]>
Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
7ae1e1d0 GC |
13 | * Kernel Memory Controller |
14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
15 | * Authors: Glauber Costa and Suleiman Souhlal | |
16 | * | |
1575e68b JW |
17 | * Native page reclaim |
18 | * Charge lifetime sanitation | |
19 | * Lockless page tracking & accounting | |
20 | * Unified hierarchy configuration model | |
21 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
22 | * | |
8cdea7c0 BS |
23 | * This program is free software; you can redistribute it and/or modify |
24 | * it under the terms of the GNU General Public License as published by | |
25 | * the Free Software Foundation; either version 2 of the License, or | |
26 | * (at your option) any later version. | |
27 | * | |
28 | * This program is distributed in the hope that it will be useful, | |
29 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
30 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
31 | * GNU General Public License for more details. | |
32 | */ | |
33 | ||
3e32cb2e | 34 | #include <linux/page_counter.h> |
8cdea7c0 BS |
35 | #include <linux/memcontrol.h> |
36 | #include <linux/cgroup.h> | |
78fb7466 | 37 | #include <linux/mm.h> |
6e84f315 | 38 | #include <linux/sched/mm.h> |
3a4f8a0b | 39 | #include <linux/shmem_fs.h> |
4ffef5fe | 40 | #include <linux/hugetlb.h> |
d13d1443 | 41 | #include <linux/pagemap.h> |
d52aa412 | 42 | #include <linux/smp.h> |
8a9f3ccd | 43 | #include <linux/page-flags.h> |
66e1707b | 44 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
45 | #include <linux/bit_spinlock.h> |
46 | #include <linux/rcupdate.h> | |
e222432b | 47 | #include <linux/limits.h> |
b9e15baf | 48 | #include <linux/export.h> |
8c7c6e34 | 49 | #include <linux/mutex.h> |
bb4cc1a8 | 50 | #include <linux/rbtree.h> |
b6ac57d5 | 51 | #include <linux/slab.h> |
66e1707b | 52 | #include <linux/swap.h> |
02491447 | 53 | #include <linux/swapops.h> |
66e1707b | 54 | #include <linux/spinlock.h> |
2e72b634 | 55 | #include <linux/eventfd.h> |
79bd9814 | 56 | #include <linux/poll.h> |
2e72b634 | 57 | #include <linux/sort.h> |
66e1707b | 58 | #include <linux/fs.h> |
d2ceb9b7 | 59 | #include <linux/seq_file.h> |
70ddf637 | 60 | #include <linux/vmpressure.h> |
b69408e8 | 61 | #include <linux/mm_inline.h> |
5d1ea48b | 62 | #include <linux/swap_cgroup.h> |
cdec2e42 | 63 | #include <linux/cpu.h> |
158e0a2d | 64 | #include <linux/oom.h> |
0056f4e6 | 65 | #include <linux/lockdep.h> |
79bd9814 | 66 | #include <linux/file.h> |
b23afb93 | 67 | #include <linux/tracehook.h> |
08e552c6 | 68 | #include "internal.h" |
d1a4c0b3 | 69 | #include <net/sock.h> |
4bd2c1ee | 70 | #include <net/ip.h> |
f35c3a8e | 71 | #include "slab.h" |
8cdea7c0 | 72 | |
7c0f6ba6 | 73 | #include <linux/uaccess.h> |
8697d331 | 74 | |
cc8e970c KM |
75 | #include <trace/events/vmscan.h> |
76 | ||
073219e9 TH |
77 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
78 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 79 | |
7d828602 JW |
80 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
81 | ||
a181b0e8 | 82 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
8cdea7c0 | 83 | |
f7e1cb6e JW |
84 | /* Socket memory accounting disabled? */ |
85 | static bool cgroup_memory_nosocket; | |
86 | ||
04823c83 VD |
87 | /* Kernel memory accounting disabled? */ |
88 | static bool cgroup_memory_nokmem; | |
89 | ||
21afa38e | 90 | /* Whether the swap controller is active */ |
c255a458 | 91 | #ifdef CONFIG_MEMCG_SWAP |
c077719b | 92 | int do_swap_account __read_mostly; |
c077719b | 93 | #else |
a0db00fc | 94 | #define do_swap_account 0 |
c077719b KH |
95 | #endif |
96 | ||
7941d214 JW |
97 | /* Whether legacy memory+swap accounting is active */ |
98 | static bool do_memsw_account(void) | |
99 | { | |
100 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; | |
101 | } | |
102 | ||
71cd3113 | 103 | static const char *const mem_cgroup_lru_names[] = { |
58cf188e SZ |
104 | "inactive_anon", |
105 | "active_anon", | |
106 | "inactive_file", | |
107 | "active_file", | |
108 | "unevictable", | |
109 | }; | |
110 | ||
a0db00fc KS |
111 | #define THRESHOLDS_EVENTS_TARGET 128 |
112 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
113 | #define NUMAINFO_EVENTS_TARGET 1024 | |
e9f8974f | 114 | |
bb4cc1a8 AM |
115 | /* |
116 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
117 | * their hierarchy representation | |
118 | */ | |
119 | ||
ef8f2327 | 120 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 121 | struct rb_root rb_root; |
fa90b2fd | 122 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
123 | spinlock_t lock; |
124 | }; | |
125 | ||
bb4cc1a8 AM |
126 | struct mem_cgroup_tree { |
127 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
128 | }; | |
129 | ||
130 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
131 | ||
9490ff27 KH |
132 | /* for OOM */ |
133 | struct mem_cgroup_eventfd_list { | |
134 | struct list_head list; | |
135 | struct eventfd_ctx *eventfd; | |
136 | }; | |
2e72b634 | 137 | |
79bd9814 TH |
138 | /* |
139 | * cgroup_event represents events which userspace want to receive. | |
140 | */ | |
3bc942f3 | 141 | struct mem_cgroup_event { |
79bd9814 | 142 | /* |
59b6f873 | 143 | * memcg which the event belongs to. |
79bd9814 | 144 | */ |
59b6f873 | 145 | struct mem_cgroup *memcg; |
79bd9814 TH |
146 | /* |
147 | * eventfd to signal userspace about the event. | |
148 | */ | |
149 | struct eventfd_ctx *eventfd; | |
150 | /* | |
151 | * Each of these stored in a list by the cgroup. | |
152 | */ | |
153 | struct list_head list; | |
fba94807 TH |
154 | /* |
155 | * register_event() callback will be used to add new userspace | |
156 | * waiter for changes related to this event. Use eventfd_signal() | |
157 | * on eventfd to send notification to userspace. | |
158 | */ | |
59b6f873 | 159 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 160 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
161 | /* |
162 | * unregister_event() callback will be called when userspace closes | |
163 | * the eventfd or on cgroup removing. This callback must be set, | |
164 | * if you want provide notification functionality. | |
165 | */ | |
59b6f873 | 166 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 167 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
168 | /* |
169 | * All fields below needed to unregister event when | |
170 | * userspace closes eventfd. | |
171 | */ | |
172 | poll_table pt; | |
173 | wait_queue_head_t *wqh; | |
ac6424b9 | 174 | wait_queue_entry_t wait; |
79bd9814 TH |
175 | struct work_struct remove; |
176 | }; | |
177 | ||
c0ff4b85 R |
178 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
179 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 180 | |
7dc74be0 DN |
181 | /* Stuffs for move charges at task migration. */ |
182 | /* | |
1dfab5ab | 183 | * Types of charges to be moved. |
7dc74be0 | 184 | */ |
1dfab5ab JW |
185 | #define MOVE_ANON 0x1U |
186 | #define MOVE_FILE 0x2U | |
187 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 188 | |
4ffef5fe DN |
189 | /* "mc" and its members are protected by cgroup_mutex */ |
190 | static struct move_charge_struct { | |
b1dd693e | 191 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 192 | struct mm_struct *mm; |
4ffef5fe DN |
193 | struct mem_cgroup *from; |
194 | struct mem_cgroup *to; | |
1dfab5ab | 195 | unsigned long flags; |
4ffef5fe | 196 | unsigned long precharge; |
854ffa8d | 197 | unsigned long moved_charge; |
483c30b5 | 198 | unsigned long moved_swap; |
8033b97c DN |
199 | struct task_struct *moving_task; /* a task moving charges */ |
200 | wait_queue_head_t waitq; /* a waitq for other context */ | |
201 | } mc = { | |
2bd9bb20 | 202 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
203 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
204 | }; | |
4ffef5fe | 205 | |
4e416953 BS |
206 | /* |
207 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
208 | * limit reclaim to prevent infinite loops, if they ever occur. | |
209 | */ | |
a0db00fc | 210 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 211 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 212 | |
217bc319 KH |
213 | enum charge_type { |
214 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
41326c17 | 215 | MEM_CGROUP_CHARGE_TYPE_ANON, |
d13d1443 | 216 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 217 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
218 | NR_CHARGE_TYPE, |
219 | }; | |
220 | ||
8c7c6e34 | 221 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
222 | enum res_type { |
223 | _MEM, | |
224 | _MEMSWAP, | |
225 | _OOM_TYPE, | |
510fc4e1 | 226 | _KMEM, |
d55f90bf | 227 | _TCP, |
86ae53e1 GC |
228 | }; |
229 | ||
a0db00fc KS |
230 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
231 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 232 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
233 | /* Used for OOM nofiier */ |
234 | #define OOM_CONTROL (0) | |
8c7c6e34 | 235 | |
d9e9af93 TH |
236 | static inline bool should_force_charge(void) |
237 | { | |
238 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
239 | (current->flags & PF_EXITING); | |
240 | } | |
241 | ||
70ddf637 AV |
242 | /* Some nice accessors for the vmpressure. */ |
243 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
244 | { | |
245 | if (!memcg) | |
246 | memcg = root_mem_cgroup; | |
247 | return &memcg->vmpressure; | |
248 | } | |
249 | ||
250 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
251 | { | |
252 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
253 | } | |
254 | ||
7ffc0edc MH |
255 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
256 | { | |
257 | return (memcg == root_mem_cgroup); | |
258 | } | |
259 | ||
127424c8 | 260 | #ifndef CONFIG_SLOB |
55007d84 | 261 | /* |
f7ce3190 | 262 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
263 | * The main reason for not using cgroup id for this: |
264 | * this works better in sparse environments, where we have a lot of memcgs, | |
265 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
266 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
267 | * 200 entry array for that. | |
55007d84 | 268 | * |
dbcf73e2 VD |
269 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
270 | * will double each time we have to increase it. | |
55007d84 | 271 | */ |
dbcf73e2 VD |
272 | static DEFINE_IDA(memcg_cache_ida); |
273 | int memcg_nr_cache_ids; | |
749c5415 | 274 | |
05257a1a VD |
275 | /* Protects memcg_nr_cache_ids */ |
276 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
277 | ||
278 | void memcg_get_cache_ids(void) | |
279 | { | |
280 | down_read(&memcg_cache_ids_sem); | |
281 | } | |
282 | ||
283 | void memcg_put_cache_ids(void) | |
284 | { | |
285 | up_read(&memcg_cache_ids_sem); | |
286 | } | |
287 | ||
55007d84 GC |
288 | /* |
289 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
290 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
291 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
292 | * tunable, but that is strictly not necessary. | |
293 | * | |
b8627835 | 294 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
295 | * this constant directly from cgroup, but it is understandable that this is |
296 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 297 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
298 | * increase ours as well if it increases. |
299 | */ | |
300 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 301 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 302 | |
d7f25f8a GC |
303 | /* |
304 | * A lot of the calls to the cache allocation functions are expected to be | |
305 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
306 | * conditional to this static branch, we'll have to allow modules that does | |
307 | * kmem_cache_alloc and the such to see this symbol as well | |
308 | */ | |
ef12947c | 309 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 310 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 311 | |
17cc4dfe TH |
312 | struct workqueue_struct *memcg_kmem_cache_wq; |
313 | ||
127424c8 | 314 | #endif /* !CONFIG_SLOB */ |
a8964b9b | 315 | |
ad7fa852 TH |
316 | /** |
317 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
318 | * @page: page of interest | |
319 | * | |
320 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
321 | * with @page is returned. The returned css remains associated with @page | |
322 | * until it is released. | |
323 | * | |
324 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
325 | * is returned. | |
ad7fa852 TH |
326 | */ |
327 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
328 | { | |
329 | struct mem_cgroup *memcg; | |
330 | ||
ad7fa852 TH |
331 | memcg = page->mem_cgroup; |
332 | ||
9e10a130 | 333 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
334 | memcg = root_mem_cgroup; |
335 | ||
ad7fa852 TH |
336 | return &memcg->css; |
337 | } | |
338 | ||
2fc04524 VD |
339 | /** |
340 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
341 | * @page: the page | |
342 | * | |
343 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
344 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
345 | * is safe to call this function without holding a reference to @page. | |
346 | * | |
347 | * Note, this function is inherently racy, because there is nothing to prevent | |
348 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
349 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
350 | * do not care (such as procfs interfaces). | |
351 | */ | |
352 | ino_t page_cgroup_ino(struct page *page) | |
353 | { | |
354 | struct mem_cgroup *memcg; | |
355 | unsigned long ino = 0; | |
356 | ||
357 | rcu_read_lock(); | |
358 | memcg = READ_ONCE(page->mem_cgroup); | |
359 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
360 | memcg = parent_mem_cgroup(memcg); | |
361 | if (memcg) | |
362 | ino = cgroup_ino(memcg->css.cgroup); | |
363 | rcu_read_unlock(); | |
364 | return ino; | |
365 | } | |
366 | ||
ef8f2327 MG |
367 | static struct mem_cgroup_per_node * |
368 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 369 | { |
97a6c37b | 370 | int nid = page_to_nid(page); |
f64c3f54 | 371 | |
ef8f2327 | 372 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
373 | } |
374 | ||
ef8f2327 MG |
375 | static struct mem_cgroup_tree_per_node * |
376 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 377 | { |
ef8f2327 | 378 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
379 | } |
380 | ||
ef8f2327 | 381 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
382 | soft_limit_tree_from_page(struct page *page) |
383 | { | |
384 | int nid = page_to_nid(page); | |
bb4cc1a8 | 385 | |
ef8f2327 | 386 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
387 | } |
388 | ||
ef8f2327 MG |
389 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
390 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 391 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
392 | { |
393 | struct rb_node **p = &mctz->rb_root.rb_node; | |
394 | struct rb_node *parent = NULL; | |
ef8f2327 | 395 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 396 | bool rightmost = true; |
bb4cc1a8 AM |
397 | |
398 | if (mz->on_tree) | |
399 | return; | |
400 | ||
401 | mz->usage_in_excess = new_usage_in_excess; | |
402 | if (!mz->usage_in_excess) | |
403 | return; | |
404 | while (*p) { | |
405 | parent = *p; | |
ef8f2327 | 406 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 407 | tree_node); |
fa90b2fd | 408 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 409 | p = &(*p)->rb_left; |
fa90b2fd DB |
410 | rightmost = false; |
411 | } | |
412 | ||
bb4cc1a8 AM |
413 | /* |
414 | * We can't avoid mem cgroups that are over their soft | |
415 | * limit by the same amount | |
416 | */ | |
417 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
418 | p = &(*p)->rb_right; | |
419 | } | |
fa90b2fd DB |
420 | |
421 | if (rightmost) | |
422 | mctz->rb_rightmost = &mz->tree_node; | |
423 | ||
bb4cc1a8 AM |
424 | rb_link_node(&mz->tree_node, parent, p); |
425 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
426 | mz->on_tree = true; | |
427 | } | |
428 | ||
ef8f2327 MG |
429 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
430 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
431 | { |
432 | if (!mz->on_tree) | |
433 | return; | |
fa90b2fd DB |
434 | |
435 | if (&mz->tree_node == mctz->rb_rightmost) | |
436 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
437 | ||
bb4cc1a8 AM |
438 | rb_erase(&mz->tree_node, &mctz->rb_root); |
439 | mz->on_tree = false; | |
440 | } | |
441 | ||
ef8f2327 MG |
442 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
443 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 444 | { |
0a31bc97 JW |
445 | unsigned long flags; |
446 | ||
447 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 448 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 449 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
450 | } |
451 | ||
3e32cb2e JW |
452 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
453 | { | |
454 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 455 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
456 | unsigned long excess = 0; |
457 | ||
458 | if (nr_pages > soft_limit) | |
459 | excess = nr_pages - soft_limit; | |
460 | ||
461 | return excess; | |
462 | } | |
bb4cc1a8 AM |
463 | |
464 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
465 | { | |
3e32cb2e | 466 | unsigned long excess; |
ef8f2327 MG |
467 | struct mem_cgroup_per_node *mz; |
468 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 469 | |
e231875b | 470 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
471 | if (!mctz) |
472 | return; | |
bb4cc1a8 AM |
473 | /* |
474 | * Necessary to update all ancestors when hierarchy is used. | |
475 | * because their event counter is not touched. | |
476 | */ | |
477 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 478 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 479 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
480 | /* |
481 | * We have to update the tree if mz is on RB-tree or | |
482 | * mem is over its softlimit. | |
483 | */ | |
484 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
485 | unsigned long flags; |
486 | ||
487 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
488 | /* if on-tree, remove it */ |
489 | if (mz->on_tree) | |
cf2c8127 | 490 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
491 | /* |
492 | * Insert again. mz->usage_in_excess will be updated. | |
493 | * If excess is 0, no tree ops. | |
494 | */ | |
cf2c8127 | 495 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 496 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
497 | } |
498 | } | |
499 | } | |
500 | ||
501 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
502 | { | |
ef8f2327 MG |
503 | struct mem_cgroup_tree_per_node *mctz; |
504 | struct mem_cgroup_per_node *mz; | |
505 | int nid; | |
bb4cc1a8 | 506 | |
e231875b | 507 | for_each_node(nid) { |
ef8f2327 MG |
508 | mz = mem_cgroup_nodeinfo(memcg, nid); |
509 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
510 | if (mctz) |
511 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
512 | } |
513 | } | |
514 | ||
ef8f2327 MG |
515 | static struct mem_cgroup_per_node * |
516 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 517 | { |
ef8f2327 | 518 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
519 | |
520 | retry: | |
521 | mz = NULL; | |
fa90b2fd | 522 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
523 | goto done; /* Nothing to reclaim from */ |
524 | ||
fa90b2fd DB |
525 | mz = rb_entry(mctz->rb_rightmost, |
526 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
527 | /* |
528 | * Remove the node now but someone else can add it back, | |
529 | * we will to add it back at the end of reclaim to its correct | |
530 | * position in the tree. | |
531 | */ | |
cf2c8127 | 532 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 533 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 534 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
535 | goto retry; |
536 | done: | |
537 | return mz; | |
538 | } | |
539 | ||
ef8f2327 MG |
540 | static struct mem_cgroup_per_node * |
541 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 542 | { |
ef8f2327 | 543 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 544 | |
0a31bc97 | 545 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 546 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 547 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
548 | return mz; |
549 | } | |
550 | ||
711d3d2c | 551 | /* |
484ebb3b GT |
552 | * Return page count for single (non recursive) @memcg. |
553 | * | |
711d3d2c KH |
554 | * Implementation Note: reading percpu statistics for memcg. |
555 | * | |
556 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
557 | * synchronization to implement "quick" read. There are trade-off between | |
558 | * reading cost and precision of value. Then, we may have a chance to implement | |
484ebb3b | 559 | * a periodic synchronization of counter in memcg's counter. |
711d3d2c KH |
560 | * |
561 | * But this _read() function is used for user interface now. The user accounts | |
562 | * memory usage by memory cgroup and he _always_ requires exact value because | |
563 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
564 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
565 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
566 | * | |
567 | * If there are kernel internal actions which can make use of some not-exact | |
568 | * value, and reading all cpu value can be performance bottleneck in some | |
484ebb3b | 569 | * common workload, threshold and synchronization as vmstat[] should be |
711d3d2c | 570 | * implemented. |
04fecbf5 MK |
571 | * |
572 | * The parameter idx can be of type enum memcg_event_item or vm_event_item. | |
711d3d2c | 573 | */ |
c62b1a3b | 574 | |
ccda7f43 | 575 | static unsigned long memcg_sum_events(struct mem_cgroup *memcg, |
04fecbf5 | 576 | int event) |
e9f8974f JW |
577 | { |
578 | unsigned long val = 0; | |
579 | int cpu; | |
580 | ||
733a572e | 581 | for_each_possible_cpu(cpu) |
df0e53d0 | 582 | val += per_cpu(memcg->stat->events[event], cpu); |
e9f8974f JW |
583 | return val; |
584 | } | |
585 | ||
c0ff4b85 | 586 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 587 | struct page *page, |
f627c2f5 | 588 | bool compound, int nr_pages) |
d52aa412 | 589 | { |
b2402857 KH |
590 | /* |
591 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
592 | * counted as CACHE even if it's on ANON LRU. | |
593 | */ | |
0a31bc97 | 594 | if (PageAnon(page)) |
71cd3113 | 595 | __this_cpu_add(memcg->stat->count[MEMCG_RSS], nr_pages); |
9a4caf1e | 596 | else { |
71cd3113 | 597 | __this_cpu_add(memcg->stat->count[MEMCG_CACHE], nr_pages); |
9a4caf1e | 598 | if (PageSwapBacked(page)) |
71cd3113 | 599 | __this_cpu_add(memcg->stat->count[NR_SHMEM], nr_pages); |
9a4caf1e | 600 | } |
55e462b0 | 601 | |
f627c2f5 KS |
602 | if (compound) { |
603 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
71cd3113 | 604 | __this_cpu_add(memcg->stat->count[MEMCG_RSS_HUGE], nr_pages); |
f627c2f5 | 605 | } |
b070e65c | 606 | |
e401f176 KH |
607 | /* pagein of a big page is an event. So, ignore page size */ |
608 | if (nr_pages > 0) | |
df0e53d0 | 609 | __this_cpu_inc(memcg->stat->events[PGPGIN]); |
3751d604 | 610 | else { |
df0e53d0 | 611 | __this_cpu_inc(memcg->stat->events[PGPGOUT]); |
3751d604 KH |
612 | nr_pages = -nr_pages; /* for event */ |
613 | } | |
e401f176 | 614 | |
13114716 | 615 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
6d12e2d8 KH |
616 | } |
617 | ||
0a6b76dd VD |
618 | unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
619 | int nid, unsigned int lru_mask) | |
bb2a0de9 | 620 | { |
b4536f0c | 621 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); |
e231875b | 622 | unsigned long nr = 0; |
ef8f2327 | 623 | enum lru_list lru; |
889976db | 624 | |
e231875b | 625 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 626 | |
ef8f2327 MG |
627 | for_each_lru(lru) { |
628 | if (!(BIT(lru) & lru_mask)) | |
629 | continue; | |
b4536f0c | 630 | nr += mem_cgroup_get_lru_size(lruvec, lru); |
e231875b JZ |
631 | } |
632 | return nr; | |
889976db | 633 | } |
bb2a0de9 | 634 | |
c0ff4b85 | 635 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 636 | unsigned int lru_mask) |
6d12e2d8 | 637 | { |
e231875b | 638 | unsigned long nr = 0; |
889976db | 639 | int nid; |
6d12e2d8 | 640 | |
31aaea4a | 641 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
642 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
643 | return nr; | |
d52aa412 KH |
644 | } |
645 | ||
f53d7ce3 JW |
646 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
647 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
648 | { |
649 | unsigned long val, next; | |
650 | ||
13114716 | 651 | val = __this_cpu_read(memcg->stat->nr_page_events); |
4799401f | 652 | next = __this_cpu_read(memcg->stat->targets[target]); |
7a159cc9 | 653 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 654 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
655 | switch (target) { |
656 | case MEM_CGROUP_TARGET_THRESH: | |
657 | next = val + THRESHOLDS_EVENTS_TARGET; | |
658 | break; | |
bb4cc1a8 AM |
659 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
660 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
661 | break; | |
f53d7ce3 JW |
662 | case MEM_CGROUP_TARGET_NUMAINFO: |
663 | next = val + NUMAINFO_EVENTS_TARGET; | |
664 | break; | |
665 | default: | |
666 | break; | |
667 | } | |
668 | __this_cpu_write(memcg->stat->targets[target], next); | |
669 | return true; | |
7a159cc9 | 670 | } |
f53d7ce3 | 671 | return false; |
d2265e6f KH |
672 | } |
673 | ||
674 | /* | |
675 | * Check events in order. | |
676 | * | |
677 | */ | |
c0ff4b85 | 678 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
679 | { |
680 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
681 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
682 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 683 | bool do_softlimit; |
82b3f2a7 | 684 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 685 | |
bb4cc1a8 AM |
686 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
687 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
688 | #if MAX_NUMNODES > 1 |
689 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
690 | MEM_CGROUP_TARGET_NUMAINFO); | |
691 | #endif | |
c0ff4b85 | 692 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
693 | if (unlikely(do_softlimit)) |
694 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 695 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 696 | if (unlikely(do_numainfo)) |
c0ff4b85 | 697 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 698 | #endif |
0a31bc97 | 699 | } |
d2265e6f KH |
700 | } |
701 | ||
cf475ad2 | 702 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 703 | { |
31a78f23 BS |
704 | /* |
705 | * mm_update_next_owner() may clear mm->owner to NULL | |
706 | * if it races with swapoff, page migration, etc. | |
707 | * So this can be called with p == NULL. | |
708 | */ | |
709 | if (unlikely(!p)) | |
710 | return NULL; | |
711 | ||
073219e9 | 712 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 713 | } |
33398cf2 | 714 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 715 | |
df381975 | 716 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 717 | { |
c0ff4b85 | 718 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 719 | |
54595fe2 KH |
720 | rcu_read_lock(); |
721 | do { | |
6f6acb00 MH |
722 | /* |
723 | * Page cache insertions can happen withou an | |
724 | * actual mm context, e.g. during disk probing | |
725 | * on boot, loopback IO, acct() writes etc. | |
726 | */ | |
727 | if (unlikely(!mm)) | |
df381975 | 728 | memcg = root_mem_cgroup; |
6f6acb00 MH |
729 | else { |
730 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
731 | if (unlikely(!memcg)) | |
732 | memcg = root_mem_cgroup; | |
733 | } | |
b7ebcc92 | 734 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 735 | rcu_read_unlock(); |
c0ff4b85 | 736 | return memcg; |
54595fe2 KH |
737 | } |
738 | ||
5660048c JW |
739 | /** |
740 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
741 | * @root: hierarchy root | |
742 | * @prev: previously returned memcg, NULL on first invocation | |
743 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
744 | * | |
745 | * Returns references to children of the hierarchy below @root, or | |
746 | * @root itself, or %NULL after a full round-trip. | |
747 | * | |
748 | * Caller must pass the return value in @prev on subsequent | |
749 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
750 | * to cancel a hierarchy walk before the round-trip is complete. | |
751 | * | |
752 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
753 | * divide up the memcgs in the hierarchy among all concurrent | |
754 | * reclaimers operating on the same zone and priority. | |
755 | */ | |
694fbc0f | 756 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 757 | struct mem_cgroup *prev, |
694fbc0f | 758 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 759 | { |
33398cf2 | 760 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 761 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 762 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 763 | struct mem_cgroup *pos = NULL; |
711d3d2c | 764 | |
694fbc0f AM |
765 | if (mem_cgroup_disabled()) |
766 | return NULL; | |
5660048c | 767 | |
9f3a0d09 JW |
768 | if (!root) |
769 | root = root_mem_cgroup; | |
7d74b06f | 770 | |
9f3a0d09 | 771 | if (prev && !reclaim) |
5ac8fb31 | 772 | pos = prev; |
14067bb3 | 773 | |
9f3a0d09 JW |
774 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
775 | if (prev) | |
5ac8fb31 | 776 | goto out; |
694fbc0f | 777 | return root; |
9f3a0d09 | 778 | } |
14067bb3 | 779 | |
542f85f9 | 780 | rcu_read_lock(); |
5f578161 | 781 | |
