) |
1540 | return true; | |
1541 | /* | |
1542 | * A few threads which were not waiting at mutex_lock_killable() can | |
1543 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1544 | */ | |
1545 | ret = should_force_charge() || out_of_memory(&oc); | |
dc56401f | 1546 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1547 | return ret; |
9cbb78bb DR |
1548 | } |
1549 | ||
ae6e71d3 MC |
1550 | #if MAX_NUMNODES > 1 |
1551 | ||
4d0c066d KH |
1552 | /** |
1553 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1554 | * @memcg: the target memcg |
4d0c066d KH |
1555 | * @nid: the node ID to be checked. |
1556 | * @noswap : specify true here if the user wants flle only information. | |
1557 | * | |
1558 | * This function returns whether the specified memcg contains any | |
1559 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1560 | * pages in the node. | |
1561 | */ | |
c0ff4b85 | 1562 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1563 | int nid, bool noswap) |
1564 | { | |
2b487e59 JW |
1565 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); |
1566 | ||
def0fdae JW |
1567 | if (lruvec_page_state(lruvec, NR_INACTIVE_FILE) || |
1568 | lruvec_page_state(lruvec, NR_ACTIVE_FILE)) | |
4d0c066d KH |
1569 | return true; |
1570 | if (noswap || !total_swap_pages) | |
1571 | return false; | |
def0fdae JW |
1572 | if (lruvec_page_state(lruvec, NR_INACTIVE_ANON) || |
1573 | lruvec_page_state(lruvec, NR_ACTIVE_ANON)) | |
4d0c066d KH |
1574 | return true; |
1575 | return false; | |
1576 | ||
1577 | } | |
889976db YH |
1578 | |
1579 | /* | |
1580 | * Always updating the nodemask is not very good - even if we have an empty | |
1581 | * list or the wrong list here, we can start from some node and traverse all | |
1582 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1583 | * | |
1584 | */ | |
c0ff4b85 | 1585 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1586 | { |
1587 | int nid; | |
453a9bf3 KH |
1588 | /* |
1589 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1590 | * pagein/pageout changes since the last update. | |
1591 | */ | |
c0ff4b85 | 1592 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1593 | return; |
c0ff4b85 | 1594 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1595 | return; |
1596 | ||
889976db | 1597 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1598 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1599 | |
31aaea4a | 1600 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1601 | |
c0ff4b85 R |
1602 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1603 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1604 | } |
453a9bf3 | 1605 | |
c0ff4b85 R |
1606 | atomic_set(&memcg->numainfo_events, 0); |
1607 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1608 | } |
1609 | ||
1610 | /* | |
1611 | * Selecting a node where we start reclaim from. Because what we need is just | |
1612 | * reducing usage counter, start from anywhere is O,K. Considering | |
1613 | * memory reclaim from current node, there are pros. and cons. | |
1614 | * | |
1615 | * Freeing memory from current node means freeing memory from a node which | |
1616 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1617 | * hit limits, it will see a contention on a node. But freeing from remote | |
1618 | * node means more costs for memory reclaim because of memory latency. | |
1619 | * | |
1620 | * Now, we use round-robin. Better algorithm is welcomed. | |
1621 | */ | |
c0ff4b85 | 1622 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1623 | { |
1624 | int node; | |
1625 | ||
c0ff4b85 R |
1626 | mem_cgroup_may_update_nodemask(memcg); |
1627 | node = memcg->last_scanned_node; | |
889976db | 1628 | |
0edaf86c | 1629 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1630 | /* |
fda3d69b MH |
1631 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1632 | * last time it really checked all the LRUs due to rate limiting. | |
1633 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1634 | */ |
1635 | if (unlikely(node == MAX_NUMNODES)) | |
1636 | node = numa_node_id(); | |
1637 | ||
c0ff4b85 | 1638 | memcg->last_scanned_node = node; |
889976db YH |
1639 | return node; |
1640 | } | |
889976db | 1641 | #else |
c0ff4b85 | 1642 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1643 | { |
1644 | return 0; | |
1645 | } | |
1646 | #endif | |
1647 | ||
0608f43d | 1648 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1649 | pg_data_t *pgdat, |
0608f43d AM |
1650 | gfp_t gfp_mask, |
1651 | unsigned long *total_scanned) | |
1652 | { | |
1653 | struct mem_cgroup *victim = NULL; | |
1654 | int total = 0; | |
1655 | int loop = 0; | |
1656 | unsigned long excess; | |
1657 | unsigned long nr_scanned; | |
1658 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1659 | .pgdat = pgdat, |
0608f43d AM |
1660 | .priority = 0, |
1661 | }; | |
1662 | ||
3e32cb2e | 1663 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1664 | |
1665 | while (1) { | |
1666 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1667 | if (!victim) { | |
1668 | loop++; | |
1669 | if (loop >= 2) { | |
1670 | /* | |
1671 | * If we have not been able to reclaim | |
1672 | * anything, it might because there are | |
1673 | * no reclaimable pages under this hierarchy | |
1674 | */ | |
1675 | if (!total) | |
1676 | break; | |
1677 | /* | |
1678 | * We want to do more targeted reclaim. | |
1679 | * excess >> 2 is not to excessive so as to | |
1680 | * reclaim too much, nor too less that we keep | |
1681 | * coming back to reclaim from this cgroup | |
1682 | */ | |
1683 | if (total >= (excess >> 2) || | |
1684 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1685 | break; | |
1686 | } | |
1687 | continue; | |
1688 | } | |
a9dd0a83 | 1689 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1690 | pgdat, &nr_scanned); |
0608f43d | 1691 | *total_scanned += nr_scanned; |
3e32cb2e | 1692 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1693 | break; |
6d61ef40 | 1694 | } |
0608f43d AM |
1695 | mem_cgroup_iter_break(root_memcg, victim); |
1696 | return total; | |
6d61ef40 BS |
1697 | } |
1698 | ||
0056f4e6 JW |
1699 | #ifdef CONFIG_LOCKDEP |
1700 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1701 | .name = "memcg_oom_lock", | |
1702 | }; | |
1703 | #endif | |
1704 | ||
fb2a6fc5 JW |
1705 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1706 | ||
867578cb KH |
1707 | /* |
1708 | * Check OOM-Killer is already running under our hierarchy. | |
1709 | * If someone is running, return false. | |
1710 | */ | |
fb2a6fc5 | 1711 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1712 | { |
79dfdacc | 1713 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1714 | |
fb2a6fc5 JW |
1715 | spin_lock(&memcg_oom_lock); |
1716 | ||
9f3a0d09 | 1717 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1718 | if (iter->oom_lock) { |
79dfdacc MH |
1719 | /* |
1720 | * this subtree of our hierarchy is already locked | |
1721 | * so we cannot give a lock. | |
1722 | */ | |
79dfdacc | 1723 | failed = iter; |
9f3a0d09 JW |
1724 | mem_cgroup_iter_break(memcg, iter); |
1725 | break; | |
23751be0 JW |
1726 | } else |
1727 | iter->oom_lock = true; | |
7d74b06f | 1728 | } |
867578cb | 1729 | |
fb2a6fc5 JW |
1730 | if (failed) { |
1731 | /* | |
1732 | * OK, we failed to lock the whole subtree so we have | |
1733 | * to clean up what we set up to the failing subtree | |
1734 | */ | |
1735 | for_each_mem_cgroup_tree(iter, memcg) { | |
1736 | if (iter == failed) { | |
1737 | mem_cgroup_iter_break(memcg, iter); | |
1738 | break; | |
1739 | } | |
1740 | iter->oom_lock = false; | |
79dfdacc | 1741 | } |
0056f4e6 JW |
1742 | } else |
1743 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1744 | |
1745 | spin_unlock(&memcg_oom_lock); | |
1746 | ||
1747 | return !failed; | |
a636b327 | 1748 | } |
0b7f569e | 1749 | |
fb2a6fc5 | 1750 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1751 | { |
7d74b06f KH |
1752 | struct mem_cgroup *iter; |
1753 | ||
fb2a6fc5 | 1754 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1755 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1756 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1757 | iter->oom_lock = false; |
fb2a6fc5 | 1758 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1759 | } |
1760 | ||
c0ff4b85 | 1761 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1762 | { |
1763 | struct mem_cgroup *iter; | |
1764 | ||
c2b42d3c | 1765 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1766 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1767 | iter->under_oom++; |
1768 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1769 | } |
1770 | ||
c0ff4b85 | 1771 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1772 | { |
1773 | struct mem_cgroup *iter; | |
1774 | ||
867578cb KH |
1775 | /* |
1776 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1777 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1778 | */ |
c2b42d3c | 1779 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1780 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1781 | if (iter->under_oom > 0) |
1782 | iter->under_oom--; | |
1783 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1784 | } |
1785 | ||
867578cb KH |
1786 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1787 | ||
dc98df5a | 1788 | struct oom_wait_info { |
d79154bb | 1789 | struct mem_cgroup *memcg; |
ac6424b9 | 1790 | wait_queue_entry_t wait; |
dc98df5a KH |
1791 | }; |
1792 | ||
ac6424b9 | 1793 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1794 | unsigned mode, int sync, void *arg) |
1795 | { | |
d79154bb HD |
1796 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1797 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1798 | struct oom_wait_info *oom_wait_info; |
1799 | ||
1800 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1801 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1802 | |
2314b42d JW |
1803 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1804 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1805 | return 0; |
dc98df5a KH |
1806 | return autoremove_wake_function(wait, mode, sync, arg); |
1807 | } | |
1808 | ||
c0ff4b85 | 1809 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1810 | { |
c2b42d3c TH |
1811 | /* |
1812 | * For the following lockless ->under_oom test, the only required | |
1813 | * guarantee is that it must see the state asserted by an OOM when | |
1814 | * this function is called as a result of userland actions | |
1815 | * triggered by the notification of the OOM. This is trivially | |
1816 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1817 | * triggering notification. | |
1818 | */ | |
1819 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1820 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1821 | } |
1822 | ||
29ef680a MH |
1823 | enum oom_status { |
1824 | OOM_SUCCESS, | |
1825 | OOM_FAILED, | |
1826 | OOM_ASYNC, | |
1827 | OOM_SKIPPED | |
1828 | }; | |
1829 | ||
1830 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1831 | { |
7056d3a3 MH |
1832 | enum oom_status ret; |
1833 | bool locked; | |
1834 | ||
29ef680a MH |
1835 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1836 | return OOM_SKIPPED; | |
1837 | ||
7a1adfdd RG |
1838 | memcg_memory_event(memcg, MEMCG_OOM); |
1839 | ||
867578cb | 1840 | /* |
49426420 JW |
1841 | * We are in the middle of the charge context here, so we |
1842 | * don't want to block when potentially sitting on a callstack | |
1843 | * that holds all kinds of filesystem and mm locks. | |
1844 | * | |
29ef680a MH |
1845 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1846 | * handling until the charge can succeed; remember the context and put | |
1847 | * the task to sleep at the end of the page fault when all locks are | |
1848 | * released. | |
49426420 | 1849 | * |
29ef680a MH |
1850 | * On the other hand, in-kernel OOM killer allows for an async victim |
1851 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1852 | * relying on the oom victim to make a forward progress and we can | |
1853 | * invoke the oom killer here. | |
1854 | * | |
1855 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1856 | * victim and then we have to bail out from the charge path. | |
867578cb | 1857 | */ |
29ef680a MH |
1858 | if (memcg->oom_kill_disable) { |
1859 | if (!current->in_user_fault) | |
1860 | return OOM_SKIPPED; | |
1861 | css_get(&memcg->css); | |
1862 | current->memcg_in_oom = memcg; | |
1863 | current->memcg_oom_gfp_mask = mask; | |
1864 | current->memcg_oom_order = order; | |
1865 | ||
1866 | return OOM_ASYNC; | |
1867 | } | |
1868 | ||
7056d3a3 MH |
1869 | mem_cgroup_mark_under_oom(memcg); |
1870 | ||
1871 | locked = mem_cgroup_oom_trylock(memcg); | |
1872 | ||
1873 | if (locked) | |
1874 | mem_cgroup_oom_notify(memcg); | |
1875 | ||
1876 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1877 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1878 | ret = OOM_SUCCESS; |
1879 | else | |
1880 | ret = OOM_FAILED; | |
1881 | ||
1882 | if (locked) | |
1883 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1884 | |
7056d3a3 | 1885 | return ret; |
3812c8c8 JW |
1886 | } |
1887 | ||
1888 | /** | |
1889 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1890 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1891 | * |
49426420 JW |
1892 | * This has to be called at the end of a page fault if the memcg OOM |
1893 | * handler was enabled. | |
3812c8c8 | 1894 | * |
49426420 | 1895 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1896 | * sleep on a waitqueue until the userspace task resolves the |
1897 | * situation. Sleeping directly in the charge context with all kinds | |
1898 | * of locks held is not a good idea, instead we remember an OOM state | |
1899 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1900 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1901 | * |
1902 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1903 | * completed, %false otherwise. |
3812c8c8 | 1904 | */ |
49426420 | 1905 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1906 | { |
626ebc41 | 1907 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1908 | struct oom_wait_info owait; |
49426420 | 1909 | bool locked; |
3812c8c8 JW |
1910 | |
1911 | /* OOM is global, do not handle */ | |
3812c8c8 | 1912 | if (!memcg) |
49426420 | 1913 | return false; |
3812c8c8 | 1914 | |
7c5f64f8 | 1915 | if (!handle) |
49426420 | 1916 | goto cleanup; |
3812c8c8 JW |
1917 | |
1918 | owait.memcg = memcg; | |
1919 | owait.wait.flags = 0; | |
1920 | owait.wait.func = memcg_oom_wake_function; | |
1921 | owait.wait.private = current; | |
2055da97 | 1922 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1923 | |
3812c8c8 | 1924 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1925 | mem_cgroup_mark_under_oom(memcg); |
1926 | ||
1927 | locked = mem_cgroup_oom_trylock(memcg); | |
1928 | ||
1929 | if (locked) | |
1930 | mem_cgroup_oom_notify(memcg); | |
1931 | ||
1932 | if (locked && !memcg->oom_kill_disable) { | |
1933 | mem_cgroup_unmark_under_oom(memcg); | |
1934 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1935 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1936 | current->memcg_oom_order); | |
49426420 | 1937 | } else { |
3812c8c8 | 1938 | schedule(); |
49426420 JW |
1939 | mem_cgroup_unmark_under_oom(memcg); |
1940 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1941 | } | |
1942 | ||
1943 | if (locked) { | |
fb2a6fc5 JW |
1944 | mem_cgroup_oom_unlock(memcg); |
1945 | /* | |
1946 | * There is no guarantee that an OOM-lock contender | |
1947 | * sees the wakeups triggered by the OOM kill | |
1948 | * uncharges. Wake any sleepers explicitely. | |
1949 | */ | |
1950 | memcg_oom_recover(memcg); | |
1951 | } | |
49426420 | 1952 | cleanup: |
626ebc41 | 1953 | current->memcg_in_oom = NULL; |
3812c8c8 | 1954 | css_put(&memcg->css); |
867578cb | 1955 | return true; |
0b7f569e KH |
1956 | } |
1957 | ||
3d8b38eb RG |
1958 | /** |
1959 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
1960 | * @victim: task to be killed by the OOM killer | |
1961 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
1962 | * | |
1963 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
1964 | * by killing all belonging OOM-killable tasks. | |
1965 | * | |
1966 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
1967 | */ | |
1968 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
1969 | struct mem_cgroup *oom_domain) | |
1970 | { | |
1971 | struct mem_cgroup *oom_group = NULL; | |
1972 | struct mem_cgroup *memcg; | |
1973 | ||
1974 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
1975 | return NULL; | |
1976 | ||
1977 | if (!oom_domain) | |
1978 | oom_domain = root_mem_cgroup; | |
1979 | ||
1980 | rcu_read_lock(); | |
1981 | ||
1982 | memcg = mem_cgroup_from_task(victim); | |
1983 | if (memcg == root_mem_cgroup) | |
1984 | goto out; | |
1985 | ||
1986 | /* | |
1987 | * Traverse the memory cgroup hierarchy from the victim task's | |
1988 | * cgroup up to the OOMing cgroup (or root) to find the | |
1989 | * highest-level memory cgroup with oom.group set. | |
1990 | */ | |
1991 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
1992 | if (memcg->oom_group) | |
1993 | oom_group = memcg; | |
1994 | ||
1995 | if (memcg == oom_domain) | |
1996 | break; | |
1997 | } | |
1998 | ||
1999 | if (oom_group) | |
2000 | css_get(&oom_group->css); | |
2001 | out: | |
2002 | rcu_read_unlock(); | |
2003 | ||
2004 | return oom_group; | |
2005 | } | |
2006 | ||
2007 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
2008 | { | |
2009 | pr_info("Tasks in "); | |
2010 | pr_cont_cgroup_path(memcg->css.cgroup); | |
2011 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
2012 | } | |
2013 | ||
d7365e78 | 2014 | /** |
81f8c3a4 JW |
2015 | * lock_page_memcg - lock a page->mem_cgroup binding |
2016 | * @page: the page | |
32047e2a | 2017 | * |
81f8c3a4 | 2018 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
2019 | * another cgroup. |
2020 | * | |
2021 | * It ensures lifetime of the returned memcg. Caller is responsible | |
2022 | * for the lifetime of the page; __unlock_page_memcg() is available | |
2023 | * when @page might get freed inside the locked section. | |
d69b042f | 2024 | */ |
739f79fc | 2025 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 KH |
2026 | { |
2027 | struct mem_cgroup *memcg; | |
6de22619 | 2028 | unsigned long flags; |
89c06bd5 | 2029 | |
6de22619 JW |
2030 | /* |
2031 | * The RCU lock is held throughout the transaction. The fast | |
2032 | * path can get away without acquiring the memcg->move_lock | |
2033 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
2034 | * |
2035 | * The RCU lock also protects the memcg from being freed when | |
2036 | * the page state that is going to change is the only thing | |
2037 | * preventing the page itself from being freed. E.g. writeback | |
2038 | * doesn't hold a page reference and relies on PG_writeback to | |
2039 | * keep off truncation, migration and so forth. | |
2040 | */ | |
d7365e78 JW |
2041 | rcu_read_lock(); |
2042 | ||
2043 | if (mem_cgroup_disabled()) | |
739f79fc | 2044 | return NULL; |
89c06bd5 | 2045 | again: |
1306a85a | 2046 | memcg = page->mem_cgroup; |
29833315 | 2047 | if (unlikely(!memcg)) |
739f79fc | 2048 | return NULL; |
d7365e78 | 2049 | |
bdcbb659 | 2050 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 2051 | return memcg; |
89c06bd5 | 2052 | |
6de22619 | 2053 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 2054 | if (memcg != page->mem_cgroup) { |
6de22619 | 2055 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2056 | goto again; |
2057 | } | |
6de22619 JW |
2058 | |
2059 | /* | |
2060 | * When charge migration first begins, we can have locked and | |
2061 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 2062 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
2063 | */ |
2064 | memcg->move_lock_task = current; | |
2065 | memcg->move_lock_flags = flags; | |
d7365e78 | 2066 | |
739f79fc | 2067 | return memcg; |
89c06bd5 | 2068 | } |
81f8c3a4 | 2069 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2070 | |
d7365e78 | 2071 | /** |
739f79fc JW |
2072 | * __unlock_page_memcg - unlock and unpin a memcg |
2073 | * @memcg: the memcg | |
2074 | * | |
2075 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 2076 | */ |
739f79fc | 2077 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2078 | { |
6de22619 JW |
2079 | if (memcg && memcg->move_lock_task == current) { |
2080 | unsigned long flags = memcg->move_lock_flags; | |
2081 | ||
2082 | memcg->move_lock_task = NULL; | |
2083 | memcg->move_lock_flags = 0; | |
2084 | ||
2085 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2086 | } | |
89c06bd5 | 2087 | |
d7365e78 | 2088 | rcu_read_unlock(); |
89c06bd5 | 2089 | } |
739f79fc JW |
2090 | |
2091 | /** | |
2092 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
2093 | * @page: the page | |
2094 | */ | |
2095 | void unlock_page_memcg(struct page *page) | |
2096 | { | |
2097 | __unlock_page_memcg(page->mem_cgroup); | |
2098 | } | |
81f8c3a4 | 2099 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2100 | |
cdec2e42 KH |
2101 | struct memcg_stock_pcp { |
2102 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2103 | unsigned int nr_pages; |
cdec2e42 | 2104 | struct work_struct work; |
26fe6168 | 2105 | unsigned long flags; |
a0db00fc | 2106 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2107 | }; |
2108 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2109 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2110 | |
a0956d54 SS |
2111 | /** |
2112 | * consume_stock: Try to consume stocked charge on this cpu. | |
2113 | * @memcg: memcg to consume from. | |
2114 | * @nr_pages: how many pages to charge. | |
2115 | * | |
2116 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2117 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2118 | * service an allocation will refill the stock. | |
2119 | * | |
2120 | * returns true if successful, false otherwise. | |
cdec2e42 | 2121 | */ |
a0956d54 | 2122 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2123 | { |
2124 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2125 | unsigned long flags; |
3e32cb2e | 2126 | bool ret = false; |
cdec2e42 | 2127 | |
a983b5eb | 2128 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2129 | return ret; |
a0956d54 | 2130 | |
db2ba40c JW |
2131 | local_irq_save(flags); |
2132 | ||
2133 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2134 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2135 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2136 | ret = true; |
2137 | } | |
db2ba40c JW |
2138 | |
2139 | local_irq_restore(flags); | |
2140 | ||
cdec2e42 KH |
2141 | return ret; |
2142 | } | |
2143 | ||
2144 | /* | |
3e32cb2e | 2145 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2146 | */ |
2147 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2148 | { | |
2149 | struct mem_cgroup *old = stock->cached; | |
2150 | ||
11c9ea4e | 2151 | if (stock->nr_pages) { |
3e32cb2e | 2152 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2153 | if (do_memsw_account()) |
3e32cb2e | 2154 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 2155 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 2156 | stock->nr_pages = 0; |
cdec2e42 KH |
2157 | } |
2158 | stock->cached = NULL; | |
cdec2e42 KH |
2159 | } |
2160 | ||
cdec2e42 KH |
2161 | static void drain_local_stock(struct work_struct *dummy) |
2162 | { | |
db2ba40c JW |
2163 | struct memcg_stock_pcp *stock; |
2164 | unsigned long flags; | |
2165 | ||
72f0184c MH |
2166 | /* |
2167 | * The only protection from memory hotplug vs. drain_stock races is | |
2168 | * that we always operate on local CPU stock here with IRQ disabled | |
2169 | */ | |
db2ba40c JW |
2170 | local_irq_save(flags); |
2171 | ||
2172 | stock = this_cpu_ptr(&memcg_stock); | |
cdec2e42 | 2173 | drain_stock(stock); |
26fe6168 | 2174 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2175 | |
2176 | local_irq_restore(flags); | |
cdec2e42 KH |
2177 | } |
2178 | ||
2179 | /* | |
3e32cb2e | 2180 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2181 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2182 | */ |
c0ff4b85 | 2183 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2184 | { |
db2ba40c JW |
2185 | struct memcg_stock_pcp *stock; |
2186 | unsigned long flags; | |
2187 | ||
2188 | local_irq_save(flags); | |
cdec2e42 | 2189 | |
db2ba40c | 2190 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2191 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2192 | drain_stock(stock); |
c0ff4b85 | 2193 | stock->cached = memcg; |
cdec2e42 | 2194 | } |
11c9ea4e | 2195 | stock->nr_pages += nr_pages; |
db2ba40c | 2196 | |
a983b5eb | 2197 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2198 | drain_stock(stock); |
2199 | ||
db2ba40c | 2200 | local_irq_restore(flags); |
cdec2e42 KH |
2201 | } |
2202 | ||
2203 | /* | |
c0ff4b85 | 2204 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2205 | * of the hierarchy under it. |
