]>
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 | |
70ddf637 AV |
236 | /* Some nice accessors for the vmpressure. */ |
237 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
238 | { | |
239 | if (!memcg) | |
240 | memcg = root_mem_cgroup; | |
241 | return &memcg->vmpressure; | |
242 | } | |
243 | ||
244 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
245 | { | |
246 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
247 | } | |
248 | ||
7ffc0edc MH |
249 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
250 | { | |
251 | return (memcg == root_mem_cgroup); | |
252 | } | |
253 | ||
127424c8 | 254 | #ifndef CONFIG_SLOB |
55007d84 | 255 | /* |
f7ce3190 | 256 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
b8627835 LZ |
257 | * The main reason for not using cgroup id for this: |
258 | * this works better in sparse environments, where we have a lot of memcgs, | |
259 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
260 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
261 | * 200 entry array for that. | |
55007d84 | 262 | * |
dbcf73e2 VD |
263 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
264 | * will double each time we have to increase it. | |
55007d84 | 265 | */ |
dbcf73e2 VD |
266 | static DEFINE_IDA(memcg_cache_ida); |
267 | int memcg_nr_cache_ids; | |
749c5415 | 268 | |
05257a1a VD |
269 | /* Protects memcg_nr_cache_ids */ |
270 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
271 | ||
272 | void memcg_get_cache_ids(void) | |
273 | { | |
274 | down_read(&memcg_cache_ids_sem); | |
275 | } | |
276 | ||
277 | void memcg_put_cache_ids(void) | |
278 | { | |
279 | up_read(&memcg_cache_ids_sem); | |
280 | } | |
281 | ||
55007d84 GC |
282 | /* |
283 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
284 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
285 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
286 | * tunable, but that is strictly not necessary. | |
287 | * | |
b8627835 | 288 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
289 | * this constant directly from cgroup, but it is understandable that this is |
290 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 291 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
292 | * increase ours as well if it increases. |
293 | */ | |
294 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 295 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 296 | |
d7f25f8a GC |
297 | /* |
298 | * A lot of the calls to the cache allocation functions are expected to be | |
299 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are | |
300 | * conditional to this static branch, we'll have to allow modules that does | |
301 | * kmem_cache_alloc and the such to see this symbol as well | |
302 | */ | |
ef12947c | 303 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 304 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
a8964b9b | 305 | |
17cc4dfe TH |
306 | struct workqueue_struct *memcg_kmem_cache_wq; |
307 | ||
127424c8 | 308 | #endif /* !CONFIG_SLOB */ |
a8964b9b | 309 | |
ad7fa852 TH |
310 | /** |
311 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
312 | * @page: page of interest | |
313 | * | |
314 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
315 | * with @page is returned. The returned css remains associated with @page | |
316 | * until it is released. | |
317 | * | |
318 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
319 | * is returned. | |
ad7fa852 TH |
320 | */ |
321 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
322 | { | |
323 | struct mem_cgroup *memcg; | |
324 | ||
ad7fa852 TH |
325 | memcg = page->mem_cgroup; |
326 | ||
9e10a130 | 327 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
328 | memcg = root_mem_cgroup; |
329 | ||
ad7fa852 TH |
330 | return &memcg->css; |
331 | } | |
332 | ||
2fc04524 VD |
333 | /** |
334 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
335 | * @page: the page | |
336 | * | |
337 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
338 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
339 | * is safe to call this function without holding a reference to @page. | |
340 | * | |
341 | * Note, this function is inherently racy, because there is nothing to prevent | |
342 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
343 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
344 | * do not care (such as procfs interfaces). | |
345 | */ | |
346 | ino_t page_cgroup_ino(struct page *page) | |
347 | { | |
348 | struct mem_cgroup *memcg; | |
349 | unsigned long ino = 0; | |
350 | ||
351 | rcu_read_lock(); | |
352 | memcg = READ_ONCE(page->mem_cgroup); | |
353 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) | |
354 | memcg = parent_mem_cgroup(memcg); | |
355 | if (memcg) | |
356 | ino = cgroup_ino(memcg->css.cgroup); | |
357 | rcu_read_unlock(); | |
358 | return ino; | |
359 | } | |
360 | ||
ef8f2327 MG |
361 | static struct mem_cgroup_per_node * |
362 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 363 | { |
97a6c37b | 364 | int nid = page_to_nid(page); |
f64c3f54 | 365 | |
ef8f2327 | 366 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
367 | } |
368 | ||
ef8f2327 MG |
369 | static struct mem_cgroup_tree_per_node * |
370 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 371 | { |
ef8f2327 | 372 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
373 | } |
374 | ||
ef8f2327 | 375 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
376 | soft_limit_tree_from_page(struct page *page) |
377 | { | |
378 | int nid = page_to_nid(page); | |
bb4cc1a8 | 379 | |
ef8f2327 | 380 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
381 | } |
382 | ||
ef8f2327 MG |
383 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
384 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 385 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
386 | { |
387 | struct rb_node **p = &mctz->rb_root.rb_node; | |
388 | struct rb_node *parent = NULL; | |
ef8f2327 | 389 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 390 | bool rightmost = true; |
bb4cc1a8 AM |
391 | |
392 | if (mz->on_tree) | |
393 | return; | |
394 | ||
395 | mz->usage_in_excess = new_usage_in_excess; | |
396 | if (!mz->usage_in_excess) | |
397 | return; | |
398 | while (*p) { | |
399 | parent = *p; | |
ef8f2327 | 400 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 401 | tree_node); |
fa90b2fd | 402 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 403 | p = &(*p)->rb_left; |
fa90b2fd DB |
404 | rightmost = false; |
405 | } | |
406 | ||
bb4cc1a8 AM |
407 | /* |
408 | * We can't avoid mem cgroups that are over their soft | |
409 | * limit by the same amount | |
410 | */ | |
411 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
412 | p = &(*p)->rb_right; | |
413 | } | |
fa90b2fd DB |
414 | |
415 | if (rightmost) | |
416 | mctz->rb_rightmost = &mz->tree_node; | |
417 | ||
bb4cc1a8 AM |
418 | rb_link_node(&mz->tree_node, parent, p); |
419 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
420 | mz->on_tree = true; | |
421 | } | |
422 | ||
ef8f2327 MG |
423 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
424 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
425 | { |
426 | if (!mz->on_tree) | |
427 | return; | |
fa90b2fd DB |
428 | |
429 | if (&mz->tree_node == mctz->rb_rightmost) | |
430 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
431 | ||
bb4cc1a8 AM |
432 | rb_erase(&mz->tree_node, &mctz->rb_root); |
433 | mz->on_tree = false; | |
434 | } | |
435 | ||
ef8f2327 MG |
436 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
437 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 438 | { |
0a31bc97 JW |
439 | unsigned long flags; |
440 | ||
441 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 442 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 443 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
444 | } |
445 | ||
3e32cb2e JW |
446 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
447 | { | |
448 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 449 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
450 | unsigned long excess = 0; |
451 | ||
452 | if (nr_pages > soft_limit) | |
453 | excess = nr_pages - soft_limit; | |
454 | ||
455 | return excess; | |
456 | } | |
bb4cc1a8 AM |
457 | |
458 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
459 | { | |
3e32cb2e | 460 | unsigned long excess; |
ef8f2327 MG |
461 | struct mem_cgroup_per_node *mz; |
462 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 463 | |
e231875b | 464 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
465 | if (!mctz) |
466 | return; | |
bb4cc1a8 AM |
467 | /* |
468 | * Necessary to update all ancestors when hierarchy is used. | |
469 | * because their event counter is not touched. | |
470 | */ | |
471 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 472 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 473 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
474 | /* |
475 | * We have to update the tree if mz is on RB-tree or | |
476 | * mem is over its softlimit. | |
477 | */ | |
478 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
479 | unsigned long flags; |
480 | ||
481 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
482 | /* if on-tree, remove it */ |
483 | if (mz->on_tree) | |
cf2c8127 | 484 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
485 | /* |
486 | * Insert again. mz->usage_in_excess will be updated. | |
487 | * If excess is 0, no tree ops. | |
488 | */ | |
cf2c8127 | 489 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 490 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
491 | } |
492 | } | |
493 | } | |
494 | ||
495 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
496 | { | |
ef8f2327 MG |
497 | struct mem_cgroup_tree_per_node *mctz; |
498 | struct mem_cgroup_per_node *mz; | |
499 | int nid; | |
bb4cc1a8 | 500 | |
e231875b | 501 | for_each_node(nid) { |
ef8f2327 MG |
502 | mz = mem_cgroup_nodeinfo(memcg, nid); |
503 | mctz = soft_limit_tree_node(nid); | |
bfc7228b LD |
504 | if (mctz) |
505 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
506 | } |
507 | } | |
508 | ||
ef8f2327 MG |
509 | static struct mem_cgroup_per_node * |
510 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 511 | { |
ef8f2327 | 512 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
513 | |
514 | retry: | |
515 | mz = NULL; | |
fa90b2fd | 516 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
517 | goto done; /* Nothing to reclaim from */ |
518 | ||
fa90b2fd DB |
519 | mz = rb_entry(mctz->rb_rightmost, |
520 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
521 | /* |
522 | * Remove the node now but someone else can add it back, | |
523 | * we will to add it back at the end of reclaim to its correct | |
524 | * position in the tree. | |
525 | */ | |
cf2c8127 | 526 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 527 | if (!soft_limit_excess(mz->memcg) || |
ec903c0c | 528 | !css_tryget_online(&mz->memcg->css)) |
bb4cc1a8 AM |
529 | goto retry; |
530 | done: | |
531 | return mz; | |
532 | } | |
533 | ||
ef8f2327 MG |
534 | static struct mem_cgroup_per_node * |
535 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 536 | { |
ef8f2327 | 537 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 538 | |
0a31bc97 | 539 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 540 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 541 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
542 | return mz; |
543 | } | |
544 | ||
ccda7f43 | 545 | static unsigned long memcg_sum_events(struct mem_cgroup *memcg, |
04fecbf5 | 546 | int event) |
e9f8974f | 547 | { |
a983b5eb | 548 | return atomic_long_read(&memcg->events[event]); |
e9f8974f JW |
549 | } |
550 | ||
c0ff4b85 | 551 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 552 | struct page *page, |
f627c2f5 | 553 | bool compound, int nr_pages) |
d52aa412 | 554 | { |
b2402857 KH |
555 | /* |
556 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is | |
557 | * counted as CACHE even if it's on ANON LRU. | |
558 | */ | |
0a31bc97 | 559 | if (PageAnon(page)) |
c9019e9b | 560 | __mod_memcg_state(memcg, MEMCG_RSS, nr_pages); |
9a4caf1e | 561 | else { |
c9019e9b | 562 | __mod_memcg_state(memcg, MEMCG_CACHE, nr_pages); |
9a4caf1e | 563 | if (PageSwapBacked(page)) |
c9019e9b | 564 | __mod_memcg_state(memcg, NR_SHMEM, nr_pages); |
9a4caf1e | 565 | } |
55e462b0 | 566 | |
f627c2f5 KS |
567 | if (compound) { |
568 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); | |
c9019e9b | 569 | __mod_memcg_state(memcg, MEMCG_RSS_HUGE, nr_pages); |
f627c2f5 | 570 | } |
b070e65c | 571 | |
e401f176 KH |
572 | /* pagein of a big page is an event. So, ignore page size */ |
573 | if (nr_pages > 0) | |
c9019e9b | 574 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 575 | else { |