5ac8fb31 | 782 | if (reclaim) { |
ef8f2327 | 783 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 784 | |
ef8f2327 | 785 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
5ac8fb31 JW |
786 | iter = &mz->iter[reclaim->priority]; |
787 | ||
788 | if (prev && reclaim->generation != iter->generation) | |
789 | goto out_unlock; | |
790 | ||
6df38689 | 791 | while (1) { |
4db0c3c2 | 792 | pos = READ_ONCE(iter->position); |
6df38689 VD |
793 | if (!pos || css_tryget(&pos->css)) |
794 | break; | |
5ac8fb31 | 795 | /* |
6df38689 VD |
796 | * css reference reached zero, so iter->position will |
797 | * be cleared by ->css_released. However, we should not | |
798 | * rely on this happening soon, because ->css_released | |
799 | * is called from a work queue, and by busy-waiting we | |
800 | * might block it. So we clear iter->position right | |
801 | * away. | |
5ac8fb31 | 802 | */ |
6df38689 VD |
803 | (void)cmpxchg(&iter->position, pos, NULL); |
804 | } | |
5ac8fb31 JW |
805 | } |
806 | ||
807 | if (pos) | |
808 | css = &pos->css; | |
809 | ||
810 | for (;;) { | |
811 | css = css_next_descendant_pre(css, &root->css); | |
812 | if (!css) { | |
813 | /* | |
814 | * Reclaimers share the hierarchy walk, and a | |
815 | * new one might jump in right at the end of | |
816 | * the hierarchy - make sure they see at least | |
817 | * one group and restart from the beginning. | |
818 | */ | |
819 | if (!prev) | |
820 | continue; | |
821 | break; | |
527a5ec9 | 822 | } |
7d74b06f | 823 | |
5ac8fb31 JW |
824 | /* |
825 | * Verify the css and acquire a reference. The root | |
826 | * is provided by the caller, so we know it's alive | |
827 | * and kicking, and don't take an extra reference. | |
828 | */ | |
829 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 830 | |
5ac8fb31 JW |
831 | if (css == &root->css) |
832 | break; | |
14067bb3 | 833 | |
0b8f73e1 JW |
834 | if (css_tryget(css)) |
835 | break; | |
9f3a0d09 | 836 | |
5ac8fb31 | 837 | memcg = NULL; |
9f3a0d09 | 838 | } |
5ac8fb31 JW |
839 | |
840 | if (reclaim) { | |
5ac8fb31 | 841 | /* |
6df38689 VD |
842 | * The position could have already been updated by a competing |
843 | * thread, so check that the value hasn't changed since we read | |
844 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 845 | */ |
6df38689 VD |
846 | (void)cmpxchg(&iter->position, pos, memcg); |
847 | ||
5ac8fb31 JW |
848 | if (pos) |
849 | css_put(&pos->css); | |
850 | ||
851 | if (!memcg) | |
852 | iter->generation++; | |
853 | else if (!prev) | |
854 | reclaim->generation = iter->generation; | |
9f3a0d09 | 855 | } |
5ac8fb31 | 856 | |
542f85f9 MH |
857 | out_unlock: |
858 | rcu_read_unlock(); | |
5ac8fb31 | 859 | out: |
c40046f3 MH |
860 | if (prev && prev != root) |
861 | css_put(&prev->css); | |
862 | ||
9f3a0d09 | 863 | return memcg; |
14067bb3 | 864 | } |
7d74b06f | 865 | |
5660048c JW |
866 | /** |
867 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
868 | * @root: hierarchy root | |
869 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
870 | */ | |
871 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
872 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
873 | { |
874 | if (!root) | |
875 | root = root_mem_cgroup; | |
876 | if (prev && prev != root) | |
877 | css_put(&prev->css); | |
878 | } | |
7d74b06f | 879 | |
0cddab6a MC |
880 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
881 | struct mem_cgroup *dead_memcg) | |
6df38689 | 882 | { |
6df38689 | 883 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
884 | struct mem_cgroup_per_node *mz; |
885 | int nid; | |
6df38689 VD |
886 | int i; |
887 | ||
0cddab6a MC |
888 | for_each_node(nid) { |
889 | mz = mem_cgroup_nodeinfo(from, nid); | |
890 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
891 | iter = &mz->iter[i]; | |
892 | cmpxchg(&iter->position, | |
893 | dead_memcg, NULL); | |
6df38689 VD |
894 | } |
895 | } | |
896 | } | |
897 | ||
0cddab6a MC |
898 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
899 | { | |
900 | struct mem_cgroup *memcg = dead_memcg; | |
901 | struct mem_cgroup *last; | |
902 | ||
903 | do { | |
904 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
905 | last = memcg; | |
906 | } while ((memcg = parent_mem_cgroup(memcg))); | |
907 | ||
908 | /* | |
909 | * When cgruop1 non-hierarchy mode is used, | |
910 | * parent_mem_cgroup() does not walk all the way up to the | |
911 | * cgroup root (root_mem_cgroup). So we have to handle | |
912 | * dead_memcg from cgroup root separately. | |
913 | */ | |
914 | if (last != root_mem_cgroup) | |
915 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
916 | dead_memcg); | |
917 | } | |
918 | ||
9f3a0d09 JW |
919 | /* |
920 | * Iteration constructs for visiting all cgroups (under a tree). If | |
921 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
922 | * be used for reference counting. | |
923 | */ | |
924 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 925 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 926 | iter != NULL; \ |
527a5ec9 | 927 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 928 | |
9f3a0d09 | 929 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 930 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 931 | iter != NULL; \ |
527a5ec9 | 932 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 933 | |
7c5f64f8 VD |
934 | /** |
935 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
936 | * @memcg: hierarchy root | |
937 | * @fn: function to call for each task | |
938 | * @arg: argument passed to @fn | |
939 | * | |
940 | * This function iterates over tasks attached to @memcg or to any of its | |
941 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
942 | * value, the function breaks the iteration loop and returns the value. | |
943 | * Otherwise, it will iterate over all tasks and return 0. | |
944 | * | |
945 | * This function must not be called for the root memory cgroup. | |
946 | */ | |
947 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
948 | int (*fn)(struct task_struct *, void *), void *arg) | |
949 | { | |
950 | struct mem_cgroup *iter; | |
951 | int ret = 0; | |
952 | ||
953 | BUG_ON(memcg == root_mem_cgroup); | |
954 | ||
955 | for_each_mem_cgroup_tree(iter, memcg) { | |
956 | struct css_task_iter it; | |
957 | struct task_struct *task; | |
958 | ||
bc2fb7ed | 959 | css_task_iter_start(&iter->css, 0, &it); |
7c5f64f8 VD |
960 | while (!ret && (task = css_task_iter_next(&it))) |
961 | ret = fn(task, arg); | |
962 | css_task_iter_end(&it); | |
963 | if (ret) { | |
964 | mem_cgroup_iter_break(memcg, iter); | |
965 | break; | |
966 | } | |
967 | } | |
968 | return ret; | |
969 | } | |
970 | ||
925b7673 | 971 | /** |
dfe0e773 | 972 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 973 | * @page: the page |
fa9add64 | 974 | * @zone: zone of the page |
dfe0e773 JW |
975 | * |
976 | * This function is only safe when following the LRU page isolation | |
977 | * and putback protocol: the LRU lock must be held, and the page must | |
978 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 979 | */ |
599d0c95 | 980 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 981 | { |
ef8f2327 | 982 | struct mem_cgroup_per_node *mz; |
925b7673 | 983 | struct mem_cgroup *memcg; |
bea8c150 | 984 | struct lruvec *lruvec; |
6d12e2d8 | 985 | |
bea8c150 | 986 | if (mem_cgroup_disabled()) { |
599d0c95 | 987 | lruvec = &pgdat->lruvec; |
bea8c150 HD |
988 | goto out; |
989 | } | |
925b7673 | 990 | |
1306a85a | 991 | memcg = page->mem_cgroup; |
7512102c | 992 | /* |
dfe0e773 | 993 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 994 | * possibly migrated - before they are charged. |
7512102c | 995 | */ |
29833315 JW |
996 | if (!memcg) |
997 | memcg = root_mem_cgroup; | |
7512102c | 998 | |
ef8f2327 | 999 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
1000 | lruvec = &mz->lruvec; |
1001 | out: | |
1002 | /* | |
1003 | * Since a node can be onlined after the mem_cgroup was created, | |
1004 | * we have to be prepared to initialize lruvec->zone here; | |
1005 | * and if offlined then reonlined, we need to reinitialize it. | |
1006 | */ | |
599d0c95 MG |
1007 | if (unlikely(lruvec->pgdat != pgdat)) |
1008 | lruvec->pgdat = pgdat; | |
bea8c150 | 1009 | return lruvec; |
08e552c6 | 1010 | } |
b69408e8 | 1011 | |
925b7673 | 1012 | /** |
fa9add64 HD |
1013 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1014 | * @lruvec: mem_cgroup per zone lru vector | |
1015 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1016 | * @zid: zone id of the accounted pages |
fa9add64 | 1017 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1018 | * |
ca707239 HD |
1019 | * This function must be called under lru_lock, just before a page is added |
1020 | * to or just after a page is removed from an lru list (that ordering being | |
1021 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1022 | */ |
fa9add64 | 1023 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1024 | int zid, int nr_pages) |
3f58a829 | 1025 | { |
ef8f2327 | 1026 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1027 | unsigned long *lru_size; |
ca707239 | 1028 | long size; |
3f58a829 MK |
1029 | |
1030 | if (mem_cgroup_disabled()) | |
1031 | return; | |
1032 | ||
ef8f2327 | 1033 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1034 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1035 | |
1036 | if (nr_pages < 0) | |
1037 | *lru_size += nr_pages; | |
1038 | ||
1039 | size = *lru_size; | |
b4536f0c MH |
1040 | if (WARN_ONCE(size < 0, |
1041 | "%s(%p, %d, %d): lru_size %ld\n", | |
1042 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1043 | VM_BUG_ON(1); |
1044 | *lru_size = 0; | |
1045 | } | |
1046 | ||
1047 | if (nr_pages > 0) | |
1048 | *lru_size += nr_pages; | |
08e552c6 | 1049 | } |
544122e5 | 1050 | |
2314b42d | 1051 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 1052 | { |
2314b42d | 1053 | struct mem_cgroup *task_memcg; |
158e0a2d | 1054 | struct task_struct *p; |
ffbdccf5 | 1055 | bool ret; |
4c4a2214 | 1056 | |
158e0a2d | 1057 | p = find_lock_task_mm(task); |
de077d22 | 1058 | if (p) { |
2314b42d | 1059 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1060 | task_unlock(p); |
1061 | } else { | |
1062 | /* | |
1063 | * All threads may have already detached their mm's, but the oom | |
1064 | * killer still needs to detect if they have already been oom | |
1065 | * killed to prevent needlessly killing additional tasks. | |
1066 | */ | |
ffbdccf5 | 1067 | rcu_read_lock(); |
2314b42d JW |
1068 | task_memcg = mem_cgroup_from_task(task); |
1069 | css_get(&task_memcg->css); | |
ffbdccf5 | 1070 | rcu_read_unlock(); |
de077d22 | 1071 | } |
2314b42d JW |
1072 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1073 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1074 | return ret; |
1075 | } | |
1076 | ||
19942822 | 1077 | /** |
9d11ea9f | 1078 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1079 | * @memcg: the memory cgroup |
19942822 | 1080 | * |
9d11ea9f | 1081 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1082 | * pages. |
19942822 | 1083 | */ |
c0ff4b85 | 1084 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1085 | { |
3e32cb2e JW |
1086 | unsigned long margin = 0; |
1087 | unsigned long count; | |
1088 | unsigned long limit; | |
9d11ea9f | 1089 | |
3e32cb2e | 1090 | count = page_counter_read(&memcg->memory); |
4db0c3c2 | 1091 | limit = READ_ONCE(memcg->memory.limit); |
3e32cb2e JW |
1092 | if (count < limit) |
1093 | margin = limit - count; | |
1094 | ||
7941d214 | 1095 | if (do_memsw_account()) { |
3e32cb2e | 1096 | count = page_counter_read(&memcg->memsw); |
4db0c3c2 | 1097 | limit = READ_ONCE(memcg->memsw.limit); |
3e32cb2e JW |
1098 | if (count <= limit) |
1099 | margin = min(margin, limit - count); | |
cbedbac3 LR |
1100 | else |
1101 | margin = 0; | |
3e32cb2e JW |
1102 | } |
1103 | ||
1104 | return margin; | |
19942822 JW |
1105 | } |
1106 | ||
32047e2a | 1107 | /* |
bdcbb659 | 1108 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1109 | * |
bdcbb659 QH |
1110 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1111 | * moving cgroups. This is for waiting at high-memory pressure | |
1112 | * caused by "move". | |
32047e2a | 1113 | */ |
c0ff4b85 | 1114 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1115 | { |
2bd9bb20 KH |
1116 | struct mem_cgroup *from; |
1117 | struct mem_cgroup *to; | |
4b534334 | 1118 | bool ret = false; |
2bd9bb20 KH |
1119 | /* |
1120 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1121 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1122 | */ | |
1123 | spin_lock(&mc.lock); | |
1124 | from = mc.from; | |
1125 | to = mc.to; | |
1126 | if (!from) | |
1127 | goto unlock; | |
3e92041d | 1128 | |
2314b42d JW |
1129 | ret = mem_cgroup_is_descendant(from, memcg) || |
1130 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1131 | unlock: |
1132 | spin_unlock(&mc.lock); | |
4b534334 KH |
1133 | return ret; |
1134 | } | |
1135 | ||
c0ff4b85 | 1136 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1137 | { |
1138 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1139 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1140 | DEFINE_WAIT(wait); |
1141 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1142 | /* moving charge context might have finished. */ | |
1143 | if (mc.moving_task) | |
1144 | schedule(); | |
1145 | finish_wait(&mc.waitq, &wait); | |
1146 | return true; | |
1147 | } | |
1148 | } | |
1149 | return false; | |
1150 | } | |
1151 | ||
71cd3113 JW |
1152 | unsigned int memcg1_stats[] = { |
1153 | MEMCG_CACHE, | |
1154 | MEMCG_RSS, | |
1155 | MEMCG_RSS_HUGE, | |
1156 | NR_SHMEM, | |
1157 | NR_FILE_MAPPED, | |
1158 | NR_FILE_DIRTY, | |
1159 | NR_WRITEBACK, | |
1160 | MEMCG_SWAP, | |
1161 | }; | |
1162 | ||
1163 | static const char *const memcg1_stat_names[] = { | |
1164 | "cache", | |
1165 | "rss", | |
1166 | "rss_huge", | |
1167 | "shmem", | |
1168 | "mapped_file", | |
1169 | "dirty", | |
1170 | "writeback", | |
1171 | "swap", | |
1172 | }; | |
1173 | ||
58cf188e | 1174 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1175 | /** |
58cf188e | 1176 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1177 | * @memcg: The memory cgroup that went over limit |
1178 | * @p: Task that is going to be killed | |
1179 | * | |
1180 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1181 | * enabled | |
1182 | */ | |
1183 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1184 | { | |
58cf188e SZ |
1185 | struct mem_cgroup *iter; |
1186 | unsigned int i; | |
e222432b | 1187 | |
e222432b BS |
1188 | rcu_read_lock(); |
1189 | ||
2415b9f5 BV |
1190 | if (p) { |
1191 | pr_info("Task in "); | |
1192 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1193 | pr_cont(" killed as a result of limit of "); | |
1194 | } else { | |
1195 | pr_info("Memory limit reached of cgroup "); | |
1196 | } | |
1197 | ||
e61734c5 | 1198 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1199 | pr_cont("\n"); |
e222432b | 1200 | |
e222432b BS |
1201 | rcu_read_unlock(); |
1202 | ||
3e32cb2e JW |
1203 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1204 | K((u64)page_counter_read(&memcg->memory)), | |
1205 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1206 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1207 | K((u64)page_counter_read(&memcg->memsw)), | |
1208 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1209 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1210 | K((u64)page_counter_read(&memcg->kmem)), | |
1211 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1212 | |
1213 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1214 | pr_info("Memory cgroup stats for "); |
1215 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1216 | pr_cont(":"); |
1217 | ||
71cd3113 JW |
1218 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
1219 | if (memcg1_stats[i] == MEMCG_SWAP && !do_swap_account) | |
58cf188e | 1220 | continue; |
71cd3113 | 1221 | pr_cont(" %s:%luKB", memcg1_stat_names[i], |
ccda7f43 | 1222 | K(memcg_page_state(iter, memcg1_stats[i]))); |
58cf188e SZ |
1223 | } |
1224 | ||
1225 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1226 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1227 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1228 | ||
1229 | pr_cont("\n"); | |
1230 | } | |
e222432b BS |
1231 | } |
1232 | ||
81d39c20 KH |
1233 | /* |
1234 | * This function returns the number of memcg under hierarchy tree. Returns | |
1235 | * 1(self count) if no children. | |
1236 | */ | |
c0ff4b85 | 1237 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
81d39c20 KH |
1238 | { |
1239 | int num = 0; | |
7d74b06f KH |
1240 | struct mem_cgroup *iter; |
1241 | ||
c0ff4b85 | 1242 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 1243 | num++; |
81d39c20 KH |
1244 | return num; |
1245 | } | |
1246 | ||
a63d83f4 DR |
1247 | /* |
1248 | * Return the memory (and swap, if configured) limit for a memcg. | |
1249 | */ | |
7c5f64f8 | 1250 | unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1251 | { |
3e32cb2e | 1252 | unsigned long limit; |
f3e8eb70 | 1253 | |
3e32cb2e | 1254 | limit = memcg->memory.limit; |
9a5a8f19 | 1255 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1256 | unsigned long memsw_limit; |
37e84351 | 1257 | unsigned long swap_limit; |
9a5a8f19 | 1258 | |
3e32cb2e | 1259 | memsw_limit = memcg->memsw.limit; |
37e84351 VD |
1260 | swap_limit = memcg->swap.limit; |
1261 | swap_limit = min(swap_limit, (unsigned long)total_swap_pages); | |
1262 | limit = min(limit + swap_limit, memsw_limit); | |
9a5a8f19 | 1263 | } |
9a5a8f19 | 1264 | return limit; |
a63d83f4 DR |
1265 | } |
1266 | ||
b6e6edcf | 1267 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1268 | int order) |
9cbb78bb | 1269 | { |
6e0fc46d DR |
1270 | struct oom_control oc = { |
1271 | .zonelist = NULL, | |
1272 | .nodemask = NULL, | |
2a966b77 | 1273 | .memcg = memcg, |
6e0fc46d DR |
1274 | .gfp_mask = gfp_mask, |
1275 | .order = order, | |
6e0fc46d | 1276 | }; |
7c5f64f8 | 1277 | bool ret; |
9cbb78bb | 1278 | |
d9e9af93 TH |
1279 | if (mutex_lock_killable(&oom_lock)) |
1280 | return true; | |
1281 | /* | |
1282 | * A few threads which were not waiting at mutex_lock_killable() can | |
1283 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1284 | */ | |
1285 | ret = should_force_charge() || out_of_memory(&oc); | |
dc56401f | 1286 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1287 | return ret; |
9cbb78bb DR |
1288 | } |
1289 | ||
ae6e71d3 MC |
1290 | #if MAX_NUMNODES > 1 |
1291 | ||
4d0c066d KH |
1292 | /** |
1293 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1294 | * @memcg: the target memcg |
4d0c066d KH |
1295 | * @nid: the node ID to be checked. |
1296 | * @noswap : specify true here if the user wants flle only information. | |
1297 | * | |
1298 | * This function returns whether the specified memcg contains any | |
1299 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1300 | * pages in the node. | |
1301 | */ | |
c0ff4b85 | 1302 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1303 | int nid, bool noswap) |
1304 | { | |
c0ff4b85 | 1305 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1306 | return true; |
1307 | if (noswap || !total_swap_pages) | |
1308 | return false; | |
c0ff4b85 | 1309 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1310 | return true; |
1311 | return false; | |
1312 | ||
1313 | } | |
889976db YH |
1314 | |
1315 | /* | |
1316 | * Always updating the nodemask is not very good - even if we have an empty | |
1317 | * list or the wrong list here, we can start from some node and traverse all | |
1318 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1319 | * | |
1320 | */ | |
c0ff4b85 | 1321 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1322 | { |
1323 | int nid; | |
453a9bf3 KH |
1324 | /* |
1325 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1326 | * pagein/pageout changes since the last update. | |
1327 | */ | |
c0ff4b85 | 1328 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1329 | return; |
c0ff4b85 | 1330 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1331 | return; |
1332 | ||
889976db | 1333 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1334 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1335 | |
31aaea4a | 1336 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1337 | |
c0ff4b85 R |
1338 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1339 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1340 | } |
453a9bf3 | 1341 | |
c0ff4b85 R |
1342 | atomic_set(&memcg->numainfo_events, 0); |
1343 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1344 | } |
1345 | ||
1346 | /* | |
1347 | * Selecting a node where we start reclaim from. Because what we need is just | |
1348 | * reducing usage counter, start from anywhere is O,K. Considering | |
1349 | * memory reclaim from current node, there are pros. and cons. | |
1350 | * | |
1351 | * Freeing memory from current node means freeing memory from a node which | |
1352 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1353 | * hit limits, it will see a contention on a node. But freeing from remote | |
1354 | * node means more costs for memory reclaim because of memory latency. | |
1355 | * | |
1356 | * Now, we use round-robin. Better algorithm is welcomed. | |
1357 | */ | |
c0ff4b85 | 1358 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1359 | { |
1360 | int node; | |
1361 | ||
c0ff4b85 R |
1362 | mem_cgroup_may_update_nodemask(memcg); |
1363 | node = memcg->last_scanned_node; | |
889976db | 1364 | |
0edaf86c | 1365 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1366 | /* |
fda3d69b MH |
1367 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1368 | * last time it really checked all the LRUs due to rate limiting. | |
1369 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1370 | */ |
1371 | if (unlikely(node == MAX_NUMNODES)) | |
1372 | node = numa_node_id(); | |
1373 | ||
c0ff4b85 | 1374 | memcg->last_scanned_node = node; |
889976db YH |
1375 | return node; |
1376 | } | |
889976db | 1377 | #else |
c0ff4b85 | 1378 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1379 | { |
1380 | return 0; | |
1381 | } | |
1382 | #endif | |
1383 | ||
0608f43d | 1384 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1385 | pg_data_t *pgdat, |
0608f43d AM |
1386 | gfp_t gfp_mask, |
1387 | unsigned long *total_scanned) | |
1388 | { | |
1389 | struct mem_cgroup *victim = NULL; | |
1390 | int total = 0; | |
1391 | int loop = 0; | |
1392 | unsigned long excess; | |
1393 | unsigned long nr_scanned; | |
1394 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1395 | .pgdat = pgdat, |
0608f43d AM |
1396 | .priority = 0, |
1397 | }; | |
1398 | ||
3e32cb2e | 1399 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1400 | |
1401 | while (1) { | |
1402 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1403 | if (!victim) { | |
1404 | loop++; | |
1405 | if (loop >= 2) { | |
1406 | /* | |
1407 | * If we have not been able to reclaim | |
1408 | * anything, it might because there are | |
1409 | * no reclaimable pages under this hierarchy | |
1410 | */ | |
1411 | if (!total) | |
1412 | break; | |
1413 | /* | |
1414 | * We want to do more targeted reclaim. | |
1415 | * excess >> 2 is not to excessive so as to | |
1416 | * reclaim too much, nor too less that we keep | |
1417 | * coming back to reclaim from this cgroup | |
1418 | */ | |
1419 | if (total >= (excess >> 2) || | |
1420 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1421 | break; | |
1422 | } | |
1423 | continue; | |
1424 | } | |
a9dd0a83 | 1425 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1426 | pgdat, &nr_scanned); |
0608f43d | 1427 | *total_scanned += nr_scanned; |
3e32cb2e | 1428 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1429 | break; |
6d61ef40 | 1430 | } |
0608f43d AM |
1431 | mem_cgroup_iter_break(root_memcg, victim); |
1432 | return total; | |
6d61ef40 BS |
1433 | } |
1434 | ||
0056f4e6 JW |
1435 | #ifdef CONFIG_LOCKDEP |
1436 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1437 | .name = "memcg_oom_lock", | |
1438 | }; | |
1439 | #endif | |
1440 | ||
fb2a6fc5 JW |
1441 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1442 | ||
867578cb KH |
1443 | /* |
1444 | * Check OOM-Killer is already running under our hierarchy. | |
1445 | * If someone is running, return false. | |
1446 | */ | |
fb2a6fc5 | 1447 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1448 | { |
79dfdacc | 1449 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1450 | |
fb2a6fc5 JW |
1451 | spin_lock(&memcg_oom_lock); |
1452 | ||
9f3a0d09 | 1453 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1454 | if (iter->oom_lock) { |
79dfdacc MH |
1455 | /* |
1456 | * this subtree of our hierarchy is already locked | |
1457 | * so we cannot give a lock. | |
1458 | */ | |
79dfdacc | 1459 | failed = iter; |
9f3a0d09 JW |
1460 | mem_cgroup_iter_break(memcg, iter); |
1461 | break; | |
23751be0 JW |
1462 | } else |
1463 | iter->oom_lock = true; | |
7d74b06f | 1464 | } |
867578cb | 1465 | |
fb2a6fc5 JW |
1466 | if (failed) { |
1467 | /* | |
1468 | * OK, we failed to lock the whole subtree so we have | |
1469 | * to clean up what we set up to the failing subtree | |
1470 | */ | |
1471 | for_each_mem_cgroup_tree(iter, memcg) { | |
1472 | if (iter == failed) { | |
1473 | mem_cgroup_iter_break(memcg, iter); | |
1474 | break; | |
1475 | } | |
1476 | iter->oom_lock = false; | |
79dfdacc | 1477 | } |
0056f4e6 JW |
1478 | } else |
1479 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1480 | |
1481 | spin_unlock(&memcg_oom_lock); | |
1482 | ||
1483 | return !failed; | |
a636b327 | 1484 | } |
0b7f569e | 1485 | |
fb2a6fc5 | 1486 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1487 | { |
7d74b06f KH |
1488 | struct mem_cgroup *iter; |
1489 | ||
fb2a6fc5 | 1490 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1491 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1492 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1493 | iter->oom_lock = false; |
fb2a6fc5 | 1494 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1495 | } |
1496 | ||
c0ff4b85 | 1497 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1498 | { |
1499 | struct mem_cgroup *iter; | |
1500 | ||
c2b42d3c | 1501 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1502 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1503 | iter->under_oom++; |
1504 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1505 | } |
1506 | ||
c0ff4b85 | 1507 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1508 | { |
1509 | struct mem_cgroup *iter; | |
1510 | ||
867578cb KH |
1511 | /* |
1512 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1513 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1514 | */ |
c2b42d3c | 1515 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1516 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1517 | if (iter->under_oom > 0) |
1518 | iter->under_oom--; | |
1519 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1520 | } |
1521 | ||
867578cb KH |
1522 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1523 | ||
dc98df5a | 1524 | struct oom_wait_info { |
d79154bb | 1525 | struct mem_cgroup *memcg; |
ac6424b9 | 1526 | wait_queue_entry_t wait; |
dc98df5a KH |
1527 | }; |
1528 | ||
ac6424b9 | 1529 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1530 | unsigned mode, int sync, void *arg) |
1531 | { | |
d79154bb HD |
1532 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1533 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1534 | struct oom_wait_info *oom_wait_info; |
1535 | ||
1536 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1537 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1538 | |
2314b42d JW |
1539 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1540 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1541 | return 0; |
dc98df5a KH |
1542 | return autoremove_wake_function(wait, mode, sync, arg); |
1543 | } | |
1544 | ||
c0ff4b85 | 1545 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1546 | { |
c2b42d3c TH |
1547 | /* |
1548 | * For the following lockless ->under_oom test, the only required | |
1549 | * guarantee is that it must see the state asserted by an OOM when | |
1550 | * this function is called as a result of userland actions | |
1551 | * triggered by the notification of the OOM. This is trivially | |
1552 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1553 | * triggering notification. | |
1554 | */ | |
1555 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1556 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1557 | } |
1558 | ||
3812c8c8 | 1559 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1560 | { |
d0db7afa | 1561 | if (!current->memcg_may_oom) |
3812c8c8 | 1562 | return; |
867578cb | 1563 | /* |
49426420 JW |
1564 | * We are in the middle of the charge context here, so we |
1565 | * don't want to block when potentially sitting on a callstack | |
1566 | * that holds all kinds of filesystem and mm locks. | |
1567 | * | |
1568 | * Also, the caller may handle a failed allocation gracefully | |
1569 | * (like optional page cache readahead) and so an OOM killer | |
1570 | * invocation might not even be necessary. | |
1571 | * | |
1572 | * That's why we don't do anything here except remember the | |
1573 | * OOM context and then deal with it at the end of the page | |
1574 | * fault when the stack is unwound, the locks are released, | |
1575 | * and when we know whether the fault was overall successful. | |
867578cb | 1576 | */ |
49426420 | 1577 | css_get(&memcg->css); |
626ebc41 TH |
1578 | current->memcg_in_oom = memcg; |
1579 | current->memcg_oom_gfp_mask = mask; | |
1580 | current->memcg_oom_order = order; | |
3812c8c8 JW |
1581 | } |
1582 | ||
1583 | /** | |
1584 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1585 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1586 | * |