cdec2e42 | 2206 | */ |
6d3d6aa2 | 2207 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2208 | { |
26fe6168 | 2209 | int cpu, curcpu; |
d38144b7 | 2210 | |
6d3d6aa2 JW |
2211 | /* If someone's already draining, avoid adding running more workers. */ |
2212 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2213 | return; | |
72f0184c MH |
2214 | /* |
2215 | * Notify other cpus that system-wide "drain" is running | |
2216 | * We do not care about races with the cpu hotplug because cpu down | |
2217 | * as well as workers from this path always operate on the local | |
2218 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2219 | */ | |
5af12d0e | 2220 | curcpu = get_cpu(); |
cdec2e42 KH |
2221 | for_each_online_cpu(cpu) { |
2222 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2223 | struct mem_cgroup *memcg; |
26fe6168 | 2224 | |
c0ff4b85 | 2225 | memcg = stock->cached; |
72f0184c | 2226 | if (!memcg || !stock->nr_pages || !css_tryget(&memcg->css)) |
26fe6168 | 2227 | continue; |
72f0184c MH |
2228 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) { |
2229 | css_put(&memcg->css); | |
3e92041d | 2230 | continue; |
72f0184c | 2231 | } |
d1a05b69 MH |
2232 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2233 | if (cpu == curcpu) | |
2234 | drain_local_stock(&stock->work); | |
2235 | else | |
2236 | schedule_work_on(cpu, &stock->work); | |
2237 | } | |
72f0184c | 2238 | css_put(&memcg->css); |
cdec2e42 | 2239 | } |
5af12d0e | 2240 | put_cpu(); |
9f50fad6 | 2241 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2242 | } |
2243 | ||
308167fc | 2244 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 2245 | { |
cdec2e42 | 2246 | struct memcg_stock_pcp *stock; |
42a30035 | 2247 | struct mem_cgroup *memcg, *mi; |
cdec2e42 | 2248 | |
cdec2e42 KH |
2249 | stock = &per_cpu(memcg_stock, cpu); |
2250 | drain_stock(stock); | |
a983b5eb JW |
2251 | |
2252 | for_each_mem_cgroup(memcg) { | |
2253 | int i; | |
2254 | ||
2255 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
2256 | int nid; | |
2257 | long x; | |
2258 | ||
871789d4 | 2259 | x = this_cpu_xchg(memcg->vmstats_percpu->stat[i], 0); |
815744d7 | 2260 | if (x) |
42a30035 JW |
2261 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2262 | atomic_long_add(x, &memcg->vmstats[i]); | |
a983b5eb JW |
2263 | |
2264 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
2265 | continue; | |
2266 | ||
2267 | for_each_node(nid) { | |
2268 | struct mem_cgroup_per_node *pn; | |
2269 | ||
2270 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
2271 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
815744d7 | 2272 | if (x) |
42a30035 JW |
2273 | do { |
2274 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
2275 | } while ((pn = parent_nodeinfo(pn, nid))); | |
a983b5eb JW |
2276 | } |
2277 | } | |
2278 | ||
e27be240 | 2279 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
a983b5eb JW |
2280 | long x; |
2281 | ||
871789d4 | 2282 | x = this_cpu_xchg(memcg->vmstats_percpu->events[i], 0); |
815744d7 | 2283 | if (x) |
42a30035 JW |
2284 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) |
2285 | atomic_long_add(x, &memcg->vmevents[i]); | |
a983b5eb JW |
2286 | } |
2287 | } | |
2288 | ||
308167fc | 2289 | return 0; |
cdec2e42 KH |
2290 | } |
2291 | ||
f7e1cb6e JW |
2292 | static void reclaim_high(struct mem_cgroup *memcg, |
2293 | unsigned int nr_pages, | |
2294 | gfp_t gfp_mask) | |
2295 | { | |
2296 | do { | |
2297 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
2298 | continue; | |
e27be240 | 2299 | memcg_memory_event(memcg, MEMCG_HIGH); |
f7e1cb6e JW |
2300 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
2301 | } while ((memcg = parent_mem_cgroup(memcg))); | |
2302 | } | |
2303 | ||
2304 | static void high_work_func(struct work_struct *work) | |
2305 | { | |
2306 | struct mem_cgroup *memcg; | |
2307 | ||
2308 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2309 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2310 | } |
2311 | ||
b23afb93 TH |
2312 | /* |
2313 | * Scheduled by try_charge() to be executed from the userland return path | |
2314 | * and reclaims memory over the high limit. | |
2315 | */ | |
2316 | void mem_cgroup_handle_over_high(void) | |
2317 | { | |
2318 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 2319 | struct mem_cgroup *memcg; |
b23afb93 TH |
2320 | |
2321 | if (likely(!nr_pages)) | |
2322 | return; | |
2323 | ||
f7e1cb6e JW |
2324 | memcg = get_mem_cgroup_from_mm(current->mm); |
2325 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
2326 | css_put(&memcg->css); |
2327 | current->memcg_nr_pages_over_high = 0; | |
2328 | } | |
2329 | ||
00501b53 JW |
2330 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2331 | unsigned int nr_pages) | |
8a9f3ccd | 2332 | { |
a983b5eb | 2333 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
9b130619 | 2334 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 2335 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2336 | struct page_counter *counter; |
6539cc05 | 2337 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2338 | bool may_swap = true; |
2339 | bool drained = false; | |
29ef680a | 2340 | enum oom_status oom_status; |
a636b327 | 2341 | |
ce00a967 | 2342 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2343 | return 0; |
6539cc05 | 2344 | retry: |
b6b6cc72 | 2345 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2346 | return 0; |
8a9f3ccd | 2347 | |
7941d214 | 2348 | if (!do_memsw_account() || |
6071ca52 JW |
2349 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2350 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2351 | goto done_restock; |
7941d214 | 2352 | if (do_memsw_account()) |
3e32cb2e JW |
2353 | page_counter_uncharge(&memcg->memsw, batch); |
2354 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2355 | } else { |
3e32cb2e | 2356 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2357 | may_swap = false; |
3fbe7244 | 2358 | } |
7a81b88c | 2359 | |
6539cc05 JW |
2360 | if (batch > nr_pages) { |
2361 | batch = nr_pages; | |
2362 | goto retry; | |
2363 | } | |
6d61ef40 | 2364 | |
06b078fc JW |
2365 | /* |
2366 | * Unlike in global OOM situations, memcg is not in a physical | |
2367 | * memory shortage. Allow dying and OOM-killed tasks to | |
2368 | * bypass the last charges so that they can exit quickly and | |
2369 | * free their memory. | |
2370 | */ | |
7775face | 2371 | if (unlikely(should_force_charge())) |
10d53c74 | 2372 | goto force; |
06b078fc | 2373 | |
89a28483 JW |
2374 | /* |
2375 | * Prevent unbounded recursion when reclaim operations need to | |
2376 | * allocate memory. This might exceed the limits temporarily, | |
2377 | * but we prefer facilitating memory reclaim and getting back | |
2378 | * under the limit over triggering OOM kills in these cases. | |
2379 | */ | |
2380 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2381 | goto force; | |
2382 | ||
06b078fc JW |
2383 | if (unlikely(task_in_memcg_oom(current))) |
2384 | goto nomem; | |
2385 | ||
d0164adc | 2386 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2387 | goto nomem; |
4b534334 | 2388 | |
e27be240 | 2389 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2390 | |
b70a2a21 JW |
2391 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2392 | gfp_mask, may_swap); | |
6539cc05 | 2393 | |
61e02c74 | 2394 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2395 | goto retry; |
28c34c29 | 2396 | |
b70a2a21 | 2397 | if (!drained) { |
6d3d6aa2 | 2398 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2399 | drained = true; |
2400 | goto retry; | |
2401 | } | |
2402 | ||
28c34c29 JW |
2403 | if (gfp_mask & __GFP_NORETRY) |
2404 | goto nomem; | |
6539cc05 JW |
2405 | /* |
2406 | * Even though the limit is exceeded at this point, reclaim | |
2407 | * may have been able to free some pages. Retry the charge | |
2408 | * before killing the task. | |
2409 | * | |
2410 | * Only for regular pages, though: huge pages are rather | |
2411 | * unlikely to succeed so close to the limit, and we fall back | |
2412 | * to regular pages anyway in case of failure. | |
2413 | */ | |
61e02c74 | 2414 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2415 | goto retry; |
2416 | /* | |
2417 | * At task move, charge accounts can be doubly counted. So, it's | |
2418 | * better to wait until the end of task_move if something is going on. | |
2419 | */ | |
2420 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2421 | goto retry; | |
2422 | ||
9b130619 JW |
2423 | if (nr_retries--) |
2424 | goto retry; | |
2425 | ||
38d38493 | 2426 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2427 | goto nomem; |
2428 | ||
06b078fc | 2429 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2430 | goto force; |
06b078fc | 2431 | |
6539cc05 | 2432 | if (fatal_signal_pending(current)) |
10d53c74 | 2433 | goto force; |
6539cc05 | 2434 | |
29ef680a MH |
2435 | /* |
2436 | * keep retrying as long as the memcg oom killer is able to make | |
2437 | * a forward progress or bypass the charge if the oom killer | |
2438 | * couldn't make any progress. | |
2439 | */ | |
2440 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2441 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2442 | switch (oom_status) { |
2443 | case OOM_SUCCESS: | |
2444 | nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
29ef680a MH |
2445 | goto retry; |
2446 | case OOM_FAILED: | |
2447 | goto force; | |
2448 | default: | |
2449 | goto nomem; | |
2450 | } | |
7a81b88c | 2451 | nomem: |
6d1fdc48 | 2452 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2453 | return -ENOMEM; |
10d53c74 TH |
2454 | force: |
2455 | /* | |
2456 | * The allocation either can't fail or will lead to more memory | |
2457 | * being freed very soon. Allow memory usage go over the limit | |
2458 | * temporarily by force charging it. | |
2459 | */ | |
2460 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2461 | if (do_memsw_account()) |
10d53c74 TH |
2462 | page_counter_charge(&memcg->memsw, nr_pages); |
2463 | css_get_many(&memcg->css, nr_pages); | |
2464 | ||
2465 | return 0; | |
6539cc05 JW |
2466 | |
2467 | done_restock: | |
e8ea14cc | 2468 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2469 | if (batch > nr_pages) |
2470 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2471 | |
241994ed | 2472 | /* |
b23afb93 TH |
2473 | * If the hierarchy is above the normal consumption range, schedule |
2474 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2475 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2476 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2477 | * not recorded as it most likely matches current's and won't | |
2478 | * change in the meantime. As high limit is checked again before | |
2479 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2480 | */ |
2481 | do { | |
b23afb93 | 2482 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2483 | /* Don't bother a random interrupted task */ |
2484 | if (in_interrupt()) { | |
2485 | schedule_work(&memcg->high_work); | |
2486 | break; | |
2487 | } | |
9516a18a | 2488 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2489 | set_notify_resume(current); |
2490 | break; | |
2491 | } | |
241994ed | 2492 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2493 | |
2494 | return 0; | |
7a81b88c | 2495 | } |
8a9f3ccd | 2496 | |
00501b53 | 2497 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2498 | { |
ce00a967 JW |
2499 | if (mem_cgroup_is_root(memcg)) |
2500 | return; | |
2501 | ||
3e32cb2e | 2502 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2503 | if (do_memsw_account()) |
3e32cb2e | 2504 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2505 | |
e8ea14cc | 2506 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2507 | } |
2508 | ||
0a31bc97 JW |
2509 | static void lock_page_lru(struct page *page, int *isolated) |
2510 | { | |
f4b7e272 | 2511 | pg_data_t *pgdat = page_pgdat(page); |
0a31bc97 | 2512 | |
f4b7e272 | 2513 | spin_lock_irq(&pgdat->lru_lock); |
0a31bc97 JW |
2514 | if (PageLRU(page)) { |
2515 | struct lruvec *lruvec; | |
2516 | ||
f4b7e272 | 2517 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
0a31bc97 JW |
2518 | ClearPageLRU(page); |
2519 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2520 | *isolated = 1; | |
2521 | } else | |
2522 | *isolated = 0; | |
2523 | } | |
2524 | ||
2525 | static void unlock_page_lru(struct page *page, int isolated) | |
2526 | { | |
f4b7e272 | 2527 | pg_data_t *pgdat = page_pgdat(page); |
0a31bc97 JW |
2528 | |
2529 | if (isolated) { | |
2530 | struct lruvec *lruvec; | |
2531 | ||
f4b7e272 | 2532 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
0a31bc97 JW |
2533 | VM_BUG_ON_PAGE(PageLRU(page), page); |
2534 | SetPageLRU(page); | |
2535 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2536 | } | |
f4b7e272 | 2537 | spin_unlock_irq(&pgdat->lru_lock); |
0a31bc97 JW |
2538 | } |
2539 | ||
00501b53 | 2540 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2541 | bool lrucare) |
7a81b88c | 2542 | { |
0a31bc97 | 2543 | int isolated; |
9ce70c02 | 2544 | |
1306a85a | 2545 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2546 | |
2547 | /* | |
2548 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2549 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2550 | */ | |
0a31bc97 JW |
2551 | if (lrucare) |
2552 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2553 | |
0a31bc97 JW |
2554 | /* |
2555 | * Nobody should be changing or seriously looking at | |
1306a85a | 2556 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2557 | * |
2558 | * - the page is uncharged | |
2559 | * | |
2560 | * - the page is off-LRU | |
2561 | * | |
2562 | * - an anonymous fault has exclusive page access, except for | |
2563 | * a locked page table | |
2564 | * | |
2565 | * - a page cache insertion, a swapin fault, or a migration | |
2566 | * have the page locked | |
2567 | */ | |
1306a85a | 2568 | page->mem_cgroup = memcg; |
9ce70c02 | 2569 | |
0a31bc97 JW |
2570 | if (lrucare) |
2571 | unlock_page_lru(page, isolated); | |
7a81b88c | 2572 | } |
66e1707b | 2573 | |
84c07d11 | 2574 | #ifdef CONFIG_MEMCG_KMEM |
f3bb3043 | 2575 | static int memcg_alloc_cache_id(void) |
55007d84 | 2576 | { |
f3bb3043 VD |
2577 | int id, size; |
2578 | int err; | |
2579 | ||
dbcf73e2 | 2580 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2581 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2582 | if (id < 0) | |
2583 | return id; | |
55007d84 | 2584 | |
dbcf73e2 | 2585 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2586 | return id; |
2587 | ||
2588 | /* | |
2589 | * There's no space for the new id in memcg_caches arrays, | |
2590 | * so we have to grow them. | |
2591 | */ | |
05257a1a | 2592 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2593 | |
2594 | size = 2 * (id + 1); | |
55007d84 GC |
2595 | if (size < MEMCG_CACHES_MIN_SIZE) |
2596 | size = MEMCG_CACHES_MIN_SIZE; | |
2597 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2598 | size = MEMCG_CACHES_MAX_SIZE; | |
2599 | ||
f3bb3043 | 2600 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2601 | if (!err) |
2602 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2603 | if (!err) |
2604 | memcg_nr_cache_ids = size; | |
2605 | ||
2606 | up_write(&memcg_cache_ids_sem); | |
2607 | ||
f3bb3043 | 2608 | if (err) { |
dbcf73e2 | 2609 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2610 | return err; |
2611 | } | |
2612 | return id; | |
2613 | } | |
2614 | ||
2615 | static void memcg_free_cache_id(int id) | |
2616 | { | |
dbcf73e2 | 2617 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2618 | } |
2619 | ||
d5b3cf71 | 2620 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2621 | struct mem_cgroup *memcg; |
2622 | struct kmem_cache *cachep; | |
2623 | struct work_struct work; | |
2624 | }; | |
2625 | ||
d5b3cf71 | 2626 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2627 | { |
d5b3cf71 VD |
2628 | struct memcg_kmem_cache_create_work *cw = |
2629 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2630 | struct mem_cgroup *memcg = cw->memcg; |
2631 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2632 | |
d5b3cf71 | 2633 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2634 | |
5722d094 | 2635 | css_put(&memcg->css); |
d7f25f8a GC |
2636 | kfree(cw); |
2637 | } | |
2638 | ||
2639 | /* | |
2640 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2641 | */ |
85cfb245 | 2642 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
d5b3cf71 | 2643 | struct kmem_cache *cachep) |
d7f25f8a | 2644 | { |
d5b3cf71 | 2645 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2646 | |
f0a3a24b RG |
2647 | if (!css_tryget_online(&memcg->css)) |
2648 | return; | |
2649 | ||
c892fd82 | 2650 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT | __GFP_NOWARN); |
8135be5a | 2651 | if (!cw) |
d7f25f8a | 2652 | return; |
8135be5a | 2653 | |
d7f25f8a GC |
2654 | cw->memcg = memcg; |
2655 | cw->cachep = cachep; | |
d5b3cf71 | 2656 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2657 | |
17cc4dfe | 2658 | queue_work(memcg_kmem_cache_wq, &cw->work); |
d7f25f8a GC |
2659 | } |
2660 | ||
45264778 VD |
2661 | static inline bool memcg_kmem_bypass(void) |
2662 | { | |
2663 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
2664 | return true; | |
2665 | return false; | |
2666 | } | |
2667 | ||
2668 | /** | |
2669 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation | |
2670 | * @cachep: the original global kmem cache | |
2671 | * | |
d7f25f8a GC |
2672 | * Return the kmem_cache we're supposed to use for a slab allocation. |
2673 | * We try to use the current memcg's version of the cache. | |
2674 | * | |
45264778 VD |
2675 | * If the cache does not exist yet, if we are the first user of it, we |
2676 | * create it asynchronously in a workqueue and let the current allocation | |
2677 | * go through with the original cache. | |
d7f25f8a | 2678 | * |
45264778 VD |
2679 | * This function takes a reference to the cache it returns to assure it |
2680 | * won't get destroyed while we are working with it. Once the caller is | |
2681 | * done with it, memcg_kmem_put_cache() must be called to release the | |
2682 | * reference. | |
d7f25f8a | 2683 | */ |
45264778 | 2684 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2685 | { |
2686 | struct mem_cgroup *memcg; | |
959c8963 | 2687 | struct kmem_cache *memcg_cachep; |
f0a3a24b | 2688 | struct memcg_cache_array *arr; |
2a4db7eb | 2689 | int kmemcg_id; |
d7f25f8a | 2690 | |
f7ce3190 | 2691 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2692 | |
45264778 | 2693 | if (memcg_kmem_bypass()) |
230e9fc2 VD |
2694 | return cachep; |
2695 | ||
f0a3a24b RG |
2696 | rcu_read_lock(); |
2697 | ||
2698 | if (unlikely(current->active_memcg)) | |
2699 | memcg = current->active_memcg; | |
2700 | else | |
2701 | memcg = mem_cgroup_from_task(current); | |
2702 | ||
2703 | if (!memcg || memcg == root_mem_cgroup) | |
2704 | goto out_unlock; | |
2705 | ||
4db0c3c2 | 2706 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2707 | if (kmemcg_id < 0) |
f0a3a24b | 2708 | goto out_unlock; |
d7f25f8a | 2709 | |
f0a3a24b RG |
2710 | arr = rcu_dereference(cachep->memcg_params.memcg_caches); |
2711 | ||
2712 | /* | |
2713 | * Make sure we will access the up-to-date value. The code updating | |
2714 | * memcg_caches issues a write barrier to match the data dependency | |
2715 | * barrier inside READ_ONCE() (see memcg_create_kmem_cache()). | |
2716 | */ | |
2717 | memcg_cachep = READ_ONCE(arr->entries[kmemcg_id]); | |
ca0dde97 LZ |
2718 | |
2719 | /* | |
2720 | * If we are in a safe context (can wait, and not in interrupt | |
2721 | * context), we could be be predictable and return right away. | |
2722 | * This would guarantee that the allocation being performed | |
2723 | * already belongs in the new cache. | |
2724 | * | |
2725 | * However, there are some clashes that can arrive from locking. | |
2726 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2727 | * memcg_create_kmem_cache, this means no further allocation |
2728 | * could happen with the slab_mutex held. So it's better to | |
2729 | * defer everything. | |
f0a3a24b RG |
2730 | * |
2731 | * If the memcg is dying or memcg_cache is about to be released, | |
2732 | * don't bother creating new kmem_caches. Because memcg_cachep | |
2733 | * is ZEROed as the fist step of kmem offlining, we don't need | |
2734 | * percpu_ref_tryget_live() here. css_tryget_online() check in | |
2735 | * memcg_schedule_kmem_cache_create() will prevent us from | |
2736 | * creation of a new kmem_cache. | |
ca0dde97 | 2737 | */ |
f0a3a24b RG |
2738 | if (unlikely(!memcg_cachep)) |
2739 | memcg_schedule_kmem_cache_create(memcg, cachep); | |
2740 | else if (percpu_ref_tryget(&memcg_cachep->memcg_params.refcnt)) | |
2741 | cachep = memcg_cachep; | |
2742 | out_unlock: | |
2743 | rcu_read_unlock(); | |
ca0dde97 | 2744 | return cachep; |
d7f25f8a | 2745 | } |
d7f25f8a | 2746 | |
45264778 VD |
2747 | /** |
2748 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache | |
2749 | * @cachep: the cache returned by memcg_kmem_get_cache | |
2750 | */ | |
2751 | void memcg_kmem_put_cache(struct kmem_cache *cachep) | |
8135be5a VD |
2752 | { |
2753 | if (!is_root_cache(cachep)) | |
f0a3a24b | 2754 | percpu_ref_put(&cachep->memcg_params.refcnt); |
8135be5a VD |
2755 | } |
2756 | ||
45264778 | 2757 | /** |
60cd4bcd | 2758 | * __memcg_kmem_charge_memcg: charge a kmem page |
45264778 VD |
2759 | * @page: page to charge |
2760 | * @gfp: reclaim mode | |
2761 | * @order: allocation order | |
2762 | * @memcg: memory cgroup to charge | |
2763 | * | |
2764 | * Returns 0 on success, an error code on failure. | |
2765 | */ | |
60cd4bcd | 2766 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
45264778 | 2767 | struct mem_cgroup *memcg) |
7ae1e1d0 | 2768 | { |
f3ccb2c4 VD |
2769 | unsigned int nr_pages = 1 << order; |
2770 | struct page_counter *counter; | |
7ae1e1d0 GC |
2771 | int ret; |
2772 | ||
f3ccb2c4 | 2773 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2774 | if (ret) |
f3ccb2c4 | 2775 | return ret; |
52c29b04 JW |
2776 | |
2777 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2778 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2779 | cancel_charge(memcg, nr_pages); | |
2780 | return -ENOMEM; | |
7ae1e1d0 | 2781 | } |
f3ccb2c4 | 2782 | return 0; |
7ae1e1d0 GC |
2783 | } |
2784 | ||
45264778 | 2785 | /** |
60cd4bcd | 2786 | * __memcg_kmem_charge: charge a kmem page to the current memory cgroup |
45264778 VD |
2787 | * @page: page to charge |
2788 | * @gfp: reclaim mode | |
2789 | * @order: allocation order | |
2790 | * | |
2791 | * Returns 0 on success, an error code on failure. | |
2792 | */ | |
60cd4bcd | 2793 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2794 | { |
f3ccb2c4 | 2795 | struct mem_cgroup *memcg; |
fcff7d7e | 2796 | int ret = 0; |
7ae1e1d0 | 2797 | |
60cd4bcd | 2798 | if (memcg_kmem_bypass()) |
45264778 VD |
2799 | return 0; |
2800 | ||
d46eb14b | 2801 | memcg = get_mem_cgroup_from_current(); |
c4159a75 | 2802 | if (!mem_cgroup_is_root(memcg)) { |
60cd4bcd | 2803 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); |
4d96ba35 RG |
2804 | if (!ret) { |
2805 | page->mem_cgroup = memcg; | |
c4159a75 | 2806 | __SetPageKmemcg(page); |
4d96ba35 | 2807 | } |
c4159a75 | 2808 | } |
7ae1e1d0 | 2809 | css_put(&memcg->css); |
d05e83a6 | 2810 | return ret; |
7ae1e1d0 | 2811 | } |
49a18eae RG |
2812 | |
2813 | /** | |
2814 | * __memcg_kmem_uncharge_memcg: uncharge a kmem page | |
2815 | * @memcg: memcg to uncharge | |
2816 | * @nr_pages: number of pages to uncharge | |
2817 | */ | |
2818 | void __memcg_kmem_uncharge_memcg(struct mem_cgroup *memcg, | |
2819 | unsigned int nr_pages) | |
2820 | { | |
2821 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
2822 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2823 | ||
2824 | page_counter_uncharge(&memcg->memory, nr_pages); | |
2825 | if (do_memsw_account()) | |
2826 | page_counter_uncharge(&memcg->memsw, nr_pages); | |
2827 | } | |
45264778 | 2828 | /** |
60cd4bcd | 2829 | * __memcg_kmem_uncharge: uncharge a kmem page |
45264778 VD |
2830 | * @page: page to uncharge |
2831 | * @order: allocation order | |
2832 | */ | |
60cd4bcd | 2833 | void __memcg_kmem_uncharge(struct page *page, int order) |
7ae1e1d0 | 2834 | { |
1306a85a | 2835 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2836 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2837 | |
7ae1e1d0 GC |
2838 | if (!memcg) |
2839 | return; | |
2840 | ||
309381fe | 2841 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
49a18eae | 2842 | __memcg_kmem_uncharge_memcg(memcg, nr_pages); |
1306a85a | 2843 | page->mem_cgroup = NULL; |
c4159a75 VD |
2844 | |
2845 | /* slab pages do not have PageKmemcg flag set */ | |
2846 | if (PageKmemcg(page)) | |
2847 | __ClearPageKmemcg(page); | |
2848 | ||
f3ccb2c4 | 2849 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2850 | } |