c9019e9b | 576 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
577 | nr_pages = -nr_pages; /* for event */ |
578 | } | |
e401f176 | 579 | |
a983b5eb | 580 | __this_cpu_add(memcg->stat_cpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
581 | } |
582 | ||
0a6b76dd VD |
583 | unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
584 | int nid, unsigned int lru_mask) | |
bb2a0de9 | 585 | { |
b4536f0c | 586 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); |
e231875b | 587 | unsigned long nr = 0; |
ef8f2327 | 588 | enum lru_list lru; |
889976db | 589 | |
e231875b | 590 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
bb2a0de9 | 591 | |
ef8f2327 MG |
592 | for_each_lru(lru) { |
593 | if (!(BIT(lru) & lru_mask)) | |
594 | continue; | |
b4536f0c | 595 | nr += mem_cgroup_get_lru_size(lruvec, lru); |
e231875b JZ |
596 | } |
597 | return nr; | |
889976db | 598 | } |
bb2a0de9 | 599 | |
c0ff4b85 | 600 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
bb2a0de9 | 601 | unsigned int lru_mask) |
6d12e2d8 | 602 | { |
e231875b | 603 | unsigned long nr = 0; |
889976db | 604 | int nid; |
6d12e2d8 | 605 | |
31aaea4a | 606 | for_each_node_state(nid, N_MEMORY) |
e231875b JZ |
607 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
608 | return nr; | |
d52aa412 KH |
609 | } |
610 | ||
f53d7ce3 JW |
611 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
612 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
613 | { |
614 | unsigned long val, next; | |
615 | ||
a983b5eb JW |
616 | val = __this_cpu_read(memcg->stat_cpu->nr_page_events); |
617 | next = __this_cpu_read(memcg->stat_cpu->targets[target]); | |
7a159cc9 | 618 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 619 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
620 | switch (target) { |
621 | case MEM_CGROUP_TARGET_THRESH: | |
622 | next = val + THRESHOLDS_EVENTS_TARGET; | |
623 | break; | |
bb4cc1a8 AM |
624 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
625 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
626 | break; | |
f53d7ce3 JW |
627 | case MEM_CGROUP_TARGET_NUMAINFO: |
628 | next = val + NUMAINFO_EVENTS_TARGET; | |
629 | break; | |
630 | default: | |
631 | break; | |
632 | } | |
a983b5eb | 633 | __this_cpu_write(memcg->stat_cpu->targets[target], next); |
f53d7ce3 | 634 | return true; |
7a159cc9 | 635 | } |
f53d7ce3 | 636 | return false; |
d2265e6f KH |
637 | } |
638 | ||
639 | /* | |
640 | * Check events in order. | |
641 | * | |
642 | */ | |
c0ff4b85 | 643 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
644 | { |
645 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
646 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
647 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 648 | bool do_softlimit; |
82b3f2a7 | 649 | bool do_numainfo __maybe_unused; |
f53d7ce3 | 650 | |
bb4cc1a8 AM |
651 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
652 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
f53d7ce3 JW |
653 | #if MAX_NUMNODES > 1 |
654 | do_numainfo = mem_cgroup_event_ratelimit(memcg, | |
655 | MEM_CGROUP_TARGET_NUMAINFO); | |
656 | #endif | |
c0ff4b85 | 657 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
658 | if (unlikely(do_softlimit)) |
659 | mem_cgroup_update_tree(memcg, page); | |
453a9bf3 | 660 | #if MAX_NUMNODES > 1 |
f53d7ce3 | 661 | if (unlikely(do_numainfo)) |
c0ff4b85 | 662 | atomic_inc(&memcg->numainfo_events); |
453a9bf3 | 663 | #endif |
0a31bc97 | 664 | } |
d2265e6f KH |
665 | } |
666 | ||
cf475ad2 | 667 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 668 | { |
31a78f23 BS |
669 | /* |
670 | * mm_update_next_owner() may clear mm->owner to NULL | |
671 | * if it races with swapoff, page migration, etc. | |
672 | * So this can be called with p == NULL. | |
673 | */ | |
674 | if (unlikely(!p)) | |
675 | return NULL; | |
676 | ||
073219e9 | 677 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 678 | } |
33398cf2 | 679 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 680 | |
df381975 | 681 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 | 682 | { |
c0ff4b85 | 683 | struct mem_cgroup *memcg = NULL; |
0b7f569e | 684 | |
54595fe2 KH |
685 | rcu_read_lock(); |
686 | do { | |
6f6acb00 MH |
687 | /* |
688 | * Page cache insertions can happen withou an | |
689 | * actual mm context, e.g. during disk probing | |
690 | * on boot, loopback IO, acct() writes etc. | |
691 | */ | |
692 | if (unlikely(!mm)) | |
df381975 | 693 | memcg = root_mem_cgroup; |
6f6acb00 MH |
694 | else { |
695 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
696 | if (unlikely(!memcg)) | |
697 | memcg = root_mem_cgroup; | |
698 | } | |
ec903c0c | 699 | } while (!css_tryget_online(&memcg->css)); |
54595fe2 | 700 | rcu_read_unlock(); |
c0ff4b85 | 701 | return memcg; |
54595fe2 KH |
702 | } |
703 | ||
5660048c JW |
704 | /** |
705 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
706 | * @root: hierarchy root | |
707 | * @prev: previously returned memcg, NULL on first invocation | |
708 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
709 | * | |
710 | * Returns references to children of the hierarchy below @root, or | |
711 | * @root itself, or %NULL after a full round-trip. | |
712 | * | |
713 | * Caller must pass the return value in @prev on subsequent | |
714 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
715 | * to cancel a hierarchy walk before the round-trip is complete. | |
716 | * | |
717 | * Reclaimers can specify a zone and a priority level in @reclaim to | |
718 | * divide up the memcgs in the hierarchy among all concurrent | |
719 | * reclaimers operating on the same zone and priority. | |
720 | */ | |
694fbc0f | 721 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 722 | struct mem_cgroup *prev, |
694fbc0f | 723 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 724 | { |
33398cf2 | 725 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
5ac8fb31 | 726 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 727 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 728 | struct mem_cgroup *pos = NULL; |
711d3d2c | 729 | |
694fbc0f AM |
730 | if (mem_cgroup_disabled()) |
731 | return NULL; | |
5660048c | 732 | |
9f3a0d09 JW |
733 | if (!root) |
734 | root = root_mem_cgroup; | |
7d74b06f | 735 | |
9f3a0d09 | 736 | if (prev && !reclaim) |
5ac8fb31 | 737 | pos = prev; |
14067bb3 | 738 | |
9f3a0d09 JW |
739 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
740 | if (prev) | |
5ac8fb31 | 741 | goto out; |
694fbc0f | 742 | return root; |
9f3a0d09 | 743 | } |
14067bb3 | 744 | |
542f85f9 | 745 | rcu_read_lock(); |
5f578161 | 746 | |
5ac8fb31 | 747 | if (reclaim) { |
ef8f2327 | 748 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 749 | |
ef8f2327 | 750 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
5ac8fb31 JW |
751 | iter = &mz->iter[reclaim->priority]; |
752 | ||
753 | if (prev && reclaim->generation != iter->generation) | |
754 | goto out_unlock; | |
755 | ||
6df38689 | 756 | while (1) { |
4db0c3c2 | 757 | pos = READ_ONCE(iter->position); |
6df38689 VD |
758 | if (!pos || css_tryget(&pos->css)) |
759 | break; | |
5ac8fb31 | 760 | /* |
6df38689 VD |
761 | * css reference reached zero, so iter->position will |
762 | * be cleared by ->css_released. However, we should not | |
763 | * rely on this happening soon, because ->css_released | |
764 | * is called from a work queue, and by busy-waiting we | |
765 | * might block it. So we clear iter->position right | |
766 | * away. | |
5ac8fb31 | 767 | */ |
6df38689 VD |
768 | (void)cmpxchg(&iter->position, pos, NULL); |
769 | } | |
5ac8fb31 JW |
770 | } |
771 | ||
772 | if (pos) | |
773 | css = &pos->css; | |
774 | ||
775 | for (;;) { | |
776 | css = css_next_descendant_pre(css, &root->css); | |
777 | if (!css) { | |
778 | /* | |
779 | * Reclaimers share the hierarchy walk, and a | |
780 | * new one might jump in right at the end of | |
781 | * the hierarchy - make sure they see at least | |
782 | * one group and restart from the beginning. | |
783 | */ | |
784 | if (!prev) | |
785 | continue; | |
786 | break; | |
527a5ec9 | 787 | } |
7d74b06f | 788 | |
5ac8fb31 JW |
789 | /* |
790 | * Verify the css and acquire a reference. The root | |
791 | * is provided by the caller, so we know it's alive | |
792 | * and kicking, and don't take an extra reference. | |
793 | */ | |
794 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 795 | |
5ac8fb31 JW |
796 | if (css == &root->css) |
797 | break; | |
14067bb3 | 798 | |
0b8f73e1 JW |
799 | if (css_tryget(css)) |
800 | break; | |
9f3a0d09 | 801 | |
5ac8fb31 | 802 | memcg = NULL; |
9f3a0d09 | 803 | } |
5ac8fb31 JW |
804 | |
805 | if (reclaim) { | |
5ac8fb31 | 806 | /* |
6df38689 VD |
807 | * The position could have already been updated by a competing |
808 | * thread, so check that the value hasn't changed since we read | |
809 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 810 | */ |
6df38689 VD |
811 | (void)cmpxchg(&iter->position, pos, memcg); |
812 | ||
5ac8fb31 JW |
813 | if (pos) |
814 | css_put(&pos->css); | |
815 | ||
816 | if (!memcg) | |
817 | iter->generation++; | |
818 | else if (!prev) | |
819 | reclaim->generation = iter->generation; | |
9f3a0d09 | 820 | } |
5ac8fb31 | 821 | |
542f85f9 MH |
822 | out_unlock: |
823 | rcu_read_unlock(); | |
5ac8fb31 | 824 | out: |
c40046f3 MH |
825 | if (prev && prev != root) |
826 | css_put(&prev->css); | |
827 | ||
9f3a0d09 | 828 | return memcg; |
14067bb3 | 829 | } |
7d74b06f | 830 | |
5660048c JW |
831 | /** |
832 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
833 | * @root: hierarchy root | |
834 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
835 | */ | |
836 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
837 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
838 | { |
839 | if (!root) | |
840 | root = root_mem_cgroup; | |
841 | if (prev && prev != root) | |
842 | css_put(&prev->css); | |
843 | } | |
7d74b06f | 844 | |
6df38689 VD |
845 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
846 | { | |
847 | struct mem_cgroup *memcg = dead_memcg; | |
848 | struct mem_cgroup_reclaim_iter *iter; | |
ef8f2327 MG |
849 | struct mem_cgroup_per_node *mz; |
850 | int nid; | |
6df38689 VD |
851 | int i; |
852 | ||
853 | while ((memcg = parent_mem_cgroup(memcg))) { | |
854 | for_each_node(nid) { | |
ef8f2327 MG |
855 | mz = mem_cgroup_nodeinfo(memcg, nid); |
856 | for (i = 0; i <= DEF_PRIORITY; i++) { | |
857 | iter = &mz->iter[i]; | |
858 | cmpxchg(&iter->position, | |
859 | dead_memcg, NULL); | |
6df38689 VD |
860 | } |
861 | } | |
862 | } | |
863 | } | |
864 | ||
9f3a0d09 JW |
865 | /* |
866 | * Iteration constructs for visiting all cgroups (under a tree). If | |
867 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
868 | * be used for reference counting. | |
869 | */ | |
870 | #define for_each_mem_cgroup_tree(iter, root) \ | |
527a5ec9 | 871 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
9f3a0d09 | 872 | iter != NULL; \ |
527a5ec9 | 873 | iter = mem_cgroup_iter(root, iter, NULL)) |
711d3d2c | 874 | |
9f3a0d09 | 875 | #define for_each_mem_cgroup(iter) \ |
527a5ec9 | 876 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
9f3a0d09 | 877 | iter != NULL; \ |
527a5ec9 | 878 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
14067bb3 | 879 | |
7c5f64f8 VD |
880 | /** |
881 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
882 | * @memcg: hierarchy root | |
883 | * @fn: function to call for each task | |
884 | * @arg: argument passed to @fn | |
885 | * | |
886 | * This function iterates over tasks attached to @memcg or to any of its | |
887 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
888 | * value, the function breaks the iteration loop and returns the value. | |
889 | * Otherwise, it will iterate over all tasks and return 0. | |
890 | * | |
891 | * This function must not be called for the root memory cgroup. | |
892 | */ | |
893 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
894 | int (*fn)(struct task_struct *, void *), void *arg) | |
895 | { | |
896 | struct mem_cgroup *iter; | |
897 | int ret = 0; | |
898 | ||
899 | BUG_ON(memcg == root_mem_cgroup); | |
900 | ||
901 | for_each_mem_cgroup_tree(iter, memcg) { | |
902 | struct css_task_iter it; | |
903 | struct task_struct *task; | |
904 | ||
bc2fb7ed | 905 | css_task_iter_start(&iter->css, 0, &it); |
7c5f64f8 VD |
906 | while (!ret && (task = css_task_iter_next(&it))) |
907 | ret = fn(task, arg); | |
908 | css_task_iter_end(&it); | |
909 | if (ret) { | |
910 | mem_cgroup_iter_break(memcg, iter); | |
911 | break; | |
912 | } | |
913 | } | |
914 | return ret; | |
915 | } | |
916 | ||
925b7673 | 917 | /** |
dfe0e773 | 918 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