49426420 JW |
1587 | * This has to be called at the end of a page fault if the memcg OOM |
1588 | * handler was enabled. | |
3812c8c8 | 1589 | * |
49426420 | 1590 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1591 | * sleep on a waitqueue until the userspace task resolves the |
1592 | * situation. Sleeping directly in the charge context with all kinds | |
1593 | * of locks held is not a good idea, instead we remember an OOM state | |
1594 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1595 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1596 | * |
1597 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1598 | * completed, %false otherwise. |
3812c8c8 | 1599 | */ |
49426420 | 1600 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1601 | { |
626ebc41 | 1602 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1603 | struct oom_wait_info owait; |
49426420 | 1604 | bool locked; |
3812c8c8 JW |
1605 | |
1606 | /* OOM is global, do not handle */ | |
3812c8c8 | 1607 | if (!memcg) |
49426420 | 1608 | return false; |
3812c8c8 | 1609 | |
7c5f64f8 | 1610 | if (!handle) |
49426420 | 1611 | goto cleanup; |
3812c8c8 JW |
1612 | |
1613 | owait.memcg = memcg; | |
1614 | owait.wait.flags = 0; | |
1615 | owait.wait.func = memcg_oom_wake_function; | |
1616 | owait.wait.private = current; | |
2055da97 | 1617 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1618 | |
3812c8c8 | 1619 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1620 | mem_cgroup_mark_under_oom(memcg); |
1621 | ||
1622 | locked = mem_cgroup_oom_trylock(memcg); | |
1623 | ||
1624 | if (locked) | |
1625 | mem_cgroup_oom_notify(memcg); | |
1626 | ||
1627 | if (locked && !memcg->oom_kill_disable) { | |
1628 | mem_cgroup_unmark_under_oom(memcg); | |
1629 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1630 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1631 | current->memcg_oom_order); | |
49426420 | 1632 | } else { |
3812c8c8 | 1633 | schedule(); |
49426420 JW |
1634 | mem_cgroup_unmark_under_oom(memcg); |
1635 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1636 | } | |
1637 | ||
1638 | if (locked) { | |
fb2a6fc5 JW |
1639 | mem_cgroup_oom_unlock(memcg); |
1640 | /* | |
1641 | * There is no guarantee that an OOM-lock contender | |
1642 | * sees the wakeups triggered by the OOM kill | |
1643 | * uncharges. Wake any sleepers explicitely. | |
1644 | */ | |
1645 | memcg_oom_recover(memcg); | |
1646 | } | |
49426420 | 1647 | cleanup: |
626ebc41 | 1648 | current->memcg_in_oom = NULL; |
3812c8c8 | 1649 | css_put(&memcg->css); |
867578cb | 1650 | return true; |
0b7f569e KH |
1651 | } |
1652 | ||
d7365e78 | 1653 | /** |
81f8c3a4 JW |
1654 | * lock_page_memcg - lock a page->mem_cgroup binding |
1655 | * @page: the page | |
32047e2a | 1656 | * |
81f8c3a4 | 1657 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
1658 | * another cgroup. |
1659 | * | |
1660 | * It ensures lifetime of the returned memcg. Caller is responsible | |
1661 | * for the lifetime of the page; __unlock_page_memcg() is available | |
1662 | * when @page might get freed inside the locked section. | |
d69b042f | 1663 | */ |
739f79fc | 1664 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 KH |
1665 | { |
1666 | struct mem_cgroup *memcg; | |
6de22619 | 1667 | unsigned long flags; |
89c06bd5 | 1668 | |
6de22619 JW |
1669 | /* |
1670 | * The RCU lock is held throughout the transaction. The fast | |
1671 | * path can get away without acquiring the memcg->move_lock | |
1672 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
1673 | * |
1674 | * The RCU lock also protects the memcg from being freed when | |
1675 | * the page state that is going to change is the only thing | |
1676 | * preventing the page itself from being freed. E.g. writeback | |
1677 | * doesn't hold a page reference and relies on PG_writeback to | |
1678 | * keep off truncation, migration and so forth. | |
1679 | */ | |
d7365e78 JW |
1680 | rcu_read_lock(); |
1681 | ||
1682 | if (mem_cgroup_disabled()) | |
739f79fc | 1683 | return NULL; |
89c06bd5 | 1684 | again: |
1306a85a | 1685 | memcg = page->mem_cgroup; |
29833315 | 1686 | if (unlikely(!memcg)) |
739f79fc | 1687 | return NULL; |
d7365e78 | 1688 | |
bdcbb659 | 1689 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 1690 | return memcg; |
89c06bd5 | 1691 | |
6de22619 | 1692 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1693 | if (memcg != page->mem_cgroup) { |
6de22619 | 1694 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1695 | goto again; |
1696 | } | |
6de22619 JW |
1697 | |
1698 | /* | |
1699 | * When charge migration first begins, we can have locked and | |
1700 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 1701 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
1702 | */ |
1703 | memcg->move_lock_task = current; | |
1704 | memcg->move_lock_flags = flags; | |
d7365e78 | 1705 | |
739f79fc | 1706 | return memcg; |
89c06bd5 | 1707 | } |
81f8c3a4 | 1708 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 1709 | |
d7365e78 | 1710 | /** |
739f79fc JW |
1711 | * __unlock_page_memcg - unlock and unpin a memcg |
1712 | * @memcg: the memcg | |
1713 | * | |
1714 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 1715 | */ |
739f79fc | 1716 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 1717 | { |
6de22619 JW |
1718 | if (memcg && memcg->move_lock_task == current) { |
1719 | unsigned long flags = memcg->move_lock_flags; | |
1720 | ||
1721 | memcg->move_lock_task = NULL; | |
1722 | memcg->move_lock_flags = 0; | |
1723 | ||
1724 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1725 | } | |
89c06bd5 | 1726 | |
d7365e78 | 1727 | rcu_read_unlock(); |
89c06bd5 | 1728 | } |
739f79fc JW |
1729 | |
1730 | /** | |
1731 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
1732 | * @page: the page | |
1733 | */ | |
1734 | void unlock_page_memcg(struct page *page) | |
1735 | { | |
1736 | __unlock_page_memcg(page->mem_cgroup); | |
1737 | } | |
81f8c3a4 | 1738 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 1739 | |
cdec2e42 KH |
1740 | /* |
1741 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1742 | * TODO: maybe necessary to use big numbers in big irons. | |
1743 | */ | |
7ec99d62 | 1744 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1745 | struct memcg_stock_pcp { |
1746 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1747 | unsigned int nr_pages; |
cdec2e42 | 1748 | struct work_struct work; |
26fe6168 | 1749 | unsigned long flags; |
a0db00fc | 1750 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1751 | }; |
1752 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1753 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1754 | |
a0956d54 SS |
1755 | /** |
1756 | * consume_stock: Try to consume stocked charge on this cpu. | |
1757 | * @memcg: memcg to consume from. | |
1758 | * @nr_pages: how many pages to charge. | |
1759 | * | |
1760 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1761 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1762 | * service an allocation will refill the stock. | |
1763 | * | |
1764 | * returns true if successful, false otherwise. | |
cdec2e42 | 1765 | */ |
a0956d54 | 1766 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1767 | { |
1768 | struct memcg_stock_pcp *stock; | |
db2ba40c | 1769 | unsigned long flags; |
3e32cb2e | 1770 | bool ret = false; |
cdec2e42 | 1771 | |
a0956d54 | 1772 | if (nr_pages > CHARGE_BATCH) |
3e32cb2e | 1773 | return ret; |
a0956d54 | 1774 | |
db2ba40c JW |
1775 | local_irq_save(flags); |
1776 | ||
1777 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 1778 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1779 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1780 | ret = true; |
1781 | } | |
db2ba40c JW |
1782 | |
1783 | local_irq_restore(flags); | |
1784 | ||
cdec2e42 KH |
1785 | return ret; |
1786 | } | |
1787 | ||
1788 | /* | |
3e32cb2e | 1789 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1790 | */ |
1791 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1792 | { | |
1793 | struct mem_cgroup *old = stock->cached; | |
1794 | ||
11c9ea4e | 1795 | if (stock->nr_pages) { |
3e32cb2e | 1796 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 1797 | if (do_memsw_account()) |
3e32cb2e | 1798 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1799 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1800 | stock->nr_pages = 0; |
cdec2e42 KH |
1801 | } |
1802 | stock->cached = NULL; | |
cdec2e42 KH |
1803 | } |
1804 | ||
cdec2e42 KH |
1805 | static void drain_local_stock(struct work_struct *dummy) |
1806 | { | |
db2ba40c JW |
1807 | struct memcg_stock_pcp *stock; |
1808 | unsigned long flags; | |
1809 | ||
72f0184c MH |
1810 | /* |
1811 | * The only protection from memory hotplug vs. drain_stock races is | |
1812 | * that we always operate on local CPU stock here with IRQ disabled | |
1813 | */ | |
db2ba40c JW |
1814 | local_irq_save(flags); |
1815 | ||
1816 | stock = this_cpu_ptr(&memcg_stock); | |
cdec2e42 | 1817 | drain_stock(stock); |
26fe6168 | 1818 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
1819 | |
1820 | local_irq_restore(flags); | |
cdec2e42 KH |
1821 | } |
1822 | ||
1823 | /* | |
3e32cb2e | 1824 | * Cache charges(val) to local per_cpu area. |
320cc51d | 1825 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1826 | */ |
c0ff4b85 | 1827 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 1828 | { |
db2ba40c JW |
1829 | struct memcg_stock_pcp *stock; |
1830 | unsigned long flags; | |
1831 | ||
1832 | local_irq_save(flags); | |
cdec2e42 | 1833 | |
db2ba40c | 1834 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 1835 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1836 | drain_stock(stock); |
c0ff4b85 | 1837 | stock->cached = memcg; |
cdec2e42 | 1838 | } |
11c9ea4e | 1839 | stock->nr_pages += nr_pages; |
db2ba40c | 1840 | |
475d0487 RG |
1841 | if (stock->nr_pages > CHARGE_BATCH) |
1842 | drain_stock(stock); | |
1843 | ||
db2ba40c | 1844 | local_irq_restore(flags); |
cdec2e42 KH |
1845 | } |
1846 | ||
1847 | /* | |
c0ff4b85 | 1848 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 1849 | * of the hierarchy under it. |
cdec2e42 | 1850 | */ |
6d3d6aa2 | 1851 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 1852 | { |
26fe6168 | 1853 | int cpu, curcpu; |
d38144b7 | 1854 | |
6d3d6aa2 JW |
1855 | /* If someone's already draining, avoid adding running more workers. */ |
1856 | if (!mutex_trylock(&percpu_charge_mutex)) | |
1857 | return; | |
72f0184c MH |
1858 | /* |
1859 | * Notify other cpus that system-wide "drain" is running | |
1860 | * We do not care about races with the cpu hotplug because cpu down | |
1861 | * as well as workers from this path always operate on the local | |
1862 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
1863 | */ | |
5af12d0e | 1864 | curcpu = get_cpu(); |
cdec2e42 KH |
1865 | for_each_online_cpu(cpu) { |
1866 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1867 | struct mem_cgroup *memcg; |
26fe6168 | 1868 | |
c0ff4b85 | 1869 | memcg = stock->cached; |
72f0184c | 1870 | if (!memcg || !stock->nr_pages || !css_tryget(&memcg->css)) |
26fe6168 | 1871 | continue; |
72f0184c MH |
1872 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) { |
1873 | css_put(&memcg->css); | |
3e92041d | 1874 | continue; |
72f0184c | 1875 | } |
d1a05b69 MH |
1876 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
1877 | if (cpu == curcpu) | |
1878 | drain_local_stock(&stock->work); | |
1879 | else | |
1880 | schedule_work_on(cpu, &stock->work); | |
1881 | } | |
72f0184c | 1882 | css_put(&memcg->css); |
cdec2e42 | 1883 | } |
5af12d0e | 1884 | put_cpu(); |
9f50fad6 | 1885 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
1886 | } |
1887 | ||
308167fc | 1888 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 1889 | { |
cdec2e42 KH |
1890 | struct memcg_stock_pcp *stock; |
1891 | ||
cdec2e42 KH |
1892 | stock = &per_cpu(memcg_stock, cpu); |
1893 | drain_stock(stock); | |
308167fc | 1894 | return 0; |
cdec2e42 KH |
1895 | } |
1896 | ||
f7e1cb6e JW |
1897 | static void reclaim_high(struct mem_cgroup *memcg, |
1898 | unsigned int nr_pages, | |
1899 | gfp_t gfp_mask) | |
1900 | { | |
1901 | do { | |
1902 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
1903 | continue; | |
31176c78 | 1904 | mem_cgroup_event(memcg, MEMCG_HIGH); |
f7e1cb6e JW |
1905 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
1906 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1907 | } | |
1908 | ||
1909 | static void high_work_func(struct work_struct *work) | |
1910 | { | |
1911 | struct mem_cgroup *memcg; | |
1912 | ||
1913 | memcg = container_of(work, struct mem_cgroup, high_work); | |
1914 | reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL); | |
1915 | } | |
1916 | ||
b23afb93 TH |
1917 | /* |
1918 | * Scheduled by try_charge() to be executed from the userland return path | |
1919 | * and reclaims memory over the high limit. | |
1920 | */ | |
1921 | void mem_cgroup_handle_over_high(void) | |
1922 | { | |
1923 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 1924 | struct mem_cgroup *memcg; |
b23afb93 TH |
1925 | |
1926 | if (likely(!nr_pages)) | |
1927 | return; | |
1928 | ||
f7e1cb6e JW |
1929 | memcg = get_mem_cgroup_from_mm(current->mm); |
1930 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
1931 | css_put(&memcg->css); |
1932 | current->memcg_nr_pages_over_high = 0; | |
1933 | } | |
1934 | ||
00501b53 JW |
1935 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1936 | unsigned int nr_pages) | |
8a9f3ccd | 1937 | { |
7ec99d62 | 1938 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
9b130619 | 1939 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 1940 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 1941 | struct page_counter *counter; |
6539cc05 | 1942 | unsigned long nr_reclaimed; |
b70a2a21 JW |
1943 | bool may_swap = true; |
1944 | bool drained = false; | |
a636b327 | 1945 | |
ce00a967 | 1946 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 1947 | return 0; |
6539cc05 | 1948 | retry: |
b6b6cc72 | 1949 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 1950 | return 0; |
8a9f3ccd | 1951 | |
7941d214 | 1952 | if (!do_memsw_account() || |
6071ca52 JW |
1953 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
1954 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 1955 | goto done_restock; |
7941d214 | 1956 | if (do_memsw_account()) |
3e32cb2e JW |
1957 | page_counter_uncharge(&memcg->memsw, batch); |
1958 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 1959 | } else { |
3e32cb2e | 1960 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 1961 | may_swap = false; |
3fbe7244 | 1962 | } |
7a81b88c | 1963 | |
6539cc05 JW |
1964 | if (batch > nr_pages) { |
1965 | batch = nr_pages; | |
1966 | goto retry; | |
1967 | } | |
6d61ef40 | 1968 | |
115e02ed JW |
1969 | /* |
1970 | * Memcg doesn't have a dedicated reserve for atomic | |
1971 | * allocations. But like the global atomic pool, we need to | |
1972 | * put the burden of reclaim on regular allocation requests | |
1973 | * and let these go through as privileged allocations. | |
1974 | */ | |
1975 | if (gfp_mask & __GFP_ATOMIC) | |
1976 | goto force; | |
1977 | ||
06b078fc JW |
1978 | /* |
1979 | * Unlike in global OOM situations, memcg is not in a physical | |
1980 | * memory shortage. Allow dying and OOM-killed tasks to | |
1981 | * bypass the last charges so that they can exit quickly and | |
1982 | * free their memory. | |
1983 | */ | |
d9e9af93 | 1984 | if (unlikely(should_force_charge())) |
10d53c74 | 1985 | goto force; |
06b078fc | 1986 | |
89a28483 JW |
1987 | /* |
1988 | * Prevent unbounded recursion when reclaim operations need to | |
1989 | * allocate memory. This might exceed the limits temporarily, | |
1990 | * but we prefer facilitating memory reclaim and getting back | |
1991 | * under the limit over triggering OOM kills in these cases. | |
1992 | */ | |
1993 | if (unlikely(current->flags & PF_MEMALLOC)) | |
1994 | goto force; | |
1995 | ||
06b078fc JW |
1996 | if (unlikely(task_in_memcg_oom(current))) |
1997 | goto nomem; | |
1998 | ||
d0164adc | 1999 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2000 | goto nomem; |
4b534334 | 2001 | |
31176c78 | 2002 | mem_cgroup_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2003 | |
b70a2a21 JW |
2004 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2005 | gfp_mask, may_swap); | |
6539cc05 | 2006 | |
61e02c74 | 2007 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2008 | goto retry; |
28c34c29 | 2009 | |
b70a2a21 | 2010 | if (!drained) { |
6d3d6aa2 | 2011 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2012 | drained = true; |
2013 | goto retry; | |
2014 | } | |
2015 | ||
28c34c29 JW |
2016 | if (gfp_mask & __GFP_NORETRY) |
2017 | goto nomem; | |
6539cc05 JW |
2018 | /* |
2019 | * Even though the limit is exceeded at this point, reclaim | |
2020 | * may have been able to free some pages. Retry the charge | |
2021 | * before killing the task. | |
2022 | * | |
2023 | * Only for regular pages, though: huge pages are rather | |
2024 | * unlikely to succeed so close to the limit, and we fall back | |
2025 | * to regular pages anyway in case of failure. | |
2026 | */ | |
61e02c74 | 2027 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2028 | goto retry; |
2029 | /* | |
2030 | * At task move, charge accounts can be doubly counted. So, it's | |
2031 | * better to wait until the end of task_move if something is going on. | |
2032 | */ | |
2033 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2034 | goto retry; | |
2035 | ||
9b130619 JW |
2036 | if (nr_retries--) |
2037 | goto retry; | |
2038 | ||
06b078fc | 2039 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 2040 | goto force; |
06b078fc | 2041 | |
6539cc05 | 2042 | if (fatal_signal_pending(current)) |
10d53c74 | 2043 | goto force; |
6539cc05 | 2044 | |
31176c78 | 2045 | mem_cgroup_event(mem_over_limit, MEMCG_OOM); |
241994ed | 2046 | |
3608de07 JM |
2047 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
2048 | get_order(nr_pages * PAGE_SIZE)); | |
7a81b88c | 2049 | nomem: |
6d1fdc48 | 2050 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2051 | return -ENOMEM; |
10d53c74 TH |
2052 | force: |
2053 | /* | |
2054 | * The allocation either can't fail or will lead to more memory | |
2055 | * being freed very soon. Allow memory usage go over the limit | |
2056 | * temporarily by force charging it. | |
2057 | */ | |
2058 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2059 | if (do_memsw_account()) |
10d53c74 TH |
2060 | page_counter_charge(&memcg->memsw, nr_pages); |
2061 | css_get_many(&memcg->css, nr_pages); | |
2062 | ||
2063 | return 0; | |
6539cc05 JW |
2064 | |
2065 | done_restock: | |
e8ea14cc | 2066 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2067 | if (batch > nr_pages) |
2068 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2069 | |
241994ed | 2070 | /* |
b23afb93 TH |
2071 | * If the hierarchy is above the normal consumption range, schedule |
2072 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2073 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2074 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2075 | * not recorded as it most likely matches current's and won't | |
2076 | * change in the meantime. As high limit is checked again before | |
2077 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2078 | */ |
2079 | do { | |
b23afb93 | 2080 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2081 | /* Don't bother a random interrupted task */ |
2082 | if (in_interrupt()) { | |
2083 | schedule_work(&memcg->high_work); | |
2084 | break; | |
2085 | } | |
9516a18a | 2086 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2087 | set_notify_resume(current); |
2088 | break; | |
2089 | } | |
241994ed | 2090 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2091 | |
2092 | return 0; | |
7a81b88c | 2093 | } |
8a9f3ccd | 2094 | |
00501b53 | 2095 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2096 | { |
ce00a967 JW |
2097 | if (mem_cgroup_is_root(memcg)) |
2098 | return; | |
2099 | ||
3e32cb2e | 2100 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2101 | if (do_memsw_account()) |
3e32cb2e | 2102 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2103 | |
e8ea14cc | 2104 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2105 | } |
2106 | ||
0a31bc97 JW |
2107 | static void lock_page_lru(struct page *page, int *isolated) |
2108 | { | |
2109 | struct zone *zone = page_zone(page); | |
2110 | ||
a52633d8 | 2111 | spin_lock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2112 | if (PageLRU(page)) { |
2113 | struct lruvec *lruvec; | |
2114 | ||
599d0c95 | 2115 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2116 | ClearPageLRU(page); |
2117 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2118 | *isolated = 1; | |
2119 | } else | |
2120 | *isolated = 0; | |
2121 | } | |
2122 | ||
2123 | static void unlock_page_lru(struct page *page, int isolated) | |
2124 | { | |
2125 | struct zone *zone = page_zone(page); | |
2126 | ||
2127 | if (isolated) { | |
2128 | struct lruvec *lruvec; | |
2129 | ||
599d0c95 | 2130 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2131 | VM_BUG_ON_PAGE(PageLRU(page), page); |
2132 | SetPageLRU(page); | |
2133 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2134 | } | |
a52633d8 | 2135 | spin_unlock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2136 | } |
2137 | ||
00501b53 | 2138 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2139 | bool lrucare) |
7a81b88c | 2140 | { |
0a31bc97 | 2141 | int isolated; |
9ce70c02 | 2142 | |
1306a85a | 2143 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2144 | |
2145 | /* | |
2146 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2147 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2148 | */ | |
0a31bc97 JW |
2149 | if (lrucare) |
2150 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2151 | |
0a31bc97 JW |
2152 | /* |
2153 | * Nobody should be changing or seriously looking at | |
1306a85a | 2154 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2155 | * |
2156 | * - the page is uncharged | |
2157 | * | |
2158 | * - the page is off-LRU | |
2159 | * | |
2160 | * - an anonymous fault has exclusive page access, except for | |
2161 | * a locked page table | |
2162 | * | |
2163 | * - a page cache insertion, a swapin fault, or a migration | |
2164 | * have the page locked | |
2165 | */ | |
1306a85a | 2166 | page->mem_cgroup = memcg; |
9ce70c02 | 2167 | |
0a31bc97 JW |
2168 | if (lrucare) |
2169 | unlock_page_lru(page, isolated); | |
7a81b88c | 2170 | } |
66e1707b | 2171 | |
127424c8 | 2172 | #ifndef CONFIG_SLOB |
f3bb3043 | 2173 | static int memcg_alloc_cache_id(void) |
55007d84 | 2174 | { |
f3bb3043 VD |
2175 | int id, size; |
2176 | int err; | |
2177 | ||
dbcf73e2 | 2178 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2179 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2180 | if (id < 0) | |
2181 | return id; | |
55007d84 | 2182 | |
dbcf73e2 | 2183 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2184 | return id; |
2185 | ||
2186 | /* | |
2187 | * There's no space for the new id in memcg_caches arrays, | |
2188 | * so we have to grow them. | |
2189 | */ | |
05257a1a | 2190 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2191 | |
2192 | size = 2 * (id + 1); | |
55007d84 GC |
2193 | if (size < MEMCG_CACHES_MIN_SIZE) |
2194 | size = MEMCG_CACHES_MIN_SIZE; | |
2195 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2196 | size = MEMCG_CACHES_MAX_SIZE; | |
2197 | ||
f3bb3043 | 2198 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2199 | if (!err) |
2200 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2201 | if (!err) |
2202 | memcg_nr_cache_ids = size; | |
2203 | ||
2204 | up_write(&memcg_cache_ids_sem); | |
2205 | ||
f3bb3043 | 2206 | if (err) { |
dbcf73e2 | 2207 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2208 | return err; |
2209 | } | |
2210 | return id; | |
2211 | } | |
2212 | ||
2213 | static void memcg_free_cache_id(int id) | |
2214 | { | |
dbcf73e2 | 2215 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2216 | } |
2217 | ||
d5b3cf71 | 2218 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2219 | struct mem_cgroup *memcg; |
2220 | struct kmem_cache *cachep; | |
2221 | struct work_struct work; | |
2222 | }; | |
2223 | ||
d5b3cf71 | 2224 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2225 | { |
d5b3cf71 VD |
2226 | struct memcg_kmem_cache_create_work *cw = |
2227 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2228 | struct mem_cgroup *memcg = cw->memcg; |
2229 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2230 | |
d5b3cf71 | 2231 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2232 | |
5722d094 | 2233 | css_put(&memcg->css); |
d7f25f8a GC |
2234 | kfree(cw); |
2235 | } | |
2236 | ||
2237 | /* | |
2238 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2239 | */ |
d5b3cf71 VD |
2240 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2241 | struct kmem_cache *cachep) | |
d7f25f8a | 2242 | { |
d5b3cf71 | 2243 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2244 | |
8954327d | 2245 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT | __GFP_NOWARN); |
8135be5a | 2246 | if (!cw) |
d7f25f8a | 2247 | return; |
8135be5a VD |
2248 | |
2249 | css_get(&memcg->css); | |
d7f25f8a GC |
2250 | |
2251 | cw->memcg = memcg; | |
2252 | cw->cachep = cachep; | |
d5b3cf71 | 2253 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2254 | |
17cc4dfe | 2255 | queue_work(memcg_kmem_cache_wq, &cw->work); |
d7f25f8a GC |
2256 | } |
2257 | ||
d5b3cf71 VD |
2258 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2259 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2260 | { |
2261 | /* | |
2262 | * We need to stop accounting when we kmalloc, because if the | |
2263 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2264 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2265 | * |
2266 | * However, it is better to enclose the whole function. Depending on | |
2267 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2268 | * trigger an allocation. This too, will make us recurse. Because at | |
2269 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2270 | * the safest choice is to do it like this, wrapping the whole function. | |
2271 | */ | |
6f185c29 | 2272 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2273 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2274 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2275 | } |
c67a8a68 | 2276 | |
45264778 VD |
2277 | static inline bool memcg_kmem_bypass(void) |
2278 | { | |
2279 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
2280 | return true; | |
2281 | return false; | |
2282 | } | |
2283 | ||
2284 | /** | |
2285 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation | |
2286 | * @cachep: the original global kmem cache | |
2287 | * | |
d7f25f8a GC |
2288 | * Return the kmem_cache we're supposed to use for a slab allocation. |
2289 | * We try to use the current memcg's version of the cache. | |
2290 | * | |
45264778 VD |
2291 | * If the cache does not exist yet, if we are the first user of it, we |
2292 | * create it asynchronously in a workqueue and let the current allocation | |
2293 | * go through with the original cache. | |
d7f25f8a | 2294 | * |
45264778 VD |
2295 | * This function takes a reference to the cache it returns to assure it |
2296 | * won't get destroyed while we are working with it. Once the caller is | |
2297 | * done with it, memcg_kmem_put_cache() must be called to release the | |
2298 | * reference. | |
d7f25f8a | 2299 | */ |
45264778 | 2300 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2301 | { |
2302 | struct mem_cgroup *memcg; | |
959c8963 | 2303 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2304 | int kmemcg_id; |
d7f25f8a | 2305 | |
f7ce3190 | 2306 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2307 | |
45264778 | 2308 | if (memcg_kmem_bypass()) |
230e9fc2 VD |
2309 | return cachep; |
2310 | ||
9d100c5e | 2311 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2312 | return cachep; |
2313 | ||
8135be5a | 2314 | memcg = get_mem_cgroup_from_mm(current->mm); |
4db0c3c2 | 2315 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2316 | if (kmemcg_id < 0) |
ca0dde97 | 2317 | goto out; |
d7f25f8a | 2318 | |
2a4db7eb | 2319 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2320 | if (likely(memcg_cachep)) |
2321 | return memcg_cachep; | |
ca0dde97 LZ |
2322 | |
2323 | /* | |
2324 | * If we are in a safe context (can wait, and not in interrupt | |
2325 | * context), we could be be predictable and return right away. | |
2326 | * This would guarantee that the allocation being performed | |
2327 | * already belongs in the new cache. | |
2328 | * | |
2329 | * However, there are some clashes that can arrive from locking. | |
2330 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2331 | * memcg_create_kmem_cache, this means no further allocation |
2332 | * could happen with the slab_mutex held. So it's better to | |
2333 | * defer everything. | |
ca0dde97 | 2334 | */ |
d5b3cf71 | 2335 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2336 | out: |
8135be5a | 2337 | css_put(&memcg->css); |
ca0dde97 | 2338 | return cachep; |
d7f25f8a | 2339 | } |
d7f25f8a | 2340 | |
45264778 VD |
2341 | /** |
2342 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache | |
2343 | * @cachep: the cache returned by memcg_kmem_get_cache | |
2344 | */ | |
2345 | void memcg_kmem_put_cache(struct kmem_cache *cachep) | |
8135be5a VD |
2346 | { |
2347 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2348 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2349 | } |
2350 | ||
45264778 VD |
2351 | /** |
2352 | * memcg_kmem_charge: charge a kmem page | |
2353 | * @page: page to charge | |
2354 | * @gfp: reclaim mode | |
2355 | * @order: allocation order | |