84c07d11 | 2851 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 2852 | |
ca3e0214 KH |
2853 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2854 | ||
ca3e0214 KH |
2855 | /* |
2856 | * Because tail pages are not marked as "used", set it. We're under | |
f4b7e272 | 2857 | * pgdat->lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2858 | */ |
e94c8a9c | 2859 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2860 | { |
e94c8a9c | 2861 | int i; |
ca3e0214 | 2862 | |
3d37c4a9 KH |
2863 | if (mem_cgroup_disabled()) |
2864 | return; | |
b070e65c | 2865 | |
29833315 | 2866 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2867 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2868 | |
c9019e9b | 2869 | __mod_memcg_state(head->mem_cgroup, MEMCG_RSS_HUGE, -HPAGE_PMD_NR); |
ca3e0214 | 2870 | } |
12d27107 | 2871 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2872 | |
c255a458 | 2873 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
2874 | /** |
2875 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2876 | * @entry: swap entry to be moved | |
2877 | * @from: mem_cgroup which the entry is moved from | |
2878 | * @to: mem_cgroup which the entry is moved to | |
2879 | * | |
2880 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2881 | * as the mem_cgroup's id of @from. | |
2882 | * | |
2883 | * Returns 0 on success, -EINVAL on failure. | |
2884 | * | |
3e32cb2e | 2885 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2886 | * both res and memsw, and called css_get(). |
2887 | */ | |
2888 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2889 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2890 | { |
2891 | unsigned short old_id, new_id; | |
2892 | ||
34c00c31 LZ |
2893 | old_id = mem_cgroup_id(from); |
2894 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2895 | |
2896 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
2897 | mod_memcg_state(from, MEMCG_SWAP, -1); |
2898 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
2899 | return 0; |
2900 | } | |
2901 | return -EINVAL; | |
2902 | } | |
2903 | #else | |
2904 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2905 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2906 | { |
2907 | return -EINVAL; | |
2908 | } | |
8c7c6e34 | 2909 | #endif |
d13d1443 | 2910 | |
bbec2e15 | 2911 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 2912 | |
bbec2e15 RG |
2913 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
2914 | unsigned long max, bool memsw) | |
628f4235 | 2915 | { |
3e32cb2e | 2916 | bool enlarge = false; |
bb4a7ea2 | 2917 | bool drained = false; |
3e32cb2e | 2918 | int ret; |
c054a78c YZ |
2919 | bool limits_invariant; |
2920 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 2921 | |
3e32cb2e | 2922 | do { |
628f4235 KH |
2923 | if (signal_pending(current)) { |
2924 | ret = -EINTR; | |
2925 | break; | |
2926 | } | |
3e32cb2e | 2927 | |
bbec2e15 | 2928 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
2929 | /* |
2930 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 2931 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 2932 | */ |
bbec2e15 RG |
2933 | limits_invariant = memsw ? max >= memcg->memory.max : |
2934 | max <= memcg->memsw.max; | |
c054a78c | 2935 | if (!limits_invariant) { |
bbec2e15 | 2936 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 2937 | ret = -EINVAL; |
8c7c6e34 KH |
2938 | break; |
2939 | } | |
bbec2e15 | 2940 | if (max > counter->max) |
3e32cb2e | 2941 | enlarge = true; |
bbec2e15 RG |
2942 | ret = page_counter_set_max(counter, max); |
2943 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
2944 | |
2945 | if (!ret) | |
2946 | break; | |
2947 | ||
bb4a7ea2 SB |
2948 | if (!drained) { |
2949 | drain_all_stock(memcg); | |
2950 | drained = true; | |
2951 | continue; | |
2952 | } | |
2953 | ||
1ab5c056 AR |
2954 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
2955 | GFP_KERNEL, !memsw)) { | |
2956 | ret = -EBUSY; | |
2957 | break; | |
2958 | } | |
2959 | } while (true); | |
3e32cb2e | 2960 | |
3c11ecf4 KH |
2961 | if (!ret && enlarge) |
2962 | memcg_oom_recover(memcg); | |
3e32cb2e | 2963 | |
628f4235 KH |
2964 | return ret; |
2965 | } | |
2966 | ||
ef8f2327 | 2967 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
2968 | gfp_t gfp_mask, |
2969 | unsigned long *total_scanned) | |
2970 | { | |
2971 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 2972 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
2973 | unsigned long reclaimed; |
2974 | int loop = 0; | |
ef8f2327 | 2975 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 2976 | unsigned long excess; |
0608f43d AM |
2977 | unsigned long nr_scanned; |
2978 | ||
2979 | if (order > 0) | |
2980 | return 0; | |
2981 | ||
ef8f2327 | 2982 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
2983 | |
2984 | /* | |
2985 | * Do not even bother to check the largest node if the root | |
2986 | * is empty. Do it lockless to prevent lock bouncing. Races | |
2987 | * are acceptable as soft limit is best effort anyway. | |
2988 | */ | |
bfc7228b | 2989 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
2990 | return 0; |
2991 | ||
0608f43d AM |
2992 | /* |
2993 | * This loop can run a while, specially if mem_cgroup's continuously | |
2994 | * keep exceeding their soft limit and putting the system under | |
2995 | * pressure | |
2996 | */ | |
2997 | do { | |
2998 | if (next_mz) | |
2999 | mz = next_mz; | |
3000 | else | |
3001 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3002 | if (!mz) | |
3003 | break; | |
3004 | ||
3005 | nr_scanned = 0; | |
ef8f2327 | 3006 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
3007 | gfp_mask, &nr_scanned); |
3008 | nr_reclaimed += reclaimed; | |
3009 | *total_scanned += nr_scanned; | |
0a31bc97 | 3010 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3011 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3012 | |
3013 | /* | |
3014 | * If we failed to reclaim anything from this memory cgroup | |
3015 | * it is time to move on to the next cgroup | |
3016 | */ | |
3017 | next_mz = NULL; | |
bc2f2e7f VD |
3018 | if (!reclaimed) |
3019 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3020 | ||
3e32cb2e | 3021 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3022 | /* |
3023 | * One school of thought says that we should not add | |
3024 | * back the node to the tree if reclaim returns 0. | |
3025 | * But our reclaim could return 0, simply because due | |
3026 | * to priority we are exposing a smaller subset of | |
3027 | * memory to reclaim from. Consider this as a longer | |
3028 | * term TODO. | |
3029 | */ | |
3030 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3031 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3032 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3033 | css_put(&mz->memcg->css); |
3034 | loop++; | |
3035 | /* | |
3036 | * Could not reclaim anything and there are no more | |
3037 | * mem cgroups to try or we seem to be looping without | |
3038 | * reclaiming anything. | |
3039 | */ | |
3040 | if (!nr_reclaimed && | |
3041 | (next_mz == NULL || | |
3042 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3043 | break; | |
3044 | } while (!nr_reclaimed); | |
3045 | if (next_mz) | |
3046 | css_put(&next_mz->memcg->css); | |
3047 | return nr_reclaimed; | |
3048 | } | |
3049 | ||
ea280e7b TH |
3050 | /* |
3051 | * Test whether @memcg has children, dead or alive. Note that this | |
3052 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
3053 | * returns %true if there are child csses according to the cgroup | |
3054 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
3055 | */ | |
b5f99b53 GC |
3056 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
3057 | { | |
ea280e7b TH |
3058 | bool ret; |
3059 | ||
ea280e7b TH |
3060 | rcu_read_lock(); |
3061 | ret = css_next_child(NULL, &memcg->css); | |
3062 | rcu_read_unlock(); | |
3063 | return ret; | |
b5f99b53 GC |
3064 | } |
3065 | ||
c26251f9 | 3066 | /* |
51038171 | 3067 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3068 | * |
3069 | * Caller is responsible for holding css reference for memcg. | |
3070 | */ | |
3071 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3072 | { | |
3073 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 3074 | |
c1e862c1 KH |
3075 | /* we call try-to-free pages for make this cgroup empty */ |
3076 | lru_add_drain_all(); | |
d12c60f6 JS |
3077 | |
3078 | drain_all_stock(memcg); | |
3079 | ||
f817ed48 | 3080 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3081 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3082 | int progress; |
c1e862c1 | 3083 | |
c26251f9 MH |
3084 | if (signal_pending(current)) |
3085 | return -EINTR; | |
3086 | ||
b70a2a21 JW |
3087 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3088 | GFP_KERNEL, true); | |
c1e862c1 | 3089 | if (!progress) { |
f817ed48 | 3090 | nr_retries--; |
c1e862c1 | 3091 | /* maybe some writeback is necessary */ |
8aa7e847 | 3092 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3093 | } |
f817ed48 KH |
3094 | |
3095 | } | |
ab5196c2 MH |
3096 | |
3097 | return 0; | |
cc847582 KH |
3098 | } |
3099 | ||
6770c64e TH |
3100 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3101 | char *buf, size_t nbytes, | |
3102 | loff_t off) | |
c1e862c1 | 3103 | { |
6770c64e | 3104 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3105 | |
d8423011 MH |
3106 | if (mem_cgroup_is_root(memcg)) |
3107 | return -EINVAL; | |
6770c64e | 3108 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3109 | } |
3110 | ||
182446d0 TH |
3111 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3112 | struct cftype *cft) | |
18f59ea7 | 3113 | { |
182446d0 | 3114 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
3115 | } |
3116 | ||
182446d0 TH |
3117 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3118 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
3119 | { |
3120 | int retval = 0; | |
182446d0 | 3121 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 3122 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 3123 | |
567fb435 | 3124 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 3125 | return 0; |
567fb435 | 3126 | |
18f59ea7 | 3127 | /* |
af901ca1 | 3128 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3129 | * in the child subtrees. If it is unset, then the change can |
3130 | * occur, provided the current cgroup has no children. | |
3131 | * | |
3132 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3133 | * set if there are no children. | |
3134 | */ | |
c0ff4b85 | 3135 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 3136 | (val == 1 || val == 0)) { |
ea280e7b | 3137 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 3138 | memcg->use_hierarchy = val; |
18f59ea7 BS |
3139 | else |
3140 | retval = -EBUSY; | |
3141 | } else | |
3142 | retval = -EINVAL; | |
567fb435 | 3143 | |
18f59ea7 BS |
3144 | return retval; |
3145 | } | |
3146 | ||
6f646156 | 3147 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3148 | { |
42a30035 | 3149 | unsigned long val; |
ce00a967 | 3150 | |
3e32cb2e | 3151 | if (mem_cgroup_is_root(memcg)) { |
42a30035 JW |
3152 | val = memcg_page_state(memcg, MEMCG_CACHE) + |
3153 | memcg_page_state(memcg, MEMCG_RSS); | |
3154 | if (swap) | |
3155 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3156 | } else { |
ce00a967 | 3157 | if (!swap) |
3e32cb2e | 3158 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3159 | else |
3e32cb2e | 3160 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3161 | } |
c12176d3 | 3162 | return val; |
ce00a967 JW |
3163 | } |
3164 | ||
3e32cb2e JW |
3165 | enum { |
3166 | RES_USAGE, | |
3167 | RES_LIMIT, | |
3168 | RES_MAX_USAGE, | |
3169 | RES_FAILCNT, | |
3170 | RES_SOFT_LIMIT, | |
3171 | }; | |
ce00a967 | 3172 | |
791badbd | 3173 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3174 | struct cftype *cft) |
8cdea7c0 | 3175 | { |
182446d0 | 3176 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3177 | struct page_counter *counter; |
af36f906 | 3178 | |
3e32cb2e | 3179 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3180 | case _MEM: |
3e32cb2e JW |
3181 | counter = &memcg->memory; |
3182 | break; | |
8c7c6e34 | 3183 | case _MEMSWAP: |
3e32cb2e JW |
3184 | counter = &memcg->memsw; |
3185 | break; | |
510fc4e1 | 3186 | case _KMEM: |
3e32cb2e | 3187 | counter = &memcg->kmem; |
510fc4e1 | 3188 | break; |
d55f90bf | 3189 | case _TCP: |
0db15298 | 3190 | counter = &memcg->tcpmem; |
d55f90bf | 3191 | break; |
8c7c6e34 KH |
3192 | default: |
3193 | BUG(); | |
8c7c6e34 | 3194 | } |
3e32cb2e JW |
3195 | |
3196 | switch (MEMFILE_ATTR(cft->private)) { | |
3197 | case RES_USAGE: | |
3198 | if (counter == &memcg->memory) | |
c12176d3 | 3199 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3200 | if (counter == &memcg->memsw) |
c12176d3 | 3201 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3202 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3203 | case RES_LIMIT: | |
bbec2e15 | 3204 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3205 | case RES_MAX_USAGE: |
3206 | return (u64)counter->watermark * PAGE_SIZE; | |
3207 | case RES_FAILCNT: | |
3208 | return counter->failcnt; | |
3209 | case RES_SOFT_LIMIT: | |
3210 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3211 | default: | |
3212 | BUG(); | |
3213 | } | |
8cdea7c0 | 3214 | } |
510fc4e1 | 3215 | |
84c07d11 | 3216 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3217 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3218 | { |
d6441637 VD |
3219 | int memcg_id; |
3220 | ||
b313aeee VD |
3221 | if (cgroup_memory_nokmem) |
3222 | return 0; | |
3223 | ||
2a4db7eb | 3224 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3225 | BUG_ON(memcg->kmem_state); |
d6441637 | 3226 | |
f3bb3043 | 3227 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3228 | if (memcg_id < 0) |
3229 | return memcg_id; | |
d6441637 | 3230 | |
ef12947c | 3231 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 3232 | /* |
567e9ab2 | 3233 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 3234 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
3235 | * guarantee no one starts accounting before all call sites are |
3236 | * patched. | |
3237 | */ | |
900a38f0 | 3238 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3239 | memcg->kmem_state = KMEM_ONLINE; |
bc2791f8 | 3240 | INIT_LIST_HEAD(&memcg->kmem_caches); |
0b8f73e1 JW |
3241 | |
3242 | return 0; | |
d6441637 VD |
3243 | } |
3244 | ||
8e0a8912 JW |
3245 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3246 | { | |
3247 | struct cgroup_subsys_state *css; | |
3248 | struct mem_cgroup *parent, *child; | |
3249 | int kmemcg_id; | |
3250 | ||
3251 | if (memcg->kmem_state != KMEM_ONLINE) | |
3252 | return; | |
3253 | /* | |
3254 | * Clear the online state before clearing memcg_caches array | |
3255 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
3256 | * guarantees that no cache will be created for this cgroup | |
3257 | * after we are done (see memcg_create_kmem_cache()). | |
3258 | */ | |
3259 | memcg->kmem_state = KMEM_ALLOCATED; | |
3260 | ||
8e0a8912 JW |
3261 | parent = parent_mem_cgroup(memcg); |
3262 | if (!parent) | |
3263 | parent = root_mem_cgroup; | |
3264 | ||
fb2f2b0a RG |
3265 | memcg_deactivate_kmem_caches(memcg, parent); |
3266 | ||
3267 | kmemcg_id = memcg->kmemcg_id; | |
3268 | BUG_ON(kmemcg_id < 0); | |
3269 | ||
8e0a8912 JW |
3270 | /* |
3271 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3272 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3273 | * to ones of the parent. After we have finished, all list_lrus | |
3274 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3275 | * ordering is imposed by list_lru_node->lock taken by | |
3276 | * memcg_drain_all_list_lrus(). | |
3277 | */ | |
3a06bb78 | 3278 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3279 | css_for_each_descendant_pre(css, &memcg->css) { |
3280 | child = mem_cgroup_from_css(css); | |
3281 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3282 | child->kmemcg_id = parent->kmemcg_id; | |
3283 | if (!memcg->use_hierarchy) | |
3284 | break; | |
3285 | } | |
3a06bb78 TH |
3286 | rcu_read_unlock(); |
3287 | ||
9bec5c35 | 3288 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3289 | |
3290 | memcg_free_cache_id(kmemcg_id); | |
3291 | } | |
3292 | ||
3293 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3294 | { | |
0b8f73e1 JW |
3295 | /* css_alloc() failed, offlining didn't happen */ |
3296 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3297 | memcg_offline_kmem(memcg); | |
3298 | ||
8e0a8912 | 3299 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
f0a3a24b | 3300 | WARN_ON(!list_empty(&memcg->kmem_caches)); |
8e0a8912 | 3301 | static_branch_dec(&memcg_kmem_enabled_key); |
8e0a8912 | 3302 | } |
8e0a8912 | 3303 | } |
d6441637 | 3304 | #else |
0b8f73e1 | 3305 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3306 | { |
3307 | return 0; | |
3308 | } | |
3309 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3310 | { | |
3311 | } | |
3312 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3313 | { | |
3314 | } | |
84c07d11 | 3315 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3316 | |
bbec2e15 RG |
3317 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3318 | unsigned long max) | |
d6441637 | 3319 | { |
b313aeee | 3320 | int ret; |
127424c8 | 3321 | |
bbec2e15 RG |
3322 | mutex_lock(&memcg_max_mutex); |
3323 | ret = page_counter_set_max(&memcg->kmem, max); | |
3324 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3325 | return ret; |
d6441637 | 3326 | } |
510fc4e1 | 3327 | |
bbec2e15 | 3328 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3329 | { |
3330 | int ret; | |
3331 | ||
bbec2e15 | 3332 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3333 | |
bbec2e15 | 3334 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3335 | if (ret) |
3336 | goto out; | |
3337 | ||
0db15298 | 3338 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3339 | /* |
3340 | * The active flag needs to be written after the static_key | |
3341 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3342 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3343 | * for details, and note that we don't mark any socket as | |
3344 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3345 | * |
3346 | * We need to do this, because static_keys will span multiple | |
3347 | * sites, but we can't control their order. If we mark a socket | |
3348 | * as accounted, but the accounting functions are not patched in | |
3349 | * yet, we'll lose accounting. | |
3350 | * | |
2d758073 | 3351 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3352 | * because when this value change, the code to process it is not |
3353 | * patched in yet. | |
3354 | */ | |
3355 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3356 | memcg->tcpmem_active = true; |
d55f90bf VD |
3357 | } |
3358 | out: | |
bbec2e15 | 3359 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3360 | return ret; |
3361 | } | |
d55f90bf | 3362 | |
628f4235 KH |
3363 | /* |
3364 | * The user of this function is... | |
3365 | * RES_LIMIT. | |
3366 | */ | |
451af504 TH |
3367 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3368 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3369 | { |
451af504 | 3370 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3371 | unsigned long nr_pages; |
628f4235 KH |
3372 | int ret; |
3373 | ||
451af504 | 3374 | buf = strstrip(buf); |
650c5e56 | 3375 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3376 | if (ret) |
3377 | return ret; | |
af36f906 | 3378 | |
3e32cb2e | 3379 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3380 | case RES_LIMIT: |
4b3bde4c BS |
3381 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3382 | ret = -EINVAL; | |
3383 | break; | |
3384 | } | |
3e32cb2e JW |
3385 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3386 | case _MEM: | |
bbec2e15 | 3387 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3388 | break; |
3e32cb2e | 3389 | case _MEMSWAP: |
bbec2e15 | 3390 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3391 | break; |
3e32cb2e | 3392 | case _KMEM: |
bbec2e15 | 3393 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3394 | break; |
d55f90bf | 3395 | case _TCP: |
bbec2e15 | 3396 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3397 | break; |
3e32cb2e | 3398 | } |
296c81d8 | 3399 | break; |
3e32cb2e JW |
3400 | case RES_SOFT_LIMIT: |
3401 | memcg->soft_limit = nr_pages; | |
3402 | ret = 0; | |
628f4235 KH |
3403 | break; |
3404 | } | |
451af504 | 3405 | return ret ?: nbytes; |
8cdea7c0 BS |
3406 | } |
3407 | ||
6770c64e TH |
3408 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3409 | size_t nbytes, loff_t off) | |
c84872e1 | 3410 | { |
6770c64e | 3411 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3412 | struct page_counter *counter; |
c84872e1 | 3413 | |
3e32cb2e JW |
3414 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3415 | case _MEM: | |
3416 | counter = &memcg->memory; | |
3417 | break; | |
3418 | case _MEMSWAP: | |
3419 | counter = &memcg->memsw; | |
3420 | break; | |
3421 | case _KMEM: | |
3422 | counter = &memcg->kmem; | |
3423 | break; | |
d55f90bf | 3424 | case _TCP: |
0db15298 | 3425 | counter = &memcg->tcpmem; |
d55f90bf | 3426 | break; |
3e32cb2e JW |
3427 | default: |
3428 | BUG(); | |
3429 | } | |
af36f906 | 3430 | |
3e32cb2e | 3431 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3432 | case RES_MAX_USAGE: |
3e32cb2e | 3433 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3434 | break; |
3435 | case RES_FAILCNT: | |
3e32cb2e | 3436 | counter->failcnt = 0; |
29f2a4da | 3437 | break; |
3e32cb2e JW |
3438 | default: |
3439 | BUG(); | |
29f2a4da | 3440 | } |
f64c3f54 | 3441 | |
6770c64e | 3442 | return nbytes; |
c84872e1 PE |
3443 | } |
3444 | ||
182446d0 | 3445 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3446 | struct cftype *cft) |
3447 | { | |
182446d0 | 3448 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3449 | } |
3450 | ||
02491447 | 3451 | #ifdef CONFIG_MMU |
182446d0 | 3452 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3453 | struct cftype *cft, u64 val) |
3454 | { | |
182446d0 | 3455 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3456 | |
1dfab5ab | 3457 | if (val & ~MOVE_MASK) |
7dc74be0 | 3458 | return -EINVAL; |
ee5e8472 | 3459 | |
7dc74be0 | 3460 | /* |
ee5e8472 GC |
3461 | * No kind of locking is needed in here, because ->can_attach() will |
3462 | * check this value once in the beginning of the process, and then carry | |
3463 | * on with stale data. This means that changes to this value will only | |
3464 | * affect task migrations starting after the change. | |
7dc74be0 | 3465 | */ |
c0ff4b85 | 3466 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3467 | return 0; |
3468 | } | |
02491447 | 3469 | #else |
182446d0 | 3470 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3471 | struct cftype *cft, u64 val) |
3472 | { | |
3473 | return -ENOSYS; | |
3474 | } | |
3475 | #endif | |
7dc74be0 | 3476 | |
406eb0c9 | 3477 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3478 | |
3479 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3480 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3481 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3482 | ||
3483 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
3484 | int nid, unsigned int lru_mask) | |
3485 | { | |
3486 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); | |
3487 | unsigned long nr = 0; | |
3488 | enum lru_list lru; | |
3489 | ||
3490 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3491 | ||
3492 | for_each_lru(lru) { | |
3493 | if (!(BIT(lru) & lru_mask)) | |
3494 | continue; | |
205b20cc | 3495 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); |
113b7dfd JW |
3496 | } |
3497 | return nr; | |
3498 | } | |
3499 | ||
3500 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
3501 | unsigned int lru_mask) | |
3502 | { | |
3503 | unsigned long nr = 0; | |
3504 | enum lru_list lru; | |
3505 | ||
3506 | for_each_lru(lru) { | |
3507 | if (!(BIT(lru) & lru_mask)) | |
3508 | continue; | |
205b20cc | 3509 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); |
113b7dfd JW |
3510 | } |
3511 | return nr; | |
3512 | } | |
3513 | ||
2da8ca82 | 3514 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3515 | { |
25485de6 GT |
3516 | struct numa_stat { |
3517 | const char *name; | |
3518 | unsigned int lru_mask; | |
3519 | }; | |
3520 | ||
3521 | static const struct numa_stat stats[] = { | |
3522 | { "total", LRU_ALL }, | |
3523 | { "file", LRU_ALL_FILE }, | |
3524 | { "anon", LRU_ALL_ANON }, | |
3525 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3526 | }; | |
3527 | const struct numa_stat *stat; | |
406eb0c9 | 3528 | int nid; |
25485de6 | 3529 | unsigned long nr; |
aa9694bb | 3530 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3531 | |
25485de6 GT |