925b7673 | 919 | * @page: the page |
fa9add64 | 920 | * @zone: zone of the page |
dfe0e773 JW |
921 | * |
922 | * This function is only safe when following the LRU page isolation | |
923 | * and putback protocol: the LRU lock must be held, and the page must | |
924 | * either be PageLRU() or the caller must have isolated/allocated it. | |
925b7673 | 925 | */ |
599d0c95 | 926 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
08e552c6 | 927 | { |
ef8f2327 | 928 | struct mem_cgroup_per_node *mz; |
925b7673 | 929 | struct mem_cgroup *memcg; |
bea8c150 | 930 | struct lruvec *lruvec; |
6d12e2d8 | 931 | |
bea8c150 | 932 | if (mem_cgroup_disabled()) { |
599d0c95 | 933 | lruvec = &pgdat->lruvec; |
bea8c150 HD |
934 | goto out; |
935 | } | |
925b7673 | 936 | |
1306a85a | 937 | memcg = page->mem_cgroup; |
7512102c | 938 | /* |
dfe0e773 | 939 | * Swapcache readahead pages are added to the LRU - and |
29833315 | 940 | * possibly migrated - before they are charged. |
7512102c | 941 | */ |
29833315 JW |
942 | if (!memcg) |
943 | memcg = root_mem_cgroup; | |
7512102c | 944 | |
ef8f2327 | 945 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
bea8c150 HD |
946 | lruvec = &mz->lruvec; |
947 | out: | |
948 | /* | |
949 | * Since a node can be onlined after the mem_cgroup was created, | |
950 | * we have to be prepared to initialize lruvec->zone here; | |
951 | * and if offlined then reonlined, we need to reinitialize it. | |
952 | */ | |
599d0c95 MG |
953 | if (unlikely(lruvec->pgdat != pgdat)) |
954 | lruvec->pgdat = pgdat; | |
bea8c150 | 955 | return lruvec; |
08e552c6 | 956 | } |
b69408e8 | 957 | |
925b7673 | 958 | /** |
fa9add64 HD |
959 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
960 | * @lruvec: mem_cgroup per zone lru vector | |
961 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 962 | * @zid: zone id of the accounted pages |
fa9add64 | 963 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 964 | * |
ca707239 HD |
965 | * This function must be called under lru_lock, just before a page is added |
966 | * to or just after a page is removed from an lru list (that ordering being | |
967 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 968 | */ |
fa9add64 | 969 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 970 | int zid, int nr_pages) |
3f58a829 | 971 | { |
ef8f2327 | 972 | struct mem_cgroup_per_node *mz; |
fa9add64 | 973 | unsigned long *lru_size; |
ca707239 | 974 | long size; |
3f58a829 MK |
975 | |
976 | if (mem_cgroup_disabled()) | |
977 | return; | |
978 | ||
ef8f2327 | 979 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 980 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
981 | |
982 | if (nr_pages < 0) | |
983 | *lru_size += nr_pages; | |
984 | ||
985 | size = *lru_size; | |
b4536f0c MH |
986 | if (WARN_ONCE(size < 0, |
987 | "%s(%p, %d, %d): lru_size %ld\n", | |
988 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
989 | VM_BUG_ON(1); |
990 | *lru_size = 0; | |
991 | } | |
992 | ||
993 | if (nr_pages > 0) | |
994 | *lru_size += nr_pages; | |
08e552c6 | 995 | } |
544122e5 | 996 | |
2314b42d | 997 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
c3ac9a8a | 998 | { |
2314b42d | 999 | struct mem_cgroup *task_memcg; |
158e0a2d | 1000 | struct task_struct *p; |
ffbdccf5 | 1001 | bool ret; |
4c4a2214 | 1002 | |
158e0a2d | 1003 | p = find_lock_task_mm(task); |
de077d22 | 1004 | if (p) { |
2314b42d | 1005 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
de077d22 DR |
1006 | task_unlock(p); |
1007 | } else { | |
1008 | /* | |
1009 | * All threads may have already detached their mm's, but the oom | |
1010 | * killer still needs to detect if they have already been oom | |
1011 | * killed to prevent needlessly killing additional tasks. | |
1012 | */ | |
ffbdccf5 | 1013 | rcu_read_lock(); |
2314b42d JW |
1014 | task_memcg = mem_cgroup_from_task(task); |
1015 | css_get(&task_memcg->css); | |
ffbdccf5 | 1016 | rcu_read_unlock(); |
de077d22 | 1017 | } |
2314b42d JW |
1018 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
1019 | css_put(&task_memcg->css); | |
4c4a2214 DR |
1020 | return ret; |
1021 | } | |
1022 | ||
19942822 | 1023 | /** |
9d11ea9f | 1024 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1025 | * @memcg: the memory cgroup |
19942822 | 1026 | * |
9d11ea9f | 1027 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1028 | * pages. |
19942822 | 1029 | */ |
c0ff4b85 | 1030 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1031 | { |
3e32cb2e JW |
1032 | unsigned long margin = 0; |
1033 | unsigned long count; | |
1034 | unsigned long limit; | |
9d11ea9f | 1035 | |
3e32cb2e | 1036 | count = page_counter_read(&memcg->memory); |
4db0c3c2 | 1037 | limit = READ_ONCE(memcg->memory.limit); |
3e32cb2e JW |
1038 | if (count < limit) |
1039 | margin = limit - count; | |
1040 | ||
7941d214 | 1041 | if (do_memsw_account()) { |
3e32cb2e | 1042 | count = page_counter_read(&memcg->memsw); |
4db0c3c2 | 1043 | limit = READ_ONCE(memcg->memsw.limit); |
3e32cb2e JW |
1044 | if (count <= limit) |
1045 | margin = min(margin, limit - count); | |
cbedbac3 LR |
1046 | else |
1047 | margin = 0; | |
3e32cb2e JW |
1048 | } |
1049 | ||
1050 | return margin; | |
19942822 JW |
1051 | } |
1052 | ||
32047e2a | 1053 | /* |
bdcbb659 | 1054 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1055 | * |
bdcbb659 QH |
1056 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1057 | * moving cgroups. This is for waiting at high-memory pressure | |
1058 | * caused by "move". | |
32047e2a | 1059 | */ |
c0ff4b85 | 1060 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1061 | { |
2bd9bb20 KH |
1062 | struct mem_cgroup *from; |
1063 | struct mem_cgroup *to; | |
4b534334 | 1064 | bool ret = false; |
2bd9bb20 KH |
1065 | /* |
1066 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1067 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1068 | */ | |
1069 | spin_lock(&mc.lock); | |
1070 | from = mc.from; | |
1071 | to = mc.to; | |
1072 | if (!from) | |
1073 | goto unlock; | |
3e92041d | 1074 | |
2314b42d JW |
1075 | ret = mem_cgroup_is_descendant(from, memcg) || |
1076 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1077 | unlock: |
1078 | spin_unlock(&mc.lock); | |
4b534334 KH |
1079 | return ret; |
1080 | } | |
1081 | ||
c0ff4b85 | 1082 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1083 | { |
1084 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1085 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1086 | DEFINE_WAIT(wait); |
1087 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1088 | /* moving charge context might have finished. */ | |
1089 | if (mc.moving_task) | |
1090 | schedule(); | |
1091 | finish_wait(&mc.waitq, &wait); | |
1092 | return true; | |
1093 | } | |
1094 | } | |
1095 | return false; | |
1096 | } | |
1097 | ||
8ad6e404 | 1098 | static const unsigned int memcg1_stats[] = { |
71cd3113 JW |
1099 | MEMCG_CACHE, |
1100 | MEMCG_RSS, | |
1101 | MEMCG_RSS_HUGE, | |
1102 | NR_SHMEM, | |
1103 | NR_FILE_MAPPED, | |
1104 | NR_FILE_DIRTY, | |
1105 | NR_WRITEBACK, | |
1106 | MEMCG_SWAP, | |
1107 | }; | |
1108 | ||
1109 | static const char *const memcg1_stat_names[] = { | |
1110 | "cache", | |
1111 | "rss", | |
1112 | "rss_huge", | |
1113 | "shmem", | |
1114 | "mapped_file", | |
1115 | "dirty", | |
1116 | "writeback", | |
1117 | "swap", | |
1118 | }; | |
1119 | ||
58cf188e | 1120 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1121 | /** |
58cf188e | 1122 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
e222432b BS |
1123 | * @memcg: The memory cgroup that went over limit |
1124 | * @p: Task that is going to be killed | |
1125 | * | |
1126 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1127 | * enabled | |
1128 | */ | |
1129 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1130 | { | |
58cf188e SZ |
1131 | struct mem_cgroup *iter; |
1132 | unsigned int i; | |
e222432b | 1133 | |
e222432b BS |
1134 | rcu_read_lock(); |
1135 | ||
2415b9f5 BV |
1136 | if (p) { |
1137 | pr_info("Task in "); | |
1138 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); | |
1139 | pr_cont(" killed as a result of limit of "); | |
1140 | } else { | |
1141 | pr_info("Memory limit reached of cgroup "); | |
1142 | } | |
1143 | ||
e61734c5 | 1144 | pr_cont_cgroup_path(memcg->css.cgroup); |
0346dadb | 1145 | pr_cont("\n"); |
e222432b | 1146 | |
e222432b BS |
1147 | rcu_read_unlock(); |
1148 | ||
3e32cb2e JW |
1149 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1150 | K((u64)page_counter_read(&memcg->memory)), | |
1151 | K((u64)memcg->memory.limit), memcg->memory.failcnt); | |
1152 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1153 | K((u64)page_counter_read(&memcg->memsw)), | |
1154 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); | |
1155 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1156 | K((u64)page_counter_read(&memcg->kmem)), | |
1157 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); | |
58cf188e SZ |
1158 | |
1159 | for_each_mem_cgroup_tree(iter, memcg) { | |
e61734c5 TH |
1160 | pr_info("Memory cgroup stats for "); |
1161 | pr_cont_cgroup_path(iter->css.cgroup); | |
58cf188e SZ |
1162 | pr_cont(":"); |
1163 | ||
71cd3113 JW |
1164 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
1165 | if (memcg1_stats[i] == MEMCG_SWAP && !do_swap_account) | |
58cf188e | 1166 | continue; |
71cd3113 | 1167 | pr_cont(" %s:%luKB", memcg1_stat_names[i], |
ccda7f43 | 1168 | K(memcg_page_state(iter, memcg1_stats[i]))); |
58cf188e SZ |
1169 | } |
1170 | ||
1171 | for (i = 0; i < NR_LRU_LISTS; i++) | |
1172 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], | |
1173 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); | |
1174 | ||
1175 | pr_cont("\n"); | |
1176 | } | |
e222432b BS |
1177 | } |
1178 | ||
a63d83f4 DR |
1179 | /* |
1180 | * Return the memory (and swap, if configured) limit for a memcg. | |
1181 | */ | |
7c5f64f8 | 1182 | unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
a63d83f4 | 1183 | { |
3e32cb2e | 1184 | unsigned long limit; |
f3e8eb70 | 1185 | |
3e32cb2e | 1186 | limit = memcg->memory.limit; |
9a5a8f19 | 1187 | if (mem_cgroup_swappiness(memcg)) { |
3e32cb2e | 1188 | unsigned long memsw_limit; |
37e84351 | 1189 | unsigned long swap_limit; |
9a5a8f19 | 1190 | |
3e32cb2e | 1191 | memsw_limit = memcg->memsw.limit; |
37e84351 VD |
1192 | swap_limit = memcg->swap.limit; |
1193 | swap_limit = min(swap_limit, (unsigned long)total_swap_pages); | |
1194 | limit = min(limit + swap_limit, memsw_limit); | |
9a5a8f19 | 1195 | } |
9a5a8f19 | 1196 | return limit; |
a63d83f4 DR |
1197 | } |
1198 | ||
b6e6edcf | 1199 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1200 | int order) |
9cbb78bb | 1201 | { |
6e0fc46d DR |
1202 | struct oom_control oc = { |
1203 | .zonelist = NULL, | |
1204 | .nodemask = NULL, | |
2a966b77 | 1205 | .memcg = memcg, |
6e0fc46d DR |
1206 | .gfp_mask = gfp_mask, |
1207 | .order = order, | |
6e0fc46d | 1208 | }; |
7c5f64f8 | 1209 | bool ret; |
9cbb78bb | 1210 | |
dc56401f | 1211 | mutex_lock(&oom_lock); |
7c5f64f8 | 1212 | ret = out_of_memory(&oc); |
dc56401f | 1213 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1214 | return ret; |
9cbb78bb DR |
1215 | } |
1216 | ||
ae6e71d3 MC |
1217 | #if MAX_NUMNODES > 1 |
1218 | ||
4d0c066d KH |
1219 | /** |
1220 | * test_mem_cgroup_node_reclaimable | |
dad7557e | 1221 | * @memcg: the target memcg |
4d0c066d KH |
1222 | * @nid: the node ID to be checked. |
1223 | * @noswap : specify true here if the user wants flle only information. | |
1224 | * | |
1225 | * This function returns whether the specified memcg contains any | |
1226 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1227 | * pages in the node. | |
1228 | */ | |
c0ff4b85 | 1229 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
4d0c066d KH |
1230 | int nid, bool noswap) |
1231 | { | |
c0ff4b85 | 1232 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1233 | return true; |
1234 | if (noswap || !total_swap_pages) | |
1235 | return false; | |
c0ff4b85 | 1236 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1237 | return true; |
1238 | return false; | |
1239 | ||
1240 | } | |
889976db YH |
1241 | |
1242 | /* | |
1243 | * Always updating the nodemask is not very good - even if we have an empty | |
1244 | * list or the wrong list here, we can start from some node and traverse all | |
1245 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1246 | * | |
1247 | */ | |
c0ff4b85 | 1248 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
889976db YH |
1249 | { |
1250 | int nid; | |
453a9bf3 KH |
1251 | /* |
1252 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1253 | * pagein/pageout changes since the last update. | |
1254 | */ | |
c0ff4b85 | 1255 | if (!atomic_read(&memcg->numainfo_events)) |
453a9bf3 | 1256 | return; |