2356 | * @memcg: memory cgroup to charge | |
2357 | * | |
2358 | * Returns 0 on success, an error code on failure. | |
2359 | */ | |
2360 | int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, | |
2361 | struct mem_cgroup *memcg) | |
7ae1e1d0 | 2362 | { |
f3ccb2c4 VD |
2363 | unsigned int nr_pages = 1 << order; |
2364 | struct page_counter *counter; | |
7ae1e1d0 GC |
2365 | int ret; |
2366 | ||
f3ccb2c4 | 2367 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2368 | if (ret) |
f3ccb2c4 | 2369 | return ret; |
52c29b04 JW |
2370 | |
2371 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2372 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
a6c47d49 MH |
2373 | |
2374 | /* | |
2375 | * Enforce __GFP_NOFAIL allocation because callers are not | |
2376 | * prepared to see failures and likely do not have any failure | |
2377 | * handling code. | |
2378 | */ | |
2379 | if (gfp & __GFP_NOFAIL) { | |
2380 | page_counter_charge(&memcg->kmem, nr_pages); | |
2381 | return 0; | |
2382 | } | |
52c29b04 JW |
2383 | cancel_charge(memcg, nr_pages); |
2384 | return -ENOMEM; | |
7ae1e1d0 GC |
2385 | } |
2386 | ||
f3ccb2c4 | 2387 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2388 | |
f3ccb2c4 | 2389 | return 0; |
7ae1e1d0 GC |
2390 | } |
2391 | ||
45264778 VD |
2392 | /** |
2393 | * memcg_kmem_charge: charge a kmem page to the current memory cgroup | |
2394 | * @page: page to charge | |
2395 | * @gfp: reclaim mode | |
2396 | * @order: allocation order | |
2397 | * | |
2398 | * Returns 0 on success, an error code on failure. | |
2399 | */ | |
2400 | int memcg_kmem_charge(struct page *page, gfp_t gfp, int order) | |
7ae1e1d0 | 2401 | { |
f3ccb2c4 | 2402 | struct mem_cgroup *memcg; |
fcff7d7e | 2403 | int ret = 0; |
7ae1e1d0 | 2404 | |
b1850bcd | 2405 | if (mem_cgroup_disabled() || memcg_kmem_bypass()) |
45264778 VD |
2406 | return 0; |
2407 | ||
f3ccb2c4 | 2408 | memcg = get_mem_cgroup_from_mm(current->mm); |
c4159a75 | 2409 | if (!mem_cgroup_is_root(memcg)) { |
45264778 | 2410 | ret = memcg_kmem_charge_memcg(page, gfp, order, memcg); |
c4159a75 VD |
2411 | if (!ret) |
2412 | __SetPageKmemcg(page); | |
2413 | } | |
7ae1e1d0 | 2414 | css_put(&memcg->css); |
d05e83a6 | 2415 | return ret; |
7ae1e1d0 | 2416 | } |
45264778 VD |
2417 | /** |
2418 | * memcg_kmem_uncharge: uncharge a kmem page | |
2419 | * @page: page to uncharge | |
2420 | * @order: allocation order | |
2421 | */ | |
2422 | void memcg_kmem_uncharge(struct page *page, int order) | |
7ae1e1d0 | 2423 | { |
1306a85a | 2424 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2425 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2426 | |
7ae1e1d0 GC |
2427 | if (!memcg) |
2428 | return; | |
2429 | ||
309381fe | 2430 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2431 | |
52c29b04 JW |
2432 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2433 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2434 | ||
f3ccb2c4 | 2435 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2436 | if (do_memsw_account()) |
f3ccb2c4 | 2437 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2438 | |
1306a85a | 2439 | page->mem_cgroup = NULL; |
c4159a75 VD |
2440 | |
2441 | /* slab pages do not have PageKmemcg flag set */ | |
2442 | if (PageKmemcg(page)) | |
2443 | __ClearPageKmemcg(page); | |
2444 | ||
f3ccb2c4 | 2445 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2446 | } |
127424c8 | 2447 | #endif /* !CONFIG_SLOB */ |
7ae1e1d0 | 2448 | |
ca3e0214 KH |
2449 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2450 | ||
ca3e0214 KH |
2451 | /* |
2452 | * Because tail pages are not marked as "used", set it. We're under | |
a52633d8 | 2453 | * zone_lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2454 | */ |
e94c8a9c | 2455 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2456 | { |
e94c8a9c | 2457 | int i; |
ca3e0214 | 2458 | |
3d37c4a9 KH |
2459 | if (mem_cgroup_disabled()) |
2460 | return; | |
b070e65c | 2461 | |
29833315 | 2462 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2463 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2464 | |
71cd3113 | 2465 | __this_cpu_sub(head->mem_cgroup->stat->count[MEMCG_RSS_HUGE], |
b070e65c | 2466 | HPAGE_PMD_NR); |
ca3e0214 | 2467 | } |
12d27107 | 2468 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2469 | |
c255a458 | 2470 | #ifdef CONFIG_MEMCG_SWAP |
0a31bc97 | 2471 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
38d8b4e6 | 2472 | int nr_entries) |
d13d1443 | 2473 | { |
38d8b4e6 | 2474 | this_cpu_add(memcg->stat->count[MEMCG_SWAP], nr_entries); |
d13d1443 | 2475 | } |
02491447 DN |
2476 | |
2477 | /** | |
2478 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2479 | * @entry: swap entry to be moved | |
2480 | * @from: mem_cgroup which the entry is moved from | |
2481 | * @to: mem_cgroup which the entry is moved to | |
2482 | * | |
2483 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2484 | * as the mem_cgroup's id of @from. | |
2485 | * | |
2486 | * Returns 0 on success, -EINVAL on failure. | |
2487 | * | |
3e32cb2e | 2488 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2489 | * both res and memsw, and called css_get(). |
2490 | */ | |
2491 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2492 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2493 | { |
2494 | unsigned short old_id, new_id; | |
2495 | ||
34c00c31 LZ |
2496 | old_id = mem_cgroup_id(from); |
2497 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2498 | |
2499 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
38d8b4e6 HY |
2500 | mem_cgroup_swap_statistics(from, -1); |
2501 | mem_cgroup_swap_statistics(to, 1); | |
02491447 DN |
2502 | return 0; |
2503 | } | |
2504 | return -EINVAL; | |
2505 | } | |
2506 | #else | |
2507 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2508 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2509 | { |
2510 | return -EINVAL; | |
2511 | } | |
8c7c6e34 | 2512 | #endif |
d13d1443 | 2513 | |
3e32cb2e | 2514 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2515 | |
d38d2a75 | 2516 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2517 | unsigned long limit) |
628f4235 | 2518 | { |
3e32cb2e JW |
2519 | unsigned long curusage; |
2520 | unsigned long oldusage; | |
2521 | bool enlarge = false; | |
81d39c20 | 2522 | int retry_count; |
3e32cb2e | 2523 | int ret; |
81d39c20 KH |
2524 | |
2525 | /* | |
2526 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2527 | * is depends on callers. We set our retry-count to be function | |
2528 | * of # of children which we should visit in this loop. | |
2529 | */ | |
3e32cb2e JW |
2530 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2531 | mem_cgroup_count_children(memcg); | |
81d39c20 | 2532 | |
3e32cb2e | 2533 | oldusage = page_counter_read(&memcg->memory); |
628f4235 | 2534 | |
3e32cb2e | 2535 | do { |
628f4235 KH |
2536 | if (signal_pending(current)) { |
2537 | ret = -EINTR; | |
2538 | break; | |
2539 | } | |
3e32cb2e JW |
2540 | |
2541 | mutex_lock(&memcg_limit_mutex); | |
2542 | if (limit > memcg->memsw.limit) { | |
2543 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2544 | ret = -EINVAL; |
628f4235 KH |
2545 | break; |
2546 | } | |
3e32cb2e JW |
2547 | if (limit > memcg->memory.limit) |
2548 | enlarge = true; | |
2549 | ret = page_counter_limit(&memcg->memory, limit); | |
2550 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2551 | |
2552 | if (!ret) | |
2553 | break; | |
2554 | ||
b70a2a21 JW |
2555 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
2556 | ||
3e32cb2e | 2557 | curusage = page_counter_read(&memcg->memory); |
81d39c20 | 2558 | /* Usage is reduced ? */ |
f894ffa8 | 2559 | if (curusage >= oldusage) |
81d39c20 KH |
2560 | retry_count--; |
2561 | else | |
2562 | oldusage = curusage; | |
3e32cb2e JW |
2563 | } while (retry_count); |
2564 | ||
3c11ecf4 KH |
2565 | if (!ret && enlarge) |
2566 | memcg_oom_recover(memcg); | |
14797e23 | 2567 | |
8c7c6e34 KH |
2568 | return ret; |
2569 | } | |
2570 | ||
338c8431 | 2571 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2572 | unsigned long limit) |
8c7c6e34 | 2573 | { |
3e32cb2e JW |
2574 | unsigned long curusage; |
2575 | unsigned long oldusage; | |
2576 | bool enlarge = false; | |
81d39c20 | 2577 | int retry_count; |
3e32cb2e | 2578 | int ret; |
8c7c6e34 | 2579 | |
81d39c20 | 2580 | /* see mem_cgroup_resize_res_limit */ |
3e32cb2e JW |
2581 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
2582 | mem_cgroup_count_children(memcg); | |
2583 | ||
2584 | oldusage = page_counter_read(&memcg->memsw); | |
2585 | ||
2586 | do { | |
8c7c6e34 KH |
2587 | if (signal_pending(current)) { |
2588 | ret = -EINTR; | |
2589 | break; | |
2590 | } | |
3e32cb2e JW |
2591 | |
2592 | mutex_lock(&memcg_limit_mutex); | |
2593 | if (limit < memcg->memory.limit) { | |
2594 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 | 2595 | ret = -EINVAL; |
8c7c6e34 KH |
2596 | break; |
2597 | } | |
3e32cb2e JW |
2598 | if (limit > memcg->memsw.limit) |
2599 | enlarge = true; | |
2600 | ret = page_counter_limit(&memcg->memsw, limit); | |
2601 | mutex_unlock(&memcg_limit_mutex); | |
8c7c6e34 KH |
2602 | |
2603 | if (!ret) | |
2604 | break; | |
2605 | ||
b70a2a21 JW |
2606 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
2607 | ||
3e32cb2e | 2608 | curusage = page_counter_read(&memcg->memsw); |
81d39c20 | 2609 | /* Usage is reduced ? */ |
8c7c6e34 | 2610 | if (curusage >= oldusage) |
628f4235 | 2611 | retry_count--; |
81d39c20 KH |
2612 | else |
2613 | oldusage = curusage; | |
3e32cb2e JW |
2614 | } while (retry_count); |
2615 | ||
3c11ecf4 KH |
2616 | if (!ret && enlarge) |
2617 | memcg_oom_recover(memcg); | |
3e32cb2e | 2618 | |
628f4235 KH |
2619 | return ret; |
2620 | } | |
2621 | ||
ef8f2327 | 2622 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
2623 | gfp_t gfp_mask, |
2624 | unsigned long *total_scanned) | |
2625 | { | |
2626 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 2627 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
2628 | unsigned long reclaimed; |
2629 | int loop = 0; | |
ef8f2327 | 2630 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 2631 | unsigned long excess; |
0608f43d AM |
2632 | unsigned long nr_scanned; |
2633 | ||
2634 | if (order > 0) | |
2635 | return 0; | |
2636 | ||
ef8f2327 | 2637 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
2638 | |
2639 | /* | |
2640 | * Do not even bother to check the largest node if the root | |
2641 | * is empty. Do it lockless to prevent lock bouncing. Races | |
2642 | * are acceptable as soft limit is best effort anyway. | |
2643 | */ | |
bfc7228b | 2644 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
2645 | return 0; |
2646 | ||
0608f43d AM |
2647 | /* |
2648 | * This loop can run a while, specially if mem_cgroup's continuously | |
2649 | * keep exceeding their soft limit and putting the system under | |
2650 | * pressure | |
2651 | */ | |
2652 | do { | |
2653 | if (next_mz) | |
2654 | mz = next_mz; | |
2655 | else | |
2656 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2657 | if (!mz) | |
2658 | break; | |
2659 | ||
2660 | nr_scanned = 0; | |
ef8f2327 | 2661 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
2662 | gfp_mask, &nr_scanned); |
2663 | nr_reclaimed += reclaimed; | |
2664 | *total_scanned += nr_scanned; | |
0a31bc97 | 2665 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2666 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2667 | |
2668 | /* | |
2669 | * If we failed to reclaim anything from this memory cgroup | |
2670 | * it is time to move on to the next cgroup | |
2671 | */ | |
2672 | next_mz = NULL; | |
bc2f2e7f VD |
2673 | if (!reclaimed) |
2674 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2675 | ||
3e32cb2e | 2676 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2677 | /* |
2678 | * One school of thought says that we should not add | |
2679 | * back the node to the tree if reclaim returns 0. | |
2680 | * But our reclaim could return 0, simply because due | |
2681 | * to priority we are exposing a smaller subset of | |
2682 | * memory to reclaim from. Consider this as a longer | |
2683 | * term TODO. | |
2684 | */ | |
2685 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2686 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2687 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2688 | css_put(&mz->memcg->css); |
2689 | loop++; | |
2690 | /* | |
2691 | * Could not reclaim anything and there are no more | |
2692 | * mem cgroups to try or we seem to be looping without | |
2693 | * reclaiming anything. | |
2694 | */ | |
2695 | if (!nr_reclaimed && | |
2696 | (next_mz == NULL || | |
2697 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2698 | break; | |
2699 | } while (!nr_reclaimed); | |
2700 | if (next_mz) | |
2701 | css_put(&next_mz->memcg->css); | |
2702 | return nr_reclaimed; | |
2703 | } | |
2704 | ||
ea280e7b TH |
2705 | /* |
2706 | * Test whether @memcg has children, dead or alive. Note that this | |
2707 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2708 | * returns %true if there are child csses according to the cgroup | |
2709 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2710 | */ | |
b5f99b53 GC |
2711 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2712 | { | |
ea280e7b TH |
2713 | bool ret; |
2714 | ||
ea280e7b TH |
2715 | rcu_read_lock(); |
2716 | ret = css_next_child(NULL, &memcg->css); | |
2717 | rcu_read_unlock(); | |
2718 | return ret; | |
b5f99b53 GC |
2719 | } |
2720 | ||
c26251f9 | 2721 | /* |
51038171 | 2722 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
2723 | * |
2724 | * Caller is responsible for holding css reference for memcg. | |
2725 | */ | |
2726 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2727 | { | |
2728 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2729 | |
c1e862c1 KH |
2730 | /* we call try-to-free pages for make this cgroup empty */ |
2731 | lru_add_drain_all(); | |
f817ed48 | 2732 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2733 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2734 | int progress; |
c1e862c1 | 2735 | |
c26251f9 MH |
2736 | if (signal_pending(current)) |
2737 | return -EINTR; | |
2738 | ||
b70a2a21 JW |
2739 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2740 | GFP_KERNEL, true); | |
c1e862c1 | 2741 | if (!progress) { |
f817ed48 | 2742 | nr_retries--; |
c1e862c1 | 2743 | /* maybe some writeback is necessary */ |
8aa7e847 | 2744 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2745 | } |
f817ed48 KH |
2746 | |
2747 | } | |
ab5196c2 MH |
2748 | |
2749 | return 0; | |
cc847582 KH |
2750 | } |
2751 | ||
6770c64e TH |
2752 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2753 | char *buf, size_t nbytes, | |
2754 | loff_t off) | |
c1e862c1 | 2755 | { |
6770c64e | 2756 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2757 | |
d8423011 MH |
2758 | if (mem_cgroup_is_root(memcg)) |
2759 | return -EINVAL; | |
6770c64e | 2760 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2761 | } |
2762 | ||
182446d0 TH |
2763 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2764 | struct cftype *cft) | |
18f59ea7 | 2765 | { |
182446d0 | 2766 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2767 | } |
2768 | ||
182446d0 TH |
2769 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2770 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2771 | { |
2772 | int retval = 0; | |
182446d0 | 2773 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2774 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2775 | |
567fb435 | 2776 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 2777 | return 0; |
567fb435 | 2778 | |
18f59ea7 | 2779 | /* |
af901ca1 | 2780 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2781 | * in the child subtrees. If it is unset, then the change can |
2782 | * occur, provided the current cgroup has no children. | |
2783 | * | |
2784 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2785 | * set if there are no children. | |
2786 | */ | |
c0ff4b85 | 2787 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2788 | (val == 1 || val == 0)) { |
ea280e7b | 2789 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2790 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2791 | else |
2792 | retval = -EBUSY; | |
2793 | } else | |
2794 | retval = -EINVAL; | |
567fb435 | 2795 | |
18f59ea7 BS |
2796 | return retval; |
2797 | } | |
2798 | ||
72b54e73 | 2799 | static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat) |
ce00a967 JW |
2800 | { |
2801 | struct mem_cgroup *iter; | |
72b54e73 | 2802 | int i; |
ce00a967 | 2803 | |
72b54e73 | 2804 | memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT); |
ce00a967 | 2805 | |
72b54e73 VD |
2806 | for_each_mem_cgroup_tree(iter, memcg) { |
2807 | for (i = 0; i < MEMCG_NR_STAT; i++) | |
ccda7f43 | 2808 | stat[i] += memcg_page_state(iter, i); |
72b54e73 | 2809 | } |
ce00a967 JW |
2810 | } |
2811 | ||
72b54e73 | 2812 | static void tree_events(struct mem_cgroup *memcg, unsigned long *events) |
587d9f72 JW |
2813 | { |
2814 | struct mem_cgroup *iter; | |
72b54e73 | 2815 | int i; |
587d9f72 | 2816 | |
72b54e73 | 2817 | memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS); |
587d9f72 | 2818 | |
72b54e73 VD |
2819 | for_each_mem_cgroup_tree(iter, memcg) { |
2820 | for (i = 0; i < MEMCG_NR_EVENTS; i++) | |
ccda7f43 | 2821 | events[i] += memcg_sum_events(iter, i); |
72b54e73 | 2822 | } |
587d9f72 JW |
2823 | } |
2824 | ||
6f646156 | 2825 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2826 | { |
72b54e73 | 2827 | unsigned long val = 0; |
ce00a967 | 2828 | |
3e32cb2e | 2829 | if (mem_cgroup_is_root(memcg)) { |
72b54e73 VD |
2830 | struct mem_cgroup *iter; |
2831 | ||
2832 | for_each_mem_cgroup_tree(iter, memcg) { | |
ccda7f43 JW |
2833 | val += memcg_page_state(iter, MEMCG_CACHE); |
2834 | val += memcg_page_state(iter, MEMCG_RSS); | |
72b54e73 | 2835 | if (swap) |
ccda7f43 | 2836 | val += memcg_page_state(iter, MEMCG_SWAP); |
72b54e73 | 2837 | } |
3e32cb2e | 2838 | } else { |
ce00a967 | 2839 | if (!swap) |
3e32cb2e | 2840 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2841 | else |
3e32cb2e | 2842 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2843 | } |
c12176d3 | 2844 | return val; |
ce00a967 JW |
2845 | } |
2846 | ||
3e32cb2e JW |
2847 | enum { |
2848 | RES_USAGE, | |
2849 | RES_LIMIT, | |
2850 | RES_MAX_USAGE, | |
2851 | RES_FAILCNT, | |
2852 | RES_SOFT_LIMIT, | |
2853 | }; | |
ce00a967 | 2854 | |
791badbd | 2855 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2856 | struct cftype *cft) |
8cdea7c0 | 2857 | { |
182446d0 | 2858 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2859 | struct page_counter *counter; |
af36f906 | 2860 | |
3e32cb2e | 2861 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 2862 | case _MEM: |
3e32cb2e JW |
2863 | counter = &memcg->memory; |
2864 | break; | |
8c7c6e34 | 2865 | case _MEMSWAP: |
3e32cb2e JW |
2866 | counter = &memcg->memsw; |
2867 | break; | |
510fc4e1 | 2868 | case _KMEM: |
3e32cb2e | 2869 | counter = &memcg->kmem; |
510fc4e1 | 2870 | break; |
d55f90bf | 2871 | case _TCP: |
0db15298 | 2872 | counter = &memcg->tcpmem; |
d55f90bf | 2873 | break; |
8c7c6e34 KH |
2874 | default: |
2875 | BUG(); | |
8c7c6e34 | 2876 | } |
3e32cb2e JW |
2877 | |
2878 | switch (MEMFILE_ATTR(cft->private)) { | |
2879 | case RES_USAGE: | |
2880 | if (counter == &memcg->memory) | |
c12176d3 | 2881 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 2882 | if (counter == &memcg->memsw) |
c12176d3 | 2883 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
2884 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
2885 | case RES_LIMIT: | |
2886 | return (u64)counter->limit * PAGE_SIZE; | |
2887 | case RES_MAX_USAGE: | |
2888 | return (u64)counter->watermark * PAGE_SIZE; | |
2889 | case RES_FAILCNT: | |
2890 | return counter->failcnt; | |
2891 | case RES_SOFT_LIMIT: | |
2892 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
2893 | default: | |
2894 | BUG(); | |
2895 | } | |
8cdea7c0 | 2896 | } |
510fc4e1 | 2897 | |
127424c8 | 2898 | #ifndef CONFIG_SLOB |
567e9ab2 | 2899 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 2900 | { |
d6441637 VD |
2901 | int memcg_id; |
2902 | ||
b313aeee VD |
2903 | if (cgroup_memory_nokmem) |
2904 | return 0; | |
2905 | ||
2a4db7eb | 2906 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 2907 | BUG_ON(memcg->kmem_state); |
d6441637 | 2908 | |
f3bb3043 | 2909 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
2910 | if (memcg_id < 0) |
2911 | return memcg_id; | |
d6441637 | 2912 | |
ef12947c | 2913 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 2914 | /* |
567e9ab2 | 2915 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 2916 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
2917 | * guarantee no one starts accounting before all call sites are |
2918 | * patched. | |
2919 | */ | |
900a38f0 | 2920 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 2921 | memcg->kmem_state = KMEM_ONLINE; |
bc2791f8 | 2922 | INIT_LIST_HEAD(&memcg->kmem_caches); |
0b8f73e1 JW |
2923 | |
2924 | return 0; | |
d6441637 VD |
2925 | } |
2926 | ||
8e0a8912 JW |
2927 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
2928 | { | |
2929 | struct cgroup_subsys_state *css; | |
2930 | struct mem_cgroup *parent, *child; | |
2931 | int kmemcg_id; | |
2932 | ||
2933 | if (memcg->kmem_state != KMEM_ONLINE) | |
2934 | return; | |
2935 | /* | |
2936 | * Clear the online state before clearing memcg_caches array | |
2937 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
2938 | * guarantees that no cache will be created for this cgroup | |
2939 | * after we are done (see memcg_create_kmem_cache()). | |
2940 | */ | |
2941 | memcg->kmem_state = KMEM_ALLOCATED; | |
2942 | ||
2943 | memcg_deactivate_kmem_caches(memcg); | |
2944 | ||
2945 | kmemcg_id = memcg->kmemcg_id; | |
2946 | BUG_ON(kmemcg_id < 0); | |
2947 | ||
2948 | parent = parent_mem_cgroup(memcg); | |
2949 | if (!parent) | |
2950 | parent = root_mem_cgroup; | |
2951 | ||
2952 | /* | |
2953 | * Change kmemcg_id of this cgroup and all its descendants to the | |
2954 | * parent's id, and then move all entries from this cgroup's list_lrus | |
2955 | * to ones of the parent. After we have finished, all list_lrus | |
2956 | * corresponding to this cgroup are guaranteed to remain empty. The | |
2957 | * ordering is imposed by list_lru_node->lock taken by | |
2958 | * memcg_drain_all_list_lrus(). | |
2959 | */ | |
3a06bb78 | 2960 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
2961 | css_for_each_descendant_pre(css, &memcg->css) { |
2962 | child = mem_cgroup_from_css(css); | |
2963 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
2964 | child->kmemcg_id = parent->kmemcg_id; | |
2965 | if (!memcg->use_hierarchy) | |
2966 | break; | |
2967 | } | |
3a06bb78 TH |
2968 | rcu_read_unlock(); |
2969 | ||
8e0a8912 JW |
2970 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); |
2971 | ||
2972 | memcg_free_cache_id(kmemcg_id); | |
2973 | } | |
2974 | ||
2975 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2976 | { | |
0b8f73e1 JW |
2977 | /* css_alloc() failed, offlining didn't happen */ |
2978 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
2979 | memcg_offline_kmem(memcg); | |
2980 | ||
8e0a8912 JW |
2981 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
2982 | memcg_destroy_kmem_caches(memcg); | |
2983 | static_branch_dec(&memcg_kmem_enabled_key); | |
2984 | WARN_ON(page_counter_read(&memcg->kmem)); | |
2985 | } | |
8e0a8912 | 2986 | } |
d6441637 | 2987 | #else |
0b8f73e1 | 2988 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
2989 | { |
2990 | return 0; | |
2991 | } | |
2992 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
2993 | { | |
2994 | } | |
2995 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2996 | { | |
2997 | } | |
2998 | #endif /* !CONFIG_SLOB */ | |
2999 | ||
d6441637 | 3000 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 3001 | unsigned long limit) |
d6441637 | 3002 | { |
b313aeee | 3003 | int ret; |
127424c8 JW |
3004 | |
3005 | mutex_lock(&memcg_limit_mutex); | |
127424c8 | 3006 | ret = page_counter_limit(&memcg->kmem, limit); |
127424c8 JW |
3007 | mutex_unlock(&memcg_limit_mutex); |
3008 | return ret; | |
d6441637 | 3009 | } |
510fc4e1 | 3010 | |
d55f90bf VD |
3011 | static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit) |
3012 | { | |
3013 | int ret; | |
3014 | ||
3015 | mutex_lock(&memcg_limit_mutex); | |
3016 | ||
0db15298 | 3017 | ret = page_counter_limit(&memcg->tcpmem, limit); |
d55f90bf VD |
3018 | if (ret) |
3019 | goto out; | |
3020 | ||
0db15298 | 3021 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3022 | /* |
3023 | * The active flag needs to be written after the static_key | |
3024 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3025 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3026 | * for details, and note that we don't mark any socket as | |
3027 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3028 | * |
3029 | * We need to do this, because static_keys will span multiple | |
3030 | * sites, but we can't control their order. If we mark a socket | |
3031 | * as accounted, but the accounting functions are not patched in | |
3032 | * yet, we'll lose accounting. | |
3033 | * | |
2d758073 | 3034 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3035 | * because when this value change, the code to process it is not |
3036 | * patched in yet. | |
3037 | */ | |
3038 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3039 | memcg->tcpmem_active = true; |
d55f90bf VD |
3040 | } |
3041 | out: | |
3042 | mutex_unlock(&memcg_limit_mutex); | |
3043 | return ret; | |
3044 | } | |
d55f90bf | 3045 | |
628f4235 KH |
3046 | /* |
3047 | * The user of this function is... | |
3048 | * RES_LIMIT. | |
3049 | */ | |
451af504 TH |
3050 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3051 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3052 | { |
451af504 | 3053 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3054 | unsigned long nr_pages; |
628f4235 KH |
3055 | int ret; |
3056 | ||
451af504 | 3057 | buf = strstrip(buf); |
650c5e56 | 3058 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3059 | if (ret) |
3060 | return ret; | |
af36f906 | 3061 | |
3e32cb2e | 3062 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3063 | case RES_LIMIT: |
4b3bde4c BS |
3064 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3065 | ret = -EINVAL; | |
3066 | break; | |
3067 | } | |
3e32cb2e JW |
3068 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3069 | case _MEM: | |
3070 | ret = mem_cgroup_resize_limit(memcg, nr_pages); | |
8c7c6e34 | 3071 | break; |
3e32cb2e JW |
3072 | case _MEMSWAP: |
3073 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); | |
296c81d8 | 3074 | break; |
3e32cb2e JW |
3075 | case _KMEM: |
3076 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
3077 | break; | |
d55f90bf VD |
3078 | case _TCP: |
3079 | ret = memcg_update_tcp_limit(memcg, nr_pages); | |
3080 | break; | |
3e32cb2e | 3081 | } |
296c81d8 | 3082 | break; |
3e32cb2e JW |
3083 | case RES_SOFT_LIMIT: |
3084 | memcg->soft_limit = nr_pages; | |
3085 | ret = 0; | |
628f4235 KH |
3086 | break; |
3087 | } | |
451af504 | 3088 | return ret ?: nbytes; |
8cdea7c0 BS |
3089 | } |
3090 | ||
6770c64e TH |
3091 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3092 | size_t nbytes, loff_t off) | |
c84872e1 | 3093 | { |
6770c64e | 3094 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3095 | struct page_counter *counter; |
c84872e1 | 3096 | |
3e32cb2e JW |
3097 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3098 | case _MEM: | |
3099 | counter = &memcg->memory; | |
3100 | break; | |
3101 | case _MEMSWAP: | |
3102 | counter = &memcg->memsw; | |
3103 | break; | |
3104 | case _KMEM: | |
3105 | counter = &memcg->kmem; | |
3106 | break; | |
d55f90bf | 3107 | case _TCP: |
0db15298 | 3108 | counter = &memcg->tcpmem; |
d55f90bf | 3109 | break; |
3e32cb2e JW |
3110 | default: |
3111 | BUG(); | |
3112 | } | |
af36f906 | 3113 | |
3e32cb2e | 3114 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3115 | case RES_MAX_USAGE: |
3e32cb2e | 3116 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3117 | break; |
3118 | case RES_FAILCNT: | |
3e32cb2e | 3119 | counter->failcnt = 0; |
29f2a4da | 3120 | break; |
3e32cb2e JW |
3121 | default: |
3122 | BUG(); | |
29f2a4da | 3123 | } |
f64c3f54 | 3124 | |
6770c64e | 3125 | return nbytes; |
c84872e1 PE |
3126 | } |
3127 | ||
182446d0 | 3128 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3129 | struct cftype *cft) |
3130 | { | |
182446d0 | 3131 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3132 | } |
3133 | ||
02491447 | 3134 | #ifdef CONFIG_MMU |
182446d0 | 3135 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3136 | struct cftype *cft, u64 val) |
3137 | { | |
182446d0 | 3138 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3139 | |
1dfab5ab | 3140 | if (val & ~MOVE_MASK) |
7dc74be0 | 3141 | return -EINVAL; |
ee5e8472 | 3142 | |
7dc74be0 | 3143 | /* |
ee5e8472 GC |
3144 | * No kind of locking is needed in here, because ->can_attach() will |
3145 | * check this value once in the beginning of the process, and then carry | |
3146 | * on with stale data. This means that changes to this value will only | |
3147 | * affect task migrations starting after the change. | |
7dc74be0 | 3148 | */ |
c0ff4b85 | 3149 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3150 | return 0; |
3151 | } | |
02491447 | 3152 | #else |
182446d0 | 3153 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3154 | struct cftype *cft, u64 val) |
3155 | { | |
3156 | return -ENOSYS; | |
3157 | } | |
3158 | #endif | |
7dc74be0 | 3159 | |
406eb0c9 | 3160 | #ifdef CONFIG_NUMA |
2da8ca82 | 3161 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3162 | { |
25485de6 GT |
3163 | struct numa_stat { |
3164 | const char *name; | |
3165 | unsigned int lru_mask; | |