3532 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3533 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3534 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3535 | for_each_node_state(nid, N_MEMORY) { | |
3536 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3537 | stat->lru_mask); | |
3538 | seq_printf(m, " N%d=%lu", nid, nr); | |
3539 | } | |
3540 | seq_putc(m, '\n'); | |
406eb0c9 | 3541 | } |
406eb0c9 | 3542 | |
071aee13 YH |
3543 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3544 | struct mem_cgroup *iter; | |
3545 | ||
3546 | nr = 0; | |
3547 | for_each_mem_cgroup_tree(iter, memcg) | |
3548 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3549 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3550 | for_each_node_state(nid, N_MEMORY) { | |
3551 | nr = 0; | |
3552 | for_each_mem_cgroup_tree(iter, memcg) | |
3553 | nr += mem_cgroup_node_nr_lru_pages( | |
3554 | iter, nid, stat->lru_mask); | |
3555 | seq_printf(m, " N%d=%lu", nid, nr); | |
3556 | } | |
3557 | seq_putc(m, '\n'); | |
406eb0c9 | 3558 | } |
406eb0c9 | 3559 | |
406eb0c9 YH |
3560 | return 0; |
3561 | } | |
3562 | #endif /* CONFIG_NUMA */ | |
3563 | ||
c8713d0b JW |
3564 | static const unsigned int memcg1_stats[] = { |
3565 | MEMCG_CACHE, | |
3566 | MEMCG_RSS, | |
3567 | MEMCG_RSS_HUGE, | |
3568 | NR_SHMEM, | |
3569 | NR_FILE_MAPPED, | |
3570 | NR_FILE_DIRTY, | |
3571 | NR_WRITEBACK, | |
3572 | MEMCG_SWAP, | |
3573 | }; | |
3574 | ||
3575 | static const char *const memcg1_stat_names[] = { | |
3576 | "cache", | |
3577 | "rss", | |
3578 | "rss_huge", | |
3579 | "shmem", | |
3580 | "mapped_file", | |
3581 | "dirty", | |
3582 | "writeback", | |
3583 | "swap", | |
3584 | }; | |
3585 | ||
df0e53d0 | 3586 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3587 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
3588 | PGPGIN, |
3589 | PGPGOUT, | |
3590 | PGFAULT, | |
3591 | PGMAJFAULT, | |
3592 | }; | |
3593 | ||
3594 | static const char *const memcg1_event_names[] = { | |
3595 | "pgpgin", | |
3596 | "pgpgout", | |
3597 | "pgfault", | |
3598 | "pgmajfault", | |
3599 | }; | |
3600 | ||
2da8ca82 | 3601 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3602 | { |
aa9694bb | 3603 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 3604 | unsigned long memory, memsw; |
af7c4b0e JW |
3605 | struct mem_cgroup *mi; |
3606 | unsigned int i; | |
406eb0c9 | 3607 | |
71cd3113 | 3608 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c RS |
3609 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3610 | ||
71cd3113 JW |
3611 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
3612 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) | |
1dd3a273 | 3613 | continue; |
71cd3113 | 3614 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], |
205b20cc | 3615 | memcg_page_state_local(memcg, memcg1_stats[i]) * |
71cd3113 | 3616 | PAGE_SIZE); |
1dd3a273 | 3617 | } |
7b854121 | 3618 | |
df0e53d0 JW |
3619 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3620 | seq_printf(m, "%s %lu\n", memcg1_event_names[i], | |
205b20cc | 3621 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
3622 | |
3623 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3624 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
205b20cc | 3625 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 3626 | PAGE_SIZE); |
af7c4b0e | 3627 | |
14067bb3 | 3628 | /* Hierarchical information */ |
3e32cb2e JW |
3629 | memory = memsw = PAGE_COUNTER_MAX; |
3630 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
bbec2e15 RG |
3631 | memory = min(memory, mi->memory.max); |
3632 | memsw = min(memsw, mi->memsw.max); | |
fee7b548 | 3633 | } |
3e32cb2e JW |
3634 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3635 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3636 | if (do_memsw_account()) |
3e32cb2e JW |
3637 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3638 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3639 | |
8de7ecc6 | 3640 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
71cd3113 | 3641 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3642 | continue; |
8de7ecc6 | 3643 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
dd923990 YS |
3644 | (u64)memcg_page_state(memcg, memcg1_stats[i]) * |
3645 | PAGE_SIZE); | |
af7c4b0e JW |
3646 | } |
3647 | ||
8de7ecc6 SB |
3648 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3649 | seq_printf(m, "total_%s %llu\n", memcg1_event_names[i], | |
dd923990 | 3650 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 3651 | |
8de7ecc6 SB |
3652 | for (i = 0; i < NR_LRU_LISTS; i++) |
3653 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], | |
42a30035 JW |
3654 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
3655 | PAGE_SIZE); | |
14067bb3 | 3656 | |
7f016ee8 | 3657 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 3658 | { |
ef8f2327 MG |
3659 | pg_data_t *pgdat; |
3660 | struct mem_cgroup_per_node *mz; | |
89abfab1 | 3661 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3662 | unsigned long recent_rotated[2] = {0, 0}; |
3663 | unsigned long recent_scanned[2] = {0, 0}; | |
3664 | ||
ef8f2327 MG |
3665 | for_each_online_pgdat(pgdat) { |
3666 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
3667 | rstat = &mz->lruvec.reclaim_stat; | |
7f016ee8 | 3668 | |
ef8f2327 MG |
3669 | recent_rotated[0] += rstat->recent_rotated[0]; |
3670 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3671 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3672 | recent_scanned[1] += rstat->recent_scanned[1]; | |
3673 | } | |
78ccf5b5 JW |
3674 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3675 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3676 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3677 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3678 | } |
3679 | #endif | |
3680 | ||
d2ceb9b7 KH |
3681 | return 0; |
3682 | } | |
3683 | ||
182446d0 TH |
3684 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3685 | struct cftype *cft) | |
a7885eb8 | 3686 | { |
182446d0 | 3687 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3688 | |
1f4c025b | 3689 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3690 | } |
3691 | ||
182446d0 TH |
3692 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3693 | struct cftype *cft, u64 val) | |
a7885eb8 | 3694 | { |
182446d0 | 3695 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3696 | |
3dae7fec | 3697 | if (val > 100) |
a7885eb8 KM |
3698 | return -EINVAL; |
3699 | ||
14208b0e | 3700 | if (css->parent) |
3dae7fec JW |
3701 | memcg->swappiness = val; |
3702 | else | |
3703 | vm_swappiness = val; | |
068b38c1 | 3704 | |
a7885eb8 KM |
3705 | return 0; |
3706 | } | |
3707 | ||
2e72b634 KS |
3708 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3709 | { | |
3710 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3711 | unsigned long usage; |
2e72b634 KS |
3712 | int i; |
3713 | ||
3714 | rcu_read_lock(); | |
3715 | if (!swap) | |
2c488db2 | 3716 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3717 | else |
2c488db2 | 3718 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3719 | |
3720 | if (!t) | |
3721 | goto unlock; | |
3722 | ||
ce00a967 | 3723 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3724 | |
3725 | /* | |
748dad36 | 3726 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3727 | * If it's not true, a threshold was crossed after last |
3728 | * call of __mem_cgroup_threshold(). | |
3729 | */ | |
5407a562 | 3730 | i = t->current_threshold; |
2e72b634 KS |
3731 | |
3732 | /* | |
3733 | * Iterate backward over array of thresholds starting from | |
3734 | * current_threshold and check if a threshold is crossed. | |
3735 | * If none of thresholds below usage is crossed, we read | |
3736 | * only one element of the array here. | |
3737 | */ | |
3738 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3739 | eventfd_signal(t->entries[i].eventfd, 1); | |
3740 | ||
3741 | /* i = current_threshold + 1 */ | |
3742 | i++; | |
3743 | ||
3744 | /* | |
3745 | * Iterate forward over array of thresholds starting from | |
3746 | * current_threshold+1 and check if a threshold is crossed. | |
3747 | * If none of thresholds above usage is crossed, we read | |
3748 | * only one element of the array here. | |
3749 | */ | |
3750 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3751 | eventfd_signal(t->entries[i].eventfd, 1); | |
3752 | ||
3753 | /* Update current_threshold */ | |
5407a562 | 3754 | t->current_threshold = i - 1; |
2e72b634 KS |
3755 | unlock: |
3756 | rcu_read_unlock(); | |
3757 | } | |
3758 | ||
3759 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3760 | { | |
ad4ca5f4 KS |
3761 | while (memcg) { |
3762 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3763 | if (do_memsw_account()) |
ad4ca5f4 KS |
3764 | __mem_cgroup_threshold(memcg, true); |
3765 | ||
3766 | memcg = parent_mem_cgroup(memcg); | |
3767 | } | |
2e72b634 KS |
3768 | } |
3769 | ||
3770 | static int compare_thresholds(const void *a, const void *b) | |
3771 | { | |
3772 | const struct mem_cgroup_threshold *_a = a; | |
3773 | const struct mem_cgroup_threshold *_b = b; | |
3774 | ||
2bff24a3 GT |
3775 | if (_a->threshold > _b->threshold) |
3776 | return 1; | |
3777 | ||
3778 | if (_a->threshold < _b->threshold) | |
3779 | return -1; | |
3780 | ||
3781 | return 0; | |
2e72b634 KS |
3782 | } |
3783 | ||
c0ff4b85 | 3784 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3785 | { |
3786 | struct mem_cgroup_eventfd_list *ev; | |
3787 | ||
2bcf2e92 MH |
3788 | spin_lock(&memcg_oom_lock); |
3789 | ||
c0ff4b85 | 3790 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3791 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3792 | |
3793 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3794 | return 0; |
3795 | } | |
3796 | ||
c0ff4b85 | 3797 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3798 | { |
7d74b06f KH |
3799 | struct mem_cgroup *iter; |
3800 | ||
c0ff4b85 | 3801 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3802 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3803 | } |
3804 | ||
59b6f873 | 3805 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3806 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3807 | { |
2c488db2 KS |
3808 | struct mem_cgroup_thresholds *thresholds; |
3809 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3810 | unsigned long threshold; |
3811 | unsigned long usage; | |
2c488db2 | 3812 | int i, size, ret; |
2e72b634 | 3813 | |
650c5e56 | 3814 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3815 | if (ret) |
3816 | return ret; | |
3817 | ||
3818 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3819 | |
05b84301 | 3820 | if (type == _MEM) { |
2c488db2 | 3821 | thresholds = &memcg->thresholds; |
ce00a967 | 3822 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3823 | } else if (type == _MEMSWAP) { |
2c488db2 | 3824 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3825 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3826 | } else |
2e72b634 KS |
3827 | BUG(); |
3828 | ||
2e72b634 | 3829 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3830 | if (thresholds->primary) |
2e72b634 KS |
3831 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3832 | ||
2c488db2 | 3833 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3834 | |
3835 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 3836 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 3837 | if (!new) { |
2e72b634 KS |
3838 | ret = -ENOMEM; |
3839 | goto unlock; | |
3840 | } | |
2c488db2 | 3841 | new->size = size; |
2e72b634 KS |
3842 | |
3843 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3844 | if (thresholds->primary) { |
3845 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3846 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3847 | } |
3848 | ||
2e72b634 | 3849 | /* Add new threshold */ |
2c488db2 KS |
3850 | new->entries[size - 1].eventfd = eventfd; |
3851 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3852 | |
3853 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3854 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3855 | compare_thresholds, NULL); |
3856 | ||
3857 | /* Find current threshold */ | |
2c488db2 | 3858 | new->current_threshold = -1; |
2e72b634 | 3859 | for (i = 0; i < size; i++) { |
748dad36 | 3860 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3861 | /* |
2c488db2 KS |
3862 | * new->current_threshold will not be used until |
3863 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3864 | * it here. |
3865 | */ | |
2c488db2 | 3866 | ++new->current_threshold; |
748dad36 SZ |
3867 | } else |
3868 | break; | |
2e72b634 KS |
3869 | } |
3870 | ||
2c488db2 KS |
3871 | /* Free old spare buffer and save old primary buffer as spare */ |
3872 | kfree(thresholds->spare); | |
3873 | thresholds->spare = thresholds->primary; | |
3874 | ||
3875 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3876 | |
907860ed | 3877 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3878 | synchronize_rcu(); |
3879 | ||
2e72b634 KS |
3880 | unlock: |
3881 | mutex_unlock(&memcg->thresholds_lock); | |
3882 | ||
3883 | return ret; | |
3884 | } | |
3885 | ||
59b6f873 | 3886 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3887 | struct eventfd_ctx *eventfd, const char *args) |
3888 | { | |
59b6f873 | 3889 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3890 | } |
3891 | ||
59b6f873 | 3892 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3893 | struct eventfd_ctx *eventfd, const char *args) |
3894 | { | |
59b6f873 | 3895 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3896 | } |
3897 | ||
59b6f873 | 3898 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3899 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3900 | { |
2c488db2 KS |
3901 | struct mem_cgroup_thresholds *thresholds; |
3902 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3903 | unsigned long usage; |
2c488db2 | 3904 | int i, j, size; |
2e72b634 KS |
3905 | |
3906 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3907 | |
3908 | if (type == _MEM) { | |
2c488db2 | 3909 | thresholds = &memcg->thresholds; |
ce00a967 | 3910 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3911 | } else if (type == _MEMSWAP) { |
2c488db2 | 3912 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3913 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3914 | } else |
2e72b634 KS |
3915 | BUG(); |
3916 | ||
371528ca AV |
3917 | if (!thresholds->primary) |
3918 | goto unlock; | |
3919 | ||
2e72b634 KS |
3920 | /* Check if a threshold crossed before removing */ |
3921 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3922 | ||
3923 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3924 | size = 0; |
3925 | for (i = 0; i < thresholds->primary->size; i++) { | |
3926 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3927 | size++; |
3928 | } | |
3929 | ||
2c488db2 | 3930 | new = thresholds->spare; |
907860ed | 3931 | |
2e72b634 KS |
3932 | /* Set thresholds array to NULL if we don't have thresholds */ |
3933 | if (!size) { | |
2c488db2 KS |
3934 | kfree(new); |
3935 | new = NULL; | |
907860ed | 3936 | goto swap_buffers; |
2e72b634 KS |
3937 | } |
3938 | ||
2c488db2 | 3939 | new->size = size; |
2e72b634 KS |
3940 | |
3941 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3942 | new->current_threshold = -1; |
3943 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3944 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3945 | continue; |
3946 | ||
2c488db2 | 3947 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3948 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3949 | /* |
2c488db2 | 3950 | * new->current_threshold will not be used |
2e72b634 KS |
3951 | * until rcu_assign_pointer(), so it's safe to increment |
3952 | * it here. | |
3953 | */ | |
2c488db2 | 3954 | ++new->current_threshold; |
2e72b634 KS |
3955 | } |
3956 | j++; | |
3957 | } | |
3958 | ||
907860ed | 3959 | swap_buffers: |
2c488db2 KS |
3960 | /* Swap primary and spare array */ |
3961 | thresholds->spare = thresholds->primary; | |
8c757763 | 3962 | |
2c488db2 | 3963 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3964 | |
907860ed | 3965 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3966 | synchronize_rcu(); |
6611d8d7 MC |
3967 | |
3968 | /* If all events are unregistered, free the spare array */ | |
3969 | if (!new) { | |
3970 | kfree(thresholds->spare); | |
3971 | thresholds->spare = NULL; | |
3972 | } | |
371528ca | 3973 | unlock: |
2e72b634 | 3974 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3975 | } |
c1e862c1 | 3976 | |
59b6f873 | 3977 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3978 | struct eventfd_ctx *eventfd) |
3979 | { | |
59b6f873 | 3980 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3981 | } |
3982 | ||
59b6f873 | 3983 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3984 | struct eventfd_ctx *eventfd) |
3985 | { | |
59b6f873 | 3986 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3987 | } |
3988 | ||
59b6f873 | 3989 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3990 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3991 | { |
9490ff27 | 3992 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3993 | |
9490ff27 KH |
3994 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3995 | if (!event) | |
3996 | return -ENOMEM; | |
3997 | ||
1af8efe9 | 3998 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3999 | |
4000 | event->eventfd = eventfd; | |
4001 | list_add(&event->list, &memcg->oom_notify); | |
4002 | ||
4003 | /* already in OOM ? */ | |
c2b42d3c | 4004 | if (memcg->under_oom) |
9490ff27 | 4005 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4006 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4007 | |
4008 | return 0; | |
4009 | } | |
4010 | ||
59b6f873 | 4011 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4012 | struct eventfd_ctx *eventfd) |
9490ff27 | 4013 | { |
9490ff27 | 4014 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4015 | |
1af8efe9 | 4016 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4017 | |
c0ff4b85 | 4018 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4019 | if (ev->eventfd == eventfd) { |
4020 | list_del(&ev->list); | |
4021 | kfree(ev); | |
4022 | } | |
4023 | } | |
4024 | ||
1af8efe9 | 4025 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4026 | } |
4027 | ||
2da8ca82 | 4028 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4029 | { |
aa9694bb | 4030 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4031 | |
791badbd | 4032 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4033 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4034 | seq_printf(sf, "oom_kill %lu\n", |
4035 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4036 | return 0; |
4037 | } | |
4038 | ||
182446d0 | 4039 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4040 | struct cftype *cft, u64 val) |
4041 | { | |
182446d0 | 4042 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4043 | |
4044 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 4045 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4046 | return -EINVAL; |
4047 | ||
c0ff4b85 | 4048 | memcg->oom_kill_disable = val; |
4d845ebf | 4049 | if (!val) |
c0ff4b85 | 4050 | memcg_oom_recover(memcg); |
3dae7fec | 4051 | |
3c11ecf4 KH |
4052 | return 0; |
4053 | } | |
4054 | ||
52ebea74 TH |
4055 | #ifdef CONFIG_CGROUP_WRITEBACK |
4056 | ||
841710aa TH |
4057 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4058 | { | |
4059 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4060 | } | |
4061 | ||
4062 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4063 | { | |
4064 | wb_domain_exit(&memcg->cgwb_domain); | |
4065 | } | |
4066 | ||
2529bb3a TH |
4067 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4068 | { | |
4069 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4070 | } | |
4071 | ||
841710aa TH |
4072 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4073 | { | |
4074 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4075 | ||
4076 | if (!memcg->css.parent) | |
4077 | return NULL; | |
4078 | ||
4079 | return &memcg->cgwb_domain; | |
4080 | } | |
4081 | ||
0b3d6e6f GT |
4082 | /* |
4083 | * idx can be of type enum memcg_stat_item or node_stat_item. | |
4084 | * Keep in sync with memcg_exact_page(). | |
4085 | */ | |
4086 | static unsigned long memcg_exact_page_state(struct mem_cgroup *memcg, int idx) | |
4087 | { | |
871789d4 | 4088 | long x = atomic_long_read(&memcg->vmstats[idx]); |
0b3d6e6f GT |
4089 | int cpu; |
4090 | ||
4091 | for_each_online_cpu(cpu) | |
871789d4 | 4092 | x += per_cpu_ptr(memcg->vmstats_percpu, cpu)->stat[idx]; |
0b3d6e6f GT |
4093 | if (x < 0) |
4094 | x = 0; | |
4095 | return x; | |
4096 | } | |
4097 | ||
c2aa723a TH |
4098 | /** |
4099 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4100 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4101 | * @pfilepages: out parameter for number of file pages |
4102 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4103 | * @pdirty: out parameter for number of dirty pages |
4104 | * @pwriteback: out parameter for number of pages under writeback | |
4105 | * | |
c5edf9cd TH |
4106 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4107 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4108 | * is a bit more involved. | |
c2aa723a | 4109 | * |
c5edf9cd TH |
4110 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4111 | * headroom is calculated as the lowest headroom of itself and the | |
4112 | * ancestors. Note that this doesn't consider the actual amount of | |
4113 | * available memory in the system. The caller should further cap | |
4114 | * *@pheadroom accordingly. | |
c2aa723a | 4115 | */ |
c5edf9cd TH |
4116 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4117 | unsigned long *pheadroom, unsigned long *pdirty, | |
4118 | unsigned long *pwriteback) | |
c2aa723a TH |
4119 | { |
4120 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4121 | struct mem_cgroup *parent; | |
c2aa723a | 4122 | |
0b3d6e6f | 4123 | *pdirty = memcg_exact_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a TH |
4124 | |
4125 | /* this should eventually include NR_UNSTABLE_NFS */ | |
0b3d6e6f | 4126 | *pwriteback = memcg_exact_page_state(memcg, NR_WRITEBACK); |
21d89d15 JW |
4127 | *pfilepages = memcg_exact_page_state(memcg, NR_INACTIVE_FILE) + |
4128 | memcg_exact_page_state(memcg, NR_ACTIVE_FILE); | |
c5edf9cd | 4129 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4130 | |
c2aa723a | 4131 | while ((parent = parent_mem_cgroup(memcg))) { |
bbec2e15 | 4132 | unsigned long ceiling = min(memcg->memory.max, memcg->high); |
c2aa723a TH |
4133 | unsigned long used = page_counter_read(&memcg->memory); |
4134 | ||
c5edf9cd | 4135 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4136 | memcg = parent; |
4137 | } | |
c2aa723a TH |
4138 | } |
4139 | ||
841710aa TH |
4140 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4141 | ||
4142 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4143 | { | |
4144 | return 0; | |
4145 | } | |
4146 | ||
4147 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4148 | { | |
4149 | } | |
4150 | ||
2529bb3a TH |
4151 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4152 | { | |
4153 | } | |
4154 | ||
52ebea74 TH |
4155 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4156 | ||
3bc942f3 TH |
4157 | /* |
4158 | * DO NOT USE IN NEW FILES. | |
4159 | * | |
4160 | * "cgroup.event_control" implementation. | |
4161 | * | |
4162 | * This is way over-engineered. It tries to support fully configurable | |
4163 | * events for each user. Such level of flexibility is completely | |
4164 | * unnecessary especially in the light of the planned unified hierarchy. | |
4165 | * | |
4166 | * Please deprecate this and replace with something simpler if at all | |
4167 | * possible. | |
4168 | */ | |
4169 | ||
79bd9814 TH |
4170 | /* |
4171 | * Unregister event and free resources. | |
4172 | * | |
4173 | * Gets called from workqueue. | |
4174 | */ | |
3bc942f3 | 4175 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4176 | { |
3bc942f3 TH |
4177 | struct mem_cgroup_event *event = |
4178 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4179 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4180 | |
4181 | remove_wait_queue(event->wqh, &event->wait); | |
4182 | ||
59b6f873 | 4183 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4184 | |
4185 | /* Notify userspace the event is going away. */ | |
4186 | eventfd_signal(event->eventfd, 1); | |
4187 | ||
4188 | eventfd_ctx_put(event->eventfd); | |
4189 | kfree(event); | |
59b6f873 | 4190 | css_put(&memcg->css); |
79bd9814 TH |
4191 | } |
4192 | ||
4193 | /* | |
a9a08845 | 4194 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4195 | * |
4196 | * Called with wqh->lock held and interrupts disabled. | |
4197 | */ | |
ac6424b9 | 4198 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4199 | int sync, void *key) |
79bd9814 | 4200 | { |
3bc942f3 TH |
4201 | struct mem_cgroup_event *event = |