c0ff4b85 | 1257 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
889976db YH |
1258 | return; |
1259 | ||
889976db | 1260 | /* make a nodemask where this memcg uses memory from */ |
31aaea4a | 1261 | memcg->scan_nodes = node_states[N_MEMORY]; |
889976db | 1262 | |
31aaea4a | 1263 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
889976db | 1264 | |
c0ff4b85 R |
1265 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
1266 | node_clear(nid, memcg->scan_nodes); | |
889976db | 1267 | } |
453a9bf3 | 1268 | |
c0ff4b85 R |
1269 | atomic_set(&memcg->numainfo_events, 0); |
1270 | atomic_set(&memcg->numainfo_updating, 0); | |
889976db YH |
1271 | } |
1272 | ||
1273 | /* | |
1274 | * Selecting a node where we start reclaim from. Because what we need is just | |
1275 | * reducing usage counter, start from anywhere is O,K. Considering | |
1276 | * memory reclaim from current node, there are pros. and cons. | |
1277 | * | |
1278 | * Freeing memory from current node means freeing memory from a node which | |
1279 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1280 | * hit limits, it will see a contention on a node. But freeing from remote | |
1281 | * node means more costs for memory reclaim because of memory latency. | |
1282 | * | |
1283 | * Now, we use round-robin. Better algorithm is welcomed. | |
1284 | */ | |
c0ff4b85 | 1285 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1286 | { |
1287 | int node; | |
1288 | ||
c0ff4b85 R |
1289 | mem_cgroup_may_update_nodemask(memcg); |
1290 | node = memcg->last_scanned_node; | |
889976db | 1291 | |
0edaf86c | 1292 | node = next_node_in(node, memcg->scan_nodes); |
889976db | 1293 | /* |
fda3d69b MH |
1294 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
1295 | * last time it really checked all the LRUs due to rate limiting. | |
1296 | * Fallback to the current node in that case for simplicity. | |
889976db YH |
1297 | */ |
1298 | if (unlikely(node == MAX_NUMNODES)) | |
1299 | node = numa_node_id(); | |
1300 | ||
c0ff4b85 | 1301 | memcg->last_scanned_node = node; |
889976db YH |
1302 | return node; |
1303 | } | |
889976db | 1304 | #else |
c0ff4b85 | 1305 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
889976db YH |
1306 | { |
1307 | return 0; | |
1308 | } | |
1309 | #endif | |
1310 | ||
0608f43d | 1311 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1312 | pg_data_t *pgdat, |
0608f43d AM |
1313 | gfp_t gfp_mask, |
1314 | unsigned long *total_scanned) | |
1315 | { | |
1316 | struct mem_cgroup *victim = NULL; | |
1317 | int total = 0; | |
1318 | int loop = 0; | |
1319 | unsigned long excess; | |
1320 | unsigned long nr_scanned; | |
1321 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1322 | .pgdat = pgdat, |
0608f43d AM |
1323 | .priority = 0, |
1324 | }; | |
1325 | ||
3e32cb2e | 1326 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1327 | |
1328 | while (1) { | |
1329 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1330 | if (!victim) { | |
1331 | loop++; | |
1332 | if (loop >= 2) { | |
1333 | /* | |
1334 | * If we have not been able to reclaim | |
1335 | * anything, it might because there are | |
1336 | * no reclaimable pages under this hierarchy | |
1337 | */ | |
1338 | if (!total) | |
1339 | break; | |
1340 | /* | |
1341 | * We want to do more targeted reclaim. | |
1342 | * excess >> 2 is not to excessive so as to | |
1343 | * reclaim too much, nor too less that we keep | |
1344 | * coming back to reclaim from this cgroup | |
1345 | */ | |
1346 | if (total >= (excess >> 2) || | |
1347 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1348 | break; | |
1349 | } | |
1350 | continue; | |
1351 | } | |
a9dd0a83 | 1352 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1353 | pgdat, &nr_scanned); |
0608f43d | 1354 | *total_scanned += nr_scanned; |
3e32cb2e | 1355 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1356 | break; |
6d61ef40 | 1357 | } |
0608f43d AM |
1358 | mem_cgroup_iter_break(root_memcg, victim); |
1359 | return total; | |
6d61ef40 BS |
1360 | } |
1361 | ||
0056f4e6 JW |
1362 | #ifdef CONFIG_LOCKDEP |
1363 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1364 | .name = "memcg_oom_lock", | |
1365 | }; | |
1366 | #endif | |
1367 | ||
fb2a6fc5 JW |
1368 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1369 | ||
867578cb KH |
1370 | /* |
1371 | * Check OOM-Killer is already running under our hierarchy. | |
1372 | * If someone is running, return false. | |
1373 | */ | |
fb2a6fc5 | 1374 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1375 | { |
79dfdacc | 1376 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1377 | |
fb2a6fc5 JW |
1378 | spin_lock(&memcg_oom_lock); |
1379 | ||
9f3a0d09 | 1380 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1381 | if (iter->oom_lock) { |
79dfdacc MH |
1382 | /* |
1383 | * this subtree of our hierarchy is already locked | |
1384 | * so we cannot give a lock. | |
1385 | */ | |
79dfdacc | 1386 | failed = iter; |
9f3a0d09 JW |
1387 | mem_cgroup_iter_break(memcg, iter); |
1388 | break; | |
23751be0 JW |
1389 | } else |
1390 | iter->oom_lock = true; | |
7d74b06f | 1391 | } |
867578cb | 1392 | |
fb2a6fc5 JW |
1393 | if (failed) { |
1394 | /* | |
1395 | * OK, we failed to lock the whole subtree so we have | |
1396 | * to clean up what we set up to the failing subtree | |
1397 | */ | |
1398 | for_each_mem_cgroup_tree(iter, memcg) { | |
1399 | if (iter == failed) { | |
1400 | mem_cgroup_iter_break(memcg, iter); | |
1401 | break; | |
1402 | } | |
1403 | iter->oom_lock = false; | |
79dfdacc | 1404 | } |
0056f4e6 JW |
1405 | } else |
1406 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1407 | |
1408 | spin_unlock(&memcg_oom_lock); | |
1409 | ||
1410 | return !failed; | |
a636b327 | 1411 | } |
0b7f569e | 1412 | |
fb2a6fc5 | 1413 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1414 | { |
7d74b06f KH |
1415 | struct mem_cgroup *iter; |
1416 | ||
fb2a6fc5 | 1417 | spin_lock(&memcg_oom_lock); |
0056f4e6 | 1418 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
c0ff4b85 | 1419 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1420 | iter->oom_lock = false; |
fb2a6fc5 | 1421 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1422 | } |
1423 | ||
c0ff4b85 | 1424 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1425 | { |
1426 | struct mem_cgroup *iter; | |
1427 | ||
c2b42d3c | 1428 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1429 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1430 | iter->under_oom++; |
1431 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1432 | } |
1433 | ||
c0ff4b85 | 1434 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1435 | { |
1436 | struct mem_cgroup *iter; | |
1437 | ||
867578cb KH |
1438 | /* |
1439 | * When a new child is created while the hierarchy is under oom, | |
c2b42d3c | 1440 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
867578cb | 1441 | */ |
c2b42d3c | 1442 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1443 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1444 | if (iter->under_oom > 0) |
1445 | iter->under_oom--; | |
1446 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1447 | } |
1448 | ||
867578cb KH |
1449 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1450 | ||
dc98df5a | 1451 | struct oom_wait_info { |
d79154bb | 1452 | struct mem_cgroup *memcg; |
ac6424b9 | 1453 | wait_queue_entry_t wait; |
dc98df5a KH |
1454 | }; |
1455 | ||
ac6424b9 | 1456 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1457 | unsigned mode, int sync, void *arg) |
1458 | { | |
d79154bb HD |
1459 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1460 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1461 | struct oom_wait_info *oom_wait_info; |
1462 | ||
1463 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1464 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1465 | |
2314b42d JW |
1466 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1467 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1468 | return 0; |
dc98df5a KH |
1469 | return autoremove_wake_function(wait, mode, sync, arg); |
1470 | } | |
1471 | ||
c0ff4b85 | 1472 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1473 | { |
c2b42d3c TH |
1474 | /* |
1475 | * For the following lockless ->under_oom test, the only required | |
1476 | * guarantee is that it must see the state asserted by an OOM when | |
1477 | * this function is called as a result of userland actions | |
1478 | * triggered by the notification of the OOM. This is trivially | |
1479 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1480 | * triggering notification. | |
1481 | */ | |
1482 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1483 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1484 | } |
1485 | ||
3812c8c8 | 1486 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
0b7f569e | 1487 | { |
d0db7afa | 1488 | if (!current->memcg_may_oom) |
3812c8c8 | 1489 | return; |
867578cb | 1490 | /* |
49426420 JW |
1491 | * We are in the middle of the charge context here, so we |
1492 | * don't want to block when potentially sitting on a callstack | |
1493 | * that holds all kinds of filesystem and mm locks. | |
1494 | * | |
1495 | * Also, the caller may handle a failed allocation gracefully | |
1496 | * (like optional page cache readahead) and so an OOM killer | |
1497 | * invocation might not even be necessary. | |
1498 | * | |
1499 | * That's why we don't do anything here except remember the | |
1500 | * OOM context and then deal with it at the end of the page | |
1501 | * fault when the stack is unwound, the locks are released, | |
1502 | * and when we know whether the fault was overall successful. | |
867578cb | 1503 | */ |
49426420 | 1504 | css_get(&memcg->css); |
626ebc41 TH |
1505 | current->memcg_in_oom = memcg; |
1506 | current->memcg_oom_gfp_mask = mask; | |
1507 | current->memcg_oom_order = order; | |
3812c8c8 JW |
1508 | } |
1509 | ||
1510 | /** | |
1511 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1512 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1513 | * |
49426420 JW |
1514 | * This has to be called at the end of a page fault if the memcg OOM |
1515 | * handler was enabled. | |
3812c8c8 | 1516 | * |
49426420 | 1517 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1518 | * sleep on a waitqueue until the userspace task resolves the |
1519 | * situation. Sleeping directly in the charge context with all kinds | |
1520 | * of locks held is not a good idea, instead we remember an OOM state | |
1521 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1522 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1523 | * |
1524 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1525 | * completed, %false otherwise. |
3812c8c8 | 1526 | */ |
49426420 | 1527 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1528 | { |
626ebc41 | 1529 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1530 | struct oom_wait_info owait; |
49426420 | 1531 | bool locked; |
3812c8c8 JW |
1532 | |
1533 | /* OOM is global, do not handle */ | |
3812c8c8 | 1534 | if (!memcg) |
49426420 | 1535 | return false; |
3812c8c8 | 1536 | |
7c5f64f8 | 1537 | if (!handle) |
49426420 | 1538 | goto cleanup; |
3812c8c8 JW |
1539 | |
1540 | owait.memcg = memcg; | |
1541 | owait.wait.flags = 0; | |
1542 | owait.wait.func = memcg_oom_wake_function; | |
1543 | owait.wait.private = current; | |
2055da97 | 1544 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1545 | |
3812c8c8 | 1546 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1547 | mem_cgroup_mark_under_oom(memcg); |
1548 | ||
1549 | locked = mem_cgroup_oom_trylock(memcg); | |
1550 | ||
1551 | if (locked) | |
1552 | mem_cgroup_oom_notify(memcg); | |
1553 | ||
1554 | if (locked && !memcg->oom_kill_disable) { | |
1555 | mem_cgroup_unmark_under_oom(memcg); | |
1556 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1557 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1558 | current->memcg_oom_order); | |
49426420 | 1559 | } else { |
3812c8c8 | 1560 | schedule(); |
49426420 JW |
1561 | mem_cgroup_unmark_under_oom(memcg); |
1562 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1563 | } | |
1564 | ||
1565 | if (locked) { | |
fb2a6fc5 JW |
1566 | mem_cgroup_oom_unlock(memcg); |
1567 | /* | |
1568 | * There is no guarantee that an OOM-lock contender | |
1569 | * sees the wakeups triggered by the OOM kill | |
1570 | * uncharges. Wake any sleepers explicitely. | |
1571 | */ | |
1572 | memcg_oom_recover(memcg); | |
1573 | } | |
49426420 | 1574 | cleanup: |
626ebc41 | 1575 | current->memcg_in_oom = NULL; |
3812c8c8 | 1576 | css_put(&memcg->css); |
867578cb | 1577 | return true; |
0b7f569e KH |
1578 | } |
1579 | ||
d7365e78 | 1580 | /** |
81f8c3a4 JW |
1581 | * lock_page_memcg - lock a page->mem_cgroup binding |