3166 | }; | |
3167 | ||
3168 | static const struct numa_stat stats[] = { | |
3169 | { "total", LRU_ALL }, | |
3170 | { "file", LRU_ALL_FILE }, | |
3171 | { "anon", LRU_ALL_ANON }, | |
3172 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3173 | }; | |
3174 | const struct numa_stat *stat; | |
406eb0c9 | 3175 | int nid; |
25485de6 | 3176 | unsigned long nr; |
2da8ca82 | 3177 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3178 | |
25485de6 GT |
3179 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3180 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3181 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3182 | for_each_node_state(nid, N_MEMORY) { | |
3183 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3184 | stat->lru_mask); | |
3185 | seq_printf(m, " N%d=%lu", nid, nr); | |
3186 | } | |
3187 | seq_putc(m, '\n'); | |
406eb0c9 | 3188 | } |
406eb0c9 | 3189 | |
071aee13 YH |
3190 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3191 | struct mem_cgroup *iter; | |
3192 | ||
3193 | nr = 0; | |
3194 | for_each_mem_cgroup_tree(iter, memcg) | |
3195 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3196 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3197 | for_each_node_state(nid, N_MEMORY) { | |
3198 | nr = 0; | |
3199 | for_each_mem_cgroup_tree(iter, memcg) | |
3200 | nr += mem_cgroup_node_nr_lru_pages( | |
3201 | iter, nid, stat->lru_mask); | |
3202 | seq_printf(m, " N%d=%lu", nid, nr); | |
3203 | } | |
3204 | seq_putc(m, '\n'); | |
406eb0c9 | 3205 | } |
406eb0c9 | 3206 | |
406eb0c9 YH |
3207 | return 0; |
3208 | } | |
3209 | #endif /* CONFIG_NUMA */ | |
3210 | ||
df0e53d0 JW |
3211 | /* Universal VM events cgroup1 shows, original sort order */ |
3212 | unsigned int memcg1_events[] = { | |
3213 | PGPGIN, | |
3214 | PGPGOUT, | |
3215 | PGFAULT, | |
3216 | PGMAJFAULT, | |
3217 | }; | |
3218 | ||
3219 | static const char *const memcg1_event_names[] = { | |
3220 | "pgpgin", | |
3221 | "pgpgout", | |
3222 | "pgfault", | |
3223 | "pgmajfault", | |
3224 | }; | |
3225 | ||
2da8ca82 | 3226 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3227 | { |
2da8ca82 | 3228 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3229 | unsigned long memory, memsw; |
af7c4b0e JW |
3230 | struct mem_cgroup *mi; |
3231 | unsigned int i; | |
406eb0c9 | 3232 | |
71cd3113 | 3233 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c RS |
3234 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3235 | ||
71cd3113 JW |
3236 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
3237 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) | |
1dd3a273 | 3238 | continue; |
71cd3113 | 3239 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], |
ccda7f43 | 3240 | memcg_page_state(memcg, memcg1_stats[i]) * |
71cd3113 | 3241 | PAGE_SIZE); |
1dd3a273 | 3242 | } |
7b854121 | 3243 | |
df0e53d0 JW |
3244 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3245 | seq_printf(m, "%s %lu\n", memcg1_event_names[i], | |
ccda7f43 | 3246 | memcg_sum_events(memcg, memcg1_events[i])); |
af7c4b0e JW |
3247 | |
3248 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3249 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3250 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3251 | ||
14067bb3 | 3252 | /* Hierarchical information */ |
3e32cb2e JW |
3253 | memory = memsw = PAGE_COUNTER_MAX; |
3254 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3255 | memory = min(memory, mi->memory.limit); | |
3256 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3257 | } |
3e32cb2e JW |
3258 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3259 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3260 | if (do_memsw_account()) |
3e32cb2e JW |
3261 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3262 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3263 | |
71cd3113 | 3264 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
484ebb3b | 3265 | unsigned long long val = 0; |
af7c4b0e | 3266 | |
71cd3113 | 3267 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3268 | continue; |
af7c4b0e | 3269 | for_each_mem_cgroup_tree(mi, memcg) |
ccda7f43 | 3270 | val += memcg_page_state(mi, memcg1_stats[i]) * |
71cd3113 JW |
3271 | PAGE_SIZE; |
3272 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], val); | |
af7c4b0e JW |
3273 | } |
3274 | ||
df0e53d0 | 3275 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) { |
af7c4b0e JW |
3276 | unsigned long long val = 0; |
3277 | ||
3278 | for_each_mem_cgroup_tree(mi, memcg) | |
ccda7f43 | 3279 | val += memcg_sum_events(mi, memcg1_events[i]); |
df0e53d0 | 3280 | seq_printf(m, "total_%s %llu\n", memcg1_event_names[i], val); |
af7c4b0e JW |
3281 | } |
3282 | ||
3283 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3284 | unsigned long long val = 0; | |
3285 | ||
3286 | for_each_mem_cgroup_tree(mi, memcg) | |
3287 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3288 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3289 | } |
14067bb3 | 3290 | |
7f016ee8 | 3291 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 3292 | { |
ef8f2327 MG |
3293 | pg_data_t *pgdat; |
3294 | struct mem_cgroup_per_node *mz; | |
89abfab1 | 3295 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3296 | unsigned long recent_rotated[2] = {0, 0}; |
3297 | unsigned long recent_scanned[2] = {0, 0}; | |
3298 | ||
ef8f2327 MG |
3299 | for_each_online_pgdat(pgdat) { |
3300 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
3301 | rstat = &mz->lruvec.reclaim_stat; | |
7f016ee8 | 3302 | |
ef8f2327 MG |
3303 | recent_rotated[0] += rstat->recent_rotated[0]; |
3304 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3305 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3306 | recent_scanned[1] += rstat->recent_scanned[1]; | |
3307 | } | |
78ccf5b5 JW |
3308 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3309 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3310 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3311 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3312 | } |
3313 | #endif | |
3314 | ||
d2ceb9b7 KH |
3315 | return 0; |
3316 | } | |
3317 | ||
182446d0 TH |
3318 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3319 | struct cftype *cft) | |
a7885eb8 | 3320 | { |
182446d0 | 3321 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3322 | |
1f4c025b | 3323 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3324 | } |
3325 | ||
182446d0 TH |
3326 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3327 | struct cftype *cft, u64 val) | |
a7885eb8 | 3328 | { |
182446d0 | 3329 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3330 | |
3dae7fec | 3331 | if (val > 100) |
a7885eb8 KM |
3332 | return -EINVAL; |
3333 | ||
14208b0e | 3334 | if (css->parent) |
3dae7fec JW |
3335 | memcg->swappiness = val; |
3336 | else | |
3337 | vm_swappiness = val; | |
068b38c1 | 3338 | |
a7885eb8 KM |
3339 | return 0; |
3340 | } | |
3341 | ||
2e72b634 KS |
3342 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3343 | { | |
3344 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3345 | unsigned long usage; |
2e72b634 KS |
3346 | int i; |
3347 | ||
3348 | rcu_read_lock(); | |
3349 | if (!swap) | |
2c488db2 | 3350 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3351 | else |
2c488db2 | 3352 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3353 | |
3354 | if (!t) | |
3355 | goto unlock; | |
3356 | ||
ce00a967 | 3357 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3358 | |
3359 | /* | |
748dad36 | 3360 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3361 | * If it's not true, a threshold was crossed after last |
3362 | * call of __mem_cgroup_threshold(). | |
3363 | */ | |
5407a562 | 3364 | i = t->current_threshold; |
2e72b634 KS |
3365 | |
3366 | /* | |
3367 | * Iterate backward over array of thresholds starting from | |
3368 | * current_threshold and check if a threshold is crossed. | |
3369 | * If none of thresholds below usage is crossed, we read | |
3370 | * only one element of the array here. | |
3371 | */ | |
3372 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3373 | eventfd_signal(t->entries[i].eventfd, 1); | |
3374 | ||
3375 | /* i = current_threshold + 1 */ | |
3376 | i++; | |
3377 | ||
3378 | /* | |
3379 | * Iterate forward over array of thresholds starting from | |
3380 | * current_threshold+1 and check if a threshold is crossed. | |
3381 | * If none of thresholds above usage is crossed, we read | |
3382 | * only one element of the array here. | |
3383 | */ | |
3384 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3385 | eventfd_signal(t->entries[i].eventfd, 1); | |
3386 | ||
3387 | /* Update current_threshold */ | |
5407a562 | 3388 | t->current_threshold = i - 1; |
2e72b634 KS |
3389 | unlock: |
3390 | rcu_read_unlock(); | |
3391 | } | |
3392 | ||
3393 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3394 | { | |
ad4ca5f4 KS |
3395 | while (memcg) { |
3396 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3397 | if (do_memsw_account()) |
ad4ca5f4 KS |
3398 | __mem_cgroup_threshold(memcg, true); |
3399 | ||
3400 | memcg = parent_mem_cgroup(memcg); | |
3401 | } | |
2e72b634 KS |
3402 | } |
3403 | ||
3404 | static int compare_thresholds(const void *a, const void *b) | |
3405 | { | |
3406 | const struct mem_cgroup_threshold *_a = a; | |
3407 | const struct mem_cgroup_threshold *_b = b; | |
3408 | ||
2bff24a3 GT |
3409 | if (_a->threshold > _b->threshold) |
3410 | return 1; | |
3411 | ||
3412 | if (_a->threshold < _b->threshold) | |
3413 | return -1; | |
3414 | ||
3415 | return 0; | |
2e72b634 KS |
3416 | } |
3417 | ||
c0ff4b85 | 3418 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3419 | { |
3420 | struct mem_cgroup_eventfd_list *ev; | |
3421 | ||
2bcf2e92 MH |
3422 | spin_lock(&memcg_oom_lock); |
3423 | ||
c0ff4b85 | 3424 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3425 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3426 | |
3427 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3428 | return 0; |
3429 | } | |
3430 | ||
c0ff4b85 | 3431 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3432 | { |
7d74b06f KH |
3433 | struct mem_cgroup *iter; |
3434 | ||
c0ff4b85 | 3435 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3436 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3437 | } |
3438 | ||
59b6f873 | 3439 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3440 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3441 | { |
2c488db2 KS |
3442 | struct mem_cgroup_thresholds *thresholds; |
3443 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3444 | unsigned long threshold; |
3445 | unsigned long usage; | |
2c488db2 | 3446 | int i, size, ret; |
2e72b634 | 3447 | |
650c5e56 | 3448 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3449 | if (ret) |
3450 | return ret; | |
3451 | ||
3452 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3453 | |
05b84301 | 3454 | if (type == _MEM) { |
2c488db2 | 3455 | thresholds = &memcg->thresholds; |
ce00a967 | 3456 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3457 | } else if (type == _MEMSWAP) { |
2c488db2 | 3458 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3459 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3460 | } else |
2e72b634 KS |
3461 | BUG(); |
3462 | ||
2e72b634 | 3463 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3464 | if (thresholds->primary) |
2e72b634 KS |
3465 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3466 | ||
2c488db2 | 3467 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3468 | |
3469 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3470 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3471 | GFP_KERNEL); |
2c488db2 | 3472 | if (!new) { |
2e72b634 KS |
3473 | ret = -ENOMEM; |
3474 | goto unlock; | |
3475 | } | |
2c488db2 | 3476 | new->size = size; |
2e72b634 KS |
3477 | |
3478 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3479 | if (thresholds->primary) { |
3480 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3481 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3482 | } |
3483 | ||
2e72b634 | 3484 | /* Add new threshold */ |
2c488db2 KS |
3485 | new->entries[size - 1].eventfd = eventfd; |
3486 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3487 | |
3488 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3489 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3490 | compare_thresholds, NULL); |
3491 | ||
3492 | /* Find current threshold */ | |
2c488db2 | 3493 | new->current_threshold = -1; |
2e72b634 | 3494 | for (i = 0; i < size; i++) { |
748dad36 | 3495 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3496 | /* |
2c488db2 KS |
3497 | * new->current_threshold will not be used until |
3498 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3499 | * it here. |
3500 | */ | |
2c488db2 | 3501 | ++new->current_threshold; |
748dad36 SZ |
3502 | } else |
3503 | break; | |
2e72b634 KS |
3504 | } |
3505 | ||
2c488db2 KS |
3506 | /* Free old spare buffer and save old primary buffer as spare */ |
3507 | kfree(thresholds->spare); | |
3508 | thresholds->spare = thresholds->primary; | |
3509 | ||
3510 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3511 | |
907860ed | 3512 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3513 | synchronize_rcu(); |
3514 | ||
2e72b634 KS |
3515 | unlock: |
3516 | mutex_unlock(&memcg->thresholds_lock); | |
3517 | ||
3518 | return ret; | |
3519 | } | |
3520 | ||
59b6f873 | 3521 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3522 | struct eventfd_ctx *eventfd, const char *args) |
3523 | { | |
59b6f873 | 3524 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3525 | } |
3526 | ||
59b6f873 | 3527 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3528 | struct eventfd_ctx *eventfd, const char *args) |
3529 | { | |
59b6f873 | 3530 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3531 | } |
3532 | ||
59b6f873 | 3533 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3534 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3535 | { |
2c488db2 KS |
3536 | struct mem_cgroup_thresholds *thresholds; |
3537 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3538 | unsigned long usage; |
2c488db2 | 3539 | int i, j, size; |
2e72b634 KS |
3540 | |
3541 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3542 | |
3543 | if (type == _MEM) { | |
2c488db2 | 3544 | thresholds = &memcg->thresholds; |
ce00a967 | 3545 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3546 | } else if (type == _MEMSWAP) { |
2c488db2 | 3547 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3548 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3549 | } else |
2e72b634 KS |
3550 | BUG(); |
3551 | ||
371528ca AV |
3552 | if (!thresholds->primary) |
3553 | goto unlock; | |
3554 | ||
2e72b634 KS |
3555 | /* Check if a threshold crossed before removing */ |
3556 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3557 | ||
3558 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3559 | size = 0; |
3560 | for (i = 0; i < thresholds->primary->size; i++) { | |
3561 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3562 | size++; |
3563 | } | |
3564 | ||
2c488db2 | 3565 | new = thresholds->spare; |
907860ed | 3566 | |
2e72b634 KS |
3567 | /* Set thresholds array to NULL if we don't have thresholds */ |
3568 | if (!size) { | |
2c488db2 KS |
3569 | kfree(new); |
3570 | new = NULL; | |
907860ed | 3571 | goto swap_buffers; |
2e72b634 KS |
3572 | } |
3573 | ||
2c488db2 | 3574 | new->size = size; |
2e72b634 KS |
3575 | |
3576 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3577 | new->current_threshold = -1; |
3578 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3579 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3580 | continue; |
3581 | ||
2c488db2 | 3582 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3583 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3584 | /* |
2c488db2 | 3585 | * new->current_threshold will not be used |
2e72b634 KS |
3586 | * until rcu_assign_pointer(), so it's safe to increment |
3587 | * it here. | |
3588 | */ | |
2c488db2 | 3589 | ++new->current_threshold; |
2e72b634 KS |
3590 | } |
3591 | j++; | |
3592 | } | |
3593 | ||
907860ed | 3594 | swap_buffers: |
2c488db2 KS |
3595 | /* Swap primary and spare array */ |
3596 | thresholds->spare = thresholds->primary; | |
8c757763 | 3597 | |
2c488db2 | 3598 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3599 | |
907860ed | 3600 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3601 | synchronize_rcu(); |
6611d8d7 MC |
3602 | |
3603 | /* If all events are unregistered, free the spare array */ | |
3604 | if (!new) { | |
3605 | kfree(thresholds->spare); | |
3606 | thresholds->spare = NULL; | |
3607 | } | |
371528ca | 3608 | unlock: |
2e72b634 | 3609 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3610 | } |
c1e862c1 | 3611 | |
59b6f873 | 3612 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3613 | struct eventfd_ctx *eventfd) |
3614 | { | |
59b6f873 | 3615 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3616 | } |
3617 | ||
59b6f873 | 3618 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3619 | struct eventfd_ctx *eventfd) |
3620 | { | |
59b6f873 | 3621 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3622 | } |
3623 | ||
59b6f873 | 3624 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3625 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3626 | { |
9490ff27 | 3627 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3628 | |
9490ff27 KH |
3629 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3630 | if (!event) | |
3631 | return -ENOMEM; | |
3632 | ||
1af8efe9 | 3633 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3634 | |
3635 | event->eventfd = eventfd; | |
3636 | list_add(&event->list, &memcg->oom_notify); | |
3637 | ||
3638 | /* already in OOM ? */ | |
c2b42d3c | 3639 | if (memcg->under_oom) |
9490ff27 | 3640 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3641 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3642 | |
3643 | return 0; | |
3644 | } | |
3645 | ||
59b6f873 | 3646 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3647 | struct eventfd_ctx *eventfd) |
9490ff27 | 3648 | { |
9490ff27 | 3649 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3650 | |
1af8efe9 | 3651 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3652 | |
c0ff4b85 | 3653 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3654 | if (ev->eventfd == eventfd) { |
3655 | list_del(&ev->list); | |
3656 | kfree(ev); | |
3657 | } | |
3658 | } | |
3659 | ||
1af8efe9 | 3660 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3661 | } |
3662 | ||
2da8ca82 | 3663 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3664 | { |
2da8ca82 | 3665 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3666 | |
791badbd | 3667 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3668 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
8e675f7a | 3669 | seq_printf(sf, "oom_kill %lu\n", memcg_sum_events(memcg, OOM_KILL)); |
3c11ecf4 KH |
3670 | return 0; |
3671 | } | |
3672 | ||
182446d0 | 3673 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3674 | struct cftype *cft, u64 val) |
3675 | { | |
182446d0 | 3676 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3677 | |
3678 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3679 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3680 | return -EINVAL; |
3681 | ||
c0ff4b85 | 3682 | memcg->oom_kill_disable = val; |
4d845ebf | 3683 | if (!val) |
c0ff4b85 | 3684 | memcg_oom_recover(memcg); |
3dae7fec | 3685 | |
3c11ecf4 KH |
3686 | return 0; |
3687 | } | |
3688 | ||
52ebea74 TH |
3689 | #ifdef CONFIG_CGROUP_WRITEBACK |
3690 | ||
3691 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) | |
3692 | { | |
3693 | return &memcg->cgwb_list; | |
3694 | } | |
3695 | ||
841710aa TH |
3696 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3697 | { | |
3698 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3699 | } | |
3700 | ||
3701 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3702 | { | |
3703 | wb_domain_exit(&memcg->cgwb_domain); | |
3704 | } | |
3705 | ||
2529bb3a TH |
3706 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3707 | { | |
3708 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3709 | } | |
3710 | ||
841710aa TH |
3711 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3712 | { | |
3713 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3714 | ||
3715 | if (!memcg->css.parent) | |
3716 | return NULL; | |
3717 | ||
3718 | return &memcg->cgwb_domain; | |
3719 | } | |
3720 | ||
c2aa723a TH |
3721 | /** |
3722 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3723 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3724 | * @pfilepages: out parameter for number of file pages |
3725 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3726 | * @pdirty: out parameter for number of dirty pages |
3727 | * @pwriteback: out parameter for number of pages under writeback | |
3728 | * | |
c5edf9cd TH |
3729 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3730 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3731 | * is a bit more involved. | |
c2aa723a | 3732 | * |
c5edf9cd TH |
3733 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3734 | * headroom is calculated as the lowest headroom of itself and the | |
3735 | * ancestors. Note that this doesn't consider the actual amount of | |
3736 | * available memory in the system. The caller should further cap | |
3737 | * *@pheadroom accordingly. | |
c2aa723a | 3738 | */ |
c5edf9cd TH |
3739 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3740 | unsigned long *pheadroom, unsigned long *pdirty, | |
3741 | unsigned long *pwriteback) | |
c2aa723a TH |
3742 | { |
3743 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3744 | struct mem_cgroup *parent; | |
c2aa723a | 3745 | |
ccda7f43 | 3746 | *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a TH |
3747 | |
3748 | /* this should eventually include NR_UNSTABLE_NFS */ | |
ccda7f43 | 3749 | *pwriteback = memcg_page_state(memcg, NR_WRITEBACK); |
c5edf9cd TH |
3750 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3751 | (1 << LRU_ACTIVE_FILE)); | |
3752 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3753 | |
c2aa723a TH |
3754 | while ((parent = parent_mem_cgroup(memcg))) { |
3755 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); | |
3756 | unsigned long used = page_counter_read(&memcg->memory); | |
3757 | ||
c5edf9cd | 3758 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3759 | memcg = parent; |
3760 | } | |
c2aa723a TH |
3761 | } |
3762 | ||
841710aa TH |
3763 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3764 | ||
3765 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3766 | { | |
3767 | return 0; | |
3768 | } | |
3769 | ||
3770 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3771 | { | |
3772 | } | |
3773 | ||
2529bb3a TH |
3774 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3775 | { | |
3776 | } | |
3777 | ||
52ebea74 TH |
3778 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3779 | ||
3bc942f3 TH |
3780 | /* |
3781 | * DO NOT USE IN NEW FILES. | |
3782 | * | |
3783 | * "cgroup.event_control" implementation. | |
3784 | * | |
3785 | * This is way over-engineered. It tries to support fully configurable | |
3786 | * events for each user. Such level of flexibility is completely | |
3787 | * unnecessary especially in the light of the planned unified hierarchy. | |
3788 | * | |
3789 | * Please deprecate this and replace with something simpler if at all | |
3790 | * possible. | |
3791 | */ | |
3792 | ||
79bd9814 TH |
3793 | /* |
3794 | * Unregister event and free resources. | |
3795 | * | |
3796 | * Gets called from workqueue. | |
3797 | */ | |
3bc942f3 | 3798 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3799 | { |
3bc942f3 TH |
3800 | struct mem_cgroup_event *event = |
3801 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3802 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3803 | |
3804 | remove_wait_queue(event->wqh, &event->wait); | |
3805 | ||
59b6f873 | 3806 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3807 | |
3808 | /* Notify userspace the event is going away. */ | |
3809 | eventfd_signal(event->eventfd, 1); | |
3810 | ||
3811 | eventfd_ctx_put(event->eventfd); | |
3812 | kfree(event); | |
59b6f873 | 3813 | css_put(&memcg->css); |
79bd9814 TH |
3814 | } |
3815 | ||
3816 | /* | |
3817 | * Gets called on POLLHUP on eventfd when user closes it. | |
3818 | * | |
3819 | * Called with wqh->lock held and interrupts disabled. | |
3820 | */ | |
ac6424b9 | 3821 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 3822 | int sync, void *key) |
79bd9814 | 3823 | { |
3bc942f3 TH |
3824 | struct mem_cgroup_event *event = |
3825 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3826 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3827 | unsigned long flags = (unsigned long)key; |
3828 | ||
3829 | if (flags & POLLHUP) { | |
3830 | /* | |
3831 | * If the event has been detached at cgroup removal, we | |
3832 | * can simply return knowing the other side will cleanup | |
3833 | * for us. | |
3834 | * | |
3835 | * We can't race against event freeing since the other | |
3836 | * side will require wqh->lock via remove_wait_queue(), | |
3837 | * which we hold. | |
3838 | */ | |
fba94807 | 3839 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3840 | if (!list_empty(&event->list)) { |
3841 | list_del_init(&event->list); | |
3842 | /* | |
3843 | * We are in atomic context, but cgroup_event_remove() | |
3844 | * may sleep, so we have to call it in workqueue. | |
3845 | */ | |
3846 | schedule_work(&event->remove); | |
3847 | } | |
fba94807 | 3848 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3849 | } |
3850 | ||
3851 | return 0; | |
3852 | } | |
3853 | ||
3bc942f3 | 3854 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3855 | wait_queue_head_t *wqh, poll_table *pt) |
3856 | { | |
3bc942f3 TH |
3857 | struct mem_cgroup_event *event = |
3858 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3859 | |
3860 | event->wqh = wqh; | |
3861 | add_wait_queue(wqh, &event->wait); | |
3862 | } | |
3863 | ||
3864 | /* | |
3bc942f3 TH |
3865 | * DO NOT USE IN NEW FILES. |
3866 | * | |
79bd9814 TH |
3867 | * Parse input and register new cgroup event handler. |
3868 | * | |
3869 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3870 | * Interpretation of args is defined by control file implementation. | |
3871 | */ | |
451af504 TH |
3872 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
3873 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 3874 | { |
451af504 | 3875 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 3876 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 3877 | struct mem_cgroup_event *event; |
79bd9814 TH |
3878 | struct cgroup_subsys_state *cfile_css; |
3879 | unsigned int efd, cfd; | |
3880 | struct fd efile; | |
3881 | struct fd cfile; | |
fba94807 | 3882 | const char *name; |
79bd9814 TH |
3883 | char *endp; |
3884 | int ret; | |
3885 | ||
451af504 TH |
3886 | buf = strstrip(buf); |
3887 | ||
3888 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
3889 | if (*endp != ' ') |
3890 | return -EINVAL; | |
451af504 | 3891 | buf = endp + 1; |
79bd9814 | 3892 | |
451af504 | 3893 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
3894 | if ((*endp != ' ') && (*endp != '\0')) |
3895 | return -EINVAL; | |
451af504 | 3896 | buf = endp + 1; |
79bd9814 TH |
3897 | |
3898 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3899 | if (!event) | |
3900 | return -ENOMEM; | |
3901 | ||
59b6f873 | 3902 | event->memcg = memcg; |
79bd9814 | 3903 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
3904 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
3905 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
3906 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
3907 | |
3908 | efile = fdget(efd); | |
3909 | if (!efile.file) { | |
3910 | ret = -EBADF; | |
3911 | goto out_kfree; | |
3912 | } | |
3913 | ||
3914 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
3915 | if (IS_ERR(event->eventfd)) { | |
3916 | ret = PTR_ERR(event->eventfd); | |
3917 | goto out_put_efile; | |
3918 | } | |
3919 | ||
3920 | cfile = fdget(cfd); | |
3921 | if (!cfile.file) { | |
3922 | ret = -EBADF; | |
3923 | goto out_put_eventfd; | |
3924 | } | |
3925 | ||
3926 | /* the process need read permission on control file */ | |
3927 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
3928 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
3929 | if (ret < 0) | |
3930 | goto out_put_cfile; | |
3931 | ||
fba94807 TH |
3932 | /* |
3933 | * Determine the event callbacks and set them in @event. This used | |
3934 | * to be done via struct cftype but cgroup core no longer knows | |
3935 | * about these events. The following is crude but the whole thing | |
3936 | * is for compatibility anyway. | |
3bc942f3 TH |
3937 | * |
3938 | * DO NOT ADD NEW FILES. | |
fba94807 | 3939 | */ |
b583043e | 3940 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
3941 | |
3942 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
3943 | event->register_event = mem_cgroup_usage_register_event; | |
3944 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
3945 | } else if (!strcmp(name, "memory.oom_control")) { | |
3946 | event->register_event = mem_cgroup_oom_register_event; | |
3947 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
3948 | } else if (!strcmp(name, "memory.pressure_level")) { | |
3949 | event->register_event = vmpressure_register_event; | |
3950 | event->unregister_event = vmpressure_unregister_event; | |
3951 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
3952 | event->register_event = memsw_cgroup_usage_register_event; |
3953 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
3954 | } else { |
3955 | ret = -EINVAL; | |
3956 | goto out_put_cfile; | |
3957 | } | |
3958 | ||
79bd9814 | 3959 | /* |
b5557c4c TH |