4202 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4203 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4204 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4205 | |
a9a08845 | 4206 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4207 | /* |
4208 | * If the event has been detached at cgroup removal, we | |
4209 | * can simply return knowing the other side will cleanup | |
4210 | * for us. | |
4211 | * | |
4212 | * We can't race against event freeing since the other | |
4213 | * side will require wqh->lock via remove_wait_queue(), | |
4214 | * which we hold. | |
4215 | */ | |
fba94807 | 4216 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4217 | if (!list_empty(&event->list)) { |
4218 | list_del_init(&event->list); | |
4219 | /* | |
4220 | * We are in atomic context, but cgroup_event_remove() | |
4221 | * may sleep, so we have to call it in workqueue. | |
4222 | */ | |
4223 | schedule_work(&event->remove); | |
4224 | } | |
fba94807 | 4225 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4226 | } |
4227 | ||
4228 | return 0; | |
4229 | } | |
4230 | ||
3bc942f3 | 4231 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4232 | wait_queue_head_t *wqh, poll_table *pt) |
4233 | { | |
3bc942f3 TH |
4234 | struct mem_cgroup_event *event = |
4235 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4236 | |
4237 | event->wqh = wqh; | |
4238 | add_wait_queue(wqh, &event->wait); | |
4239 | } | |
4240 | ||
4241 | /* | |
3bc942f3 TH |
4242 | * DO NOT USE IN NEW FILES. |
4243 | * | |
79bd9814 TH |
4244 | * Parse input and register new cgroup event handler. |
4245 | * | |
4246 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4247 | * Interpretation of args is defined by control file implementation. | |
4248 | */ | |
451af504 TH |
4249 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4250 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4251 | { |
451af504 | 4252 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4253 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4254 | struct mem_cgroup_event *event; |
79bd9814 TH |
4255 | struct cgroup_subsys_state *cfile_css; |
4256 | unsigned int efd, cfd; | |
4257 | struct fd efile; | |
4258 | struct fd cfile; | |
fba94807 | 4259 | const char *name; |
79bd9814 TH |
4260 | char *endp; |
4261 | int ret; | |
4262 | ||
451af504 TH |
4263 | buf = strstrip(buf); |
4264 | ||
4265 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4266 | if (*endp != ' ') |
4267 | return -EINVAL; | |
451af504 | 4268 | buf = endp + 1; |
79bd9814 | 4269 | |
451af504 | 4270 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4271 | if ((*endp != ' ') && (*endp != '\0')) |
4272 | return -EINVAL; | |
451af504 | 4273 | buf = endp + 1; |
79bd9814 TH |
4274 | |
4275 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4276 | if (!event) | |
4277 | return -ENOMEM; | |
4278 | ||
59b6f873 | 4279 | event->memcg = memcg; |
79bd9814 | 4280 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4281 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4282 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4283 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4284 | |
4285 | efile = fdget(efd); | |
4286 | if (!efile.file) { | |
4287 | ret = -EBADF; | |
4288 | goto out_kfree; | |
4289 | } | |
4290 | ||
4291 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4292 | if (IS_ERR(event->eventfd)) { | |
4293 | ret = PTR_ERR(event->eventfd); | |
4294 | goto out_put_efile; | |
4295 | } | |
4296 | ||
4297 | cfile = fdget(cfd); | |
4298 | if (!cfile.file) { | |
4299 | ret = -EBADF; | |
4300 | goto out_put_eventfd; | |
4301 | } | |
4302 | ||
4303 | /* the process need read permission on control file */ | |
4304 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
4305 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
4306 | if (ret < 0) | |
4307 | goto out_put_cfile; | |
4308 | ||
fba94807 TH |
4309 | /* |
4310 | * Determine the event callbacks and set them in @event. This used | |
4311 | * to be done via struct cftype but cgroup core no longer knows | |
4312 | * about these events. The following is crude but the whole thing | |
4313 | * is for compatibility anyway. | |
3bc942f3 TH |
4314 | * |
4315 | * DO NOT ADD NEW FILES. | |
fba94807 | 4316 | */ |
b583043e | 4317 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4318 | |
4319 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4320 | event->register_event = mem_cgroup_usage_register_event; | |
4321 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4322 | } else if (!strcmp(name, "memory.oom_control")) { | |
4323 | event->register_event = mem_cgroup_oom_register_event; | |
4324 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4325 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4326 | event->register_event = vmpressure_register_event; | |
4327 | event->unregister_event = vmpressure_unregister_event; | |
4328 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4329 | event->register_event = memsw_cgroup_usage_register_event; |
4330 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4331 | } else { |
4332 | ret = -EINVAL; | |
4333 | goto out_put_cfile; | |
4334 | } | |
4335 | ||
79bd9814 | 4336 | /* |
b5557c4c TH |
4337 | * Verify @cfile should belong to @css. Also, remaining events are |
4338 | * automatically removed on cgroup destruction but the removal is | |
4339 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4340 | */ |
b583043e | 4341 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4342 | &memory_cgrp_subsys); |
79bd9814 | 4343 | ret = -EINVAL; |
5a17f543 | 4344 | if (IS_ERR(cfile_css)) |
79bd9814 | 4345 | goto out_put_cfile; |
5a17f543 TH |
4346 | if (cfile_css != css) { |
4347 | css_put(cfile_css); | |
79bd9814 | 4348 | goto out_put_cfile; |
5a17f543 | 4349 | } |
79bd9814 | 4350 | |
451af504 | 4351 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4352 | if (ret) |
4353 | goto out_put_css; | |
4354 | ||
9965ed17 | 4355 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4356 | |
fba94807 TH |
4357 | spin_lock(&memcg->event_list_lock); |
4358 | list_add(&event->list, &memcg->event_list); | |
4359 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4360 | |
4361 | fdput(cfile); | |
4362 | fdput(efile); | |
4363 | ||
451af504 | 4364 | return nbytes; |
79bd9814 TH |
4365 | |
4366 | out_put_css: | |
b5557c4c | 4367 | css_put(css); |
79bd9814 TH |
4368 | out_put_cfile: |
4369 | fdput(cfile); | |
4370 | out_put_eventfd: | |
4371 | eventfd_ctx_put(event->eventfd); | |
4372 | out_put_efile: | |
4373 | fdput(efile); | |
4374 | out_kfree: | |
4375 | kfree(event); | |
4376 | ||
4377 | return ret; | |
4378 | } | |
4379 | ||
241994ed | 4380 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4381 | { |
0eea1030 | 4382 | .name = "usage_in_bytes", |
8c7c6e34 | 4383 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4384 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4385 | }, |
c84872e1 PE |
4386 | { |
4387 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4388 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4389 | .write = mem_cgroup_reset, |
791badbd | 4390 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4391 | }, |
8cdea7c0 | 4392 | { |
0eea1030 | 4393 | .name = "limit_in_bytes", |
8c7c6e34 | 4394 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4395 | .write = mem_cgroup_write, |
791badbd | 4396 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4397 | }, |
296c81d8 BS |
4398 | { |
4399 | .name = "soft_limit_in_bytes", | |
4400 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4401 | .write = mem_cgroup_write, |
791badbd | 4402 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4403 | }, |
8cdea7c0 BS |
4404 | { |
4405 | .name = "failcnt", | |
8c7c6e34 | 4406 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4407 | .write = mem_cgroup_reset, |
791badbd | 4408 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4409 | }, |
d2ceb9b7 KH |
4410 | { |
4411 | .name = "stat", | |
2da8ca82 | 4412 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4413 | }, |
c1e862c1 KH |
4414 | { |
4415 | .name = "force_empty", | |
6770c64e | 4416 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4417 | }, |
18f59ea7 BS |
4418 | { |
4419 | .name = "use_hierarchy", | |
4420 | .write_u64 = mem_cgroup_hierarchy_write, | |
4421 | .read_u64 = mem_cgroup_hierarchy_read, | |
4422 | }, | |
79bd9814 | 4423 | { |
3bc942f3 | 4424 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4425 | .write = memcg_write_event_control, |
7dbdb199 | 4426 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4427 | }, |
a7885eb8 KM |
4428 | { |
4429 | .name = "swappiness", | |
4430 | .read_u64 = mem_cgroup_swappiness_read, | |
4431 | .write_u64 = mem_cgroup_swappiness_write, | |
4432 | }, | |
7dc74be0 DN |
4433 | { |
4434 | .name = "move_charge_at_immigrate", | |
4435 | .read_u64 = mem_cgroup_move_charge_read, | |
4436 | .write_u64 = mem_cgroup_move_charge_write, | |
4437 | }, | |
9490ff27 KH |
4438 | { |
4439 | .name = "oom_control", | |
2da8ca82 | 4440 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4441 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4442 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4443 | }, | |
70ddf637 AV |
4444 | { |
4445 | .name = "pressure_level", | |
70ddf637 | 4446 | }, |
406eb0c9 YH |
4447 | #ifdef CONFIG_NUMA |
4448 | { | |
4449 | .name = "numa_stat", | |
2da8ca82 | 4450 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4451 | }, |
4452 | #endif | |
510fc4e1 GC |
4453 | { |
4454 | .name = "kmem.limit_in_bytes", | |
4455 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4456 | .write = mem_cgroup_write, |
791badbd | 4457 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4458 | }, |
4459 | { | |
4460 | .name = "kmem.usage_in_bytes", | |
4461 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4462 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4463 | }, |
4464 | { | |
4465 | .name = "kmem.failcnt", | |
4466 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4467 | .write = mem_cgroup_reset, |
791badbd | 4468 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4469 | }, |
4470 | { | |
4471 | .name = "kmem.max_usage_in_bytes", | |
4472 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4473 | .write = mem_cgroup_reset, |
791badbd | 4474 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4475 | }, |
5b365771 | 4476 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
749c5415 GC |
4477 | { |
4478 | .name = "kmem.slabinfo", | |
bc2791f8 TH |
4479 | .seq_start = memcg_slab_start, |
4480 | .seq_next = memcg_slab_next, | |
4481 | .seq_stop = memcg_slab_stop, | |
b047501c | 4482 | .seq_show = memcg_slab_show, |
749c5415 GC |
4483 | }, |
4484 | #endif | |
d55f90bf VD |
4485 | { |
4486 | .name = "kmem.tcp.limit_in_bytes", | |
4487 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4488 | .write = mem_cgroup_write, | |
4489 | .read_u64 = mem_cgroup_read_u64, | |
4490 | }, | |
4491 | { | |
4492 | .name = "kmem.tcp.usage_in_bytes", | |
4493 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4494 | .read_u64 = mem_cgroup_read_u64, | |
4495 | }, | |
4496 | { | |
4497 | .name = "kmem.tcp.failcnt", | |
4498 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4499 | .write = mem_cgroup_reset, | |
4500 | .read_u64 = mem_cgroup_read_u64, | |
4501 | }, | |
4502 | { | |
4503 | .name = "kmem.tcp.max_usage_in_bytes", | |
4504 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4505 | .write = mem_cgroup_reset, | |
4506 | .read_u64 = mem_cgroup_read_u64, | |
4507 | }, | |
6bc10349 | 4508 | { }, /* terminate */ |
af36f906 | 4509 | }; |
8c7c6e34 | 4510 | |
73f576c0 JW |
4511 | /* |
4512 | * Private memory cgroup IDR | |
4513 | * | |
4514 | * Swap-out records and page cache shadow entries need to store memcg | |
4515 | * references in constrained space, so we maintain an ID space that is | |
4516 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4517 | * memory-controlled cgroups to 64k. | |
4518 | * | |
4519 | * However, there usually are many references to the oflline CSS after | |
4520 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4521 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4522 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4523 | * relatively small ID space and prevent the creation of new cgroups | |
4524 | * even when there are much fewer than 64k cgroups - possibly none. | |
4525 | * | |
4526 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4527 | * be freed and recycled when it's no longer needed, which is usually | |
4528 | * when the CSS is offlined. | |
4529 | * | |
4530 | * The only exception to that are records of swapped out tmpfs/shmem | |
4531 | * pages that need to be attributed to live ancestors on swapin. But | |
4532 | * those references are manageable from userspace. | |
4533 | */ | |
4534 | ||
4535 | static DEFINE_IDR(mem_cgroup_idr); | |
4536 | ||
7e97de0b KT |
4537 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
4538 | { | |
4539 | if (memcg->id.id > 0) { | |
4540 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
4541 | memcg->id.id = 0; | |
4542 | } | |
4543 | } | |
4544 | ||
615d66c3 | 4545 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4546 | { |
1c2d479a | 4547 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
4548 | } |
4549 | ||
615d66c3 | 4550 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4551 | { |
1c2d479a | 4552 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 4553 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
4554 | |
4555 | /* Memcg ID pins CSS */ | |
4556 | css_put(&memcg->css); | |
4557 | } | |
4558 | } | |
4559 | ||
615d66c3 VD |
4560 | static inline void mem_cgroup_id_get(struct mem_cgroup *memcg) |
4561 | { | |
4562 | mem_cgroup_id_get_many(memcg, 1); | |
4563 | } | |
4564 | ||
4565 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) | |
4566 | { | |
4567 | mem_cgroup_id_put_many(memcg, 1); | |
4568 | } | |
4569 | ||
73f576c0 JW |
4570 | /** |
4571 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4572 | * @id: the memcg id to look up | |
4573 | * | |
4574 | * Caller must hold rcu_read_lock(). | |
4575 | */ | |
4576 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4577 | { | |
4578 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4579 | return idr_find(&mem_cgroup_idr, id); | |
4580 | } | |
4581 | ||
ef8f2327 | 4582 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4583 | { |
4584 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 4585 | int tmp = node; |
1ecaab2b KH |
4586 | /* |
4587 | * This routine is called against possible nodes. | |
4588 | * But it's BUG to call kmalloc() against offline node. | |
4589 | * | |
4590 | * TODO: this routine can waste much memory for nodes which will | |
4591 | * never be onlined. It's better to use memory hotplug callback | |
4592 | * function. | |
4593 | */ | |
41e3355d KH |
4594 | if (!node_state(node, N_NORMAL_MEMORY)) |
4595 | tmp = -1; | |
17295c88 | 4596 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4597 | if (!pn) |
4598 | return 1; | |
1ecaab2b | 4599 | |
815744d7 JW |
4600 | pn->lruvec_stat_local = alloc_percpu(struct lruvec_stat); |
4601 | if (!pn->lruvec_stat_local) { | |
4602 | kfree(pn); | |
4603 | return 1; | |
4604 | } | |
4605 | ||
a983b5eb JW |
4606 | pn->lruvec_stat_cpu = alloc_percpu(struct lruvec_stat); |
4607 | if (!pn->lruvec_stat_cpu) { | |
815744d7 | 4608 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c JW |
4609 | kfree(pn); |
4610 | return 1; | |
4611 | } | |
4612 | ||
ef8f2327 MG |
4613 | lruvec_init(&pn->lruvec); |
4614 | pn->usage_in_excess = 0; | |
4615 | pn->on_tree = false; | |
4616 | pn->memcg = memcg; | |
4617 | ||
54f72fe0 | 4618 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4619 | return 0; |
4620 | } | |
4621 | ||
ef8f2327 | 4622 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4623 | { |
00f3ca2c JW |
4624 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
4625 | ||
4eaf431f MH |
4626 | if (!pn) |
4627 | return; | |
4628 | ||
a983b5eb | 4629 | free_percpu(pn->lruvec_stat_cpu); |
815744d7 | 4630 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c | 4631 | kfree(pn); |
1ecaab2b KH |
4632 | } |
4633 | ||
40e952f9 | 4634 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4635 | { |
c8b2a36f | 4636 | int node; |
59927fb9 | 4637 | |
c8b2a36f | 4638 | for_each_node(node) |
ef8f2327 | 4639 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 4640 | free_percpu(memcg->vmstats_percpu); |
815744d7 | 4641 | free_percpu(memcg->vmstats_local); |
8ff69e2c | 4642 | kfree(memcg); |
59927fb9 | 4643 | } |
3afe36b1 | 4644 | |
40e952f9 TE |
4645 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
4646 | { | |
4647 | memcg_wb_domain_exit(memcg); | |
4648 | __mem_cgroup_free(memcg); | |
4649 | } | |
4650 | ||
0b8f73e1 | 4651 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4652 | { |
d142e3e6 | 4653 | struct mem_cgroup *memcg; |
b9726c26 | 4654 | unsigned int size; |
6d12e2d8 | 4655 | int node; |
8cdea7c0 | 4656 | |
0b8f73e1 JW |
4657 | size = sizeof(struct mem_cgroup); |
4658 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4659 | ||
4660 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4661 | if (!memcg) |
0b8f73e1 JW |
4662 | return NULL; |
4663 | ||
73f576c0 JW |
4664 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
4665 | 1, MEM_CGROUP_ID_MAX, | |
4666 | GFP_KERNEL); | |
4667 | if (memcg->id.id < 0) | |
4668 | goto fail; | |
4669 | ||
815744d7 JW |
4670 | memcg->vmstats_local = alloc_percpu(struct memcg_vmstats_percpu); |
4671 | if (!memcg->vmstats_local) | |
4672 | goto fail; | |
4673 | ||
871789d4 CD |
4674 | memcg->vmstats_percpu = alloc_percpu(struct memcg_vmstats_percpu); |
4675 | if (!memcg->vmstats_percpu) | |
0b8f73e1 | 4676 | goto fail; |
78fb7466 | 4677 | |
3ed28fa1 | 4678 | for_each_node(node) |
ef8f2327 | 4679 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 4680 | goto fail; |
f64c3f54 | 4681 | |
0b8f73e1 JW |
4682 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4683 | goto fail; | |
28dbc4b6 | 4684 | |
f7e1cb6e | 4685 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4686 | memcg->last_scanned_node = MAX_NUMNODES; |
4687 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4688 | mutex_init(&memcg->thresholds_lock); |
4689 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4690 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4691 | INIT_LIST_HEAD(&memcg->event_list); |
4692 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4693 | memcg->socket_pressure = jiffies; |
84c07d11 | 4694 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 4695 | memcg->kmemcg_id = -1; |
900a38f0 | 4696 | #endif |
52ebea74 TH |
4697 | #ifdef CONFIG_CGROUP_WRITEBACK |
4698 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4699 | #endif | |
73f576c0 | 4700 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
4701 | return memcg; |
4702 | fail: | |
7e97de0b | 4703 | mem_cgroup_id_remove(memcg); |
40e952f9 | 4704 | __mem_cgroup_free(memcg); |
0b8f73e1 | 4705 | return NULL; |
d142e3e6 GC |
4706 | } |
4707 | ||
0b8f73e1 JW |
4708 | static struct cgroup_subsys_state * __ref |
4709 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4710 | { |
0b8f73e1 JW |
4711 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4712 | struct mem_cgroup *memcg; | |
4713 | long error = -ENOMEM; | |
d142e3e6 | 4714 | |
0b8f73e1 JW |
4715 | memcg = mem_cgroup_alloc(); |
4716 | if (!memcg) | |
4717 | return ERR_PTR(error); | |
d142e3e6 | 4718 | |
0b8f73e1 JW |
4719 | memcg->high = PAGE_COUNTER_MAX; |
4720 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4721 | if (parent) { | |
4722 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4723 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4724 | } | |
4725 | if (parent && parent->use_hierarchy) { | |
4726 | memcg->use_hierarchy = true; | |
3e32cb2e | 4727 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4728 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4729 | page_counter_init(&memcg->memsw, &parent->memsw); |
4730 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4731 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4732 | } else { |
3e32cb2e | 4733 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4734 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4735 | page_counter_init(&memcg->memsw, NULL); |
4736 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4737 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4738 | /* |
4739 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4740 | * much sense so let cgroup subsystem know about this | |
4741 | * unfortunate state in our controller. | |
4742 | */ | |
d142e3e6 | 4743 | if (parent != root_mem_cgroup) |
073219e9 | 4744 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4745 | } |
d6441637 | 4746 | |
0b8f73e1 JW |
4747 | /* The following stuff does not apply to the root */ |
4748 | if (!parent) { | |
fb2f2b0a RG |
4749 | #ifdef CONFIG_MEMCG_KMEM |
4750 | INIT_LIST_HEAD(&memcg->kmem_caches); | |
4751 | #endif | |
0b8f73e1 JW |
4752 | root_mem_cgroup = memcg; |
4753 | return &memcg->css; | |
4754 | } | |
4755 | ||
b313aeee | 4756 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
4757 | if (error) |
4758 | goto fail; | |
127424c8 | 4759 | |
f7e1cb6e | 4760 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4761 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4762 | |
0b8f73e1 JW |
4763 | return &memcg->css; |
4764 | fail: | |
7e97de0b | 4765 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 4766 | mem_cgroup_free(memcg); |
ea3a9645 | 4767 | return ERR_PTR(-ENOMEM); |
0b8f73e1 JW |
4768 | } |
4769 | ||
73f576c0 | 4770 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 4771 | { |
58fa2a55 VD |
4772 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
4773 | ||
0a4465d3 KT |
4774 | /* |
4775 | * A memcg must be visible for memcg_expand_shrinker_maps() | |
4776 | * by the time the maps are allocated. So, we allocate maps | |
4777 | * here, when for_each_mem_cgroup() can't skip it. | |
4778 | */ | |
4779 | if (memcg_alloc_shrinker_maps(memcg)) { | |
4780 | mem_cgroup_id_remove(memcg); | |
4781 | return -ENOMEM; | |
4782 | } | |
4783 | ||
73f576c0 | 4784 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 4785 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 4786 | css_get(css); |
2f7dd7a4 | 4787 | return 0; |
8cdea7c0 BS |
4788 | } |
4789 | ||
eb95419b | 4790 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4791 | { |
eb95419b | 4792 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4793 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4794 | |
4795 | /* | |
4796 | * Unregister events and notify userspace. | |
4797 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4798 | * directory to avoid race between userspace and kernelspace. | |
4799 | */ | |
fba94807 TH |
4800 | spin_lock(&memcg->event_list_lock); |
4801 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4802 | list_del_init(&event->list); |
4803 | schedule_work(&event->remove); | |
4804 | } | |
fba94807 | 4805 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4806 | |
bf8d5d52 | 4807 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4808 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 4809 | |
567e9ab2 | 4810 | memcg_offline_kmem(memcg); |
52ebea74 | 4811 | wb_memcg_offline(memcg); |
73f576c0 | 4812 | |
591edfb1 RG |
4813 | drain_all_stock(memcg); |
4814 | ||
73f576c0 | 4815 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
4816 | } |
4817 | ||
6df38689 VD |
4818 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4819 | { | |
4820 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4821 | ||
4822 | invalidate_reclaim_iterators(memcg); | |
4823 | } | |
4824 | ||
eb95419b | 4825 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4826 | { |
eb95419b | 4827 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4828 | |
f7e1cb6e | 4829 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4830 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4831 | |
0db15298 | 4832 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4833 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4834 | |
0b8f73e1 JW |
4835 | vmpressure_cleanup(&memcg->vmpressure); |
4836 | cancel_work_sync(&memcg->high_work); | |
4837 | mem_cgroup_remove_from_trees(memcg); | |
0a4465d3 | 4838 | memcg_free_shrinker_maps(memcg); |
d886f4e4 | 4839 | memcg_free_kmem(memcg); |
0b8f73e1 | 4840 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4841 | } |
4842 | ||
1ced953b TH |
4843 | /** |
4844 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4845 | * @css: the target css | |
4846 | * | |
4847 | * Reset the states of the mem_cgroup associated with @css. This is | |
4848 | * invoked when the userland requests disabling on the default hierarchy | |
4849 | * but the memcg is pinned through dependency. The memcg should stop | |
4850 | * applying policies and should revert to the vanilla state as it may be | |