1582 | * @page: the page | |
32047e2a | 1583 | * |
81f8c3a4 | 1584 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
1585 | * another cgroup. |
1586 | * | |
1587 | * It ensures lifetime of the returned memcg. Caller is responsible | |
1588 | * for the lifetime of the page; __unlock_page_memcg() is available | |
1589 | * when @page might get freed inside the locked section. | |
d69b042f | 1590 | */ |
739f79fc | 1591 | struct mem_cgroup *lock_page_memcg(struct page *page) |
89c06bd5 KH |
1592 | { |
1593 | struct mem_cgroup *memcg; | |
6de22619 | 1594 | unsigned long flags; |
89c06bd5 | 1595 | |
6de22619 JW |
1596 | /* |
1597 | * The RCU lock is held throughout the transaction. The fast | |
1598 | * path can get away without acquiring the memcg->move_lock | |
1599 | * because page moving starts with an RCU grace period. | |
739f79fc JW |
1600 | * |
1601 | * The RCU lock also protects the memcg from being freed when | |
1602 | * the page state that is going to change is the only thing | |
1603 | * preventing the page itself from being freed. E.g. writeback | |
1604 | * doesn't hold a page reference and relies on PG_writeback to | |
1605 | * keep off truncation, migration and so forth. | |
1606 | */ | |
d7365e78 JW |
1607 | rcu_read_lock(); |
1608 | ||
1609 | if (mem_cgroup_disabled()) | |
739f79fc | 1610 | return NULL; |
89c06bd5 | 1611 | again: |
1306a85a | 1612 | memcg = page->mem_cgroup; |
29833315 | 1613 | if (unlikely(!memcg)) |
739f79fc | 1614 | return NULL; |
d7365e78 | 1615 | |
bdcbb659 | 1616 | if (atomic_read(&memcg->moving_account) <= 0) |
739f79fc | 1617 | return memcg; |
89c06bd5 | 1618 | |
6de22619 | 1619 | spin_lock_irqsave(&memcg->move_lock, flags); |
1306a85a | 1620 | if (memcg != page->mem_cgroup) { |
6de22619 | 1621 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
1622 | goto again; |
1623 | } | |
6de22619 JW |
1624 | |
1625 | /* | |
1626 | * When charge migration first begins, we can have locked and | |
1627 | * unlocked page stat updates happening concurrently. Track | |
81f8c3a4 | 1628 | * the task who has the lock for unlock_page_memcg(). |
6de22619 JW |
1629 | */ |
1630 | memcg->move_lock_task = current; | |
1631 | memcg->move_lock_flags = flags; | |
d7365e78 | 1632 | |
739f79fc | 1633 | return memcg; |
89c06bd5 | 1634 | } |
81f8c3a4 | 1635 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 1636 | |
d7365e78 | 1637 | /** |
739f79fc JW |
1638 | * __unlock_page_memcg - unlock and unpin a memcg |
1639 | * @memcg: the memcg | |
1640 | * | |
1641 | * Unlock and unpin a memcg returned by lock_page_memcg(). | |
d7365e78 | 1642 | */ |
739f79fc | 1643 | void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 1644 | { |
6de22619 JW |
1645 | if (memcg && memcg->move_lock_task == current) { |
1646 | unsigned long flags = memcg->move_lock_flags; | |
1647 | ||
1648 | memcg->move_lock_task = NULL; | |
1649 | memcg->move_lock_flags = 0; | |
1650 | ||
1651 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
1652 | } | |
89c06bd5 | 1653 | |
d7365e78 | 1654 | rcu_read_unlock(); |
89c06bd5 | 1655 | } |
739f79fc JW |
1656 | |
1657 | /** | |
1658 | * unlock_page_memcg - unlock a page->mem_cgroup binding | |
1659 | * @page: the page | |
1660 | */ | |
1661 | void unlock_page_memcg(struct page *page) | |
1662 | { | |
1663 | __unlock_page_memcg(page->mem_cgroup); | |
1664 | } | |
81f8c3a4 | 1665 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 1666 | |
cdec2e42 KH |
1667 | struct memcg_stock_pcp { |
1668 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1669 | unsigned int nr_pages; |
cdec2e42 | 1670 | struct work_struct work; |
26fe6168 | 1671 | unsigned long flags; |
a0db00fc | 1672 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
1673 | }; |
1674 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1675 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 1676 | |
a0956d54 SS |
1677 | /** |
1678 | * consume_stock: Try to consume stocked charge on this cpu. | |
1679 | * @memcg: memcg to consume from. | |
1680 | * @nr_pages: how many pages to charge. | |
1681 | * | |
1682 | * The charges will only happen if @memcg matches the current cpu's memcg | |
1683 | * stock, and at least @nr_pages are available in that stock. Failure to | |
1684 | * service an allocation will refill the stock. | |
1685 | * | |
1686 | * returns true if successful, false otherwise. | |
cdec2e42 | 1687 | */ |
a0956d54 | 1688 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
1689 | { |
1690 | struct memcg_stock_pcp *stock; | |
db2ba40c | 1691 | unsigned long flags; |
3e32cb2e | 1692 | bool ret = false; |
cdec2e42 | 1693 | |
a983b5eb | 1694 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 1695 | return ret; |
a0956d54 | 1696 | |
db2ba40c JW |
1697 | local_irq_save(flags); |
1698 | ||
1699 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 1700 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 1701 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
1702 | ret = true; |
1703 | } | |
db2ba40c JW |
1704 | |
1705 | local_irq_restore(flags); | |
1706 | ||
cdec2e42 KH |
1707 | return ret; |
1708 | } | |
1709 | ||
1710 | /* | |
3e32cb2e | 1711 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
1712 | */ |
1713 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1714 | { | |
1715 | struct mem_cgroup *old = stock->cached; | |
1716 | ||
11c9ea4e | 1717 | if (stock->nr_pages) { |
3e32cb2e | 1718 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 1719 | if (do_memsw_account()) |
3e32cb2e | 1720 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
e8ea14cc | 1721 | css_put_many(&old->css, stock->nr_pages); |
11c9ea4e | 1722 | stock->nr_pages = 0; |
cdec2e42 KH |
1723 | } |
1724 | stock->cached = NULL; | |
cdec2e42 KH |
1725 | } |
1726 | ||
cdec2e42 KH |
1727 | static void drain_local_stock(struct work_struct *dummy) |
1728 | { | |
db2ba40c JW |
1729 | struct memcg_stock_pcp *stock; |
1730 | unsigned long flags; | |
1731 | ||
72f0184c MH |
1732 | /* |
1733 | * The only protection from memory hotplug vs. drain_stock races is | |
1734 | * that we always operate on local CPU stock here with IRQ disabled | |
1735 | */ | |
db2ba40c JW |
1736 | local_irq_save(flags); |
1737 | ||
1738 | stock = this_cpu_ptr(&memcg_stock); | |
cdec2e42 | 1739 | drain_stock(stock); |
26fe6168 | 1740 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
1741 | |
1742 | local_irq_restore(flags); | |
cdec2e42 KH |
1743 | } |
1744 | ||
1745 | /* | |
3e32cb2e | 1746 | * Cache charges(val) to local per_cpu area. |
320cc51d | 1747 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1748 | */ |
c0ff4b85 | 1749 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 1750 | { |
db2ba40c JW |
1751 | struct memcg_stock_pcp *stock; |
1752 | unsigned long flags; | |
1753 | ||
1754 | local_irq_save(flags); | |
cdec2e42 | 1755 | |
db2ba40c | 1756 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 1757 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 1758 | drain_stock(stock); |
c0ff4b85 | 1759 | stock->cached = memcg; |
cdec2e42 | 1760 | } |
11c9ea4e | 1761 | stock->nr_pages += nr_pages; |
db2ba40c | 1762 | |
a983b5eb | 1763 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
1764 | drain_stock(stock); |
1765 | ||
db2ba40c | 1766 | local_irq_restore(flags); |
cdec2e42 KH |
1767 | } |
1768 | ||
1769 | /* | |
c0ff4b85 | 1770 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 1771 | * of the hierarchy under it. |
cdec2e42 | 1772 | */ |
6d3d6aa2 | 1773 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 1774 | { |
26fe6168 | 1775 | int cpu, curcpu; |
d38144b7 | 1776 | |
6d3d6aa2 JW |
1777 | /* If someone's already draining, avoid adding running more workers. */ |
1778 | if (!mutex_trylock(&percpu_charge_mutex)) | |
1779 | return; | |
72f0184c MH |
1780 | /* |
1781 | * Notify other cpus that system-wide "drain" is running | |
1782 | * We do not care about races with the cpu hotplug because cpu down | |
1783 | * as well as workers from this path always operate on the local | |
1784 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
1785 | */ | |
5af12d0e | 1786 | curcpu = get_cpu(); |
cdec2e42 KH |
1787 | for_each_online_cpu(cpu) { |
1788 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 1789 | struct mem_cgroup *memcg; |
26fe6168 | 1790 | |
c0ff4b85 | 1791 | memcg = stock->cached; |
72f0184c | 1792 | if (!memcg || !stock->nr_pages || !css_tryget(&memcg->css)) |
26fe6168 | 1793 | continue; |
72f0184c MH |
1794 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) { |
1795 | css_put(&memcg->css); | |
3e92041d | 1796 | continue; |
72f0184c | 1797 | } |
d1a05b69 MH |
1798 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
1799 | if (cpu == curcpu) | |
1800 | drain_local_stock(&stock->work); | |
1801 | else | |
1802 | schedule_work_on(cpu, &stock->work); | |
1803 | } | |
72f0184c | 1804 | css_put(&memcg->css); |
cdec2e42 | 1805 | } |
5af12d0e | 1806 | put_cpu(); |
9f50fad6 | 1807 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
1808 | } |
1809 | ||
308167fc | 1810 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
cdec2e42 | 1811 | { |
cdec2e42 | 1812 | struct memcg_stock_pcp *stock; |
a983b5eb | 1813 | struct mem_cgroup *memcg; |
cdec2e42 | 1814 | |
cdec2e42 KH |
1815 | stock = &per_cpu(memcg_stock, cpu); |
1816 | drain_stock(stock); | |
a983b5eb JW |
1817 | |
1818 | for_each_mem_cgroup(memcg) { | |
1819 | int i; | |
1820 | ||
1821 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
1822 | int nid; | |
1823 | long x; | |
1824 | ||
1825 | x = this_cpu_xchg(memcg->stat_cpu->count[i], 0); | |
1826 | if (x) | |
1827 | atomic_long_add(x, &memcg->stat[i]); | |
1828 | ||
1829 | if (i >= NR_VM_NODE_STAT_ITEMS) | |
1830 | continue; | |
1831 | ||
1832 | for_each_node(nid) { | |
1833 | struct mem_cgroup_per_node *pn; | |
1834 | ||
1835 | pn = mem_cgroup_nodeinfo(memcg, nid); | |
1836 | x = this_cpu_xchg(pn->lruvec_stat_cpu->count[i], 0); | |
1837 | if (x) | |
1838 | atomic_long_add(x, &pn->lruvec_stat[i]); | |
1839 | } | |
1840 | } | |
1841 | ||
1842 | for (i = 0; i < MEMCG_NR_EVENTS; i++) { | |
1843 | long x; | |
1844 | ||
1845 | x = this_cpu_xchg(memcg->stat_cpu->events[i], 0); | |
1846 | if (x) | |
1847 | atomic_long_add(x, &memcg->events[i]); | |
1848 | } | |
1849 | } | |
1850 | ||
308167fc | 1851 | return 0; |
cdec2e42 KH |
1852 | } |
1853 | ||
f7e1cb6e JW |
1854 | static void reclaim_high(struct mem_cgroup *memcg, |
1855 | unsigned int nr_pages, | |
1856 | gfp_t gfp_mask) | |
1857 | { | |
1858 | do { | |
1859 | if (page_counter_read(&memcg->memory) <= memcg->high) | |
1860 | continue; | |
31176c78 | 1861 | mem_cgroup_event(memcg, MEMCG_HIGH); |
f7e1cb6e JW |
1862 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
1863 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1864 | } | |
1865 | ||
1866 | static void high_work_func(struct work_struct *work) | |
1867 | { | |
1868 | struct mem_cgroup *memcg; | |
1869 | ||
1870 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 1871 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
1872 | } |
1873 | ||
b23afb93 TH |
1874 | /* |
1875 | * Scheduled by try_charge() to be executed from the userland return path | |
1876 | * and reclaims memory over the high limit. | |
1877 | */ | |
1878 | void mem_cgroup_handle_over_high(void) | |
1879 | { | |
1880 | unsigned int nr_pages = current->memcg_nr_pages_over_high; | |
f7e1cb6e | 1881 | struct mem_cgroup *memcg; |
b23afb93 TH |
1882 | |
1883 | if (likely(!nr_pages)) | |
1884 | return; | |
1885 | ||
f7e1cb6e JW |
1886 | memcg = get_mem_cgroup_from_mm(current->mm); |
1887 | reclaim_high(memcg, nr_pages, GFP_KERNEL); | |
b23afb93 TH |
1888 | css_put(&memcg->css); |
1889 | current->memcg_nr_pages_over_high = 0; | |
1890 | } | |
1891 | ||
00501b53 JW |
1892 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
1893 | unsigned int nr_pages) | |
8a9f3ccd | 1894 | { |
a983b5eb | 1895 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
9b130619 | 1896 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
6539cc05 | 1897 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 1898 | struct page_counter *counter; |
6539cc05 | 1899 | unsigned long nr_reclaimed; |
b70a2a21 JW |
1900 | bool may_swap = true; |
1901 | bool drained = false; | |
a636b327 | 1902 | |
ce00a967 | 1903 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 1904 | return 0; |
6539cc05 | 1905 | retry: |
b6b6cc72 | 1906 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 1907 | return 0; |
8a9f3ccd | 1908 | |
7941d214 | 1909 | if (!do_memsw_account() || |
6071ca52 JW |
1910 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
1911 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 1912 | goto done_restock; |
7941d214 | 1913 | if (do_memsw_account()) |
3e32cb2e JW |