3960 | * Verify @cfile should belong to @css. Also, remaining events are |
3961 | * automatically removed on cgroup destruction but the removal is | |
3962 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 3963 | */ |
b583043e | 3964 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 3965 | &memory_cgrp_subsys); |
79bd9814 | 3966 | ret = -EINVAL; |
5a17f543 | 3967 | if (IS_ERR(cfile_css)) |
79bd9814 | 3968 | goto out_put_cfile; |
5a17f543 TH |
3969 | if (cfile_css != css) { |
3970 | css_put(cfile_css); | |
79bd9814 | 3971 | goto out_put_cfile; |
5a17f543 | 3972 | } |
79bd9814 | 3973 | |
451af504 | 3974 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
3975 | if (ret) |
3976 | goto out_put_css; | |
3977 | ||
3978 | efile.file->f_op->poll(efile.file, &event->pt); | |
3979 | ||
fba94807 TH |
3980 | spin_lock(&memcg->event_list_lock); |
3981 | list_add(&event->list, &memcg->event_list); | |
3982 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
3983 | |
3984 | fdput(cfile); | |
3985 | fdput(efile); | |
3986 | ||
451af504 | 3987 | return nbytes; |
79bd9814 TH |
3988 | |
3989 | out_put_css: | |
b5557c4c | 3990 | css_put(css); |
79bd9814 TH |
3991 | out_put_cfile: |
3992 | fdput(cfile); | |
3993 | out_put_eventfd: | |
3994 | eventfd_ctx_put(event->eventfd); | |
3995 | out_put_efile: | |
3996 | fdput(efile); | |
3997 | out_kfree: | |
3998 | kfree(event); | |
3999 | ||
4000 | return ret; | |
4001 | } | |
4002 | ||
241994ed | 4003 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4004 | { |
0eea1030 | 4005 | .name = "usage_in_bytes", |
8c7c6e34 | 4006 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4007 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4008 | }, |
c84872e1 PE |
4009 | { |
4010 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4011 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4012 | .write = mem_cgroup_reset, |
791badbd | 4013 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4014 | }, |
8cdea7c0 | 4015 | { |
0eea1030 | 4016 | .name = "limit_in_bytes", |
8c7c6e34 | 4017 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4018 | .write = mem_cgroup_write, |
791badbd | 4019 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4020 | }, |
296c81d8 BS |
4021 | { |
4022 | .name = "soft_limit_in_bytes", | |
4023 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4024 | .write = mem_cgroup_write, |
791badbd | 4025 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4026 | }, |
8cdea7c0 BS |
4027 | { |
4028 | .name = "failcnt", | |
8c7c6e34 | 4029 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4030 | .write = mem_cgroup_reset, |
791badbd | 4031 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4032 | }, |
d2ceb9b7 KH |
4033 | { |
4034 | .name = "stat", | |
2da8ca82 | 4035 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4036 | }, |
c1e862c1 KH |
4037 | { |
4038 | .name = "force_empty", | |
6770c64e | 4039 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4040 | }, |
18f59ea7 BS |
4041 | { |
4042 | .name = "use_hierarchy", | |
4043 | .write_u64 = mem_cgroup_hierarchy_write, | |
4044 | .read_u64 = mem_cgroup_hierarchy_read, | |
4045 | }, | |
79bd9814 | 4046 | { |
3bc942f3 | 4047 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4048 | .write = memcg_write_event_control, |
7dbdb199 | 4049 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4050 | }, |
a7885eb8 KM |
4051 | { |
4052 | .name = "swappiness", | |
4053 | .read_u64 = mem_cgroup_swappiness_read, | |
4054 | .write_u64 = mem_cgroup_swappiness_write, | |
4055 | }, | |
7dc74be0 DN |
4056 | { |
4057 | .name = "move_charge_at_immigrate", | |
4058 | .read_u64 = mem_cgroup_move_charge_read, | |
4059 | .write_u64 = mem_cgroup_move_charge_write, | |
4060 | }, | |
9490ff27 KH |
4061 | { |
4062 | .name = "oom_control", | |
2da8ca82 | 4063 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4064 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4065 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4066 | }, | |
70ddf637 AV |
4067 | { |
4068 | .name = "pressure_level", | |
70ddf637 | 4069 | }, |
406eb0c9 YH |
4070 | #ifdef CONFIG_NUMA |
4071 | { | |
4072 | .name = "numa_stat", | |
2da8ca82 | 4073 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4074 | }, |
4075 | #endif | |
510fc4e1 GC |
4076 | { |
4077 | .name = "kmem.limit_in_bytes", | |
4078 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4079 | .write = mem_cgroup_write, |
791badbd | 4080 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4081 | }, |
4082 | { | |
4083 | .name = "kmem.usage_in_bytes", | |
4084 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4085 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4086 | }, |
4087 | { | |
4088 | .name = "kmem.failcnt", | |
4089 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4090 | .write = mem_cgroup_reset, |
791badbd | 4091 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4092 | }, |
4093 | { | |
4094 | .name = "kmem.max_usage_in_bytes", | |
4095 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4096 | .write = mem_cgroup_reset, |
791badbd | 4097 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4098 | }, |
5b365771 | 4099 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
749c5415 GC |
4100 | { |
4101 | .name = "kmem.slabinfo", | |
bc2791f8 TH |
4102 | .seq_start = memcg_slab_start, |
4103 | .seq_next = memcg_slab_next, | |
4104 | .seq_stop = memcg_slab_stop, | |
b047501c | 4105 | .seq_show = memcg_slab_show, |
749c5415 GC |
4106 | }, |
4107 | #endif | |
d55f90bf VD |
4108 | { |
4109 | .name = "kmem.tcp.limit_in_bytes", | |
4110 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4111 | .write = mem_cgroup_write, | |
4112 | .read_u64 = mem_cgroup_read_u64, | |
4113 | }, | |
4114 | { | |
4115 | .name = "kmem.tcp.usage_in_bytes", | |
4116 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4117 | .read_u64 = mem_cgroup_read_u64, | |
4118 | }, | |
4119 | { | |
4120 | .name = "kmem.tcp.failcnt", | |
4121 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4122 | .write = mem_cgroup_reset, | |
4123 | .read_u64 = mem_cgroup_read_u64, | |
4124 | }, | |
4125 | { | |
4126 | .name = "kmem.tcp.max_usage_in_bytes", | |
4127 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4128 | .write = mem_cgroup_reset, | |
4129 | .read_u64 = mem_cgroup_read_u64, | |
4130 | }, | |
6bc10349 | 4131 | { }, /* terminate */ |
af36f906 | 4132 | }; |
8c7c6e34 | 4133 | |
73f576c0 JW |
4134 | /* |
4135 | * Private memory cgroup IDR | |
4136 | * | |
4137 | * Swap-out records and page cache shadow entries need to store memcg | |
4138 | * references in constrained space, so we maintain an ID space that is | |
4139 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4140 | * memory-controlled cgroups to 64k. | |
4141 | * | |
4142 | * However, there usually are many references to the oflline CSS after | |
4143 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4144 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4145 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4146 | * relatively small ID space and prevent the creation of new cgroups | |
4147 | * even when there are much fewer than 64k cgroups - possibly none. | |
4148 | * | |
4149 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4150 | * be freed and recycled when it's no longer needed, which is usually | |
4151 | * when the CSS is offlined. | |
4152 | * | |
4153 | * The only exception to that are records of swapped out tmpfs/shmem | |
4154 | * pages that need to be attributed to live ancestors on swapin. But | |
4155 | * those references are manageable from userspace. | |
4156 | */ | |
4157 | ||
4158 | static DEFINE_IDR(mem_cgroup_idr); | |
4159 | ||
403a1579 KT |
4160 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
4161 | { | |
4162 | if (memcg->id.id > 0) { | |
4163 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
4164 | memcg->id.id = 0; | |
4165 | } | |
4166 | } | |
4167 | ||
615d66c3 | 4168 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4169 | { |
58fa2a55 | 4170 | VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0); |
615d66c3 | 4171 | atomic_add(n, &memcg->id.ref); |
73f576c0 JW |
4172 | } |
4173 | ||
615d66c3 | 4174 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4175 | { |
58fa2a55 | 4176 | VM_BUG_ON(atomic_read(&memcg->id.ref) < n); |
615d66c3 | 4177 | if (atomic_sub_and_test(n, &memcg->id.ref)) { |
403a1579 | 4178 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
4179 | |
4180 | /* Memcg ID pins CSS */ | |
4181 | css_put(&memcg->css); | |
4182 | } | |
4183 | } | |
4184 | ||
615d66c3 VD |
4185 | static inline void mem_cgroup_id_get(struct mem_cgroup *memcg) |
4186 | { | |
4187 | mem_cgroup_id_get_many(memcg, 1); | |
4188 | } | |
4189 | ||
4190 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) | |
4191 | { | |
4192 | mem_cgroup_id_put_many(memcg, 1); | |
4193 | } | |
4194 | ||
73f576c0 JW |
4195 | /** |
4196 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4197 | * @id: the memcg id to look up | |
4198 | * | |
4199 | * Caller must hold rcu_read_lock(). | |
4200 | */ | |
4201 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4202 | { | |
4203 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4204 | return idr_find(&mem_cgroup_idr, id); | |
4205 | } | |
4206 | ||
ef8f2327 | 4207 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4208 | { |
4209 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 4210 | int tmp = node; |
1ecaab2b KH |
4211 | /* |
4212 | * This routine is called against possible nodes. | |
4213 | * But it's BUG to call kmalloc() against offline node. | |
4214 | * | |
4215 | * TODO: this routine can waste much memory for nodes which will | |
4216 | * never be onlined. It's better to use memory hotplug callback | |
4217 | * function. | |
4218 | */ | |
41e3355d KH |
4219 | if (!node_state(node, N_NORMAL_MEMORY)) |
4220 | tmp = -1; | |
17295c88 | 4221 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4222 | if (!pn) |
4223 | return 1; | |
1ecaab2b | 4224 | |
00f3ca2c JW |
4225 | pn->lruvec_stat = alloc_percpu(struct lruvec_stat); |
4226 | if (!pn->lruvec_stat) { | |
4227 | kfree(pn); | |
4228 | return 1; | |
4229 | } | |
4230 | ||
ef8f2327 MG |
4231 | lruvec_init(&pn->lruvec); |
4232 | pn->usage_in_excess = 0; | |
4233 | pn->on_tree = false; | |
4234 | pn->memcg = memcg; | |
4235 | ||
54f72fe0 | 4236 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4237 | return 0; |
4238 | } | |
4239 | ||
ef8f2327 | 4240 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4241 | { |
00f3ca2c JW |
4242 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
4243 | ||
6d3d5ee1 MH |
4244 | if (!pn) |
4245 | return; | |
4246 | ||
00f3ca2c JW |
4247 | free_percpu(pn->lruvec_stat); |
4248 | kfree(pn); | |
1ecaab2b KH |
4249 | } |
4250 | ||
40e952f9 | 4251 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4252 | { |
c8b2a36f | 4253 | int node; |
59927fb9 | 4254 | |
c8b2a36f | 4255 | for_each_node(node) |
ef8f2327 | 4256 | free_mem_cgroup_per_node_info(memcg, node); |
c8b2a36f | 4257 | free_percpu(memcg->stat); |
8ff69e2c | 4258 | kfree(memcg); |
59927fb9 | 4259 | } |
3afe36b1 | 4260 | |
40e952f9 TE |
4261 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
4262 | { | |
4263 | memcg_wb_domain_exit(memcg); | |
4264 | __mem_cgroup_free(memcg); | |
4265 | } | |
4266 | ||
0b8f73e1 | 4267 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4268 | { |
d142e3e6 | 4269 | struct mem_cgroup *memcg; |
0b8f73e1 | 4270 | size_t size; |
6d12e2d8 | 4271 | int node; |
8cdea7c0 | 4272 | |
0b8f73e1 JW |
4273 | size = sizeof(struct mem_cgroup); |
4274 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4275 | ||
4276 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4277 | if (!memcg) |
0b8f73e1 JW |
4278 | return NULL; |
4279 | ||
73f576c0 JW |
4280 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
4281 | 1, MEM_CGROUP_ID_MAX, | |
4282 | GFP_KERNEL); | |
4283 | if (memcg->id.id < 0) | |
4284 | goto fail; | |
4285 | ||
0b8f73e1 JW |
4286 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
4287 | if (!memcg->stat) | |
4288 | goto fail; | |
78fb7466 | 4289 | |
3ed28fa1 | 4290 | for_each_node(node) |
ef8f2327 | 4291 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 4292 | goto fail; |
f64c3f54 | 4293 | |
0b8f73e1 JW |
4294 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4295 | goto fail; | |
28dbc4b6 | 4296 | |
f7e1cb6e | 4297 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4298 | memcg->last_scanned_node = MAX_NUMNODES; |
4299 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4300 | mutex_init(&memcg->thresholds_lock); |
4301 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4302 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4303 | INIT_LIST_HEAD(&memcg->event_list); |
4304 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4305 | memcg->socket_pressure = jiffies; |
127424c8 | 4306 | #ifndef CONFIG_SLOB |
900a38f0 | 4307 | memcg->kmemcg_id = -1; |
900a38f0 | 4308 | #endif |
52ebea74 TH |
4309 | #ifdef CONFIG_CGROUP_WRITEBACK |
4310 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4311 | #endif | |
73f576c0 | 4312 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
4313 | return memcg; |
4314 | fail: | |
403a1579 | 4315 | mem_cgroup_id_remove(memcg); |
40e952f9 | 4316 | __mem_cgroup_free(memcg); |
0b8f73e1 | 4317 | return NULL; |
d142e3e6 GC |
4318 | } |
4319 | ||
0b8f73e1 JW |
4320 | static struct cgroup_subsys_state * __ref |
4321 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4322 | { |
0b8f73e1 JW |
4323 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4324 | struct mem_cgroup *memcg; | |
4325 | long error = -ENOMEM; | |
d142e3e6 | 4326 | |
0b8f73e1 JW |
4327 | memcg = mem_cgroup_alloc(); |
4328 | if (!memcg) | |
4329 | return ERR_PTR(error); | |
d142e3e6 | 4330 | |
0b8f73e1 JW |
4331 | memcg->high = PAGE_COUNTER_MAX; |
4332 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4333 | if (parent) { | |
4334 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4335 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4336 | } | |
4337 | if (parent && parent->use_hierarchy) { | |
4338 | memcg->use_hierarchy = true; | |
3e32cb2e | 4339 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4340 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4341 | page_counter_init(&memcg->memsw, &parent->memsw); |
4342 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4343 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4344 | } else { |
3e32cb2e | 4345 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4346 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4347 | page_counter_init(&memcg->memsw, NULL); |
4348 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4349 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4350 | /* |
4351 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4352 | * much sense so let cgroup subsystem know about this | |
4353 | * unfortunate state in our controller. | |
4354 | */ | |
d142e3e6 | 4355 | if (parent != root_mem_cgroup) |
073219e9 | 4356 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4357 | } |
d6441637 | 4358 | |
0b8f73e1 JW |
4359 | /* The following stuff does not apply to the root */ |
4360 | if (!parent) { | |
4361 | root_mem_cgroup = memcg; | |
4362 | return &memcg->css; | |
4363 | } | |
4364 | ||
b313aeee | 4365 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
4366 | if (error) |
4367 | goto fail; | |
127424c8 | 4368 | |
f7e1cb6e | 4369 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4370 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4371 | |
0b8f73e1 JW |
4372 | return &memcg->css; |
4373 | fail: | |
403a1579 | 4374 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 4375 | mem_cgroup_free(memcg); |
ea3a9645 | 4376 | return ERR_PTR(-ENOMEM); |
0b8f73e1 JW |
4377 | } |
4378 | ||
73f576c0 | 4379 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 4380 | { |
58fa2a55 VD |
4381 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
4382 | ||
73f576c0 | 4383 | /* Online state pins memcg ID, memcg ID pins CSS */ |
58fa2a55 | 4384 | atomic_set(&memcg->id.ref, 1); |
73f576c0 | 4385 | css_get(css); |
2f7dd7a4 | 4386 | return 0; |
8cdea7c0 BS |
4387 | } |
4388 | ||
eb95419b | 4389 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4390 | { |
eb95419b | 4391 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4392 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4393 | |
4394 | /* | |
4395 | * Unregister events and notify userspace. | |
4396 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4397 | * directory to avoid race between userspace and kernelspace. | |
4398 | */ | |
fba94807 TH |
4399 | spin_lock(&memcg->event_list_lock); |
4400 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4401 | list_del_init(&event->list); |
4402 | schedule_work(&event->remove); | |
4403 | } | |
fba94807 | 4404 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4405 | |
63677c74 RG |
4406 | memcg->low = 0; |
4407 | ||
567e9ab2 | 4408 | memcg_offline_kmem(memcg); |
52ebea74 | 4409 | wb_memcg_offline(memcg); |
73f576c0 JW |
4410 | |
4411 | mem_cgroup_id_put(memcg); | |
df878fb0 KH |
4412 | } |
4413 | ||
6df38689 VD |
4414 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4415 | { | |
4416 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4417 | ||
4418 | invalidate_reclaim_iterators(memcg); | |
4419 | } | |
4420 | ||
eb95419b | 4421 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4422 | { |
eb95419b | 4423 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4424 | |
f7e1cb6e | 4425 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4426 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4427 | |
0db15298 | 4428 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4429 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4430 | |
0b8f73e1 JW |
4431 | vmpressure_cleanup(&memcg->vmpressure); |
4432 | cancel_work_sync(&memcg->high_work); | |
4433 | mem_cgroup_remove_from_trees(memcg); | |
d886f4e4 | 4434 | memcg_free_kmem(memcg); |
0b8f73e1 | 4435 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4436 | } |
4437 | ||
1ced953b TH |
4438 | /** |
4439 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4440 | * @css: the target css | |
4441 | * | |
4442 | * Reset the states of the mem_cgroup associated with @css. This is | |
4443 | * invoked when the userland requests disabling on the default hierarchy | |
4444 | * but the memcg is pinned through dependency. The memcg should stop | |
4445 | * applying policies and should revert to the vanilla state as it may be | |
4446 | * made visible again. | |
4447 | * | |
4448 | * The current implementation only resets the essential configurations. | |
4449 | * This needs to be expanded to cover all the visible parts. | |
4450 | */ | |
4451 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4452 | { | |
4453 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4454 | ||
d334c9bc VD |
4455 | page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX); |
4456 | page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX); | |
4457 | page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX); | |
4458 | page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX); | |
4459 | page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
241994ed JW |
4460 | memcg->low = 0; |
4461 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4462 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4463 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4464 | } |
4465 | ||
02491447 | 4466 | #ifdef CONFIG_MMU |
7dc74be0 | 4467 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4468 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4469 | { |
05b84301 | 4470 | int ret; |
9476db97 | 4471 | |
d0164adc MG |
4472 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4473 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4474 | if (!ret) { |
854ffa8d | 4475 | mc.precharge += count; |
854ffa8d DN |
4476 | return ret; |
4477 | } | |
9476db97 | 4478 | |
3674534b | 4479 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 4480 | while (count--) { |
3674534b | 4481 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 4482 | if (ret) |
38c5d72f | 4483 | return ret; |
854ffa8d | 4484 | mc.precharge++; |
9476db97 | 4485 | cond_resched(); |
854ffa8d | 4486 | } |
9476db97 | 4487 | return 0; |
4ffef5fe DN |
4488 | } |
4489 | ||
4ffef5fe DN |
4490 | union mc_target { |
4491 | struct page *page; | |
02491447 | 4492 | swp_entry_t ent; |
4ffef5fe DN |
4493 | }; |
4494 | ||
4ffef5fe | 4495 | enum mc_target_type { |
8d32ff84 | 4496 | MC_TARGET_NONE = 0, |
4ffef5fe | 4497 | MC_TARGET_PAGE, |
02491447 | 4498 | MC_TARGET_SWAP, |
c733a828 | 4499 | MC_TARGET_DEVICE, |
4ffef5fe DN |
4500 | }; |
4501 | ||
90254a65 DN |
4502 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4503 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4504 | { |
c733a828 | 4505 | struct page *page = _vm_normal_page(vma, addr, ptent, true); |
4ffef5fe | 4506 | |
90254a65 DN |
4507 | if (!page || !page_mapped(page)) |
4508 | return NULL; | |
4509 | if (PageAnon(page)) { | |
1dfab5ab | 4510 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4511 | return NULL; |
1dfab5ab JW |
4512 | } else { |
4513 | if (!(mc.flags & MOVE_FILE)) | |
4514 | return NULL; | |
4515 | } | |
90254a65 DN |
4516 | if (!get_page_unless_zero(page)) |
4517 | return NULL; | |
4518 | ||
4519 | return page; | |
4520 | } | |
4521 | ||
c733a828 | 4522 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 4523 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 4524 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 4525 | { |
90254a65 DN |
4526 | struct page *page = NULL; |
4527 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4528 | ||
1dfab5ab | 4529 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4530 | return NULL; |
c733a828 JG |
4531 | |
4532 | /* | |
4533 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
4534 | * a device and because they are not accessible by CPU they are store | |
4535 | * as special swap entry in the CPU page table. | |
4536 | */ | |
4537 | if (is_device_private_entry(ent)) { | |
4538 | page = device_private_entry_to_page(ent); | |
4539 | /* | |
4540 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
4541 | * a refcount of 1 when free (unlike normal page) | |
4542 | */ | |
4543 | if (!page_ref_add_unless(page, 1, 1)) | |
4544 | return NULL; | |
4545 | return page; | |
4546 | } | |
4547 | ||
4b91355e KH |
4548 | /* |
4549 | * Because lookup_swap_cache() updates some statistics counter, | |
4550 | * we call find_get_page() with swapper_space directly. | |
4551 | */ | |
f6ab1f7f | 4552 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
7941d214 | 4553 | if (do_memsw_account()) |
90254a65 DN |
4554 | entry->val = ent.val; |
4555 | ||
4556 | return page; | |
4557 | } | |
4b91355e KH |
4558 | #else |
4559 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 4560 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
4561 | { |
4562 | return NULL; | |
4563 | } | |
4564 | #endif | |
90254a65 | 4565 | |
87946a72 DN |
4566 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4567 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4568 | { | |
4569 | struct page *page = NULL; | |
87946a72 DN |
4570 | struct address_space *mapping; |
4571 | pgoff_t pgoff; | |
4572 | ||
4573 | if (!vma->vm_file) /* anonymous vma */ | |
4574 | return NULL; | |
1dfab5ab | 4575 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4576 | return NULL; |
4577 | ||
87946a72 | 4578 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4579 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4580 | |
4581 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4582 | #ifdef CONFIG_SWAP |
4583 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4584 | if (shmem_mapping(mapping)) { |
4585 | page = find_get_entry(mapping, pgoff); | |
4586 | if (radix_tree_exceptional_entry(page)) { | |
4587 | swp_entry_t swp = radix_to_swp_entry(page); | |
7941d214 | 4588 | if (do_memsw_account()) |
139b6a6f | 4589 | *entry = swp; |
f6ab1f7f HY |
4590 | page = find_get_page(swap_address_space(swp), |
4591 | swp_offset(swp)); | |
139b6a6f JW |
4592 | } |
4593 | } else | |
4594 | page = find_get_page(mapping, pgoff); | |
4595 | #else | |
4596 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4597 | #endif |
87946a72 DN |
4598 | return page; |
4599 | } | |
4600 | ||
b1b0deab CG |
4601 | /** |
4602 | * mem_cgroup_move_account - move account of the page | |
4603 | * @page: the page | |
25843c2b | 4604 | * @compound: charge the page as compound or small page |
b1b0deab CG |
4605 | * @from: mem_cgroup which the page is moved from. |
4606 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4607 | * | |
3ac808fd | 4608 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
4609 | * |
4610 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4611 | * from old cgroup. | |
4612 | */ | |
4613 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 4614 | bool compound, |
b1b0deab CG |
4615 | struct mem_cgroup *from, |
4616 | struct mem_cgroup *to) | |
4617 | { | |
4618 | unsigned long flags; | |
f627c2f5 | 4619 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 4620 | int ret; |
c4843a75 | 4621 | bool anon; |
b1b0deab CG |
4622 | |
4623 | VM_BUG_ON(from == to); | |
4624 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 4625 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
4626 | |
4627 | /* | |
6a93ca8f | 4628 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 4629 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 4630 | */ |
f627c2f5 | 4631 | ret = -EBUSY; |
b1b0deab CG |
4632 | if (!trylock_page(page)) |
4633 | goto out; | |
4634 | ||
4635 | ret = -EINVAL; | |
4636 | if (page->mem_cgroup != from) | |
4637 | goto out_unlock; | |
4638 | ||
c4843a75 GT |
4639 | anon = PageAnon(page); |
4640 | ||
b1b0deab CG |
4641 | spin_lock_irqsave(&from->move_lock, flags); |
4642 | ||
c4843a75 | 4643 | if (!anon && page_mapped(page)) { |
71cd3113 JW |
4644 | __this_cpu_sub(from->stat->count[NR_FILE_MAPPED], nr_pages); |
4645 | __this_cpu_add(to->stat->count[NR_FILE_MAPPED], nr_pages); | |
b1b0deab CG |
4646 | } |
4647 | ||
c4843a75 GT |
4648 | /* |
4649 | * move_lock grabbed above and caller set from->moving_account, so | |
ccda7f43 | 4650 | * mod_memcg_page_state will serialize updates to PageDirty. |
c4843a75 GT |
4651 | * So mapping should be stable for dirty pages. |
4652 | */ | |
4653 | if (!anon && PageDirty(page)) { | |
4654 | struct address_space *mapping = page_mapping(page); | |
4655 | ||
4656 | if (mapping_cap_account_dirty(mapping)) { | |
71cd3113 | 4657 | __this_cpu_sub(from->stat->count[NR_FILE_DIRTY], |
c4843a75 | 4658 | nr_pages); |
71cd3113 | 4659 | __this_cpu_add(to->stat->count[NR_FILE_DIRTY], |
c4843a75 GT |
4660 | nr_pages); |
4661 | } | |
4662 | } | |
4663 | ||
b1b0deab | 4664 | if (PageWriteback(page)) { |
71cd3113 JW |
4665 | __this_cpu_sub(from->stat->count[NR_WRITEBACK], nr_pages); |
4666 | __this_cpu_add(to->stat->count[NR_WRITEBACK], nr_pages); | |
b1b0deab CG |
4667 | } |
4668 | ||
4669 | /* | |
4670 | * It is safe to change page->mem_cgroup here because the page | |
4671 | * is referenced, charged, and isolated - we can't race with | |
4672 | * uncharging, charging, migration, or LRU putback. | |
4673 | */ | |
4674 | ||
4675 | /* caller should have done css_get */ | |
4676 | page->mem_cgroup = to; | |
4677 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4678 | ||
4679 | ret = 0; | |
4680 | ||
4681 | local_irq_disable(); | |
f627c2f5 | 4682 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 4683 | memcg_check_events(to, page); |
f627c2f5 | 4684 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
4685 | memcg_check_events(from, page); |
4686 | local_irq_enable(); | |
4687 | out_unlock: | |
4688 | unlock_page(page); | |
4689 | out: | |
4690 | return ret; | |
4691 | } | |
4692 | ||
7cf7806c LR |
4693 | /** |
4694 | * get_mctgt_type - get target type of moving charge | |
4695 | * @vma: the vma the pte to be checked belongs | |
4696 | * @addr: the address corresponding to the pte to be checked | |
4697 | * @ptent: the pte to be checked | |
4698 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
4699 | * | |
4700 | * Returns | |
4701 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4702 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4703 | * move charge. if @target is not NULL, the page is stored in target->page | |
4704 | * with extra refcnt got(Callers should handle it). | |
4705 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
4706 | * target for charge migration. if @target is not NULL, the entry is stored | |
4707 | * in target->ent. | |
df6ad698 JG |
4708 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PUBLIC |
4709 | * or MEMORY_DEVICE_PRIVATE (so ZONE_DEVICE page and thus not on the lru). | |
4710 | * For now we such page is charge like a regular page would be as for all | |
4711 | * intent and purposes it is just special memory taking the place of a | |
4712 | * regular page. | |
c733a828 JG |
4713 | * |
4714 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
4715 | * |
4716 | * Called with pte lock held. | |
4717 | */ | |
4718 | ||
8d32ff84 | 4719 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4720 | unsigned long addr, pte_t ptent, union mc_target *target) |
4721 | { | |
4722 | struct page *page = NULL; | |
8d32ff84 | 4723 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4724 | swp_entry_t ent = { .val = 0 }; |
4725 | ||
4726 | if (pte_present(ptent)) | |
4727 | page = mc_handle_present_pte(vma, addr, ptent); | |
4728 | else if (is_swap_pte(ptent)) | |
48406ef8 | 4729 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 4730 | else if (pte_none(ptent)) |
87946a72 | 4731 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4732 | |
4733 | if (!page && !ent.val) | |
8d32ff84 | 4734 | return ret; |