4851 | * made visible again. | |
4852 | * | |
4853 | * The current implementation only resets the essential configurations. | |
4854 | * This needs to be expanded to cover all the visible parts. | |
4855 | */ | |
4856 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4857 | { | |
4858 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4859 | ||
bbec2e15 RG |
4860 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
4861 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
4862 | page_counter_set_max(&memcg->memsw, PAGE_COUNTER_MAX); | |
4863 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); | |
4864 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 4865 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 4866 | page_counter_set_low(&memcg->memory, 0); |
241994ed | 4867 | memcg->high = PAGE_COUNTER_MAX; |
24d404dc | 4868 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4869 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4870 | } |
4871 | ||
02491447 | 4872 | #ifdef CONFIG_MMU |
7dc74be0 | 4873 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4874 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4875 | { |
05b84301 | 4876 | int ret; |
9476db97 | 4877 | |
d0164adc MG |
4878 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4879 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4880 | if (!ret) { |
854ffa8d | 4881 | mc.precharge += count; |
854ffa8d DN |
4882 | return ret; |
4883 | } | |
9476db97 | 4884 | |
3674534b | 4885 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 4886 | while (count--) { |
3674534b | 4887 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 4888 | if (ret) |
38c5d72f | 4889 | return ret; |
854ffa8d | 4890 | mc.precharge++; |
9476db97 | 4891 | cond_resched(); |
854ffa8d | 4892 | } |
9476db97 | 4893 | return 0; |
4ffef5fe DN |
4894 | } |
4895 | ||
4ffef5fe DN |
4896 | union mc_target { |
4897 | struct page *page; | |
02491447 | 4898 | swp_entry_t ent; |
4ffef5fe DN |
4899 | }; |
4900 | ||
4ffef5fe | 4901 | enum mc_target_type { |
8d32ff84 | 4902 | MC_TARGET_NONE = 0, |
4ffef5fe | 4903 | MC_TARGET_PAGE, |
02491447 | 4904 | MC_TARGET_SWAP, |
c733a828 | 4905 | MC_TARGET_DEVICE, |
4ffef5fe DN |
4906 | }; |
4907 | ||
90254a65 DN |
4908 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4909 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4910 | { |
25b2995a | 4911 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4912 | |
90254a65 DN |
4913 | if (!page || !page_mapped(page)) |
4914 | return NULL; | |
4915 | if (PageAnon(page)) { | |
1dfab5ab | 4916 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4917 | return NULL; |
1dfab5ab JW |
4918 | } else { |
4919 | if (!(mc.flags & MOVE_FILE)) | |
4920 | return NULL; | |
4921 | } | |
90254a65 DN |
4922 | if (!get_page_unless_zero(page)) |
4923 | return NULL; | |
4924 | ||
4925 | return page; | |
4926 | } | |
4927 | ||
c733a828 | 4928 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 4929 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 4930 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 4931 | { |
90254a65 DN |
4932 | struct page *page = NULL; |
4933 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4934 | ||
1dfab5ab | 4935 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4936 | return NULL; |
c733a828 JG |
4937 | |
4938 | /* | |
4939 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
4940 | * a device and because they are not accessible by CPU they are store | |
4941 | * as special swap entry in the CPU page table. | |
4942 | */ | |
4943 | if (is_device_private_entry(ent)) { | |
4944 | page = device_private_entry_to_page(ent); | |
4945 | /* | |
4946 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
4947 | * a refcount of 1 when free (unlike normal page) | |
4948 | */ | |
4949 | if (!page_ref_add_unless(page, 1, 1)) | |
4950 | return NULL; | |
4951 | return page; | |
4952 | } | |
4953 | ||
4b91355e KH |
4954 | /* |
4955 | * Because lookup_swap_cache() updates some statistics counter, | |
4956 | * we call find_get_page() with swapper_space directly. | |
4957 | */ | |
f6ab1f7f | 4958 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
7941d214 | 4959 | if (do_memsw_account()) |
90254a65 DN |
4960 | entry->val = ent.val; |
4961 | ||
4962 | return page; | |
4963 | } | |
4b91355e KH |
4964 | #else |
4965 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 4966 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
4967 | { |
4968 | return NULL; | |
4969 | } | |
4970 | #endif | |
90254a65 | 4971 | |
87946a72 DN |
4972 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4973 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4974 | { | |
4975 | struct page *page = NULL; | |
87946a72 DN |
4976 | struct address_space *mapping; |
4977 | pgoff_t pgoff; | |
4978 | ||
4979 | if (!vma->vm_file) /* anonymous vma */ | |
4980 | return NULL; | |
1dfab5ab | 4981 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4982 | return NULL; |
4983 | ||
87946a72 | 4984 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4985 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4986 | |
4987 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4988 | #ifdef CONFIG_SWAP |
4989 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4990 | if (shmem_mapping(mapping)) { |
4991 | page = find_get_entry(mapping, pgoff); | |
3159f943 | 4992 | if (xa_is_value(page)) { |
139b6a6f | 4993 | swp_entry_t swp = radix_to_swp_entry(page); |
7941d214 | 4994 | if (do_memsw_account()) |
139b6a6f | 4995 | *entry = swp; |
f6ab1f7f HY |
4996 | page = find_get_page(swap_address_space(swp), |
4997 | swp_offset(swp)); | |
139b6a6f JW |
4998 | } |
4999 | } else | |
5000 | page = find_get_page(mapping, pgoff); | |
5001 | #else | |
5002 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 5003 | #endif |
87946a72 DN |
5004 | return page; |
5005 | } | |
5006 | ||
b1b0deab CG |
5007 | /** |
5008 | * mem_cgroup_move_account - move account of the page | |
5009 | * @page: the page | |
25843c2b | 5010 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5011 | * @from: mem_cgroup which the page is moved from. |
5012 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5013 | * | |
3ac808fd | 5014 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5015 | * |
5016 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5017 | * from old cgroup. | |
5018 | */ | |
5019 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5020 | bool compound, |
b1b0deab CG |
5021 | struct mem_cgroup *from, |
5022 | struct mem_cgroup *to) | |
5023 | { | |
5024 | unsigned long flags; | |
f627c2f5 | 5025 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 5026 | int ret; |
c4843a75 | 5027 | bool anon; |
b1b0deab CG |
5028 | |
5029 | VM_BUG_ON(from == to); | |
5030 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5031 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5032 | |
5033 | /* | |
6a93ca8f | 5034 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 5035 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 5036 | */ |
f627c2f5 | 5037 | ret = -EBUSY; |
b1b0deab CG |
5038 | if (!trylock_page(page)) |
5039 | goto out; | |
5040 | ||
5041 | ret = -EINVAL; | |
5042 | if (page->mem_cgroup != from) | |
5043 | goto out_unlock; | |
5044 | ||
c4843a75 GT |
5045 | anon = PageAnon(page); |
5046 | ||
b1b0deab CG |
5047 | spin_lock_irqsave(&from->move_lock, flags); |
5048 | ||
c4843a75 | 5049 | if (!anon && page_mapped(page)) { |
c9019e9b JW |
5050 | __mod_memcg_state(from, NR_FILE_MAPPED, -nr_pages); |
5051 | __mod_memcg_state(to, NR_FILE_MAPPED, nr_pages); | |
b1b0deab CG |
5052 | } |
5053 | ||
c4843a75 GT |
5054 | /* |
5055 | * move_lock grabbed above and caller set from->moving_account, so | |
ccda7f43 | 5056 | * mod_memcg_page_state will serialize updates to PageDirty. |
c4843a75 GT |
5057 | * So mapping should be stable for dirty pages. |
5058 | */ | |
5059 | if (!anon && PageDirty(page)) { | |
5060 | struct address_space *mapping = page_mapping(page); | |
5061 | ||
5062 | if (mapping_cap_account_dirty(mapping)) { | |
c9019e9b JW |
5063 | __mod_memcg_state(from, NR_FILE_DIRTY, -nr_pages); |
5064 | __mod_memcg_state(to, NR_FILE_DIRTY, nr_pages); | |
c4843a75 GT |
5065 | } |
5066 | } | |
5067 | ||
b1b0deab | 5068 | if (PageWriteback(page)) { |
c9019e9b JW |
5069 | __mod_memcg_state(from, NR_WRITEBACK, -nr_pages); |
5070 | __mod_memcg_state(to, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5071 | } |
5072 | ||
5073 | /* | |
5074 | * It is safe to change page->mem_cgroup here because the page | |
5075 | * is referenced, charged, and isolated - we can't race with | |
5076 | * uncharging, charging, migration, or LRU putback. | |
5077 | */ | |
5078 | ||
5079 | /* caller should have done css_get */ | |
5080 | page->mem_cgroup = to; | |
5081 | spin_unlock_irqrestore(&from->move_lock, flags); | |
5082 | ||
5083 | ret = 0; | |
5084 | ||
5085 | local_irq_disable(); | |
f627c2f5 | 5086 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 5087 | memcg_check_events(to, page); |
f627c2f5 | 5088 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
5089 | memcg_check_events(from, page); |
5090 | local_irq_enable(); | |
5091 | out_unlock: | |
5092 | unlock_page(page); | |
5093 | out: | |
5094 | return ret; | |
5095 | } | |
5096 | ||
7cf7806c LR |
5097 | /** |
5098 | * get_mctgt_type - get target type of moving charge | |
5099 | * @vma: the vma the pte to be checked belongs | |
5100 | * @addr: the address corresponding to the pte to be checked | |
5101 | * @ptent: the pte to be checked | |
5102 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5103 | * | |
5104 | * Returns | |
5105 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5106 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5107 | * move charge. if @target is not NULL, the page is stored in target->page | |
5108 | * with extra refcnt got(Callers should handle it). | |
5109 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5110 | * target for charge migration. if @target is not NULL, the entry is stored | |
5111 | * in target->ent. | |
25b2995a CH |
5112 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
5113 | * (so ZONE_DEVICE page and thus not on the lru). | |
df6ad698 JG |
5114 | * For now we such page is charge like a regular page would be as for all |
5115 | * intent and purposes it is just special memory taking the place of a | |
5116 | * regular page. | |
c733a828 JG |
5117 | * |
5118 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5119 | * |
5120 | * Called with pte lock held. | |
5121 | */ | |
5122 | ||
8d32ff84 | 5123 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5124 | unsigned long addr, pte_t ptent, union mc_target *target) |
5125 | { | |
5126 | struct page *page = NULL; | |
8d32ff84 | 5127 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5128 | swp_entry_t ent = { .val = 0 }; |
5129 | ||
5130 | if (pte_present(ptent)) | |
5131 | page = mc_handle_present_pte(vma, addr, ptent); | |
5132 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5133 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5134 | else if (pte_none(ptent)) |
87946a72 | 5135 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5136 | |
5137 | if (!page && !ent.val) | |
8d32ff84 | 5138 | return ret; |
02491447 | 5139 | if (page) { |
02491447 | 5140 | /* |
0a31bc97 | 5141 | * Do only loose check w/o serialization. |
1306a85a | 5142 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5143 | * not under LRU exclusion. |
02491447 | 5144 | */ |
1306a85a | 5145 | if (page->mem_cgroup == mc.from) { |
02491447 | 5146 | ret = MC_TARGET_PAGE; |
25b2995a | 5147 | if (is_device_private_page(page)) |
c733a828 | 5148 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5149 | if (target) |
5150 | target->page = page; | |
5151 | } | |
5152 | if (!ret || !target) | |
5153 | put_page(page); | |
5154 | } | |
3e14a57b HY |
5155 | /* |
5156 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5157 | * But we cannot move a tail-page in a THP. | |
5158 | */ | |
5159 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5160 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5161 | ret = MC_TARGET_SWAP; |
5162 | if (target) | |
5163 | target->ent = ent; | |
4ffef5fe | 5164 | } |
4ffef5fe DN |
5165 | return ret; |
5166 | } | |
5167 | ||
12724850 NH |
5168 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5169 | /* | |
d6810d73 HY |
5170 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5171 | * not support them for now. | |
12724850 NH |
5172 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5173 | */ | |
5174 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5175 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5176 | { | |
5177 | struct page *page = NULL; | |
12724850 NH |
5178 | enum mc_target_type ret = MC_TARGET_NONE; |
5179 | ||
84c3fc4e ZY |
5180 | if (unlikely(is_swap_pmd(pmd))) { |
5181 | VM_BUG_ON(thp_migration_supported() && | |
5182 | !is_pmd_migration_entry(pmd)); | |
5183 | return ret; | |
5184 | } | |
12724850 | 5185 | page = pmd_page(pmd); |
309381fe | 5186 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5187 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5188 | return ret; |
1306a85a | 5189 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
5190 | ret = MC_TARGET_PAGE; |
5191 | if (target) { | |
5192 | get_page(page); | |
5193 | target->page = page; | |
5194 | } | |
5195 | } | |
5196 | return ret; | |
5197 | } | |
5198 | #else | |
5199 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5200 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5201 | { | |
5202 | return MC_TARGET_NONE; | |
5203 | } | |
5204 | #endif | |
5205 | ||
4ffef5fe DN |
5206 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5207 | unsigned long addr, unsigned long end, | |
5208 | struct mm_walk *walk) | |
5209 | { | |
26bcd64a | 5210 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5211 | pte_t *pte; |
5212 | spinlock_t *ptl; | |
5213 | ||
b6ec57f4 KS |
5214 | ptl = pmd_trans_huge_lock(pmd, vma); |
5215 | if (ptl) { | |
c733a828 JG |
5216 | /* |
5217 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5218 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5219 | * this might change. | |
c733a828 | 5220 | */ |
12724850 NH |
5221 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5222 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5223 | spin_unlock(ptl); |
1a5a9906 | 5224 | return 0; |
12724850 | 5225 | } |
03319327 | 5226 | |
45f83cef AA |
5227 | if (pmd_trans_unstable(pmd)) |
5228 | return 0; | |
4ffef5fe DN |
5229 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5230 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5231 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5232 | mc.precharge++; /* increment precharge temporarily */ |
5233 | pte_unmap_unlock(pte - 1, ptl); | |
5234 | cond_resched(); | |
5235 | ||
7dc74be0 DN |
5236 | return 0; |
5237 | } | |
5238 | ||
4ffef5fe DN |
5239 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5240 | { | |
5241 | unsigned long precharge; | |
4ffef5fe | 5242 | |
26bcd64a NH |
5243 | struct mm_walk mem_cgroup_count_precharge_walk = { |
5244 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5245 | .mm = mm, | |
5246 | }; | |
dfe076b0 | 5247 | down_read(&mm->mmap_sem); |
0247f3f4 JM |
5248 | walk_page_range(0, mm->highest_vm_end, |
5249 | &mem_cgroup_count_precharge_walk); | |
dfe076b0 | 5250 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5251 | |
5252 | precharge = mc.precharge; | |
5253 | mc.precharge = 0; | |
5254 | ||
5255 | return precharge; | |
5256 | } | |
5257 | ||
4ffef5fe DN |
5258 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5259 | { | |
dfe076b0 DN |
5260 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5261 | ||
5262 | VM_BUG_ON(mc.moving_task); | |
5263 | mc.moving_task = current; | |
5264 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5265 | } |
5266 | ||
dfe076b0 DN |
5267 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5268 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5269 | { |
2bd9bb20 KH |
5270 | struct mem_cgroup *from = mc.from; |
5271 | struct mem_cgroup *to = mc.to; | |
5272 | ||
4ffef5fe | 5273 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5274 | if (mc.precharge) { |
00501b53 | 5275 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5276 | mc.precharge = 0; |
5277 | } | |
5278 | /* | |
5279 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5280 | * we must uncharge here. | |
5281 | */ | |
5282 | if (mc.moved_charge) { | |
00501b53 | 5283 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5284 | mc.moved_charge = 0; |
4ffef5fe | 5285 | } |
483c30b5 DN |
5286 | /* we must fixup refcnts and charges */ |
5287 | if (mc.moved_swap) { | |
483c30b5 | 5288 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5289 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5290 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5291 | |
615d66c3 VD |
5292 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5293 | ||
05b84301 | 5294 | /* |
3e32cb2e JW |
5295 | * we charged both to->memory and to->memsw, so we |
5296 | * should uncharge to->memory. | |
05b84301 | 5297 | */ |
ce00a967 | 5298 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5299 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5300 | ||
615d66c3 VD |
5301 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
5302 | css_put_many(&mc.to->css, mc.moved_swap); | |
3e32cb2e | 5303 | |
483c30b5 DN |
5304 | mc.moved_swap = 0; |
5305 | } | |
dfe076b0 DN |
5306 | memcg_oom_recover(from); |
5307 | memcg_oom_recover(to); | |
5308 | wake_up_all(&mc.waitq); | |
5309 | } | |
5310 | ||
5311 | static void mem_cgroup_clear_mc(void) | |
5312 | { | |
264a0ae1 TH |
5313 | struct mm_struct *mm = mc.mm; |
5314 | ||
dfe076b0 DN |
5315 | /* |
5316 | * we must clear moving_task before waking up waiters at the end of | |
5317 | * task migration. | |
5318 | */ | |
5319 | mc.moving_task = NULL; | |
5320 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5321 | spin_lock(&mc.lock); |
4ffef5fe DN |
5322 | mc.from = NULL; |
5323 | mc.to = NULL; | |
264a0ae1 | 5324 | mc.mm = NULL; |
2bd9bb20 | 5325 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5326 | |
5327 | mmput(mm); | |
4ffef5fe DN |
5328 | } |
5329 | ||
1f7dd3e5 | 5330 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5331 | { |
1f7dd3e5 | 5332 | struct cgroup_subsys_state *css; |
eed67d75 | 5333 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5334 | struct mem_cgroup *from; |
4530eddb | 5335 | struct task_struct *leader, *p; |
9f2115f9 | 5336 | struct mm_struct *mm; |
1dfab5ab | 5337 | unsigned long move_flags; |
9f2115f9 | 5338 | int ret = 0; |
7dc74be0 | 5339 | |
1f7dd3e5 TH |
5340 | /* charge immigration isn't supported on the default hierarchy */ |
5341 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5342 | return 0; |
5343 | ||
4530eddb TH |
5344 | /* |
5345 | * Multi-process migrations only happen on the default hierarchy | |
5346 | * where charge immigration is not used. Perform charge | |
5347 | * immigration if @tset contains a leader and whine if there are | |
5348 | * multiple. | |
5349 | */ | |
5350 | p = NULL; | |
1f7dd3e5 | 5351 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5352 | WARN_ON_ONCE(p); |
5353 | p = leader; | |
1f7dd3e5 | 5354 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5355 | } |
5356 | if (!p) | |
5357 | return 0; | |
5358 | ||
1f7dd3e5 TH |
5359 | /* |
5360 | * We are now commited to this value whatever it is. Changes in this | |
5361 | * tunable will only affect upcoming migrations, not the current one. | |
5362 | * So we need to save it, and keep it going. | |
5363 | */ | |
5364 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5365 | if (!move_flags) | |
5366 | return 0; | |
5367 | ||
9f2115f9 TH |
5368 | from = mem_cgroup_from_task(p); |
5369 | ||
5370 | VM_BUG_ON(from == memcg); | |
5371 | ||
5372 | mm = get_task_mm(p); | |
5373 | if (!mm) | |
5374 | return 0; | |
5375 | /* We move charges only when we move a owner of the mm */ | |
5376 | if (mm->owner == p) { | |
5377 | VM_BUG_ON(mc.from); | |
5378 | VM_BUG_ON(mc.to); | |
5379 | VM_BUG_ON(mc.precharge); | |
5380 | VM_BUG_ON(mc.moved_charge); | |
5381 | VM_BUG_ON(mc.moved_swap); | |
5382 | ||
5383 | spin_lock(&mc.lock); | |
264a0ae1 | 5384 | mc.mm = mm; |
9f2115f9 TH |
5385 | mc.from = from; |
5386 | mc.to = memcg; | |
5387 | mc.flags = move_flags; | |
5388 | spin_unlock(&mc.lock); | |
5389 | /* We set mc.moving_task later */ | |
5390 | ||
5391 | ret = mem_cgroup_precharge_mc(mm); | |
5392 | if (ret) | |
5393 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5394 | } else { |
5395 | mmput(mm); | |
7dc74be0 DN |
5396 | } |
5397 | return ret; | |
5398 | } | |
5399 | ||
1f7dd3e5 | 5400 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5401 | { |
4e2f245d JW |
5402 | if (mc.to) |
5403 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5404 | } |
5405 | ||
4ffef5fe DN |
5406 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5407 | unsigned long addr, unsigned long end, | |
5408 | struct mm_walk *walk) | |
7dc74be0 | 5409 | { |
4ffef5fe | 5410 | int ret = 0; |
26bcd64a | 5411 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5412 | pte_t *pte; |
5413 | spinlock_t *ptl; | |
12724850 NH |
5414 | enum mc_target_type target_type; |
5415 | union mc_target target; | |
5416 | struct page *page; | |
4ffef5fe | 5417 | |
b6ec57f4 KS |
5418 | ptl = pmd_trans_huge_lock(pmd, vma); |
5419 | if (ptl) { | |
62ade86a | 5420 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5421 | spin_unlock(ptl); |
12724850 NH |
5422 | return 0; |
5423 | } | |
5424 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5425 | if (target_type == MC_TARGET_PAGE) { | |
5426 | page = target.page; | |
5427 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5428 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5429 | mc.from, mc.to)) { |
12724850 NH |
5430 | mc.precharge -= HPAGE_PMD_NR; |
5431 | mc.moved_charge += HPAGE_PMD_NR; | |
5432 | } | |
5433 | putback_lru_page(page); | |
5434 | } | |
5435 | put_page(page); | |
c733a828 JG |
5436 | } else if (target_type == MC_TARGET_DEVICE) { |
5437 | page = target.page; | |
5438 | if (!mem_cgroup_move_account(page, true, | |
5439 | mc.from, mc.to)) { | |
5440 | mc.precharge -= HPAGE_PMD_NR; | |
5441 | mc.moved_charge += HPAGE_PMD_NR; | |
5442 | } | |
5443 | put_page(page); | |
12724850 | 5444 | } |
bf929152 | 5445 | spin_unlock(ptl); |
1a5a9906 | 5446 | return 0; |
12724850 NH |
5447 | } |
5448 | ||
45f83cef AA |
5449 | if (pmd_trans_unstable(pmd)) |
5450 | return 0; | |
4ffef5fe DN |
5451 | retry: |
5452 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5453 | for (; addr != end; addr += PAGE_SIZE) { | |
5454 | pte_t ptent = *(pte++); | |
c733a828 | 5455 | bool device = false; |
02491447 | 5456 | swp_entry_t ent; |
4ffef5fe DN |
5457 | |
5458 | if (!mc.precharge) | |
5459 | break; | |
5460 | ||
8d32ff84 | 5461 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
5462 | case MC_TARGET_DEVICE: |
5463 | device = true; | |
5464 | /* fall through */ | |
4ffef5fe DN |
5465 | case MC_TARGET_PAGE: |
5466 | page = target.page; | |
53f9263b KS |
5467 | /* |
5468 | * We can have a part of the split pmd here. Moving it | |
5469 | * can be done but it would be too convoluted so simply | |
5470 | * ignore such a partial THP and keep it in original | |
5471 | * memcg. There should be somebody mapping the head. | |
5472 | */ | |
5473 | if (PageTransCompound(page)) | |
5474 | goto put; | |
c733a828 | 5475 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 5476 | goto put; |
f627c2f5 KS |
5477 | if (!mem_cgroup_move_account(page, false, |
5478 | mc.from, mc.to)) { | |
4ffef5fe | 5479 | mc.precharge--; |
854ffa8d DN |
5480 | /* we uncharge from mc.from later. */ |
5481 | mc.moved_charge++; | |
4ffef5fe | 5482 | } |
c733a828 JG |
5483 | if (!device) |
5484 | putback_lru_page(page); | |
8d32ff84 | 5485 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5486 | put_page(page); |
5487 | break; | |
02491447 DN |
5488 | case MC_TARGET_SWAP: |
5489 | ent = target.ent; | |
e91cbb42 | 5490 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5491 | mc.precharge--; |
483c30b5 DN |
5492 | /* we fixup refcnts and charges later. */ |
5493 | mc.moved_swap++; | |
5494 | } | |
02491447 | 5495 | break; |
4ffef5fe DN |
5496 | default: |
5497 | break; | |
5498 | } | |
5499 | } | |
5500 | pte_unmap_unlock(pte - 1, ptl); | |
5501 | cond_resched(); | |
5502 | ||
5503 | if (addr != end) { | |
5504 | /* | |
5505 | * We have consumed all precharges we got in can_attach(). | |
5506 | * We try charge one by one, but don't do any additional | |
5507 | * charges to mc.to if we have failed in charge once in attach() | |
5508 | * phase. | |
5509 | */ | |
854ffa8d | 5510 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5511 | if (!ret) |
5512 | goto retry; | |
5513 | } | |
5514 | ||
5515 | return ret; | |
5516 | } | |
5517 | ||