1914 | page_counter_uncharge(&memcg->memsw, batch); |
1915 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 1916 | } else { |
3e32cb2e | 1917 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 1918 | may_swap = false; |
3fbe7244 | 1919 | } |
7a81b88c | 1920 | |
6539cc05 JW |
1921 | if (batch > nr_pages) { |
1922 | batch = nr_pages; | |
1923 | goto retry; | |
1924 | } | |
6d61ef40 | 1925 | |
06b078fc JW |
1926 | /* |
1927 | * Unlike in global OOM situations, memcg is not in a physical | |
1928 | * memory shortage. Allow dying and OOM-killed tasks to | |
1929 | * bypass the last charges so that they can exit quickly and | |
1930 | * free their memory. | |
1931 | */ | |
da99ecf1 | 1932 | if (unlikely(tsk_is_oom_victim(current) || |
06b078fc JW |
1933 | fatal_signal_pending(current) || |
1934 | current->flags & PF_EXITING)) | |
10d53c74 | 1935 | goto force; |
06b078fc | 1936 | |
89a28483 JW |
1937 | /* |
1938 | * Prevent unbounded recursion when reclaim operations need to | |
1939 | * allocate memory. This might exceed the limits temporarily, | |
1940 | * but we prefer facilitating memory reclaim and getting back | |
1941 | * under the limit over triggering OOM kills in these cases. | |
1942 | */ | |
1943 | if (unlikely(current->flags & PF_MEMALLOC)) | |
1944 | goto force; | |
1945 | ||
06b078fc JW |
1946 | if (unlikely(task_in_memcg_oom(current))) |
1947 | goto nomem; | |
1948 | ||
d0164adc | 1949 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 1950 | goto nomem; |
4b534334 | 1951 | |
31176c78 | 1952 | mem_cgroup_event(mem_over_limit, MEMCG_MAX); |
241994ed | 1953 | |
b70a2a21 JW |
1954 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
1955 | gfp_mask, may_swap); | |
6539cc05 | 1956 | |
61e02c74 | 1957 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 1958 | goto retry; |
28c34c29 | 1959 | |
b70a2a21 | 1960 | if (!drained) { |
6d3d6aa2 | 1961 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
1962 | drained = true; |
1963 | goto retry; | |
1964 | } | |
1965 | ||
28c34c29 JW |
1966 | if (gfp_mask & __GFP_NORETRY) |
1967 | goto nomem; | |
6539cc05 JW |
1968 | /* |
1969 | * Even though the limit is exceeded at this point, reclaim | |
1970 | * may have been able to free some pages. Retry the charge | |
1971 | * before killing the task. | |
1972 | * | |
1973 | * Only for regular pages, though: huge pages are rather | |
1974 | * unlikely to succeed so close to the limit, and we fall back | |
1975 | * to regular pages anyway in case of failure. | |
1976 | */ | |
61e02c74 | 1977 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
1978 | goto retry; |
1979 | /* | |
1980 | * At task move, charge accounts can be doubly counted. So, it's | |
1981 | * better to wait until the end of task_move if something is going on. | |
1982 | */ | |
1983 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
1984 | goto retry; | |
1985 | ||
9b130619 JW |
1986 | if (nr_retries--) |
1987 | goto retry; | |
1988 | ||
06b078fc | 1989 | if (gfp_mask & __GFP_NOFAIL) |
10d53c74 | 1990 | goto force; |
06b078fc | 1991 | |
6539cc05 | 1992 | if (fatal_signal_pending(current)) |
10d53c74 | 1993 | goto force; |
6539cc05 | 1994 | |
31176c78 | 1995 | mem_cgroup_event(mem_over_limit, MEMCG_OOM); |
241994ed | 1996 | |
3608de07 JM |
1997 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
1998 | get_order(nr_pages * PAGE_SIZE)); | |
7a81b88c | 1999 | nomem: |
6d1fdc48 | 2000 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2001 | return -ENOMEM; |
10d53c74 TH |
2002 | force: |
2003 | /* | |
2004 | * The allocation either can't fail or will lead to more memory | |
2005 | * being freed very soon. Allow memory usage go over the limit | |
2006 | * temporarily by force charging it. | |
2007 | */ | |
2008 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2009 | if (do_memsw_account()) |
10d53c74 TH |
2010 | page_counter_charge(&memcg->memsw, nr_pages); |
2011 | css_get_many(&memcg->css, nr_pages); | |
2012 | ||
2013 | return 0; | |
6539cc05 JW |
2014 | |
2015 | done_restock: | |
e8ea14cc | 2016 | css_get_many(&memcg->css, batch); |
6539cc05 JW |
2017 | if (batch > nr_pages) |
2018 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2019 | |
241994ed | 2020 | /* |
b23afb93 TH |
2021 | * If the hierarchy is above the normal consumption range, schedule |
2022 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2023 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2024 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2025 | * not recorded as it most likely matches current's and won't | |
2026 | * change in the meantime. As high limit is checked again before | |
2027 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2028 | */ |
2029 | do { | |
b23afb93 | 2030 | if (page_counter_read(&memcg->memory) > memcg->high) { |
f7e1cb6e JW |
2031 | /* Don't bother a random interrupted task */ |
2032 | if (in_interrupt()) { | |
2033 | schedule_work(&memcg->high_work); | |
2034 | break; | |
2035 | } | |
9516a18a | 2036 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2037 | set_notify_resume(current); |
2038 | break; | |
2039 | } | |
241994ed | 2040 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2041 | |
2042 | return 0; | |
7a81b88c | 2043 | } |
8a9f3ccd | 2044 | |
00501b53 | 2045 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2046 | { |
ce00a967 JW |
2047 | if (mem_cgroup_is_root(memcg)) |
2048 | return; | |
2049 | ||
3e32cb2e | 2050 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2051 | if (do_memsw_account()) |
3e32cb2e | 2052 | page_counter_uncharge(&memcg->memsw, nr_pages); |
ce00a967 | 2053 | |
e8ea14cc | 2054 | css_put_many(&memcg->css, nr_pages); |
d01dd17f KH |
2055 | } |
2056 | ||
0a31bc97 JW |
2057 | static void lock_page_lru(struct page *page, int *isolated) |
2058 | { | |
2059 | struct zone *zone = page_zone(page); | |
2060 | ||
a52633d8 | 2061 | spin_lock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2062 | if (PageLRU(page)) { |
2063 | struct lruvec *lruvec; | |
2064 | ||
599d0c95 | 2065 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2066 | ClearPageLRU(page); |
2067 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
2068 | *isolated = 1; | |
2069 | } else | |
2070 | *isolated = 0; | |
2071 | } | |
2072 | ||
2073 | static void unlock_page_lru(struct page *page, int isolated) | |
2074 | { | |
2075 | struct zone *zone = page_zone(page); | |
2076 | ||
2077 | if (isolated) { | |
2078 | struct lruvec *lruvec; | |
2079 | ||
599d0c95 | 2080 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
0a31bc97 JW |
2081 | VM_BUG_ON_PAGE(PageLRU(page), page); |
2082 | SetPageLRU(page); | |
2083 | add_page_to_lru_list(page, lruvec, page_lru(page)); | |
2084 | } | |
a52633d8 | 2085 | spin_unlock_irq(zone_lru_lock(zone)); |
0a31bc97 JW |
2086 | } |
2087 | ||
00501b53 | 2088 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
6abb5a86 | 2089 | bool lrucare) |
7a81b88c | 2090 | { |
0a31bc97 | 2091 | int isolated; |
9ce70c02 | 2092 | |
1306a85a | 2093 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
9ce70c02 HD |
2094 | |
2095 | /* | |
2096 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page | |
2097 | * may already be on some other mem_cgroup's LRU. Take care of it. | |
2098 | */ | |
0a31bc97 JW |
2099 | if (lrucare) |
2100 | lock_page_lru(page, &isolated); | |
9ce70c02 | 2101 | |
0a31bc97 JW |
2102 | /* |
2103 | * Nobody should be changing or seriously looking at | |
1306a85a | 2104 | * page->mem_cgroup at this point: |
0a31bc97 JW |
2105 | * |
2106 | * - the page is uncharged | |
2107 | * | |
2108 | * - the page is off-LRU | |
2109 | * | |
2110 | * - an anonymous fault has exclusive page access, except for | |
2111 | * a locked page table | |
2112 | * | |
2113 | * - a page cache insertion, a swapin fault, or a migration | |
2114 | * have the page locked | |
2115 | */ | |
1306a85a | 2116 | page->mem_cgroup = memcg; |
9ce70c02 | 2117 | |
0a31bc97 JW |
2118 | if (lrucare) |
2119 | unlock_page_lru(page, isolated); | |
7a81b88c | 2120 | } |
66e1707b | 2121 | |
127424c8 | 2122 | #ifndef CONFIG_SLOB |
f3bb3043 | 2123 | static int memcg_alloc_cache_id(void) |
55007d84 | 2124 | { |
f3bb3043 VD |
2125 | int id, size; |
2126 | int err; | |
2127 | ||
dbcf73e2 | 2128 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2129 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2130 | if (id < 0) | |
2131 | return id; | |
55007d84 | 2132 | |
dbcf73e2 | 2133 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2134 | return id; |
2135 | ||
2136 | /* | |
2137 | * There's no space for the new id in memcg_caches arrays, | |
2138 | * so we have to grow them. | |
2139 | */ | |
05257a1a | 2140 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2141 | |
2142 | size = 2 * (id + 1); | |
55007d84 GC |
2143 | if (size < MEMCG_CACHES_MIN_SIZE) |
2144 | size = MEMCG_CACHES_MIN_SIZE; | |
2145 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2146 | size = MEMCG_CACHES_MAX_SIZE; | |
2147 | ||
f3bb3043 | 2148 | err = memcg_update_all_caches(size); |
60d3fd32 VD |
2149 | if (!err) |
2150 | err = memcg_update_all_list_lrus(size); | |
05257a1a VD |
2151 | if (!err) |
2152 | memcg_nr_cache_ids = size; | |
2153 | ||
2154 | up_write(&memcg_cache_ids_sem); | |
2155 | ||
f3bb3043 | 2156 | if (err) { |
dbcf73e2 | 2157 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2158 | return err; |
2159 | } | |
2160 | return id; | |
2161 | } | |
2162 | ||
2163 | static void memcg_free_cache_id(int id) | |
2164 | { | |
dbcf73e2 | 2165 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2166 | } |
2167 | ||
d5b3cf71 | 2168 | struct memcg_kmem_cache_create_work { |
5722d094 VD |
2169 | struct mem_cgroup *memcg; |
2170 | struct kmem_cache *cachep; | |
2171 | struct work_struct work; | |
2172 | }; | |
2173 | ||
d5b3cf71 | 2174 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
d7f25f8a | 2175 | { |
d5b3cf71 VD |
2176 | struct memcg_kmem_cache_create_work *cw = |
2177 | container_of(w, struct memcg_kmem_cache_create_work, work); | |
5722d094 VD |
2178 | struct mem_cgroup *memcg = cw->memcg; |
2179 | struct kmem_cache *cachep = cw->cachep; | |
d7f25f8a | 2180 | |
d5b3cf71 | 2181 | memcg_create_kmem_cache(memcg, cachep); |
bd673145 | 2182 | |
5722d094 | 2183 | css_put(&memcg->css); |
d7f25f8a GC |
2184 | kfree(cw); |
2185 | } | |
2186 | ||
2187 | /* | |
2188 | * Enqueue the creation of a per-memcg kmem_cache. | |
d7f25f8a | 2189 | */ |
d5b3cf71 VD |
2190 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2191 | struct kmem_cache *cachep) | |
d7f25f8a | 2192 | { |
d5b3cf71 | 2193 | struct memcg_kmem_cache_create_work *cw; |
d7f25f8a | 2194 | |
776ed0f0 | 2195 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
8135be5a | 2196 | if (!cw) |
d7f25f8a | 2197 | return; |
8135be5a VD |
2198 | |
2199 | css_get(&memcg->css); | |
d7f25f8a GC |
2200 | |
2201 | cw->memcg = memcg; | |
2202 | cw->cachep = cachep; | |
d5b3cf71 | 2203 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
d7f25f8a | 2204 | |
17cc4dfe | 2205 | queue_work(memcg_kmem_cache_wq, &cw->work); |
d7f25f8a GC |
2206 | } |
2207 | ||
d5b3cf71 VD |
2208 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
2209 | struct kmem_cache *cachep) | |
0e9d92f2 GC |
2210 | { |
2211 | /* | |
2212 | * We need to stop accounting when we kmalloc, because if the | |
2213 | * corresponding kmalloc cache is not yet created, the first allocation | |
d5b3cf71 | 2214 | * in __memcg_schedule_kmem_cache_create will recurse. |
0e9d92f2 GC |
2215 | * |
2216 | * However, it is better to enclose the whole function. Depending on | |
2217 | * the debugging options enabled, INIT_WORK(), for instance, can | |
2218 | * trigger an allocation. This too, will make us recurse. Because at | |
2219 | * this point we can't allow ourselves back into memcg_kmem_get_cache, | |
2220 | * the safest choice is to do it like this, wrapping the whole function. | |
2221 | */ | |
6f185c29 | 2222 | current->memcg_kmem_skip_account = 1; |
d5b3cf71 | 2223 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
6f185c29 | 2224 | current->memcg_kmem_skip_account = 0; |
0e9d92f2 | 2225 | } |
c67a8a68 | 2226 | |
45264778 VD |
2227 | static inline bool memcg_kmem_bypass(void) |
2228 | { | |
2229 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
2230 | return true; | |
2231 | return false; | |
2232 | } | |
2233 | ||
2234 | /** | |
2235 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation | |
2236 | * @cachep: the original global kmem cache | |
2237 | * | |
d7f25f8a GC |
2238 | * Return the kmem_cache we're supposed to use for a slab allocation. |
2239 | * We try to use the current memcg's version of the cache. | |
2240 | * | |
45264778 VD |
2241 | * If the cache does not exist yet, if we are the first user of it, we |