02491447 | 4735 | if (page) { |
02491447 | 4736 | /* |
0a31bc97 | 4737 | * Do only loose check w/o serialization. |
1306a85a | 4738 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4739 | * not under LRU exclusion. |
02491447 | 4740 | */ |
1306a85a | 4741 | if (page->mem_cgroup == mc.from) { |
02491447 | 4742 | ret = MC_TARGET_PAGE; |
df6ad698 JG |
4743 | if (is_device_private_page(page) || |
4744 | is_device_public_page(page)) | |
c733a828 | 4745 | ret = MC_TARGET_DEVICE; |
02491447 DN |
4746 | if (target) |
4747 | target->page = page; | |
4748 | } | |
4749 | if (!ret || !target) | |
4750 | put_page(page); | |
4751 | } | |
3e14a57b HY |
4752 | /* |
4753 | * There is a swap entry and a page doesn't exist or isn't charged. | |
4754 | * But we cannot move a tail-page in a THP. | |
4755 | */ | |
4756 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 4757 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4758 | ret = MC_TARGET_SWAP; |
4759 | if (target) | |
4760 | target->ent = ent; | |
4ffef5fe | 4761 | } |
4ffef5fe DN |
4762 | return ret; |
4763 | } | |
4764 | ||
12724850 NH |
4765 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4766 | /* | |
d6810d73 HY |
4767 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
4768 | * not support them for now. | |
12724850 NH |
4769 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
4770 | */ | |
4771 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4772 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4773 | { | |
4774 | struct page *page = NULL; | |
12724850 NH |
4775 | enum mc_target_type ret = MC_TARGET_NONE; |
4776 | ||
84c3fc4e ZY |
4777 | if (unlikely(is_swap_pmd(pmd))) { |
4778 | VM_BUG_ON(thp_migration_supported() && | |
4779 | !is_pmd_migration_entry(pmd)); | |
4780 | return ret; | |
4781 | } | |
12724850 | 4782 | page = pmd_page(pmd); |
309381fe | 4783 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4784 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4785 | return ret; |
1306a85a | 4786 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4787 | ret = MC_TARGET_PAGE; |
4788 | if (target) { | |
4789 | get_page(page); | |
4790 | target->page = page; | |
4791 | } | |
4792 | } | |
4793 | return ret; | |
4794 | } | |
4795 | #else | |
4796 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4797 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4798 | { | |
4799 | return MC_TARGET_NONE; | |
4800 | } | |
4801 | #endif | |
4802 | ||
4ffef5fe DN |
4803 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4804 | unsigned long addr, unsigned long end, | |
4805 | struct mm_walk *walk) | |
4806 | { | |
26bcd64a | 4807 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4808 | pte_t *pte; |
4809 | spinlock_t *ptl; | |
4810 | ||
b6ec57f4 KS |
4811 | ptl = pmd_trans_huge_lock(pmd, vma); |
4812 | if (ptl) { | |
c733a828 JG |
4813 | /* |
4814 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
4815 | * support transparent huge page with MEMORY_DEVICE_PUBLIC or | |
4816 | * MEMORY_DEVICE_PRIVATE but this might change. | |
4817 | */ | |
12724850 NH |
4818 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4819 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4820 | spin_unlock(ptl); |
1a5a9906 | 4821 | return 0; |
12724850 | 4822 | } |
03319327 | 4823 | |
45f83cef AA |
4824 | if (pmd_trans_unstable(pmd)) |
4825 | return 0; | |
4ffef5fe DN |
4826 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4827 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4828 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4829 | mc.precharge++; /* increment precharge temporarily */ |
4830 | pte_unmap_unlock(pte - 1, ptl); | |
4831 | cond_resched(); | |
4832 | ||
7dc74be0 DN |
4833 | return 0; |
4834 | } | |
4835 | ||
4ffef5fe DN |
4836 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4837 | { | |
4838 | unsigned long precharge; | |
4ffef5fe | 4839 | |
26bcd64a NH |
4840 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4841 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4842 | .mm = mm, | |
4843 | }; | |
dfe076b0 | 4844 | down_read(&mm->mmap_sem); |
0247f3f4 JM |
4845 | walk_page_range(0, mm->highest_vm_end, |
4846 | &mem_cgroup_count_precharge_walk); | |
dfe076b0 | 4847 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4848 | |
4849 | precharge = mc.precharge; | |
4850 | mc.precharge = 0; | |
4851 | ||
4852 | return precharge; | |
4853 | } | |
4854 | ||
4ffef5fe DN |
4855 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4856 | { | |
dfe076b0 DN |
4857 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4858 | ||
4859 | VM_BUG_ON(mc.moving_task); | |
4860 | mc.moving_task = current; | |
4861 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4862 | } |
4863 | ||
dfe076b0 DN |
4864 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4865 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4866 | { |
2bd9bb20 KH |
4867 | struct mem_cgroup *from = mc.from; |
4868 | struct mem_cgroup *to = mc.to; | |
4869 | ||
4ffef5fe | 4870 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4871 | if (mc.precharge) { |
00501b53 | 4872 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4873 | mc.precharge = 0; |
4874 | } | |
4875 | /* | |
4876 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4877 | * we must uncharge here. | |
4878 | */ | |
4879 | if (mc.moved_charge) { | |
00501b53 | 4880 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4881 | mc.moved_charge = 0; |
4ffef5fe | 4882 | } |
483c30b5 DN |
4883 | /* we must fixup refcnts and charges */ |
4884 | if (mc.moved_swap) { | |
483c30b5 | 4885 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4886 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4887 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4888 | |
615d66c3 VD |
4889 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
4890 | ||
05b84301 | 4891 | /* |
3e32cb2e JW |
4892 | * we charged both to->memory and to->memsw, so we |
4893 | * should uncharge to->memory. | |
05b84301 | 4894 | */ |
ce00a967 | 4895 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4896 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4897 | ||
615d66c3 VD |
4898 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
4899 | css_put_many(&mc.to->css, mc.moved_swap); | |
3e32cb2e | 4900 | |
483c30b5 DN |
4901 | mc.moved_swap = 0; |
4902 | } | |
dfe076b0 DN |
4903 | memcg_oom_recover(from); |
4904 | memcg_oom_recover(to); | |
4905 | wake_up_all(&mc.waitq); | |
4906 | } | |
4907 | ||
4908 | static void mem_cgroup_clear_mc(void) | |
4909 | { | |
264a0ae1 TH |
4910 | struct mm_struct *mm = mc.mm; |
4911 | ||
dfe076b0 DN |
4912 | /* |
4913 | * we must clear moving_task before waking up waiters at the end of | |
4914 | * task migration. | |
4915 | */ | |
4916 | mc.moving_task = NULL; | |
4917 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4918 | spin_lock(&mc.lock); |
4ffef5fe DN |
4919 | mc.from = NULL; |
4920 | mc.to = NULL; | |
264a0ae1 | 4921 | mc.mm = NULL; |
2bd9bb20 | 4922 | spin_unlock(&mc.lock); |
264a0ae1 TH |
4923 | |
4924 | mmput(mm); | |
4ffef5fe DN |
4925 | } |
4926 | ||
1f7dd3e5 | 4927 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4928 | { |
1f7dd3e5 | 4929 | struct cgroup_subsys_state *css; |
eed67d75 | 4930 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 4931 | struct mem_cgroup *from; |
4530eddb | 4932 | struct task_struct *leader, *p; |
9f2115f9 | 4933 | struct mm_struct *mm; |
1dfab5ab | 4934 | unsigned long move_flags; |
9f2115f9 | 4935 | int ret = 0; |
7dc74be0 | 4936 | |
1f7dd3e5 TH |
4937 | /* charge immigration isn't supported on the default hierarchy */ |
4938 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
4939 | return 0; |
4940 | ||
4530eddb TH |
4941 | /* |
4942 | * Multi-process migrations only happen on the default hierarchy | |
4943 | * where charge immigration is not used. Perform charge | |
4944 | * immigration if @tset contains a leader and whine if there are | |
4945 | * multiple. | |
4946 | */ | |
4947 | p = NULL; | |
1f7dd3e5 | 4948 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
4949 | WARN_ON_ONCE(p); |
4950 | p = leader; | |
1f7dd3e5 | 4951 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
4952 | } |
4953 | if (!p) | |
4954 | return 0; | |
4955 | ||
1f7dd3e5 TH |
4956 | /* |
4957 | * We are now commited to this value whatever it is. Changes in this | |
4958 | * tunable will only affect upcoming migrations, not the current one. | |
4959 | * So we need to save it, and keep it going. | |
4960 | */ | |
4961 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
4962 | if (!move_flags) | |
4963 | return 0; | |
4964 | ||
9f2115f9 TH |
4965 | from = mem_cgroup_from_task(p); |
4966 | ||
4967 | VM_BUG_ON(from == memcg); | |
4968 | ||
4969 | mm = get_task_mm(p); | |
4970 | if (!mm) | |
4971 | return 0; | |
4972 | /* We move charges only when we move a owner of the mm */ | |
4973 | if (mm->owner == p) { | |
4974 | VM_BUG_ON(mc.from); | |
4975 | VM_BUG_ON(mc.to); | |
4976 | VM_BUG_ON(mc.precharge); | |
4977 | VM_BUG_ON(mc.moved_charge); | |
4978 | VM_BUG_ON(mc.moved_swap); | |
4979 | ||
4980 | spin_lock(&mc.lock); | |
264a0ae1 | 4981 | mc.mm = mm; |
9f2115f9 TH |
4982 | mc.from = from; |
4983 | mc.to = memcg; | |
4984 | mc.flags = move_flags; | |
4985 | spin_unlock(&mc.lock); | |
4986 | /* We set mc.moving_task later */ | |
4987 | ||
4988 | ret = mem_cgroup_precharge_mc(mm); | |
4989 | if (ret) | |
4990 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
4991 | } else { |
4992 | mmput(mm); | |
7dc74be0 DN |
4993 | } |
4994 | return ret; | |
4995 | } | |
4996 | ||
1f7dd3e5 | 4997 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4998 | { |
4e2f245d JW |
4999 | if (mc.to) |
5000 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5001 | } |
5002 | ||
4ffef5fe DN |
5003 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5004 | unsigned long addr, unsigned long end, | |
5005 | struct mm_walk *walk) | |
7dc74be0 | 5006 | { |
4ffef5fe | 5007 | int ret = 0; |
26bcd64a | 5008 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5009 | pte_t *pte; |
5010 | spinlock_t *ptl; | |
12724850 NH |
5011 | enum mc_target_type target_type; |
5012 | union mc_target target; | |
5013 | struct page *page; | |
4ffef5fe | 5014 | |
b6ec57f4 KS |
5015 | ptl = pmd_trans_huge_lock(pmd, vma); |
5016 | if (ptl) { | |
62ade86a | 5017 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5018 | spin_unlock(ptl); |
12724850 NH |
5019 | return 0; |
5020 | } | |
5021 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5022 | if (target_type == MC_TARGET_PAGE) { | |
5023 | page = target.page; | |
5024 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5025 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5026 | mc.from, mc.to)) { |
12724850 NH |
5027 | mc.precharge -= HPAGE_PMD_NR; |
5028 | mc.moved_charge += HPAGE_PMD_NR; | |
5029 | } | |
5030 | putback_lru_page(page); | |
5031 | } | |
5032 | put_page(page); | |
c733a828 JG |
5033 | } else if (target_type == MC_TARGET_DEVICE) { |
5034 | page = target.page; | |
5035 | if (!mem_cgroup_move_account(page, true, | |
5036 | mc.from, mc.to)) { | |
5037 | mc.precharge -= HPAGE_PMD_NR; | |
5038 | mc.moved_charge += HPAGE_PMD_NR; | |
5039 | } | |
5040 | put_page(page); | |
12724850 | 5041 | } |
bf929152 | 5042 | spin_unlock(ptl); |
1a5a9906 | 5043 | return 0; |
12724850 NH |
5044 | } |
5045 | ||
45f83cef AA |
5046 | if (pmd_trans_unstable(pmd)) |
5047 | return 0; | |
4ffef5fe DN |
5048 | retry: |
5049 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5050 | for (; addr != end; addr += PAGE_SIZE) { | |
5051 | pte_t ptent = *(pte++); | |
c733a828 | 5052 | bool device = false; |
02491447 | 5053 | swp_entry_t ent; |
4ffef5fe DN |
5054 | |
5055 | if (!mc.precharge) | |
5056 | break; | |
5057 | ||
8d32ff84 | 5058 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
5059 | case MC_TARGET_DEVICE: |
5060 | device = true; | |
5061 | /* fall through */ | |
4ffef5fe DN |
5062 | case MC_TARGET_PAGE: |
5063 | page = target.page; | |
53f9263b KS |
5064 | /* |
5065 | * We can have a part of the split pmd here. Moving it | |
5066 | * can be done but it would be too convoluted so simply | |
5067 | * ignore such a partial THP and keep it in original | |
5068 | * memcg. There should be somebody mapping the head. | |
5069 | */ | |
5070 | if (PageTransCompound(page)) | |
5071 | goto put; | |
c733a828 | 5072 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 5073 | goto put; |
f627c2f5 KS |
5074 | if (!mem_cgroup_move_account(page, false, |
5075 | mc.from, mc.to)) { | |
4ffef5fe | 5076 | mc.precharge--; |
854ffa8d DN |
5077 | /* we uncharge from mc.from later. */ |
5078 | mc.moved_charge++; | |
4ffef5fe | 5079 | } |
c733a828 JG |
5080 | if (!device) |
5081 | putback_lru_page(page); | |
8d32ff84 | 5082 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5083 | put_page(page); |
5084 | break; | |
02491447 DN |
5085 | case MC_TARGET_SWAP: |
5086 | ent = target.ent; | |
e91cbb42 | 5087 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5088 | mc.precharge--; |
483c30b5 DN |
5089 | /* we fixup refcnts and charges later. */ |
5090 | mc.moved_swap++; | |
5091 | } | |
02491447 | 5092 | break; |
4ffef5fe DN |
5093 | default: |
5094 | break; | |
5095 | } | |
5096 | } | |
5097 | pte_unmap_unlock(pte - 1, ptl); | |
5098 | cond_resched(); | |
5099 | ||
5100 | if (addr != end) { | |
5101 | /* | |
5102 | * We have consumed all precharges we got in can_attach(). | |
5103 | * We try charge one by one, but don't do any additional | |
5104 | * charges to mc.to if we have failed in charge once in attach() | |
5105 | * phase. | |
5106 | */ | |
854ffa8d | 5107 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5108 | if (!ret) |
5109 | goto retry; | |
5110 | } | |
5111 | ||
5112 | return ret; | |
5113 | } | |
5114 | ||
264a0ae1 | 5115 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 5116 | { |
26bcd64a NH |
5117 | struct mm_walk mem_cgroup_move_charge_walk = { |
5118 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
264a0ae1 | 5119 | .mm = mc.mm, |
26bcd64a | 5120 | }; |
4ffef5fe DN |
5121 | |
5122 | lru_add_drain_all(); | |
312722cb | 5123 | /* |
81f8c3a4 JW |
5124 | * Signal lock_page_memcg() to take the memcg's move_lock |
5125 | * while we're moving its pages to another memcg. Then wait | |
5126 | * for already started RCU-only updates to finish. | |
312722cb JW |
5127 | */ |
5128 | atomic_inc(&mc.from->moving_account); | |
5129 | synchronize_rcu(); | |
dfe076b0 | 5130 | retry: |
264a0ae1 | 5131 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
5132 | /* |
5133 | * Someone who are holding the mmap_sem might be waiting in | |
5134 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5135 | * and retry. Because we cancel precharges, we might not be able | |
5136 | * to move enough charges, but moving charge is a best-effort | |
5137 | * feature anyway, so it wouldn't be a big problem. | |
5138 | */ | |
5139 | __mem_cgroup_clear_mc(); | |
5140 | cond_resched(); | |
5141 | goto retry; | |
5142 | } | |
26bcd64a NH |
5143 | /* |
5144 | * When we have consumed all precharges and failed in doing | |
5145 | * additional charge, the page walk just aborts. | |
5146 | */ | |
0247f3f4 JM |
5147 | walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk); |
5148 | ||
264a0ae1 | 5149 | up_read(&mc.mm->mmap_sem); |
312722cb | 5150 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5151 | } |
5152 | ||
264a0ae1 | 5153 | static void mem_cgroup_move_task(void) |
67e465a7 | 5154 | { |
264a0ae1 TH |
5155 | if (mc.to) { |
5156 | mem_cgroup_move_charge(); | |
a433658c | 5157 | mem_cgroup_clear_mc(); |
264a0ae1 | 5158 | } |
67e465a7 | 5159 | } |
5cfb80a7 | 5160 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5161 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5162 | { |
5163 | return 0; | |
5164 | } | |
1f7dd3e5 | 5165 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5166 | { |
5167 | } | |
264a0ae1 | 5168 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
5169 | { |
5170 | } | |
5171 | #endif | |
67e465a7 | 5172 | |
f00baae7 TH |
5173 | /* |
5174 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5175 | * to verify whether we're attached to the default hierarchy on each mount |
5176 | * attempt. | |
f00baae7 | 5177 | */ |
eb95419b | 5178 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5179 | { |
5180 | /* | |
aa6ec29b | 5181 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5182 | * guarantees that @root doesn't have any children, so turning it |
5183 | * on for the root memcg is enough. | |
5184 | */ | |
9e10a130 | 5185 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5186 | root_mem_cgroup->use_hierarchy = true; |
5187 | else | |
5188 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5189 | } |
5190 | ||
241994ed JW |
5191 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5192 | struct cftype *cft) | |
5193 | { | |
f5fc3c5d JW |
5194 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5195 | ||
5196 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5197 | } |
5198 | ||
5199 | static int memory_low_show(struct seq_file *m, void *v) | |
5200 | { | |
5201 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5202 | unsigned long low = READ_ONCE(memcg->low); |
241994ed JW |
5203 | |
5204 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 5205 | seq_puts(m, "max\n"); |
241994ed JW |
5206 | else |
5207 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5208 | ||
5209 | return 0; | |
5210 | } | |
5211 | ||
5212 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5213 | char *buf, size_t nbytes, loff_t off) | |
5214 | { | |
5215 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5216 | unsigned long low; | |
5217 | int err; | |
5218 | ||
5219 | buf = strstrip(buf); | |
d2973697 | 5220 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5221 | if (err) |
5222 | return err; | |
5223 | ||
5224 | memcg->low = low; | |
5225 | ||
5226 | return nbytes; | |
5227 | } | |
5228 | ||
5229 | static int memory_high_show(struct seq_file *m, void *v) | |
5230 | { | |
5231 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5232 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
5233 | |
5234 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5235 | seq_puts(m, "max\n"); |
241994ed JW |
5236 | else |
5237 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5238 | ||
5239 | return 0; | |
5240 | } | |
5241 | ||
5242 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5243 | char *buf, size_t nbytes, loff_t off) | |
5244 | { | |
5245 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 5246 | unsigned long nr_pages; |
241994ed JW |
5247 | unsigned long high; |
5248 | int err; | |
5249 | ||
5250 | buf = strstrip(buf); | |
d2973697 | 5251 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5252 | if (err) |
5253 | return err; | |
5254 | ||
5255 | memcg->high = high; | |
5256 | ||
588083bb JW |
5257 | nr_pages = page_counter_read(&memcg->memory); |
5258 | if (nr_pages > high) | |
5259 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5260 | GFP_KERNEL, true); | |
5261 | ||
2529bb3a | 5262 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5263 | return nbytes; |
5264 | } | |
5265 | ||
5266 | static int memory_max_show(struct seq_file *m, void *v) | |
5267 | { | |
5268 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5269 | unsigned long max = READ_ONCE(memcg->memory.limit); |
241994ed JW |
5270 | |
5271 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5272 | seq_puts(m, "max\n"); |
241994ed JW |
5273 | else |
5274 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5275 | ||
5276 | return 0; | |
5277 | } | |
5278 | ||
5279 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5280 | char *buf, size_t nbytes, loff_t off) | |
5281 | { | |
5282 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
5283 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5284 | bool drained = false; | |
241994ed JW |
5285 | unsigned long max; |
5286 | int err; | |
5287 | ||
5288 | buf = strstrip(buf); | |
d2973697 | 5289 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5290 | if (err) |
5291 | return err; | |
5292 | ||
b6e6edcf JW |
5293 | xchg(&memcg->memory.limit, max); |
5294 | ||
5295 | for (;;) { | |
5296 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5297 | ||
5298 | if (nr_pages <= max) | |
5299 | break; | |
5300 | ||
5301 | if (signal_pending(current)) { | |
5302 | err = -EINTR; | |
5303 | break; | |
5304 | } | |
5305 | ||
5306 | if (!drained) { | |
5307 | drain_all_stock(memcg); | |
5308 | drained = true; | |
5309 | continue; | |
5310 | } | |
5311 | ||
5312 | if (nr_reclaims) { | |
5313 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5314 | GFP_KERNEL, true)) | |
5315 | nr_reclaims--; | |
5316 | continue; | |
5317 | } | |
5318 | ||
31176c78 | 5319 | mem_cgroup_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
5320 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
5321 | break; | |
5322 | } | |
241994ed | 5323 | |
2529bb3a | 5324 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5325 | return nbytes; |
5326 | } | |
5327 | ||
5328 | static int memory_events_show(struct seq_file *m, void *v) | |
5329 | { | |
5330 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5331 | ||
ccda7f43 JW |
5332 | seq_printf(m, "low %lu\n", memcg_sum_events(memcg, MEMCG_LOW)); |
5333 | seq_printf(m, "high %lu\n", memcg_sum_events(memcg, MEMCG_HIGH)); | |
5334 | seq_printf(m, "max %lu\n", memcg_sum_events(memcg, MEMCG_MAX)); | |
5335 | seq_printf(m, "oom %lu\n", memcg_sum_events(memcg, MEMCG_OOM)); | |
8e675f7a | 5336 | seq_printf(m, "oom_kill %lu\n", memcg_sum_events(memcg, OOM_KILL)); |
241994ed JW |
5337 | |
5338 | return 0; | |
5339 | } | |
5340 | ||
587d9f72 JW |
5341 | static int memory_stat_show(struct seq_file *m, void *v) |
5342 | { | |
5343 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
72b54e73 VD |
5344 | unsigned long stat[MEMCG_NR_STAT]; |
5345 | unsigned long events[MEMCG_NR_EVENTS]; | |
587d9f72 JW |
5346 | int i; |
5347 | ||
5348 | /* | |
5349 | * Provide statistics on the state of the memory subsystem as | |
5350 | * well as cumulative event counters that show past behavior. | |
5351 | * | |
5352 | * This list is ordered following a combination of these gradients: | |
5353 | * 1) generic big picture -> specifics and details | |
5354 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
5355 | * | |
5356 | * Current memory state: | |
5357 | */ | |
5358 | ||
72b54e73 VD |
5359 | tree_stat(memcg, stat); |
5360 | tree_events(memcg, events); | |
5361 | ||
587d9f72 | 5362 | seq_printf(m, "anon %llu\n", |
71cd3113 | 5363 | (u64)stat[MEMCG_RSS] * PAGE_SIZE); |
587d9f72 | 5364 | seq_printf(m, "file %llu\n", |
71cd3113 | 5365 | (u64)stat[MEMCG_CACHE] * PAGE_SIZE); |
12580e4b | 5366 | seq_printf(m, "kernel_stack %llu\n", |
efdc9490 | 5367 | (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024); |
27ee57c9 | 5368 | seq_printf(m, "slab %llu\n", |
32049296 JW |
5369 | (u64)(stat[NR_SLAB_RECLAIMABLE] + |
5370 | stat[NR_SLAB_UNRECLAIMABLE]) * PAGE_SIZE); | |
b2807f07 | 5371 | seq_printf(m, "sock %llu\n", |
72b54e73 | 5372 | (u64)stat[MEMCG_SOCK] * PAGE_SIZE); |
587d9f72 | 5373 | |
9a4caf1e | 5374 | seq_printf(m, "shmem %llu\n", |
71cd3113 | 5375 | (u64)stat[NR_SHMEM] * PAGE_SIZE); |
587d9f72 | 5376 | seq_printf(m, "file_mapped %llu\n", |
71cd3113 | 5377 | (u64)stat[NR_FILE_MAPPED] * PAGE_SIZE); |
587d9f72 | 5378 | seq_printf(m, "file_dirty %llu\n", |
71cd3113 | 5379 | (u64)stat[NR_FILE_DIRTY] * PAGE_SIZE); |
587d9f72 | 5380 | seq_printf(m, "file_writeback %llu\n", |
71cd3113 | 5381 | (u64)stat[NR_WRITEBACK] * PAGE_SIZE); |
587d9f72 JW |
5382 | |
5383 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5384 | struct mem_cgroup *mi; | |
5385 | unsigned long val = 0; | |
5386 | ||
5387 | for_each_mem_cgroup_tree(mi, memcg) | |
5388 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)); | |
5389 | seq_printf(m, "%s %llu\n", | |
5390 | mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE); | |
5391 | } | |
5392 | ||
27ee57c9 | 5393 | seq_printf(m, "slab_reclaimable %llu\n", |
32049296 | 5394 | (u64)stat[NR_SLAB_RECLAIMABLE] * PAGE_SIZE); |
27ee57c9 | 5395 | seq_printf(m, "slab_unreclaimable %llu\n", |
32049296 | 5396 | (u64)stat[NR_SLAB_UNRECLAIMABLE] * PAGE_SIZE); |
27ee57c9 | 5397 | |
587d9f72 JW |
5398 | /* Accumulated memory events */ |
5399 | ||
df0e53d0 JW |
5400 | seq_printf(m, "pgfault %lu\n", events[PGFAULT]); |
5401 | seq_printf(m, "pgmajfault %lu\n", events[PGMAJFAULT]); | |
587d9f72 | 5402 | |
2262185c RG |
5403 | seq_printf(m, "pgrefill %lu\n", events[PGREFILL]); |
5404 | seq_printf(m, "pgscan %lu\n", events[PGSCAN_KSWAPD] + | |
5405 | events[PGSCAN_DIRECT]); | |
5406 | seq_printf(m, "pgsteal %lu\n", events[PGSTEAL_KSWAPD] + | |
5407 | events[PGSTEAL_DIRECT]); | |
5408 | seq_printf(m, "pgactivate %lu\n", events[PGACTIVATE]); | |
5409 | seq_printf(m, "pgdeactivate %lu\n", events[PGDEACTIVATE]); | |
5410 | seq_printf(m, "pglazyfree %lu\n", events[PGLAZYFREE]); | |
5411 | seq_printf(m, "pglazyfreed %lu\n", events[PGLAZYFREED]); | |
5412 | ||
2a2e4885 | 5413 | seq_printf(m, "workingset_refault %lu\n", |
71cd3113 | 5414 | stat[WORKINGSET_REFAULT]); |
2a2e4885 | 5415 | seq_printf(m, "workingset_activate %lu\n", |
71cd3113 | 5416 | stat[WORKINGSET_ACTIVATE]); |
2a2e4885 | 5417 | seq_printf(m, "workingset_nodereclaim %lu\n", |
71cd3113 | 5418 | stat[WORKINGSET_NODERECLAIM]); |
2a2e4885 | 5419 | |
587d9f72 JW |
5420 | return 0; |
5421 | } | |
5422 | ||
241994ed JW |
5423 | static struct cftype memory_files[] = { |
5424 | { | |
5425 | .name = "current", | |
f5fc3c5d | 5426 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5427 | .read_u64 = memory_current_read, |
5428 | }, | |
5429 | { | |
5430 | .name = "low", | |
5431 | .flags = CFTYPE_NOT_ON_ROOT, | |
5432 | .seq_show = memory_low_show, | |
5433 | .write = memory_low_write, | |
5434 | }, | |
5435 | { | |
5436 | .name = "high", | |
5437 | .flags = CFTYPE_NOT_ON_ROOT, | |
5438 | .seq_show = memory_high_show, | |
5439 | .write = memory_high_write, | |
5440 | }, | |
5441 | { | |
5442 | .name = "max", | |
5443 | .flags = CFTYPE_NOT_ON_ROOT, | |
5444 | .seq_show = memory_max_show, | |
5445 | .write = memory_max_write, | |
5446 | }, | |
5447 | { | |
5448 | .name = "events", | |
5449 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5450 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5451 | .seq_show = memory_events_show, |
5452 | }, | |
587d9f72 JW |
5453 | { |
5454 | .name = "stat", | |
5455 | .flags = CFTYPE_NOT_ON_ROOT, | |
5456 | .seq_show = memory_stat_show, | |
5457 | }, | |
241994ed JW |
5458 | { } /* terminate */ |
5459 | }; | |
5460 | ||
073219e9 | 5461 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5462 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5463 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5464 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5465 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5466 | .css_free = mem_cgroup_css_free, |
1ced953b | 5467 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5468 | .can_attach = mem_cgroup_can_attach, |
5469 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 5470 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5471 | .bind = mem_cgroup_bind, |
241994ed JW |
5472 | .dfl_cftypes = memory_files, |
5473 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5474 | .early_init = 0, |
8cdea7c0 | 5475 | }; |
c077719b | 5476 | |
241994ed JW |
5477 | /** |
5478 | * mem_cgroup_low - check if memory consumption is below the normal range | |
34c81057 | 5479 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
5480 | * @memcg: the memory cgroup to check |
5481 | * | |
5482 | * Returns %true if memory consumption of @memcg, and that of all | |
34c81057 SC |
5483 | * ancestors up to (but not including) @root, is below the normal range. |
5484 | * | |
5485 | * @root is exclusive; it is never low when looked at directly and isn't | |
5486 | * checked when traversing the hierarchy. | |
5487 | * | |
5488 | * Excluding @root enables using memory.low to prioritize memory usage | |
5489 | * between cgroups within a subtree of the hierarchy that is limited by | |
5490 | * memory.high or memory.max. | |
5491 | * | |
5492 | * For example, given cgroup A with children B and C: | |
5493 | * | |
5494 | * A | |
5495 | * / \ | |
5496 | * B C | |
5497 | * | |
5498 | * and | |
5499 | * | |
5500 | * 1. A/memory.current > A/memory.high | |
5501 | * 2. A/B/memory.current < A/B/memory.low | |
5502 | * 3. A/C/memory.current >= A/C/memory.low | |
5503 | * | |
5504 | * As 'A' is high, i.e. triggers reclaim from 'A', and 'B' is low, we | |
5505 | * should reclaim from 'C' until 'A' is no longer high or until we can | |
5506 | * no longer reclaim from 'C'. If 'A', i.e. @root, isn't excluded by | |
5507 | * mem_cgroup_low when reclaming from 'A', then 'B' won't be considered | |
5508 | * low and we will reclaim indiscriminately from both 'B' and 'C'. | |
241994ed JW |
5509 | */ |
5510 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5511 | { | |
5512 | if (mem_cgroup_disabled()) | |
5513 | return false; | |
5514 | ||
34c81057 SC |
5515 | if (!root) |
5516 | root = root_mem_cgroup; | |
5517 | if (memcg == root) | |
241994ed JW |
5518 | return false; |
5519 | ||
34c81057 | 5520 | for (; memcg != root; memcg = parent_mem_cgroup(memcg)) { |
4e54dede | 5521 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5522 | return false; |
5523 | } | |
34c81057 | 5524 | |
241994ed JW |
5525 | return true; |
5526 | } | |
5527 | ||
00501b53 JW |
5528 | /** |
5529 | * mem_cgroup_try_charge - try charging a page | |
5530 | * @page: page to charge | |
5531 | * @mm: mm context of the victim | |
5532 | * @gfp_mask: reclaim mode | |
5533 | * @memcgp: charged memcg return | |
25843c2b | 5534 | * @compound: charge the page as compound or small page |
00501b53 JW |
5535 | * |