264a0ae1 | 5518 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 5519 | { |
26bcd64a NH |
5520 | struct mm_walk mem_cgroup_move_charge_walk = { |
5521 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
264a0ae1 | 5522 | .mm = mc.mm, |
26bcd64a | 5523 | }; |
4ffef5fe DN |
5524 | |
5525 | lru_add_drain_all(); | |
312722cb | 5526 | /* |
81f8c3a4 JW |
5527 | * Signal lock_page_memcg() to take the memcg's move_lock |
5528 | * while we're moving its pages to another memcg. Then wait | |
5529 | * for already started RCU-only updates to finish. | |
312722cb JW |
5530 | */ |
5531 | atomic_inc(&mc.from->moving_account); | |
5532 | synchronize_rcu(); | |
dfe076b0 | 5533 | retry: |
264a0ae1 | 5534 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
5535 | /* |
5536 | * Someone who are holding the mmap_sem might be waiting in | |
5537 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5538 | * and retry. Because we cancel precharges, we might not be able | |
5539 | * to move enough charges, but moving charge is a best-effort | |
5540 | * feature anyway, so it wouldn't be a big problem. | |
5541 | */ | |
5542 | __mem_cgroup_clear_mc(); | |
5543 | cond_resched(); | |
5544 | goto retry; | |
5545 | } | |
26bcd64a NH |
5546 | /* |
5547 | * When we have consumed all precharges and failed in doing | |
5548 | * additional charge, the page walk just aborts. | |
5549 | */ | |
0247f3f4 JM |
5550 | walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk); |
5551 | ||
264a0ae1 | 5552 | up_read(&mc.mm->mmap_sem); |
312722cb | 5553 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5554 | } |
5555 | ||
264a0ae1 | 5556 | static void mem_cgroup_move_task(void) |
67e465a7 | 5557 | { |
264a0ae1 TH |
5558 | if (mc.to) { |
5559 | mem_cgroup_move_charge(); | |
a433658c | 5560 | mem_cgroup_clear_mc(); |
264a0ae1 | 5561 | } |
67e465a7 | 5562 | } |
5cfb80a7 | 5563 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5564 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5565 | { |
5566 | return 0; | |
5567 | } | |
1f7dd3e5 | 5568 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5569 | { |
5570 | } | |
264a0ae1 | 5571 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
5572 | { |
5573 | } | |
5574 | #endif | |
67e465a7 | 5575 | |
f00baae7 TH |
5576 | /* |
5577 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5578 | * to verify whether we're attached to the default hierarchy on each mount |
5579 | * attempt. | |
f00baae7 | 5580 | */ |
eb95419b | 5581 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5582 | { |
5583 | /* | |
aa6ec29b | 5584 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5585 | * guarantees that @root doesn't have any children, so turning it |
5586 | * on for the root memcg is enough. | |
5587 | */ | |
9e10a130 | 5588 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5589 | root_mem_cgroup->use_hierarchy = true; |
5590 | else | |
5591 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5592 | } |
5593 | ||
677dc973 CD |
5594 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
5595 | { | |
5596 | if (value == PAGE_COUNTER_MAX) | |
5597 | seq_puts(m, "max\n"); | |
5598 | else | |
5599 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
5600 | ||
5601 | return 0; | |
5602 | } | |
5603 | ||
241994ed JW |
5604 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5605 | struct cftype *cft) | |
5606 | { | |
f5fc3c5d JW |
5607 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5608 | ||
5609 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5610 | } |
5611 | ||
bf8d5d52 RG |
5612 | static int memory_min_show(struct seq_file *m, void *v) |
5613 | { | |
677dc973 CD |
5614 | return seq_puts_memcg_tunable(m, |
5615 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
5616 | } |
5617 | ||
5618 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
5619 | char *buf, size_t nbytes, loff_t off) | |
5620 | { | |
5621 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5622 | unsigned long min; | |
5623 | int err; | |
5624 | ||
5625 | buf = strstrip(buf); | |
5626 | err = page_counter_memparse(buf, "max", &min); | |
5627 | if (err) | |
5628 | return err; | |
5629 | ||
5630 | page_counter_set_min(&memcg->memory, min); | |
5631 | ||
5632 | return nbytes; | |
5633 | } | |
5634 | ||
241994ed JW |
5635 | static int memory_low_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.low)); | |
241994ed JW |
5639 | } |
5640 | ||
5641 | static ssize_t memory_low_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 low; | |
5646 | int err; | |
5647 | ||
5648 | buf = strstrip(buf); | |
d2973697 | 5649 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5650 | if (err) |
5651 | return err; | |
5652 | ||
23067153 | 5653 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
5654 | |
5655 | return nbytes; | |
5656 | } | |
5657 | ||
5658 | static int memory_high_show(struct seq_file *m, void *v) | |
5659 | { | |
677dc973 | 5660 | return seq_puts_memcg_tunable(m, READ_ONCE(mem_cgroup_from_seq(m)->high)); |
241994ed JW |
5661 | } |
5662 | ||
5663 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5664 | char *buf, size_t nbytes, loff_t off) | |
5665 | { | |
5666 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 5667 | unsigned long nr_pages; |
241994ed JW |
5668 | unsigned long high; |
5669 | int err; | |
5670 | ||
5671 | buf = strstrip(buf); | |
d2973697 | 5672 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5673 | if (err) |
5674 | return err; | |
5675 | ||
5676 | memcg->high = high; | |
5677 | ||
588083bb JW |
5678 | nr_pages = page_counter_read(&memcg->memory); |
5679 | if (nr_pages > high) | |
5680 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5681 | GFP_KERNEL, true); | |
5682 | ||
2529bb3a | 5683 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5684 | return nbytes; |
5685 | } | |
5686 | ||
5687 | static int memory_max_show(struct seq_file *m, void *v) | |
5688 | { | |
677dc973 CD |
5689 | return seq_puts_memcg_tunable(m, |
5690 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
5691 | } |
5692 | ||
5693 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5694 | char *buf, size_t nbytes, loff_t off) | |
5695 | { | |
5696 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
5697 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5698 | bool drained = false; | |
241994ed JW |
5699 | unsigned long max; |
5700 | int err; | |
5701 | ||
5702 | buf = strstrip(buf); | |
d2973697 | 5703 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5704 | if (err) |
5705 | return err; | |
5706 | ||
bbec2e15 | 5707 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
5708 | |
5709 | for (;;) { | |
5710 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5711 | ||
5712 | if (nr_pages <= max) | |
5713 | break; | |
5714 | ||
5715 | if (signal_pending(current)) { | |
5716 | err = -EINTR; | |
5717 | break; | |
5718 | } | |
5719 | ||
5720 | if (!drained) { | |
5721 | drain_all_stock(memcg); | |
5722 | drained = true; | |
5723 | continue; | |
5724 | } | |
5725 | ||
5726 | if (nr_reclaims) { | |
5727 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5728 | GFP_KERNEL, true)) | |
5729 | nr_reclaims--; | |
5730 | continue; | |
5731 | } | |
5732 | ||
e27be240 | 5733 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
5734 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
5735 | break; | |
5736 | } | |
241994ed | 5737 | |
2529bb3a | 5738 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5739 | return nbytes; |
5740 | } | |
5741 | ||
1e577f97 SB |
5742 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
5743 | { | |
5744 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
5745 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
5746 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
5747 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
5748 | seq_printf(m, "oom_kill %lu\n", | |
5749 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
5750 | } | |
5751 | ||
241994ed JW |
5752 | static int memory_events_show(struct seq_file *m, void *v) |
5753 | { | |
aa9694bb | 5754 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 5755 | |
1e577f97 SB |
5756 | __memory_events_show(m, memcg->memory_events); |
5757 | return 0; | |
5758 | } | |
5759 | ||
5760 | static int memory_events_local_show(struct seq_file *m, void *v) | |
5761 | { | |
5762 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 5763 | |
1e577f97 | 5764 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
5765 | return 0; |
5766 | } | |
5767 | ||
587d9f72 JW |
5768 | static int memory_stat_show(struct seq_file *m, void *v) |
5769 | { | |
aa9694bb | 5770 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 5771 | char *buf; |
1ff9e6e1 | 5772 | |
c8713d0b JW |
5773 | buf = memory_stat_format(memcg); |
5774 | if (!buf) | |
5775 | return -ENOMEM; | |
5776 | seq_puts(m, buf); | |
5777 | kfree(buf); | |
587d9f72 JW |
5778 | return 0; |
5779 | } | |
5780 | ||
3d8b38eb RG |
5781 | static int memory_oom_group_show(struct seq_file *m, void *v) |
5782 | { | |
aa9694bb | 5783 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
5784 | |
5785 | seq_printf(m, "%d\n", memcg->oom_group); | |
5786 | ||
5787 | return 0; | |
5788 | } | |
5789 | ||
5790 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
5791 | char *buf, size_t nbytes, loff_t off) | |
5792 | { | |
5793 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5794 | int ret, oom_group; | |
5795 | ||
5796 | buf = strstrip(buf); | |
5797 | if (!buf) | |
5798 | return -EINVAL; | |
5799 | ||
5800 | ret = kstrtoint(buf, 0, &oom_group); | |
5801 | if (ret) | |
5802 | return ret; | |
5803 | ||
5804 | if (oom_group != 0 && oom_group != 1) | |
5805 | return -EINVAL; | |
5806 | ||
5807 | memcg->oom_group = oom_group; | |
5808 | ||
5809 | return nbytes; | |
5810 | } | |
5811 | ||
241994ed JW |
5812 | static struct cftype memory_files[] = { |
5813 | { | |
5814 | .name = "current", | |
f5fc3c5d | 5815 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5816 | .read_u64 = memory_current_read, |
5817 | }, | |
bf8d5d52 RG |
5818 | { |
5819 | .name = "min", | |
5820 | .flags = CFTYPE_NOT_ON_ROOT, | |
5821 | .seq_show = memory_min_show, | |
5822 | .write = memory_min_write, | |
5823 | }, | |
241994ed JW |
5824 | { |
5825 | .name = "low", | |
5826 | .flags = CFTYPE_NOT_ON_ROOT, | |
5827 | .seq_show = memory_low_show, | |
5828 | .write = memory_low_write, | |
5829 | }, | |
5830 | { | |
5831 | .name = "high", | |
5832 | .flags = CFTYPE_NOT_ON_ROOT, | |
5833 | .seq_show = memory_high_show, | |
5834 | .write = memory_high_write, | |
5835 | }, | |
5836 | { | |
5837 | .name = "max", | |
5838 | .flags = CFTYPE_NOT_ON_ROOT, | |
5839 | .seq_show = memory_max_show, | |
5840 | .write = memory_max_write, | |
5841 | }, | |
5842 | { | |
5843 | .name = "events", | |
5844 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5845 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5846 | .seq_show = memory_events_show, |
5847 | }, | |
1e577f97 SB |
5848 | { |
5849 | .name = "events.local", | |
5850 | .flags = CFTYPE_NOT_ON_ROOT, | |
5851 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
5852 | .seq_show = memory_events_local_show, | |
5853 | }, | |
587d9f72 JW |
5854 | { |
5855 | .name = "stat", | |
5856 | .flags = CFTYPE_NOT_ON_ROOT, | |
5857 | .seq_show = memory_stat_show, | |
5858 | }, | |
3d8b38eb RG |
5859 | { |
5860 | .name = "oom.group", | |
5861 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
5862 | .seq_show = memory_oom_group_show, | |
5863 | .write = memory_oom_group_write, | |
5864 | }, | |
241994ed JW |
5865 | { } /* terminate */ |
5866 | }; | |
5867 | ||
073219e9 | 5868 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5869 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5870 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5871 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5872 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5873 | .css_free = mem_cgroup_css_free, |
1ced953b | 5874 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5875 | .can_attach = mem_cgroup_can_attach, |
5876 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 5877 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5878 | .bind = mem_cgroup_bind, |
241994ed JW |
5879 | .dfl_cftypes = memory_files, |
5880 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5881 | .early_init = 0, |
8cdea7c0 | 5882 | }; |
c077719b | 5883 | |
241994ed | 5884 | /** |
bf8d5d52 | 5885 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 5886 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
5887 | * @memcg: the memory cgroup to check |
5888 | * | |
23067153 RG |
5889 | * WARNING: This function is not stateless! It can only be used as part |
5890 | * of a top-down tree iteration, not for isolated queries. | |
34c81057 | 5891 | * |
bf8d5d52 RG |
5892 | * Returns one of the following: |
5893 | * MEMCG_PROT_NONE: cgroup memory is not protected | |
5894 | * MEMCG_PROT_LOW: cgroup memory is protected as long there is | |
5895 | * an unprotected supply of reclaimable memory from other cgroups. | |
5896 | * MEMCG_PROT_MIN: cgroup memory is protected | |
34c81057 | 5897 | * |
bf8d5d52 | 5898 | * @root is exclusive; it is never protected when looked at directly |
34c81057 | 5899 | * |
bf8d5d52 RG |
5900 | * To provide a proper hierarchical behavior, effective memory.min/low values |
5901 | * are used. Below is the description of how effective memory.low is calculated. | |
5902 | * Effective memory.min values is calculated in the same way. | |
34c81057 | 5903 | * |
23067153 RG |
5904 | * Effective memory.low is always equal or less than the original memory.low. |
5905 | * If there is no memory.low overcommittment (which is always true for | |
5906 | * top-level memory cgroups), these two values are equal. | |
5907 | * Otherwise, it's a part of parent's effective memory.low, | |
5908 | * calculated as a cgroup's memory.low usage divided by sum of sibling's | |
5909 | * memory.low usages, where memory.low usage is the size of actually | |
5910 | * protected memory. | |
34c81057 | 5911 | * |
23067153 RG |
5912 | * low_usage |
5913 | * elow = min( memory.low, parent->elow * ------------------ ), | |
5914 | * siblings_low_usage | |
34c81057 | 5915 | * |
23067153 RG |
5916 | * | memory.current, if memory.current < memory.low |
5917 | * low_usage = | | |
82ede7ee | 5918 | * | 0, otherwise. |
34c81057 | 5919 | * |
23067153 RG |
5920 | * |
5921 | * Such definition of the effective memory.low provides the expected | |
5922 | * hierarchical behavior: parent's memory.low value is limiting | |
5923 | * children, unprotected memory is reclaimed first and cgroups, | |
5924 | * which are not using their guarantee do not affect actual memory | |
5925 | * distribution. | |
5926 | * | |
5927 | * For example, if there are memcgs A, A/B, A/C, A/D and A/E: | |
5928 | * | |
5929 | * A A/memory.low = 2G, A/memory.current = 6G | |
5930 | * //\\ | |
5931 | * BC DE B/memory.low = 3G B/memory.current = 2G | |
5932 | * C/memory.low = 1G C/memory.current = 2G | |
5933 | * D/memory.low = 0 D/memory.current = 2G | |
5934 | * E/memory.low = 10G E/memory.current = 0 | |
5935 | * | |
5936 | * and the memory pressure is applied, the following memory distribution | |
5937 | * is expected (approximately): | |
5938 | * | |
5939 | * A/memory.current = 2G | |
5940 | * | |
5941 | * B/memory.current = 1.3G | |
5942 | * C/memory.current = 0.6G | |
5943 | * D/memory.current = 0 | |
5944 | * E/memory.current = 0 | |
5945 | * | |
5946 | * These calculations require constant tracking of the actual low usages | |
bf8d5d52 RG |
5947 | * (see propagate_protected_usage()), as well as recursive calculation of |
5948 | * effective memory.low values. But as we do call mem_cgroup_protected() | |
23067153 RG |
5949 | * path for each memory cgroup top-down from the reclaim, |
5950 | * it's possible to optimize this part, and save calculated elow | |
5951 | * for next usage. This part is intentionally racy, but it's ok, | |
5952 | * as memory.low is a best-effort mechanism. | |
241994ed | 5953 | */ |
bf8d5d52 RG |
5954 | enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root, |
5955 | struct mem_cgroup *memcg) | |
241994ed | 5956 | { |
23067153 | 5957 | struct mem_cgroup *parent; |
bf8d5d52 RG |
5958 | unsigned long emin, parent_emin; |
5959 | unsigned long elow, parent_elow; | |
5960 | unsigned long usage; | |
23067153 | 5961 | |
241994ed | 5962 | if (mem_cgroup_disabled()) |
bf8d5d52 | 5963 | return MEMCG_PROT_NONE; |
241994ed | 5964 | |
34c81057 SC |
5965 | if (!root) |
5966 | root = root_mem_cgroup; | |
5967 | if (memcg == root) | |
bf8d5d52 | 5968 | return MEMCG_PROT_NONE; |
241994ed | 5969 | |
23067153 | 5970 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 RG |
5971 | if (!usage) |
5972 | return MEMCG_PROT_NONE; | |
5973 | ||
5974 | emin = memcg->memory.min; | |
5975 | elow = memcg->memory.low; | |
34c81057 | 5976 | |
bf8d5d52 | 5977 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
5978 | /* No parent means a non-hierarchical mode on v1 memcg */ |
5979 | if (!parent) | |
5980 | return MEMCG_PROT_NONE; | |
5981 | ||
23067153 RG |
5982 | if (parent == root) |
5983 | goto exit; | |
5984 | ||
bf8d5d52 RG |
5985 | parent_emin = READ_ONCE(parent->memory.emin); |
5986 | emin = min(emin, parent_emin); | |
5987 | if (emin && parent_emin) { | |
5988 | unsigned long min_usage, siblings_min_usage; | |
5989 | ||
5990 | min_usage = min(usage, memcg->memory.min); | |
5991 | siblings_min_usage = atomic_long_read( | |
5992 | &parent->memory.children_min_usage); | |
5993 | ||
5994 | if (min_usage && siblings_min_usage) | |
5995 | emin = min(emin, parent_emin * min_usage / | |
5996 | siblings_min_usage); | |
5997 | } | |
5998 | ||
23067153 RG |
5999 | parent_elow = READ_ONCE(parent->memory.elow); |
6000 | elow = min(elow, parent_elow); | |
bf8d5d52 RG |
6001 | if (elow && parent_elow) { |
6002 | unsigned long low_usage, siblings_low_usage; | |
23067153 | 6003 | |
bf8d5d52 RG |
6004 | low_usage = min(usage, memcg->memory.low); |
6005 | siblings_low_usage = atomic_long_read( | |
6006 | &parent->memory.children_low_usage); | |
23067153 | 6007 | |
bf8d5d52 RG |
6008 | if (low_usage && siblings_low_usage) |
6009 | elow = min(elow, parent_elow * low_usage / | |
6010 | siblings_low_usage); | |
6011 | } | |
23067153 | 6012 | |
23067153 | 6013 | exit: |
bf8d5d52 | 6014 | memcg->memory.emin = emin; |
23067153 | 6015 | memcg->memory.elow = elow; |
bf8d5d52 RG |
6016 | |
6017 | if (usage <= emin) | |
6018 | return MEMCG_PROT_MIN; | |
6019 | else if (usage <= elow) | |
6020 | return MEMCG_PROT_LOW; | |
6021 | else | |
6022 | return MEMCG_PROT_NONE; | |
241994ed JW |
6023 | } |
6024 | ||
00501b53 JW |
6025 | /** |
6026 | * mem_cgroup_try_charge - try charging a page | |
6027 | * @page: page to charge | |
6028 | * @mm: mm context of the victim | |
6029 | * @gfp_mask: reclaim mode | |
6030 | * @memcgp: charged memcg return | |
25843c2b | 6031 | * @compound: charge the page as compound or small page |
00501b53 JW |
6032 | * |
6033 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6034 | * pages according to @gfp_mask if necessary. | |
6035 | * | |
6036 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
6037 | * Otherwise, an error code is returned. | |
6038 | * | |
6039 | * After page->mapping has been set up, the caller must finalize the | |
6040 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
6041 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
6042 | */ | |
6043 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
6044 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
6045 | bool compound) | |
00501b53 JW |
6046 | { |
6047 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 6048 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6049 | int ret = 0; |
6050 | ||
6051 | if (mem_cgroup_disabled()) | |
6052 | goto out; | |
6053 | ||
6054 | if (PageSwapCache(page)) { | |
00501b53 JW |
6055 | /* |
6056 | * Every swap fault against a single page tries to charge the | |
6057 | * page, bail as early as possible. shmem_unuse() encounters | |
6058 | * already charged pages, too. The USED bit is protected by | |
6059 | * the page lock, which serializes swap cache removal, which | |
6060 | * in turn serializes uncharging. | |
6061 | */ | |
e993d905 | 6062 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 6063 | if (compound_head(page)->mem_cgroup) |
00501b53 | 6064 | goto out; |
e993d905 | 6065 | |
37e84351 | 6066 | if (do_swap_account) { |
e993d905 VD |
6067 | swp_entry_t ent = { .val = page_private(page), }; |
6068 | unsigned short id = lookup_swap_cgroup_id(ent); | |
6069 | ||
6070 | rcu_read_lock(); | |
6071 | memcg = mem_cgroup_from_id(id); | |
6072 | if (memcg && !css_tryget_online(&memcg->css)) | |
6073 | memcg = NULL; | |
6074 | rcu_read_unlock(); | |
6075 | } | |
00501b53 JW |
6076 | } |
6077 | ||
00501b53 JW |
6078 | if (!memcg) |
6079 | memcg = get_mem_cgroup_from_mm(mm); | |
6080 | ||
6081 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
6082 | ||
6083 | css_put(&memcg->css); | |
00501b53 JW |
6084 | out: |
6085 | *memcgp = memcg; | |
6086 | return ret; | |
6087 | } | |
6088 | ||
2cf85583 TH |
6089 | int mem_cgroup_try_charge_delay(struct page *page, struct mm_struct *mm, |
6090 | gfp_t gfp_mask, struct mem_cgroup **memcgp, | |
6091 | bool compound) | |
6092 | { | |
6093 | struct mem_cgroup *memcg; | |
6094 | int ret; | |
6095 | ||
6096 | ret = mem_cgroup_try_charge(page, mm, gfp_mask, memcgp, compound); | |
6097 | memcg = *memcgp; | |
6098 | mem_cgroup_throttle_swaprate(memcg, page_to_nid(page), gfp_mask); | |
6099 | return ret; | |
6100 | } | |
6101 | ||
00501b53 JW |
6102 | /** |
6103 | * mem_cgroup_commit_charge - commit a page charge | |
6104 | * @page: page to charge | |
6105 | * @memcg: memcg to charge the page to | |
6106 | * @lrucare: page might be on LRU already | |
25843c2b | 6107 | * @compound: charge the page as compound or small page |
00501b53 JW |
6108 | * |
6109 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
6110 | * after page->mapping has been set up. This must happen atomically | |
6111 | * as part of the page instantiation, i.e. under the page table lock | |
6112 | * for anonymous pages, under the page lock for page and swap cache. | |
6113 | * | |
6114 | * In addition, the page must not be on the LRU during the commit, to | |
6115 | * prevent racing with task migration. If it might be, use @lrucare. | |
6116 | * | |
6117 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
6118 | */ | |
6119 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 6120 | bool lrucare, bool compound) |
00501b53 | 6121 | { |
f627c2f5 | 6122 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6123 | |
6124 | VM_BUG_ON_PAGE(!page->mapping, page); | |
6125 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
6126 | ||
6127 | if (mem_cgroup_disabled()) | |
6128 | return; | |
6129 | /* | |
6130 | * Swap faults will attempt to charge the same page multiple | |
6131 | * times. But reuse_swap_page() might have removed the page | |
6132 | * from swapcache already, so we can't check PageSwapCache(). | |
6133 | */ | |
6134 | if (!memcg) | |
6135 | return; | |
6136 | ||
6abb5a86 JW |
6137 | commit_charge(page, memcg, lrucare); |
6138 | ||
6abb5a86 | 6139 | local_irq_disable(); |
f627c2f5 | 6140 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
6141 | memcg_check_events(memcg, page); |
6142 | local_irq_enable(); | |
00501b53 | 6143 | |
7941d214 | 6144 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
6145 | swp_entry_t entry = { .val = page_private(page) }; |
6146 | /* | |
6147 | * The swap entry might not get freed for a long time, | |
6148 | * let's not wait for it. The page already received a | |
6149 | * memory+swap charge, drop the swap entry duplicate. | |
6150 | */ | |
38d8b4e6 | 6151 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 JW |
6152 | } |
6153 | } | |
6154 | ||
6155 | /** | |
6156 | * mem_cgroup_cancel_charge - cancel a page charge | |
6157 | * @page: page to charge | |
6158 | * @memcg: memcg to charge the page to | |
25843c2b | 6159 | * @compound: charge the page as compound or small page |
00501b53 JW |
6160 | * |
6161 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
6162 | */ | |
f627c2f5 KS |
6163 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
6164 | bool compound) | |
00501b53 | 6165 | { |
f627c2f5 | 6166 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
6167 | |
6168 | if (mem_cgroup_disabled()) | |
6169 | return; | |
6170 | /* | |
6171 | * Swap faults will attempt to charge the same page multiple | |
6172 | * times. But reuse_swap_page() might have removed the page | |
6173 | * from swapcache already, so we can't check PageSwapCache(). | |
6174 | */ | |
6175 | if (!memcg) | |
6176 | return; | |
6177 | ||
00501b53 JW |
6178 | cancel_charge(memcg, nr_pages); |
6179 | } | |
6180 | ||
a9d5adee JG |
6181 | struct uncharge_gather { |
6182 | struct mem_cgroup *memcg; | |
6183 | unsigned long pgpgout; | |