2242 | * create it asynchronously in a workqueue and let the current allocation | |
2243 | * go through with the original cache. | |
d7f25f8a | 2244 | * |
45264778 VD |
2245 | * This function takes a reference to the cache it returns to assure it |
2246 | * won't get destroyed while we are working with it. Once the caller is | |
2247 | * done with it, memcg_kmem_put_cache() must be called to release the | |
2248 | * reference. | |
d7f25f8a | 2249 | */ |
45264778 | 2250 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
d7f25f8a GC |
2251 | { |
2252 | struct mem_cgroup *memcg; | |
959c8963 | 2253 | struct kmem_cache *memcg_cachep; |
2a4db7eb | 2254 | int kmemcg_id; |
d7f25f8a | 2255 | |
f7ce3190 | 2256 | VM_BUG_ON(!is_root_cache(cachep)); |
d7f25f8a | 2257 | |
45264778 | 2258 | if (memcg_kmem_bypass()) |
230e9fc2 VD |
2259 | return cachep; |
2260 | ||
9d100c5e | 2261 | if (current->memcg_kmem_skip_account) |
0e9d92f2 GC |
2262 | return cachep; |
2263 | ||
8135be5a | 2264 | memcg = get_mem_cgroup_from_mm(current->mm); |
4db0c3c2 | 2265 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
2a4db7eb | 2266 | if (kmemcg_id < 0) |
ca0dde97 | 2267 | goto out; |
d7f25f8a | 2268 | |
2a4db7eb | 2269 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
8135be5a VD |
2270 | if (likely(memcg_cachep)) |
2271 | return memcg_cachep; | |
ca0dde97 LZ |
2272 | |
2273 | /* | |
2274 | * If we are in a safe context (can wait, and not in interrupt | |
2275 | * context), we could be be predictable and return right away. | |
2276 | * This would guarantee that the allocation being performed | |
2277 | * already belongs in the new cache. | |
2278 | * | |
2279 | * However, there are some clashes that can arrive from locking. | |
2280 | * For instance, because we acquire the slab_mutex while doing | |
776ed0f0 VD |
2281 | * memcg_create_kmem_cache, this means no further allocation |
2282 | * could happen with the slab_mutex held. So it's better to | |
2283 | * defer everything. | |
ca0dde97 | 2284 | */ |
d5b3cf71 | 2285 | memcg_schedule_kmem_cache_create(memcg, cachep); |
ca0dde97 | 2286 | out: |
8135be5a | 2287 | css_put(&memcg->css); |
ca0dde97 | 2288 | return cachep; |
d7f25f8a | 2289 | } |
d7f25f8a | 2290 | |
45264778 VD |
2291 | /** |
2292 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache | |
2293 | * @cachep: the cache returned by memcg_kmem_get_cache | |
2294 | */ | |
2295 | void memcg_kmem_put_cache(struct kmem_cache *cachep) | |
8135be5a VD |
2296 | { |
2297 | if (!is_root_cache(cachep)) | |
f7ce3190 | 2298 | css_put(&cachep->memcg_params.memcg->css); |
8135be5a VD |
2299 | } |
2300 | ||
45264778 VD |
2301 | /** |
2302 | * memcg_kmem_charge: charge a kmem page | |
2303 | * @page: page to charge | |
2304 | * @gfp: reclaim mode | |
2305 | * @order: allocation order | |
2306 | * @memcg: memory cgroup to charge | |
2307 | * | |
2308 | * Returns 0 on success, an error code on failure. | |
2309 | */ | |
2310 | int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, | |
2311 | struct mem_cgroup *memcg) | |
7ae1e1d0 | 2312 | { |
f3ccb2c4 VD |
2313 | unsigned int nr_pages = 1 << order; |
2314 | struct page_counter *counter; | |
7ae1e1d0 GC |
2315 | int ret; |
2316 | ||
f3ccb2c4 | 2317 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2318 | if (ret) |
f3ccb2c4 | 2319 | return ret; |
52c29b04 JW |
2320 | |
2321 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2322 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
2323 | cancel_charge(memcg, nr_pages); | |
2324 | return -ENOMEM; | |
7ae1e1d0 GC |
2325 | } |
2326 | ||
f3ccb2c4 | 2327 | page->mem_cgroup = memcg; |
7ae1e1d0 | 2328 | |
f3ccb2c4 | 2329 | return 0; |
7ae1e1d0 GC |
2330 | } |
2331 | ||
45264778 VD |
2332 | /** |
2333 | * memcg_kmem_charge: charge a kmem page to the current memory cgroup | |
2334 | * @page: page to charge | |
2335 | * @gfp: reclaim mode | |
2336 | * @order: allocation order | |
2337 | * | |
2338 | * Returns 0 on success, an error code on failure. | |
2339 | */ | |
2340 | int memcg_kmem_charge(struct page *page, gfp_t gfp, int order) | |
7ae1e1d0 | 2341 | { |
f3ccb2c4 | 2342 | struct mem_cgroup *memcg; |
fcff7d7e | 2343 | int ret = 0; |
7ae1e1d0 | 2344 | |
45264778 VD |
2345 | if (memcg_kmem_bypass()) |
2346 | return 0; | |
2347 | ||
f3ccb2c4 | 2348 | memcg = get_mem_cgroup_from_mm(current->mm); |
c4159a75 | 2349 | if (!mem_cgroup_is_root(memcg)) { |
45264778 | 2350 | ret = memcg_kmem_charge_memcg(page, gfp, order, memcg); |
c4159a75 VD |
2351 | if (!ret) |
2352 | __SetPageKmemcg(page); | |
2353 | } | |
7ae1e1d0 | 2354 | css_put(&memcg->css); |
d05e83a6 | 2355 | return ret; |
7ae1e1d0 | 2356 | } |
45264778 VD |
2357 | /** |
2358 | * memcg_kmem_uncharge: uncharge a kmem page | |
2359 | * @page: page to uncharge | |
2360 | * @order: allocation order | |
2361 | */ | |
2362 | void memcg_kmem_uncharge(struct page *page, int order) | |
7ae1e1d0 | 2363 | { |
1306a85a | 2364 | struct mem_cgroup *memcg = page->mem_cgroup; |
f3ccb2c4 | 2365 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2366 | |
7ae1e1d0 GC |
2367 | if (!memcg) |
2368 | return; | |
2369 | ||
309381fe | 2370 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
29833315 | 2371 | |
52c29b04 JW |
2372 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2373 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2374 | ||
f3ccb2c4 | 2375 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2376 | if (do_memsw_account()) |
f3ccb2c4 | 2377 | page_counter_uncharge(&memcg->memsw, nr_pages); |
60d3fd32 | 2378 | |
1306a85a | 2379 | page->mem_cgroup = NULL; |
c4159a75 VD |
2380 | |
2381 | /* slab pages do not have PageKmemcg flag set */ | |
2382 | if (PageKmemcg(page)) | |
2383 | __ClearPageKmemcg(page); | |
2384 | ||
f3ccb2c4 | 2385 | css_put_many(&memcg->css, nr_pages); |
60d3fd32 | 2386 | } |
127424c8 | 2387 | #endif /* !CONFIG_SLOB */ |
7ae1e1d0 | 2388 | |
ca3e0214 KH |
2389 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2390 | ||
ca3e0214 KH |
2391 | /* |
2392 | * Because tail pages are not marked as "used", set it. We're under | |
a52633d8 | 2393 | * zone_lru_lock and migration entries setup in all page mappings. |
ca3e0214 | 2394 | */ |
e94c8a9c | 2395 | void mem_cgroup_split_huge_fixup(struct page *head) |
ca3e0214 | 2396 | { |
e94c8a9c | 2397 | int i; |
ca3e0214 | 2398 | |
3d37c4a9 KH |
2399 | if (mem_cgroup_disabled()) |
2400 | return; | |
b070e65c | 2401 | |
29833315 | 2402 | for (i = 1; i < HPAGE_PMD_NR; i++) |
1306a85a | 2403 | head[i].mem_cgroup = head->mem_cgroup; |
b9982f8d | 2404 | |
c9019e9b | 2405 | __mod_memcg_state(head->mem_cgroup, MEMCG_RSS_HUGE, -HPAGE_PMD_NR); |
ca3e0214 | 2406 | } |
12d27107 | 2407 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
ca3e0214 | 2408 | |
c255a458 | 2409 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
2410 | /** |
2411 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2412 | * @entry: swap entry to be moved | |
2413 | * @from: mem_cgroup which the entry is moved from | |
2414 | * @to: mem_cgroup which the entry is moved to | |
2415 | * | |
2416 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2417 | * as the mem_cgroup's id of @from. | |
2418 | * | |
2419 | * Returns 0 on success, -EINVAL on failure. | |
2420 | * | |
3e32cb2e | 2421 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
2422 | * both res and memsw, and called css_get(). |
2423 | */ | |
2424 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2425 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2426 | { |
2427 | unsigned short old_id, new_id; | |
2428 | ||
34c00c31 LZ |
2429 | old_id = mem_cgroup_id(from); |
2430 | new_id = mem_cgroup_id(to); | |
02491447 DN |
2431 | |
2432 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
2433 | mod_memcg_state(from, MEMCG_SWAP, -1); |
2434 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
2435 | return 0; |
2436 | } | |
2437 | return -EINVAL; | |
2438 | } | |
2439 | #else | |
2440 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 2441 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
2442 | { |
2443 | return -EINVAL; | |
2444 | } | |
8c7c6e34 | 2445 | #endif |
d13d1443 | 2446 | |
3e32cb2e | 2447 | static DEFINE_MUTEX(memcg_limit_mutex); |
f212ad7c | 2448 | |
d38d2a75 | 2449 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
c054a78c | 2450 | unsigned long limit, bool memsw) |
628f4235 | 2451 | { |
3e32cb2e | 2452 | bool enlarge = false; |
3e32cb2e | 2453 | int ret; |
c054a78c YZ |
2454 | bool limits_invariant; |
2455 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 2456 | |
3e32cb2e | 2457 | do { |
628f4235 KH |
2458 | if (signal_pending(current)) { |
2459 | ret = -EINTR; | |
2460 | break; | |
2461 | } | |
3e32cb2e JW |
2462 | |
2463 | mutex_lock(&memcg_limit_mutex); | |
c054a78c YZ |
2464 | /* |
2465 | * Make sure that the new limit (memsw or memory limit) doesn't | |
2466 | * break our basic invariant rule memory.limit <= memsw.limit. | |
2467 | */ | |
2468 | limits_invariant = memsw ? limit >= memcg->memory.limit : | |
2469 | limit <= memcg->memsw.limit; | |
2470 | if (!limits_invariant) { | |
3e32cb2e | 2471 | mutex_unlock(&memcg_limit_mutex); |
8c7c6e34 | 2472 | ret = -EINVAL; |
8c7c6e34 KH |
2473 | break; |
2474 | } | |
c054a78c | 2475 | if (limit > counter->limit) |
3e32cb2e | 2476 | enlarge = true; |
c054a78c | 2477 | ret = page_counter_limit(counter, limit); |
3e32cb2e | 2478 | mutex_unlock(&memcg_limit_mutex); |
8c7c6e34 KH |
2479 | |
2480 | if (!ret) | |
2481 | break; | |
2482 | ||
1ab5c056 AR |
2483 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
2484 | GFP_KERNEL, !memsw)) { | |
2485 | ret = -EBUSY; | |
2486 | break; | |
2487 | } | |
2488 | } while (true); | |
3e32cb2e | 2489 | |
3c11ecf4 KH |
2490 | if (!ret && enlarge) |
2491 | memcg_oom_recover(memcg); | |
3e32cb2e | 2492 | |
628f4235 KH |
2493 | return ret; |
2494 | } | |
2495 | ||
ef8f2327 | 2496 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
2497 | gfp_t gfp_mask, |
2498 | unsigned long *total_scanned) | |
2499 | { | |
2500 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 2501 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
2502 | unsigned long reclaimed; |
2503 | int loop = 0; | |
ef8f2327 | 2504 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 2505 | unsigned long excess; |
0608f43d AM |
2506 | unsigned long nr_scanned; |
2507 | ||
2508 | if (order > 0) | |
2509 | return 0; | |
2510 | ||
ef8f2327 | 2511 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
2512 | |
2513 | /* | |
2514 | * Do not even bother to check the largest node if the root | |
2515 | * is empty. Do it lockless to prevent lock bouncing. Races | |
2516 | * are acceptable as soft limit is best effort anyway. | |
2517 | */ | |
bfc7228b | 2518 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
2519 | return 0; |
2520 | ||
0608f43d AM |
2521 | /* |
2522 | * This loop can run a while, specially if mem_cgroup's continuously | |
2523 | * keep exceeding their soft limit and putting the system under | |
2524 | * pressure | |
2525 | */ | |
2526 | do { | |
2527 | if (next_mz) | |
2528 | mz = next_mz; | |
2529 | else | |
2530 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2531 | if (!mz) | |
2532 | break; | |
2533 | ||
2534 | nr_scanned = 0; | |
ef8f2327 | 2535 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
2536 | gfp_mask, &nr_scanned); |
2537 | nr_reclaimed += reclaimed; | |
2538 | *total_scanned += nr_scanned; | |
0a31bc97 | 2539 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 2540 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
2541 | |
2542 | /* | |
2543 | * If we failed to reclaim anything from this memory cgroup | |
2544 | * it is time to move on to the next cgroup | |
2545 | */ | |
2546 | next_mz = NULL; | |
bc2f2e7f VD |
2547 | if (!reclaimed) |
2548 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
2549 | ||
3e32cb2e | 2550 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
2551 | /* |
2552 | * One school of thought says that we should not add | |
2553 | * back the node to the tree if reclaim returns 0. | |
2554 | * But our reclaim could return 0, simply because due | |
2555 | * to priority we are exposing a smaller subset of | |
2556 | * memory to reclaim from. Consider this as a longer | |
2557 | * term TODO. | |
2558 | */ | |
2559 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 2560 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 2561 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
2562 | css_put(&mz->memcg->css); |
2563 | loop++; | |
2564 | /* | |
2565 | * Could not reclaim anything and there are no more | |
2566 | * mem cgroups to try or we seem to be looping without | |
2567 | * reclaiming anything. | |
2568 | */ | |
2569 | if (!nr_reclaimed && | |
2570 | (next_mz == NULL || | |