5536 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5537 | * pages according to @gfp_mask if necessary. | |
5538 | * | |
5539 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5540 | * Otherwise, an error code is returned. | |
5541 | * | |
5542 | * After page->mapping has been set up, the caller must finalize the | |
5543 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5544 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5545 | */ | |
5546 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
5547 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
5548 | bool compound) | |
00501b53 JW |
5549 | { |
5550 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 5551 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5552 | int ret = 0; |
5553 | ||
5554 | if (mem_cgroup_disabled()) | |
5555 | goto out; | |
5556 | ||
5557 | if (PageSwapCache(page)) { | |
00501b53 JW |
5558 | /* |
5559 | * Every swap fault against a single page tries to charge the | |
5560 | * page, bail as early as possible. shmem_unuse() encounters | |
5561 | * already charged pages, too. The USED bit is protected by | |
5562 | * the page lock, which serializes swap cache removal, which | |
5563 | * in turn serializes uncharging. | |
5564 | */ | |
e993d905 | 5565 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 5566 | if (compound_head(page)->mem_cgroup) |
00501b53 | 5567 | goto out; |
e993d905 | 5568 | |
37e84351 | 5569 | if (do_swap_account) { |
e993d905 VD |
5570 | swp_entry_t ent = { .val = page_private(page), }; |
5571 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5572 | ||
5573 | rcu_read_lock(); | |
5574 | memcg = mem_cgroup_from_id(id); | |
5575 | if (memcg && !css_tryget_online(&memcg->css)) | |
5576 | memcg = NULL; | |
5577 | rcu_read_unlock(); | |
5578 | } | |
00501b53 JW |
5579 | } |
5580 | ||
00501b53 JW |
5581 | if (!memcg) |
5582 | memcg = get_mem_cgroup_from_mm(mm); | |
5583 | ||
5584 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5585 | ||
5586 | css_put(&memcg->css); | |
00501b53 JW |
5587 | out: |
5588 | *memcgp = memcg; | |
5589 | return ret; | |
5590 | } | |
5591 | ||
5592 | /** | |
5593 | * mem_cgroup_commit_charge - commit a page charge | |
5594 | * @page: page to charge | |
5595 | * @memcg: memcg to charge the page to | |
5596 | * @lrucare: page might be on LRU already | |
25843c2b | 5597 | * @compound: charge the page as compound or small page |
00501b53 JW |
5598 | * |
5599 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5600 | * after page->mapping has been set up. This must happen atomically | |
5601 | * as part of the page instantiation, i.e. under the page table lock | |
5602 | * for anonymous pages, under the page lock for page and swap cache. | |
5603 | * | |
5604 | * In addition, the page must not be on the LRU during the commit, to | |
5605 | * prevent racing with task migration. If it might be, use @lrucare. | |
5606 | * | |
5607 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5608 | */ | |
5609 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 5610 | bool lrucare, bool compound) |
00501b53 | 5611 | { |
f627c2f5 | 5612 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5613 | |
5614 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5615 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5616 | ||
5617 | if (mem_cgroup_disabled()) | |
5618 | return; | |
5619 | /* | |
5620 | * Swap faults will attempt to charge the same page multiple | |
5621 | * times. But reuse_swap_page() might have removed the page | |
5622 | * from swapcache already, so we can't check PageSwapCache(). | |
5623 | */ | |
5624 | if (!memcg) | |
5625 | return; | |
5626 | ||
6abb5a86 JW |
5627 | commit_charge(page, memcg, lrucare); |
5628 | ||
6abb5a86 | 5629 | local_irq_disable(); |
f627c2f5 | 5630 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
5631 | memcg_check_events(memcg, page); |
5632 | local_irq_enable(); | |
00501b53 | 5633 | |
7941d214 | 5634 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
5635 | swp_entry_t entry = { .val = page_private(page) }; |
5636 | /* | |
5637 | * The swap entry might not get freed for a long time, | |
5638 | * let's not wait for it. The page already received a | |
5639 | * memory+swap charge, drop the swap entry duplicate. | |
5640 | */ | |
38d8b4e6 | 5641 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 JW |
5642 | } |
5643 | } | |
5644 | ||
5645 | /** | |
5646 | * mem_cgroup_cancel_charge - cancel a page charge | |
5647 | * @page: page to charge | |
5648 | * @memcg: memcg to charge the page to | |
25843c2b | 5649 | * @compound: charge the page as compound or small page |
00501b53 JW |
5650 | * |
5651 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5652 | */ | |
f627c2f5 KS |
5653 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
5654 | bool compound) | |
00501b53 | 5655 | { |
f627c2f5 | 5656 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5657 | |
5658 | if (mem_cgroup_disabled()) | |
5659 | return; | |
5660 | /* | |
5661 | * Swap faults will attempt to charge the same page multiple | |
5662 | * times. But reuse_swap_page() might have removed the page | |
5663 | * from swapcache already, so we can't check PageSwapCache(). | |
5664 | */ | |
5665 | if (!memcg) | |
5666 | return; | |
5667 | ||
00501b53 JW |
5668 | cancel_charge(memcg, nr_pages); |
5669 | } | |
5670 | ||
a9d5adee JG |
5671 | struct uncharge_gather { |
5672 | struct mem_cgroup *memcg; | |
5673 | unsigned long pgpgout; | |
5674 | unsigned long nr_anon; | |
5675 | unsigned long nr_file; | |
5676 | unsigned long nr_kmem; | |
5677 | unsigned long nr_huge; | |
5678 | unsigned long nr_shmem; | |
5679 | struct page *dummy_page; | |
5680 | }; | |
5681 | ||
5682 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 5683 | { |
a9d5adee JG |
5684 | memset(ug, 0, sizeof(*ug)); |
5685 | } | |
5686 | ||
5687 | static void uncharge_batch(const struct uncharge_gather *ug) | |
5688 | { | |
5689 | unsigned long nr_pages = ug->nr_anon + ug->nr_file + ug->nr_kmem; | |
747db954 JW |
5690 | unsigned long flags; |
5691 | ||
a9d5adee JG |
5692 | if (!mem_cgroup_is_root(ug->memcg)) { |
5693 | page_counter_uncharge(&ug->memcg->memory, nr_pages); | |
7941d214 | 5694 | if (do_memsw_account()) |
a9d5adee JG |
5695 | page_counter_uncharge(&ug->memcg->memsw, nr_pages); |
5696 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) | |
5697 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
5698 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 5699 | } |
747db954 JW |
5700 | |
5701 | local_irq_save(flags); | |
a9d5adee JG |
5702 | __this_cpu_sub(ug->memcg->stat->count[MEMCG_RSS], ug->nr_anon); |
5703 | __this_cpu_sub(ug->memcg->stat->count[MEMCG_CACHE], ug->nr_file); | |
5704 | __this_cpu_sub(ug->memcg->stat->count[MEMCG_RSS_HUGE], ug->nr_huge); | |
5705 | __this_cpu_sub(ug->memcg->stat->count[NR_SHMEM], ug->nr_shmem); | |
5706 | __this_cpu_add(ug->memcg->stat->events[PGPGOUT], ug->pgpgout); | |
5707 | __this_cpu_add(ug->memcg->stat->nr_page_events, nr_pages); | |
5708 | memcg_check_events(ug->memcg, ug->dummy_page); | |
747db954 | 5709 | local_irq_restore(flags); |
e8ea14cc | 5710 | |
a9d5adee JG |
5711 | if (!mem_cgroup_is_root(ug->memcg)) |
5712 | css_put_many(&ug->memcg->css, nr_pages); | |
5713 | } | |
5714 | ||
5715 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
5716 | { | |
5717 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
3f2eb028 JG |
5718 | VM_BUG_ON_PAGE(page_count(page) && !is_zone_device_page(page) && |
5719 | !PageHWPoison(page) , page); | |
a9d5adee JG |
5720 | |
5721 | if (!page->mem_cgroup) | |
5722 | return; | |
5723 | ||
5724 | /* | |
5725 | * Nobody should be changing or seriously looking at | |
5726 | * page->mem_cgroup at this point, we have fully | |
5727 | * exclusive access to the page. | |
5728 | */ | |
5729 | ||
5730 | if (ug->memcg != page->mem_cgroup) { | |
5731 | if (ug->memcg) { | |
5732 | uncharge_batch(ug); | |
5733 | uncharge_gather_clear(ug); | |
5734 | } | |
5735 | ug->memcg = page->mem_cgroup; | |
5736 | } | |
5737 | ||
5738 | if (!PageKmemcg(page)) { | |
5739 | unsigned int nr_pages = 1; | |
5740 | ||
5741 | if (PageTransHuge(page)) { | |
5742 | nr_pages <<= compound_order(page); | |
5743 | ug->nr_huge += nr_pages; | |
5744 | } | |
5745 | if (PageAnon(page)) | |
5746 | ug->nr_anon += nr_pages; | |
5747 | else { | |
5748 | ug->nr_file += nr_pages; | |
5749 | if (PageSwapBacked(page)) | |
5750 | ug->nr_shmem += nr_pages; | |
5751 | } | |
5752 | ug->pgpgout++; | |
5753 | } else { | |
5754 | ug->nr_kmem += 1 << compound_order(page); | |
5755 | __ClearPageKmemcg(page); | |
5756 | } | |
5757 | ||
5758 | ug->dummy_page = page; | |
5759 | page->mem_cgroup = NULL; | |
747db954 JW |
5760 | } |
5761 | ||
5762 | static void uncharge_list(struct list_head *page_list) | |
5763 | { | |
a9d5adee | 5764 | struct uncharge_gather ug; |
747db954 | 5765 | struct list_head *next; |
a9d5adee JG |
5766 | |
5767 | uncharge_gather_clear(&ug); | |
747db954 | 5768 | |
8b592656 JW |
5769 | /* |
5770 | * Note that the list can be a single page->lru; hence the | |
5771 | * do-while loop instead of a simple list_for_each_entry(). | |
5772 | */ | |
747db954 JW |
5773 | next = page_list->next; |
5774 | do { | |
a9d5adee JG |
5775 | struct page *page; |
5776 | ||
747db954 JW |
5777 | page = list_entry(next, struct page, lru); |
5778 | next = page->lru.next; | |
5779 | ||
a9d5adee | 5780 | uncharge_page(page, &ug); |
747db954 JW |
5781 | } while (next != page_list); |
5782 | ||
a9d5adee JG |
5783 | if (ug.memcg) |
5784 | uncharge_batch(&ug); | |
747db954 JW |
5785 | } |
5786 | ||
0a31bc97 JW |
5787 | /** |
5788 | * mem_cgroup_uncharge - uncharge a page | |
5789 | * @page: page to uncharge | |
5790 | * | |
5791 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5792 | * mem_cgroup_commit_charge(). | |
5793 | */ | |
5794 | void mem_cgroup_uncharge(struct page *page) | |
5795 | { | |
a9d5adee JG |
5796 | struct uncharge_gather ug; |
5797 | ||
0a31bc97 JW |
5798 | if (mem_cgroup_disabled()) |
5799 | return; | |
5800 | ||
747db954 | 5801 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5802 | if (!page->mem_cgroup) |
0a31bc97 JW |
5803 | return; |
5804 | ||
a9d5adee JG |
5805 | uncharge_gather_clear(&ug); |
5806 | uncharge_page(page, &ug); | |
5807 | uncharge_batch(&ug); | |
747db954 | 5808 | } |
0a31bc97 | 5809 | |
747db954 JW |
5810 | /** |
5811 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5812 | * @page_list: list of pages to uncharge | |
5813 | * | |
5814 | * Uncharge a list of pages previously charged with | |
5815 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5816 | */ | |
5817 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5818 | { | |
5819 | if (mem_cgroup_disabled()) | |
5820 | return; | |
0a31bc97 | 5821 | |
747db954 JW |
5822 | if (!list_empty(page_list)) |
5823 | uncharge_list(page_list); | |
0a31bc97 JW |
5824 | } |
5825 | ||
5826 | /** | |
6a93ca8f JW |
5827 | * mem_cgroup_migrate - charge a page's replacement |
5828 | * @oldpage: currently circulating page | |
5829 | * @newpage: replacement page | |
0a31bc97 | 5830 | * |
6a93ca8f JW |
5831 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
5832 | * be uncharged upon free. | |
0a31bc97 JW |
5833 | * |
5834 | * Both pages must be locked, @newpage->mapping must be set up. | |
5835 | */ | |
6a93ca8f | 5836 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 5837 | { |
29833315 | 5838 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
5839 | unsigned int nr_pages; |
5840 | bool compound; | |
d93c4130 | 5841 | unsigned long flags; |
0a31bc97 JW |
5842 | |
5843 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5844 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 5845 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
5846 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5847 | newpage); | |
0a31bc97 JW |
5848 | |
5849 | if (mem_cgroup_disabled()) | |
5850 | return; | |
5851 | ||
5852 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5853 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5854 | return; |
5855 | ||
45637bab | 5856 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 5857 | memcg = oldpage->mem_cgroup; |
29833315 | 5858 | if (!memcg) |
0a31bc97 JW |
5859 | return; |
5860 | ||
44b7a8d3 JW |
5861 | /* Force-charge the new page. The old one will be freed soon */ |
5862 | compound = PageTransHuge(newpage); | |
5863 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
5864 | ||
5865 | page_counter_charge(&memcg->memory, nr_pages); | |
5866 | if (do_memsw_account()) | |
5867 | page_counter_charge(&memcg->memsw, nr_pages); | |
5868 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 5869 | |
9cf7666a | 5870 | commit_charge(newpage, memcg, false); |
44b7a8d3 | 5871 | |
d93c4130 | 5872 | local_irq_save(flags); |
44b7a8d3 JW |
5873 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
5874 | memcg_check_events(memcg, newpage); | |
d93c4130 | 5875 | local_irq_restore(flags); |
0a31bc97 JW |
5876 | } |
5877 | ||
ef12947c | 5878 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
5879 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
5880 | ||
2d758073 | 5881 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
5882 | { |
5883 | struct mem_cgroup *memcg; | |
5884 | ||
2d758073 JW |
5885 | if (!mem_cgroup_sockets_enabled) |
5886 | return; | |
5887 | ||
344ae441 RG |
5888 | /* |
5889 | * Socket cloning can throw us here with sk_memcg already | |
5890 | * filled. It won't however, necessarily happen from | |
5891 | * process context. So the test for root memcg given | |
5892 | * the current task's memcg won't help us in this case. | |
5893 | * | |
5894 | * Respecting the original socket's memcg is a better | |
5895 | * decision in this case. | |
5896 | */ | |
5897 | if (sk->sk_memcg) { | |
5898 | css_get(&sk->sk_memcg->css); | |
5899 | return; | |
5900 | } | |
5901 | ||
11092087 JW |
5902 | rcu_read_lock(); |
5903 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
5904 | if (memcg == root_mem_cgroup) |
5905 | goto out; | |
0db15298 | 5906 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 5907 | goto out; |
f7e1cb6e | 5908 | if (css_tryget_online(&memcg->css)) |
11092087 | 5909 | sk->sk_memcg = memcg; |
f7e1cb6e | 5910 | out: |
11092087 JW |
5911 | rcu_read_unlock(); |
5912 | } | |
11092087 | 5913 | |
2d758073 | 5914 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 5915 | { |
2d758073 JW |
5916 | if (sk->sk_memcg) |
5917 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
5918 | } |
5919 | ||
5920 | /** | |
5921 | * mem_cgroup_charge_skmem - charge socket memory | |
5922 | * @memcg: memcg to charge | |
5923 | * @nr_pages: number of pages to charge | |
5924 | * | |
5925 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
5926 | * @memcg's configured limit, %false if the charge had to be forced. | |
5927 | */ | |
5928 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5929 | { | |
f7e1cb6e | 5930 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 5931 | |
f7e1cb6e | 5932 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5933 | struct page_counter *fail; |
f7e1cb6e | 5934 | |
0db15298 JW |
5935 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
5936 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
5937 | return true; |
5938 | } | |
0db15298 JW |
5939 | page_counter_charge(&memcg->tcpmem, nr_pages); |
5940 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 5941 | return false; |
11092087 | 5942 | } |
d886f4e4 | 5943 | |
f7e1cb6e JW |
5944 | /* Don't block in the packet receive path */ |
5945 | if (in_softirq()) | |
5946 | gfp_mask = GFP_NOWAIT; | |
5947 | ||
b2807f07 JW |
5948 | this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages); |
5949 | ||
f7e1cb6e JW |
5950 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
5951 | return true; | |
5952 | ||
5953 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
5954 | return false; |
5955 | } | |
5956 | ||
5957 | /** | |
5958 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
5959 | * @memcg - memcg to uncharge | |
5960 | * @nr_pages - number of pages to uncharge | |
5961 | */ | |
5962 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5963 | { | |
f7e1cb6e | 5964 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5965 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
5966 | return; |
5967 | } | |
d886f4e4 | 5968 | |
b2807f07 JW |
5969 | this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages); |
5970 | ||
475d0487 | 5971 | refill_stock(memcg, nr_pages); |
11092087 JW |
5972 | } |
5973 | ||
f7e1cb6e JW |
5974 | static int __init cgroup_memory(char *s) |
5975 | { | |
5976 | char *token; | |
5977 | ||
5978 | while ((token = strsep(&s, ",")) != NULL) { | |
5979 | if (!*token) | |
5980 | continue; | |
5981 | if (!strcmp(token, "nosocket")) | |
5982 | cgroup_memory_nosocket = true; | |
04823c83 VD |
5983 | if (!strcmp(token, "nokmem")) |
5984 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
5985 | } |
5986 | return 0; | |
5987 | } | |
5988 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 5989 | |
2d11085e | 5990 | /* |
1081312f MH |
5991 | * subsys_initcall() for memory controller. |
5992 | * | |
308167fc SAS |
5993 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
5994 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
5995 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
5996 | * should be initialized from here. | |
2d11085e MH |
5997 | */ |
5998 | static int __init mem_cgroup_init(void) | |
5999 | { | |
95a045f6 JW |
6000 | int cpu, node; |
6001 | ||
13583c3d VD |
6002 | #ifndef CONFIG_SLOB |
6003 | /* | |
6004 | * Kmem cache creation is mostly done with the slab_mutex held, | |
17cc4dfe TH |
6005 | * so use a workqueue with limited concurrency to avoid stalling |
6006 | * all worker threads in case lots of cgroups are created and | |
6007 | * destroyed simultaneously. | |
13583c3d | 6008 | */ |
17cc4dfe TH |
6009 | memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1); |
6010 | BUG_ON(!memcg_kmem_cache_wq); | |
13583c3d VD |
6011 | #endif |
6012 | ||
308167fc SAS |
6013 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
6014 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
6015 | |
6016 | for_each_possible_cpu(cpu) | |
6017 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
6018 | drain_local_stock); | |
6019 | ||
6020 | for_each_node(node) { | |
6021 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
6022 | |
6023 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
6024 | node_online(node) ? node : NUMA_NO_NODE); | |
6025 | ||
ef8f2327 | 6026 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 6027 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 6028 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
6029 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
6030 | } | |
6031 | ||
2d11085e MH |
6032 | return 0; |
6033 | } | |
6034 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
6035 | |
6036 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
6037 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
6038 | { | |
6039 | while (!atomic_inc_not_zero(&memcg->id.ref)) { | |
6040 | /* | |
6041 | * The root cgroup cannot be destroyed, so it's refcount must | |
6042 | * always be >= 1. | |
6043 | */ | |
6044 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
6045 | VM_BUG_ON(1); | |
6046 | break; | |
6047 | } | |
6048 | memcg = parent_mem_cgroup(memcg); | |
6049 | if (!memcg) | |
6050 | memcg = root_mem_cgroup; | |
6051 | } | |
6052 | return memcg; | |
6053 | } | |
6054 | ||
21afa38e JW |
6055 | /** |
6056 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
6057 | * @page: page whose memsw charge to transfer | |
6058 | * @entry: swap entry to move the charge to | |
6059 | * | |
6060 | * Transfer the memsw charge of @page to @entry. | |
6061 | */ | |
6062 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
6063 | { | |
1f47b61f | 6064 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 6065 | unsigned int nr_entries; |
21afa38e JW |
6066 | unsigned short oldid; |
6067 | ||
6068 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
6069 | VM_BUG_ON_PAGE(page_count(page), page); | |
6070 | ||
7941d214 | 6071 | if (!do_memsw_account()) |
21afa38e JW |
6072 | return; |
6073 | ||
6074 | memcg = page->mem_cgroup; | |
6075 | ||
6076 | /* Readahead page, never charged */ | |
6077 | if (!memcg) | |
6078 | return; | |
6079 | ||
1f47b61f VD |
6080 | /* |
6081 | * In case the memcg owning these pages has been offlined and doesn't | |
6082 | * have an ID allocated to it anymore, charge the closest online | |
6083 | * ancestor for the swap instead and transfer the memory+swap charge. | |
6084 | */ | |
6085 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
d6810d73 HY |
6086 | nr_entries = hpage_nr_pages(page); |
6087 | /* Get references for the tail pages, too */ | |
6088 | if (nr_entries > 1) | |
6089 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
6090 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
6091 | nr_entries); | |
21afa38e | 6092 | VM_BUG_ON_PAGE(oldid, page); |
d6810d73 | 6093 | mem_cgroup_swap_statistics(swap_memcg, nr_entries); |
21afa38e JW |
6094 | |
6095 | page->mem_cgroup = NULL; | |
6096 | ||
6097 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 6098 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 6099 | |
1f47b61f VD |
6100 | if (memcg != swap_memcg) { |
6101 | if (!mem_cgroup_is_root(swap_memcg)) | |
d6810d73 HY |
6102 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
6103 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
6104 | } |
6105 | ||
ce9ce665 SAS |
6106 | /* |
6107 | * Interrupts should be disabled here because the caller holds the | |
6108 | * mapping->tree_lock lock which is taken with interrupts-off. It is | |
6109 | * important here to have the interrupts disabled because it is the | |
6110 | * only synchronisation we have for udpating the per-CPU variables. | |
6111 | */ | |
6112 | VM_BUG_ON(!irqs_disabled()); | |
d6810d73 HY |
6113 | mem_cgroup_charge_statistics(memcg, page, PageTransHuge(page), |
6114 | -nr_entries); | |
21afa38e | 6115 | memcg_check_events(memcg, page); |
73f576c0 JW |
6116 | |
6117 | if (!mem_cgroup_is_root(memcg)) | |
d08afa14 | 6118 | css_put_many(&memcg->css, nr_entries); |
21afa38e JW |
6119 | } |
6120 | ||
38d8b4e6 HY |
6121 | /** |
6122 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
6123 | * @page: page being added to swap |
6124 | * @entry: swap entry to charge | |
6125 | * | |
38d8b4e6 | 6126 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
6127 | * |
6128 | * Returns 0 on success, -ENOMEM on failure. | |
6129 | */ | |
6130 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
6131 | { | |
38d8b4e6 | 6132 | unsigned int nr_pages = hpage_nr_pages(page); |
37e84351 | 6133 | struct page_counter *counter; |
38d8b4e6 | 6134 | struct mem_cgroup *memcg; |
37e84351 VD |
6135 | unsigned short oldid; |
6136 | ||
6137 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
6138 | return 0; | |
6139 | ||
6140 | memcg = page->mem_cgroup; | |
6141 | ||
6142 | /* Readahead page, never charged */ | |
6143 | if (!memcg) | |
6144 | return 0; | |
6145 | ||
1f47b61f VD |
6146 | memcg = mem_cgroup_id_get_online(memcg); |
6147 | ||
37e84351 | 6148 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 6149 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
1f47b61f | 6150 | mem_cgroup_id_put(memcg); |
37e84351 | 6151 | return -ENOMEM; |
1f47b61f | 6152 | } |
37e84351 | 6153 | |
38d8b4e6 HY |
6154 | /* Get references for the tail pages, too */ |
6155 | if (nr_pages > 1) | |
6156 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
6157 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 6158 | VM_BUG_ON_PAGE(oldid, page); |
38d8b4e6 | 6159 | mem_cgroup_swap_statistics(memcg, nr_pages); |
37e84351 | 6160 | |
37e84351 VD |
6161 | return 0; |
6162 | } | |
6163 | ||
21afa38e | 6164 | /** |
38d8b4e6 | 6165 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 6166 | * @entry: swap entry to uncharge |
38d8b4e6 | 6167 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 6168 | */ |
38d8b4e6 | 6169 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
6170 | { |
6171 | struct mem_cgroup *memcg; | |
6172 | unsigned short id; | |
6173 | ||
37e84351 | 6174 | if (!do_swap_account) |
21afa38e JW |
6175 | return; |
6176 | ||
38d8b4e6 | 6177 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 6178 | rcu_read_lock(); |
adbe427b | 6179 | memcg = mem_cgroup_from_id(id); |
21afa38e | 6180 | if (memcg) { |
37e84351 VD |
6181 | if (!mem_cgroup_is_root(memcg)) { |
6182 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
38d8b4e6 | 6183 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 6184 | else |
38d8b4e6 | 6185 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 6186 | } |
38d8b4e6 HY |
6187 | mem_cgroup_swap_statistics(memcg, -nr_pages); |
6188 | mem_cgroup_id_put_many(memcg, nr_pages); | |
21afa38e JW |
6189 | } |
6190 | rcu_read_unlock(); | |
6191 | } | |
6192 | ||
d8b38438 VD |
6193 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
6194 | { | |
6195 | long nr_swap_pages = get_nr_swap_pages(); | |
6196 | ||
6197 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6198 | return nr_swap_pages; | |
6199 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
6200 | nr_swap_pages = min_t(long, nr_swap_pages, | |
6201 | READ_ONCE(memcg->swap.limit) - | |
6202 | page_counter_read(&memcg->swap)); | |
6203 | return nr_swap_pages; | |
6204 | } | |
6205 | ||
5ccc5aba VD |
6206 | bool mem_cgroup_swap_full(struct page *page) |
6207 | { | |
6208 | struct mem_cgroup *memcg; | |
6209 | ||
6210 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
6211 | ||
6212 | if (vm_swap_full()) | |
6213 | return true; | |
6214 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6215 | return false; | |
6216 | ||
6217 | memcg = page->mem_cgroup; | |
6218 | if (!memcg) | |
6219 | return false; | |
6220 | ||
6221 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
6222 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit) | |
6223 | return true; | |
6224 | ||
6225 | return false; | |
6226 | } | |
6227 | ||
21afa38e JW |
6228 | /* for remember boot option*/ |
6229 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
6230 | static int really_do_swap_account __initdata = 1; | |
6231 | #else | |
6232 | static int really_do_swap_account __initdata; | |
6233 | #endif | |
6234 | ||
6235 | static int __init enable_swap_account(char *s) | |
6236 | { | |
6237 | if (!strcmp(s, "1")) | |
6238 | really_do_swap_account = 1; | |
6239 | else if (!strcmp(s, "0")) | |
6240 | really_do_swap_account = 0; | |
6241 | return 1; | |
6242 | } | |
6243 | __setup("swapaccount=", enable_swap_account); | |
6244 | ||
37e84351 VD |
6245 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
6246 | struct cftype *cft) | |
6247 | { | |
6248 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6249 | ||
6250 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
6251 | } | |
6252 | ||
6253 | static int swap_max_show(struct seq_file *m, void *v) | |
6254 | { | |
6255 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
6256 | unsigned long max = READ_ONCE(memcg->swap.limit); | |
6257 | ||
6258 | if (max == PAGE_COUNTER_MAX) | |
6259 | seq_puts(m, "max\n"); | |
6260 | else | |
6261 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
6262 | ||
6263 | return 0; | |
6264 | } | |
6265 | ||
6266 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
6267 | char *buf, size_t nbytes, loff_t off) | |
6268 | { | |
6269 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6270 | unsigned long max; | |
6271 | int err; | |
6272 | ||
6273 | buf = strstrip(buf); | |
6274 | err = page_counter_memparse(buf, "max", &max); | |
6275 | if (err) | |
6276 | return err; | |
6277 | ||
6278 | mutex_lock(&memcg_limit_mutex); | |
6279 | err = page_counter_limit(&memcg->swap, max); | |
6280 | mutex_unlock(&memcg_limit_mutex); | |
6281 | if (err) | |
6282 | return err; | |
6283 | ||
6284 | return nbytes; | |
6285 | } | |
6286 | ||
6287 | static struct cftype swap_files[] = { | |
6288 | { | |
6289 | .name = "swap.current", | |
6290 | .flags = CFTYPE_NOT_ON_ROOT, | |
6291 | .read_u64 = swap_current_read, | |
6292 | }, | |
6293 | { | |
6294 | .name = "swap.max", | |
6295 | .flags = CFTYPE_NOT_ON_ROOT, | |
6296 | .seq_show = swap_max_show, | |
6297 | .write = swap_max_write, | |
6298 | }, | |
6299 | { } /* terminate */ | |
6300 | }; | |
6301 | ||
21afa38e JW |
6302 | static struct cftype memsw_cgroup_files[] = { |
6303 | { | |
6304 | .name = "memsw.usage_in_bytes", | |
6305 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6306 | .read_u64 = mem_cgroup_read_u64, | |
6307 | }, | |
6308 | { | |
6309 | .name = "memsw.max_usage_in_bytes", | |
6310 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6311 | .write = mem_cgroup_reset, | |
6312 | .read_u64 = mem_cgroup_read_u64, | |
6313 | }, | |
6314 | { | |
6315 | .name = "memsw.limit_in_bytes", | |
6316 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6317 | .write = mem_cgroup_write, | |
6318 | .read_u64 = mem_cgroup_read_u64, | |
6319 | }, | |
6320 | { | |
6321 | .name = "memsw.failcnt", | |
6322 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6323 | .write = mem_cgroup_reset, | |
6324 | .read_u64 = mem_cgroup_read_u64, | |
6325 | }, | |
6326 | { }, /* terminate */ | |
6327 | }; | |
6328 | ||
6329 | static int __init mem_cgroup_swap_init(void) | |
6330 | { | |
6331 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6332 | do_swap_account = 1; | |
37e84351 VD |
6333 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
6334 | swap_files)); | |
21afa38e JW |
6335 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6336 | memsw_cgroup_files)); | |
6337 | } | |
6338 | return 0; | |
6339 | } | |
6340 | subsys_initcall(mem_cgroup_swap_init); | |
6341 | ||
6342 | #endif /* CONFIG_MEMCG_SWAP */ |