6184 | unsigned long nr_anon; | |
6185 | unsigned long nr_file; | |
6186 | unsigned long nr_kmem; | |
6187 | unsigned long nr_huge; | |
6188 | unsigned long nr_shmem; | |
6189 | struct page *dummy_page; | |
6190 | }; | |
6191 | ||
6192 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6193 | { |
a9d5adee JG |
6194 | memset(ug, 0, sizeof(*ug)); |
6195 | } | |
6196 | ||
6197 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6198 | { | |
6199 | unsigned long nr_pages = ug->nr_anon + ug->nr_file + ug->nr_kmem; | |
747db954 JW |
6200 | unsigned long flags; |
6201 | ||
a9d5adee JG |
6202 | if (!mem_cgroup_is_root(ug->memcg)) { |
6203 | page_counter_uncharge(&ug->memcg->memory, nr_pages); | |
7941d214 | 6204 | if (do_memsw_account()) |
a9d5adee JG |
6205 | page_counter_uncharge(&ug->memcg->memsw, nr_pages); |
6206 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) | |
6207 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6208 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6209 | } |
747db954 JW |
6210 | |
6211 | local_irq_save(flags); | |
c9019e9b JW |
6212 | __mod_memcg_state(ug->memcg, MEMCG_RSS, -ug->nr_anon); |
6213 | __mod_memcg_state(ug->memcg, MEMCG_CACHE, -ug->nr_file); | |
6214 | __mod_memcg_state(ug->memcg, MEMCG_RSS_HUGE, -ug->nr_huge); | |
6215 | __mod_memcg_state(ug->memcg, NR_SHMEM, -ug->nr_shmem); | |
6216 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); | |
871789d4 | 6217 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, nr_pages); |
a9d5adee | 6218 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6219 | local_irq_restore(flags); |
e8ea14cc | 6220 | |
a9d5adee JG |
6221 | if (!mem_cgroup_is_root(ug->memcg)) |
6222 | css_put_many(&ug->memcg->css, nr_pages); | |
6223 | } | |
6224 | ||
6225 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6226 | { | |
6227 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
3f2eb028 JG |
6228 | VM_BUG_ON_PAGE(page_count(page) && !is_zone_device_page(page) && |
6229 | !PageHWPoison(page) , page); | |
a9d5adee JG |
6230 | |
6231 | if (!page->mem_cgroup) | |
6232 | return; | |
6233 | ||
6234 | /* | |
6235 | * Nobody should be changing or seriously looking at | |
6236 | * page->mem_cgroup at this point, we have fully | |
6237 | * exclusive access to the page. | |
6238 | */ | |
6239 | ||
6240 | if (ug->memcg != page->mem_cgroup) { | |
6241 | if (ug->memcg) { | |
6242 | uncharge_batch(ug); | |
6243 | uncharge_gather_clear(ug); | |
6244 | } | |
6245 | ug->memcg = page->mem_cgroup; | |
6246 | } | |
6247 | ||
6248 | if (!PageKmemcg(page)) { | |
6249 | unsigned int nr_pages = 1; | |
6250 | ||
6251 | if (PageTransHuge(page)) { | |
6252 | nr_pages <<= compound_order(page); | |
6253 | ug->nr_huge += nr_pages; | |
6254 | } | |
6255 | if (PageAnon(page)) | |
6256 | ug->nr_anon += nr_pages; | |
6257 | else { | |
6258 | ug->nr_file += nr_pages; | |
6259 | if (PageSwapBacked(page)) | |
6260 | ug->nr_shmem += nr_pages; | |
6261 | } | |
6262 | ug->pgpgout++; | |
6263 | } else { | |
6264 | ug->nr_kmem += 1 << compound_order(page); | |
6265 | __ClearPageKmemcg(page); | |
6266 | } | |
6267 | ||
6268 | ug->dummy_page = page; | |
6269 | page->mem_cgroup = NULL; | |
747db954 JW |
6270 | } |
6271 | ||
6272 | static void uncharge_list(struct list_head *page_list) | |
6273 | { | |
a9d5adee | 6274 | struct uncharge_gather ug; |
747db954 | 6275 | struct list_head *next; |
a9d5adee JG |
6276 | |
6277 | uncharge_gather_clear(&ug); | |
747db954 | 6278 | |
8b592656 JW |
6279 | /* |
6280 | * Note that the list can be a single page->lru; hence the | |
6281 | * do-while loop instead of a simple list_for_each_entry(). | |
6282 | */ | |
747db954 JW |
6283 | next = page_list->next; |
6284 | do { | |
a9d5adee JG |
6285 | struct page *page; |
6286 | ||
747db954 JW |
6287 | page = list_entry(next, struct page, lru); |
6288 | next = page->lru.next; | |
6289 | ||
a9d5adee | 6290 | uncharge_page(page, &ug); |
747db954 JW |
6291 | } while (next != page_list); |
6292 | ||
a9d5adee JG |
6293 | if (ug.memcg) |
6294 | uncharge_batch(&ug); | |
747db954 JW |
6295 | } |
6296 | ||
0a31bc97 JW |
6297 | /** |
6298 | * mem_cgroup_uncharge - uncharge a page | |
6299 | * @page: page to uncharge | |
6300 | * | |
6301 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
6302 | * mem_cgroup_commit_charge(). | |
6303 | */ | |
6304 | void mem_cgroup_uncharge(struct page *page) | |
6305 | { | |
a9d5adee JG |
6306 | struct uncharge_gather ug; |
6307 | ||
0a31bc97 JW |
6308 | if (mem_cgroup_disabled()) |
6309 | return; | |
6310 | ||
747db954 | 6311 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 6312 | if (!page->mem_cgroup) |
0a31bc97 JW |
6313 | return; |
6314 | ||
a9d5adee JG |
6315 | uncharge_gather_clear(&ug); |
6316 | uncharge_page(page, &ug); | |
6317 | uncharge_batch(&ug); | |
747db954 | 6318 | } |
0a31bc97 | 6319 | |
747db954 JW |
6320 | /** |
6321 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6322 | * @page_list: list of pages to uncharge | |
6323 | * | |
6324 | * Uncharge a list of pages previously charged with | |
6325 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
6326 | */ | |
6327 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6328 | { | |
6329 | if (mem_cgroup_disabled()) | |
6330 | return; | |
0a31bc97 | 6331 | |
747db954 JW |
6332 | if (!list_empty(page_list)) |
6333 | uncharge_list(page_list); | |
0a31bc97 JW |
6334 | } |
6335 | ||
6336 | /** | |
6a93ca8f JW |
6337 | * mem_cgroup_migrate - charge a page's replacement |
6338 | * @oldpage: currently circulating page | |
6339 | * @newpage: replacement page | |
0a31bc97 | 6340 | * |
6a93ca8f JW |
6341 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6342 | * be uncharged upon free. | |
0a31bc97 JW |
6343 | * |
6344 | * Both pages must be locked, @newpage->mapping must be set up. | |
6345 | */ | |
6a93ca8f | 6346 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6347 | { |
29833315 | 6348 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
6349 | unsigned int nr_pages; |
6350 | bool compound; | |
d93c4130 | 6351 | unsigned long flags; |
0a31bc97 JW |
6352 | |
6353 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6354 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6355 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6356 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6357 | newpage); | |
0a31bc97 JW |
6358 | |
6359 | if (mem_cgroup_disabled()) | |
6360 | return; | |
6361 | ||
6362 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 6363 | if (newpage->mem_cgroup) |
0a31bc97 JW |
6364 | return; |
6365 | ||
45637bab | 6366 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 6367 | memcg = oldpage->mem_cgroup; |
29833315 | 6368 | if (!memcg) |
0a31bc97 JW |
6369 | return; |
6370 | ||
44b7a8d3 JW |
6371 | /* Force-charge the new page. The old one will be freed soon */ |
6372 | compound = PageTransHuge(newpage); | |
6373 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
6374 | ||
6375 | page_counter_charge(&memcg->memory, nr_pages); | |
6376 | if (do_memsw_account()) | |
6377 | page_counter_charge(&memcg->memsw, nr_pages); | |
6378 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 6379 | |
9cf7666a | 6380 | commit_charge(newpage, memcg, false); |
44b7a8d3 | 6381 | |
d93c4130 | 6382 | local_irq_save(flags); |
44b7a8d3 JW |
6383 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
6384 | memcg_check_events(memcg, newpage); | |
d93c4130 | 6385 | local_irq_restore(flags); |
0a31bc97 JW |
6386 | } |
6387 | ||
ef12947c | 6388 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6389 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6390 | ||
2d758073 | 6391 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6392 | { |
6393 | struct mem_cgroup *memcg; | |
6394 | ||
2d758073 JW |
6395 | if (!mem_cgroup_sockets_enabled) |
6396 | return; | |
6397 | ||
edbe69ef RG |
6398 | /* |
6399 | * Socket cloning can throw us here with sk_memcg already | |
6400 | * filled. It won't however, necessarily happen from | |
6401 | * process context. So the test for root memcg given | |
6402 | * the current task's memcg won't help us in this case. | |
6403 | * | |
6404 | * Respecting the original socket's memcg is a better | |
6405 | * decision in this case. | |
6406 | */ | |
6407 | if (sk->sk_memcg) { | |
6408 | css_get(&sk->sk_memcg->css); | |
6409 | return; | |
6410 | } | |
6411 | ||
11092087 JW |
6412 | rcu_read_lock(); |
6413 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6414 | if (memcg == root_mem_cgroup) |
6415 | goto out; | |
0db15298 | 6416 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6417 | goto out; |
f7e1cb6e | 6418 | if (css_tryget_online(&memcg->css)) |
11092087 | 6419 | sk->sk_memcg = memcg; |
f7e1cb6e | 6420 | out: |
11092087 JW |
6421 | rcu_read_unlock(); |
6422 | } | |
11092087 | 6423 | |
2d758073 | 6424 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6425 | { |
2d758073 JW |
6426 | if (sk->sk_memcg) |
6427 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6428 | } |
6429 | ||
6430 | /** | |
6431 | * mem_cgroup_charge_skmem - charge socket memory | |
6432 | * @memcg: memcg to charge | |
6433 | * @nr_pages: number of pages to charge | |
6434 | * | |
6435 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6436 | * @memcg's configured limit, %false if the charge had to be forced. | |
6437 | */ | |
6438 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6439 | { | |
f7e1cb6e | 6440 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6441 | |
f7e1cb6e | 6442 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6443 | struct page_counter *fail; |
f7e1cb6e | 6444 | |
0db15298 JW |
6445 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6446 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6447 | return true; |
6448 | } | |
0db15298 JW |
6449 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6450 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6451 | return false; |
11092087 | 6452 | } |
d886f4e4 | 6453 | |
f7e1cb6e JW |
6454 | /* Don't block in the packet receive path */ |
6455 | if (in_softirq()) | |
6456 | gfp_mask = GFP_NOWAIT; | |
6457 | ||
c9019e9b | 6458 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6459 | |
f7e1cb6e JW |
6460 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6461 | return true; | |
6462 | ||
6463 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6464 | return false; |
6465 | } | |
6466 | ||
6467 | /** | |
6468 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
6469 | * @memcg: memcg to uncharge |
6470 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
6471 | */ |
6472 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6473 | { | |
f7e1cb6e | 6474 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6475 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
6476 | return; |
6477 | } | |
d886f4e4 | 6478 | |
c9019e9b | 6479 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 6480 | |
475d0487 | 6481 | refill_stock(memcg, nr_pages); |
11092087 JW |
6482 | } |
6483 | ||
f7e1cb6e JW |
6484 | static int __init cgroup_memory(char *s) |
6485 | { | |
6486 | char *token; | |
6487 | ||
6488 | while ((token = strsep(&s, ",")) != NULL) { | |
6489 | if (!*token) | |
6490 | continue; | |
6491 | if (!strcmp(token, "nosocket")) | |
6492 | cgroup_memory_nosocket = true; | |
04823c83 VD |
6493 | if (!strcmp(token, "nokmem")) |
6494 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
6495 | } |
6496 | return 0; | |
6497 | } | |
6498 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 6499 | |
2d11085e | 6500 | /* |
1081312f MH |
6501 | * subsys_initcall() for memory controller. |
6502 | * | |
308167fc SAS |
6503 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
6504 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
6505 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
6506 | * should be initialized from here. | |
2d11085e MH |
6507 | */ |
6508 | static int __init mem_cgroup_init(void) | |
6509 | { | |
95a045f6 JW |
6510 | int cpu, node; |
6511 | ||
84c07d11 | 6512 | #ifdef CONFIG_MEMCG_KMEM |
13583c3d VD |
6513 | /* |
6514 | * Kmem cache creation is mostly done with the slab_mutex held, | |
17cc4dfe TH |
6515 | * so use a workqueue with limited concurrency to avoid stalling |
6516 | * all worker threads in case lots of cgroups are created and | |
6517 | * destroyed simultaneously. | |
13583c3d | 6518 | */ |
17cc4dfe TH |
6519 | memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1); |
6520 | BUG_ON(!memcg_kmem_cache_wq); | |
13583c3d VD |
6521 | #endif |
6522 | ||
308167fc SAS |
6523 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
6524 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
6525 | |
6526 | for_each_possible_cpu(cpu) | |
6527 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
6528 | drain_local_stock); | |
6529 | ||
6530 | for_each_node(node) { | |
6531 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
6532 | |
6533 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
6534 | node_online(node) ? node : NUMA_NO_NODE); | |
6535 | ||
ef8f2327 | 6536 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 6537 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 6538 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
6539 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
6540 | } | |
6541 | ||
2d11085e MH |
6542 | return 0; |
6543 | } | |
6544 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
6545 | |
6546 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
6547 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
6548 | { | |
1c2d479a | 6549 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
6550 | /* |
6551 | * The root cgroup cannot be destroyed, so it's refcount must | |
6552 | * always be >= 1. | |
6553 | */ | |
6554 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
6555 | VM_BUG_ON(1); | |
6556 | break; | |
6557 | } | |
6558 | memcg = parent_mem_cgroup(memcg); | |
6559 | if (!memcg) | |
6560 | memcg = root_mem_cgroup; | |
6561 | } | |
6562 | return memcg; | |
6563 | } | |
6564 | ||
21afa38e JW |
6565 | /** |
6566 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
6567 | * @page: page whose memsw charge to transfer | |
6568 | * @entry: swap entry to move the charge to | |
6569 | * | |
6570 | * Transfer the memsw charge of @page to @entry. | |
6571 | */ | |
6572 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
6573 | { | |
1f47b61f | 6574 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 6575 | unsigned int nr_entries; |
21afa38e JW |
6576 | unsigned short oldid; |
6577 | ||
6578 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6579 | VM_BUG_ON_PAGE(page_count(page), page); | |
6580 | ||
7941d214 | 6581 | if (!do_memsw_account()) |
21afa38e JW |
6582 | return; |
6583 | ||
6584 | memcg = page->mem_cgroup; | |
6585 | ||
6586 | /* Readahead page, never charged */ | |
6587 | if (!memcg) | |
6588 | return; | |
6589 | ||
1f47b61f VD |
6590 | /* |
6591 | * In case the memcg owning these pages has been offlined and doesn't | |
6592 | * have an ID allocated to it anymore, charge the closest online | |
6593 | * ancestor for the swap instead and transfer the memory+swap charge. | |
6594 | */ | |
6595 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
d6810d73 HY |
6596 | nr_entries = hpage_nr_pages(page); |
6597 | /* Get references for the tail pages, too */ | |
6598 | if (nr_entries > 1) | |
6599 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
6600 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
6601 | nr_entries); | |
21afa38e | 6602 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6603 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
6604 | |
6605 | page->mem_cgroup = NULL; | |
6606 | ||
6607 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 6608 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 6609 | |
1f47b61f VD |
6610 | if (memcg != swap_memcg) { |
6611 | if (!mem_cgroup_is_root(swap_memcg)) | |
d6810d73 HY |
6612 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
6613 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
6614 | } |
6615 | ||
ce9ce665 SAS |
6616 | /* |
6617 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 6618 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 6619 | * important here to have the interrupts disabled because it is the |
b93b0163 | 6620 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
6621 | */ |
6622 | VM_BUG_ON(!irqs_disabled()); | |
d6810d73 HY |
6623 | mem_cgroup_charge_statistics(memcg, page, PageTransHuge(page), |
6624 | -nr_entries); | |
21afa38e | 6625 | memcg_check_events(memcg, page); |
73f576c0 JW |
6626 | |
6627 | if (!mem_cgroup_is_root(memcg)) | |
d08afa14 | 6628 | css_put_many(&memcg->css, nr_entries); |
21afa38e JW |
6629 | } |
6630 | ||
38d8b4e6 HY |
6631 | /** |
6632 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
6633 | * @page: page being added to swap |
6634 | * @entry: swap entry to charge | |
6635 | * | |
38d8b4e6 | 6636 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
6637 | * |
6638 | * Returns 0 on success, -ENOMEM on failure. | |
6639 | */ | |
6640 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
6641 | { | |
38d8b4e6 | 6642 | unsigned int nr_pages = hpage_nr_pages(page); |
37e84351 | 6643 | struct page_counter *counter; |
38d8b4e6 | 6644 | struct mem_cgroup *memcg; |
37e84351 VD |
6645 | unsigned short oldid; |
6646 | ||
6647 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
6648 | return 0; | |
6649 | ||
6650 | memcg = page->mem_cgroup; | |
6651 | ||
6652 | /* Readahead page, never charged */ | |
6653 | if (!memcg) | |
6654 | return 0; | |
6655 | ||
f3a53a3a TH |
6656 | if (!entry.val) { |
6657 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 6658 | return 0; |
f3a53a3a | 6659 | } |
bb98f2c5 | 6660 | |
1f47b61f VD |
6661 | memcg = mem_cgroup_id_get_online(memcg); |
6662 | ||
37e84351 | 6663 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 6664 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
6665 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
6666 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 6667 | mem_cgroup_id_put(memcg); |
37e84351 | 6668 | return -ENOMEM; |
1f47b61f | 6669 | } |
37e84351 | 6670 | |
38d8b4e6 HY |
6671 | /* Get references for the tail pages, too */ |
6672 | if (nr_pages > 1) | |
6673 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
6674 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 6675 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6676 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 6677 | |
37e84351 VD |
6678 | return 0; |
6679 | } | |
6680 | ||
21afa38e | 6681 | /** |
38d8b4e6 | 6682 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 6683 | * @entry: swap entry to uncharge |
38d8b4e6 | 6684 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 6685 | */ |
38d8b4e6 | 6686 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
6687 | { |
6688 | struct mem_cgroup *memcg; | |
6689 | unsigned short id; | |
6690 | ||
37e84351 | 6691 | if (!do_swap_account) |
21afa38e JW |
6692 | return; |
6693 | ||
38d8b4e6 | 6694 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 6695 | rcu_read_lock(); |
adbe427b | 6696 | memcg = mem_cgroup_from_id(id); |
21afa38e | 6697 | if (memcg) { |
37e84351 VD |
6698 | if (!mem_cgroup_is_root(memcg)) { |
6699 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
38d8b4e6 | 6700 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 6701 | else |
38d8b4e6 | 6702 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 6703 | } |
c9019e9b | 6704 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 6705 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
6706 | } |
6707 | rcu_read_unlock(); | |
6708 | } | |
6709 | ||
d8b38438 VD |
6710 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
6711 | { | |
6712 | long nr_swap_pages = get_nr_swap_pages(); | |
6713 | ||
6714 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6715 | return nr_swap_pages; | |
6716 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
6717 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 6718 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
6719 | page_counter_read(&memcg->swap)); |
6720 | return nr_swap_pages; | |
6721 | } | |
6722 | ||
5ccc5aba VD |
6723 | bool mem_cgroup_swap_full(struct page *page) |
6724 | { | |
6725 | struct mem_cgroup *memcg; | |
6726 | ||
6727 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
6728 | ||
6729 | if (vm_swap_full()) | |
6730 | return true; | |
6731 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6732 | return false; | |
6733 | ||
6734 | memcg = page->mem_cgroup; | |
6735 | if (!memcg) | |
6736 | return false; | |
6737 | ||
6738 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
bbec2e15 | 6739 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.max) |
5ccc5aba VD |
6740 | return true; |
6741 | ||
6742 | return false; | |
6743 | } | |
6744 | ||
21afa38e JW |
6745 | /* for remember boot option*/ |
6746 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
6747 | static int really_do_swap_account __initdata = 1; | |
6748 | #else | |
6749 | static int really_do_swap_account __initdata; | |
6750 | #endif | |
6751 | ||
6752 | static int __init enable_swap_account(char *s) | |
6753 | { | |
6754 | if (!strcmp(s, "1")) | |
6755 | really_do_swap_account = 1; | |
6756 | else if (!strcmp(s, "0")) | |
6757 | really_do_swap_account = 0; | |
6758 | return 1; | |
6759 | } | |
6760 | __setup("swapaccount=", enable_swap_account); | |
6761 | ||
37e84351 VD |
6762 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
6763 | struct cftype *cft) | |
6764 | { | |
6765 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6766 | ||
6767 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
6768 | } | |
6769 | ||
6770 | static int swap_max_show(struct seq_file *m, void *v) | |
6771 | { | |
677dc973 CD |
6772 | return seq_puts_memcg_tunable(m, |
6773 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
6774 | } |
6775 | ||
6776 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
6777 | char *buf, size_t nbytes, loff_t off) | |
6778 | { | |
6779 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6780 | unsigned long max; | |
6781 | int err; | |
6782 | ||
6783 | buf = strstrip(buf); | |
6784 | err = page_counter_memparse(buf, "max", &max); | |
6785 | if (err) | |
6786 | return err; | |
6787 | ||
be09102b | 6788 | xchg(&memcg->swap.max, max); |
37e84351 VD |
6789 | |
6790 | return nbytes; | |
6791 | } | |
6792 | ||
f3a53a3a TH |
6793 | static int swap_events_show(struct seq_file *m, void *v) |
6794 | { | |
aa9694bb | 6795 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a TH |
6796 | |
6797 | seq_printf(m, "max %lu\n", | |
6798 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
6799 | seq_printf(m, "fail %lu\n", | |
6800 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
6801 | ||
6802 | return 0; | |
6803 | } | |
6804 | ||
37e84351 VD |
6805 | static struct cftype swap_files[] = { |
6806 | { | |
6807 | .name = "swap.current", | |
6808 | .flags = CFTYPE_NOT_ON_ROOT, | |
6809 | .read_u64 = swap_current_read, | |
6810 | }, | |
6811 | { | |
6812 | .name = "swap.max", | |
6813 | .flags = CFTYPE_NOT_ON_ROOT, | |
6814 | .seq_show = swap_max_show, | |
6815 | .write = swap_max_write, | |
6816 | }, | |
f3a53a3a TH |
6817 | { |
6818 | .name = "swap.events", | |
6819 | .flags = CFTYPE_NOT_ON_ROOT, | |
6820 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
6821 | .seq_show = swap_events_show, | |
6822 | }, | |
37e84351 VD |
6823 | { } /* terminate */ |
6824 | }; | |
6825 | ||
21afa38e JW |
6826 | static struct cftype memsw_cgroup_files[] = { |
6827 | { | |
6828 | .name = "memsw.usage_in_bytes", | |
6829 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6830 | .read_u64 = mem_cgroup_read_u64, | |
6831 | }, | |
6832 | { | |
6833 | .name = "memsw.max_usage_in_bytes", | |
6834 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6835 | .write = mem_cgroup_reset, | |
6836 | .read_u64 = mem_cgroup_read_u64, | |
6837 | }, | |
6838 | { | |
6839 | .name = "memsw.limit_in_bytes", | |
6840 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6841 | .write = mem_cgroup_write, | |
6842 | .read_u64 = mem_cgroup_read_u64, | |
6843 | }, | |
6844 | { | |
6845 | .name = "memsw.failcnt", | |
6846 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6847 | .write = mem_cgroup_reset, | |
6848 | .read_u64 = mem_cgroup_read_u64, | |
6849 | }, | |
6850 | { }, /* terminate */ | |
6851 | }; | |
6852 | ||
6853 | static int __init mem_cgroup_swap_init(void) | |
6854 | { | |
6855 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6856 | do_swap_account = 1; | |
37e84351 VD |
6857 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
6858 | swap_files)); | |
21afa38e JW |
6859 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6860 | memsw_cgroup_files)); | |
6861 | } | |
6862 | return 0; | |
6863 | } | |
6864 | subsys_initcall(mem_cgroup_swap_init); | |
6865 | ||
6866 | #endif /* CONFIG_MEMCG_SWAP */ |