2571 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2572 | break; | |
2573 | } while (!nr_reclaimed); | |
2574 | if (next_mz) | |
2575 | css_put(&next_mz->memcg->css); | |
2576 | return nr_reclaimed; | |
2577 | } | |
2578 | ||
ea280e7b TH |
2579 | /* |
2580 | * Test whether @memcg has children, dead or alive. Note that this | |
2581 | * function doesn't care whether @memcg has use_hierarchy enabled and | |
2582 | * returns %true if there are child csses according to the cgroup | |
2583 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. | |
2584 | */ | |
b5f99b53 GC |
2585 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
2586 | { | |
ea280e7b TH |
2587 | bool ret; |
2588 | ||
ea280e7b TH |
2589 | rcu_read_lock(); |
2590 | ret = css_next_child(NULL, &memcg->css); | |
2591 | rcu_read_unlock(); | |
2592 | return ret; | |
b5f99b53 GC |
2593 | } |
2594 | ||
c26251f9 | 2595 | /* |
51038171 | 2596 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
2597 | * |
2598 | * Caller is responsible for holding css reference for memcg. | |
2599 | */ | |
2600 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
2601 | { | |
2602 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c26251f9 | 2603 | |
c1e862c1 KH |
2604 | /* we call try-to-free pages for make this cgroup empty */ |
2605 | lru_add_drain_all(); | |
f817ed48 | 2606 | /* try to free all pages in this cgroup */ |
3e32cb2e | 2607 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 2608 | int progress; |
c1e862c1 | 2609 | |
c26251f9 MH |
2610 | if (signal_pending(current)) |
2611 | return -EINTR; | |
2612 | ||
b70a2a21 JW |
2613 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
2614 | GFP_KERNEL, true); | |
c1e862c1 | 2615 | if (!progress) { |
f817ed48 | 2616 | nr_retries--; |
c1e862c1 | 2617 | /* maybe some writeback is necessary */ |
8aa7e847 | 2618 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2619 | } |
f817ed48 KH |
2620 | |
2621 | } | |
ab5196c2 MH |
2622 | |
2623 | return 0; | |
cc847582 KH |
2624 | } |
2625 | ||
6770c64e TH |
2626 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
2627 | char *buf, size_t nbytes, | |
2628 | loff_t off) | |
c1e862c1 | 2629 | { |
6770c64e | 2630 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 2631 | |
d8423011 MH |
2632 | if (mem_cgroup_is_root(memcg)) |
2633 | return -EINVAL; | |
6770c64e | 2634 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
2635 | } |
2636 | ||
182446d0 TH |
2637 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
2638 | struct cftype *cft) | |
18f59ea7 | 2639 | { |
182446d0 | 2640 | return mem_cgroup_from_css(css)->use_hierarchy; |
18f59ea7 BS |
2641 | } |
2642 | ||
182446d0 TH |
2643 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
2644 | struct cftype *cft, u64 val) | |
18f59ea7 BS |
2645 | { |
2646 | int retval = 0; | |
182446d0 | 2647 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5c9d535b | 2648 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
18f59ea7 | 2649 | |
567fb435 | 2650 | if (memcg->use_hierarchy == val) |
0b8f73e1 | 2651 | return 0; |
567fb435 | 2652 | |
18f59ea7 | 2653 | /* |
af901ca1 | 2654 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2655 | * in the child subtrees. If it is unset, then the change can |
2656 | * occur, provided the current cgroup has no children. | |
2657 | * | |
2658 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2659 | * set if there are no children. | |
2660 | */ | |
c0ff4b85 | 2661 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
18f59ea7 | 2662 | (val == 1 || val == 0)) { |
ea280e7b | 2663 | if (!memcg_has_children(memcg)) |
c0ff4b85 | 2664 | memcg->use_hierarchy = val; |
18f59ea7 BS |
2665 | else |
2666 | retval = -EBUSY; | |
2667 | } else | |
2668 | retval = -EINVAL; | |
567fb435 | 2669 | |
18f59ea7 BS |
2670 | return retval; |
2671 | } | |
2672 | ||
72b54e73 | 2673 | static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat) |
ce00a967 JW |
2674 | { |
2675 | struct mem_cgroup *iter; | |
72b54e73 | 2676 | int i; |
ce00a967 | 2677 | |
72b54e73 | 2678 | memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT); |
ce00a967 | 2679 | |
72b54e73 VD |
2680 | for_each_mem_cgroup_tree(iter, memcg) { |
2681 | for (i = 0; i < MEMCG_NR_STAT; i++) | |
ccda7f43 | 2682 | stat[i] += memcg_page_state(iter, i); |
72b54e73 | 2683 | } |
ce00a967 JW |
2684 | } |
2685 | ||
72b54e73 | 2686 | static void tree_events(struct mem_cgroup *memcg, unsigned long *events) |
587d9f72 JW |
2687 | { |
2688 | struct mem_cgroup *iter; | |
72b54e73 | 2689 | int i; |
587d9f72 | 2690 | |
72b54e73 | 2691 | memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS); |
587d9f72 | 2692 | |
72b54e73 VD |
2693 | for_each_mem_cgroup_tree(iter, memcg) { |
2694 | for (i = 0; i < MEMCG_NR_EVENTS; i++) | |
ccda7f43 | 2695 | events[i] += memcg_sum_events(iter, i); |
72b54e73 | 2696 | } |
587d9f72 JW |
2697 | } |
2698 | ||
6f646156 | 2699 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 2700 | { |
72b54e73 | 2701 | unsigned long val = 0; |
ce00a967 | 2702 | |
3e32cb2e | 2703 | if (mem_cgroup_is_root(memcg)) { |
72b54e73 VD |
2704 | struct mem_cgroup *iter; |
2705 | ||
2706 | for_each_mem_cgroup_tree(iter, memcg) { | |
ccda7f43 JW |
2707 | val += memcg_page_state(iter, MEMCG_CACHE); |
2708 | val += memcg_page_state(iter, MEMCG_RSS); | |
72b54e73 | 2709 | if (swap) |
ccda7f43 | 2710 | val += memcg_page_state(iter, MEMCG_SWAP); |
72b54e73 | 2711 | } |
3e32cb2e | 2712 | } else { |
ce00a967 | 2713 | if (!swap) |
3e32cb2e | 2714 | val = page_counter_read(&memcg->memory); |
ce00a967 | 2715 | else |
3e32cb2e | 2716 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 2717 | } |
c12176d3 | 2718 | return val; |
ce00a967 JW |
2719 | } |
2720 | ||
3e32cb2e JW |
2721 | enum { |
2722 | RES_USAGE, | |
2723 | RES_LIMIT, | |
2724 | RES_MAX_USAGE, | |
2725 | RES_FAILCNT, | |
2726 | RES_SOFT_LIMIT, | |
2727 | }; | |
ce00a967 | 2728 | |
791badbd | 2729 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 2730 | struct cftype *cft) |
8cdea7c0 | 2731 | { |
182446d0 | 2732 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 2733 | struct page_counter *counter; |
af36f906 | 2734 | |
3e32cb2e | 2735 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 2736 | case _MEM: |
3e32cb2e JW |
2737 | counter = &memcg->memory; |
2738 | break; | |
8c7c6e34 | 2739 | case _MEMSWAP: |
3e32cb2e JW |
2740 | counter = &memcg->memsw; |
2741 | break; | |
510fc4e1 | 2742 | case _KMEM: |
3e32cb2e | 2743 | counter = &memcg->kmem; |
510fc4e1 | 2744 | break; |
d55f90bf | 2745 | case _TCP: |
0db15298 | 2746 | counter = &memcg->tcpmem; |
d55f90bf | 2747 | break; |
8c7c6e34 KH |
2748 | default: |
2749 | BUG(); | |
8c7c6e34 | 2750 | } |
3e32cb2e JW |
2751 | |
2752 | switch (MEMFILE_ATTR(cft->private)) { | |
2753 | case RES_USAGE: | |
2754 | if (counter == &memcg->memory) | |
c12176d3 | 2755 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 2756 | if (counter == &memcg->memsw) |
c12176d3 | 2757 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
2758 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
2759 | case RES_LIMIT: | |
2760 | return (u64)counter->limit * PAGE_SIZE; | |
2761 | case RES_MAX_USAGE: | |
2762 | return (u64)counter->watermark * PAGE_SIZE; | |
2763 | case RES_FAILCNT: | |
2764 | return counter->failcnt; | |
2765 | case RES_SOFT_LIMIT: | |
2766 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
2767 | default: | |
2768 | BUG(); | |
2769 | } | |
8cdea7c0 | 2770 | } |
510fc4e1 | 2771 | |
127424c8 | 2772 | #ifndef CONFIG_SLOB |
567e9ab2 | 2773 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 2774 | { |
d6441637 VD |
2775 | int memcg_id; |
2776 | ||
b313aeee VD |
2777 | if (cgroup_memory_nokmem) |
2778 | return 0; | |
2779 | ||
2a4db7eb | 2780 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 2781 | BUG_ON(memcg->kmem_state); |
d6441637 | 2782 | |
f3bb3043 | 2783 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
2784 | if (memcg_id < 0) |
2785 | return memcg_id; | |
d6441637 | 2786 | |
ef12947c | 2787 | static_branch_inc(&memcg_kmem_enabled_key); |
d6441637 | 2788 | /* |
567e9ab2 | 2789 | * A memory cgroup is considered kmem-online as soon as it gets |
900a38f0 | 2790 | * kmemcg_id. Setting the id after enabling static branching will |
d6441637 VD |
2791 | * guarantee no one starts accounting before all call sites are |
2792 | * patched. | |
2793 | */ | |
900a38f0 | 2794 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 2795 | memcg->kmem_state = KMEM_ONLINE; |
bc2791f8 | 2796 | INIT_LIST_HEAD(&memcg->kmem_caches); |
0b8f73e1 JW |
2797 | |
2798 | return 0; | |
d6441637 VD |
2799 | } |
2800 | ||
8e0a8912 JW |
2801 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
2802 | { | |
2803 | struct cgroup_subsys_state *css; | |
2804 | struct mem_cgroup *parent, *child; | |
2805 | int kmemcg_id; | |
2806 | ||
2807 | if (memcg->kmem_state != KMEM_ONLINE) | |
2808 | return; | |
2809 | /* | |
2810 | * Clear the online state before clearing memcg_caches array | |
2811 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() | |
2812 | * guarantees that no cache will be created for this cgroup | |
2813 | * after we are done (see memcg_create_kmem_cache()). | |
2814 | */ | |
2815 | memcg->kmem_state = KMEM_ALLOCATED; | |
2816 | ||
2817 | memcg_deactivate_kmem_caches(memcg); | |
2818 | ||
2819 | kmemcg_id = memcg->kmemcg_id; | |
2820 | BUG_ON(kmemcg_id < 0); | |
2821 | ||
2822 | parent = parent_mem_cgroup(memcg); | |
2823 | if (!parent) | |
2824 | parent = root_mem_cgroup; | |
2825 | ||
2826 | /* | |
2827 | * Change kmemcg_id of this cgroup and all its descendants to the | |
2828 | * parent's id, and then move all entries from this cgroup's list_lrus | |
2829 | * to ones of the parent. After we have finished, all list_lrus | |
2830 | * corresponding to this cgroup are guaranteed to remain empty. The | |
2831 | * ordering is imposed by list_lru_node->lock taken by | |
2832 | * memcg_drain_all_list_lrus(). | |
2833 | */ | |
3a06bb78 | 2834 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
2835 | css_for_each_descendant_pre(css, &memcg->css) { |
2836 | child = mem_cgroup_from_css(css); | |
2837 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
2838 | child->kmemcg_id = parent->kmemcg_id; | |
2839 | if (!memcg->use_hierarchy) | |
2840 | break; | |
2841 | } | |
3a06bb78 TH |
2842 | rcu_read_unlock(); |
2843 | ||
8e0a8912 JW |
2844 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); |
2845 | ||
2846 | memcg_free_cache_id(kmemcg_id); | |
2847 | } | |
2848 | ||
2849 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2850 | { | |
0b8f73e1 JW |
2851 | /* css_alloc() failed, offlining didn't happen */ |
2852 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
2853 | memcg_offline_kmem(memcg); | |
2854 | ||
8e0a8912 JW |
2855 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
2856 | memcg_destroy_kmem_caches(memcg); | |
2857 | static_branch_dec(&memcg_kmem_enabled_key); | |
2858 | WARN_ON(page_counter_read(&memcg->kmem)); | |
2859 | } | |
8e0a8912 | 2860 | } |
d6441637 | 2861 | #else |
0b8f73e1 | 2862 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
2863 | { |
2864 | return 0; | |
2865 | } | |
2866 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
2867 | { | |
2868 | } | |
2869 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
2870 | { | |
2871 | } | |
2872 | #endif /* !CONFIG_SLOB */ | |
2873 | ||
d6441637 | 2874 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
3e32cb2e | 2875 | unsigned long limit) |
d6441637 | 2876 | { |
b313aeee | 2877 | int ret; |
127424c8 JW |
2878 | |
2879 | mutex_lock(&memcg_limit_mutex); | |
127424c8 | 2880 | ret = page_counter_limit(&memcg->kmem, limit); |
127424c8 JW |
2881 | mutex_unlock(&memcg_limit_mutex); |
2882 | return ret; | |
d6441637 | 2883 | } |
510fc4e1 | 2884 | |
d55f90bf VD |
2885 | static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit) |
2886 | { | |
2887 | int ret; | |
2888 | ||
2889 | mutex_lock(&memcg_limit_mutex); | |
2890 | ||
0db15298 | 2891 | ret = page_counter_limit(&memcg->tcpmem, limit); |
d55f90bf VD |
2892 | if (ret) |
2893 | goto out; | |
2894 | ||
0db15298 | 2895 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
2896 | /* |
2897 | * The active flag needs to be written after the static_key | |
2898 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
2899 | * function is the last one to run. See mem_cgroup_sk_alloc() |
2900 | * for details, and note that we don't mark any socket as | |
2901 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
2902 | * |
2903 | * We need to do this, because static_keys will span multiple | |
2904 | * sites, but we can't control their order. If we mark a socket | |
2905 | * as accounted, but the accounting functions are not patched in | |
2906 | * yet, we'll lose accounting. | |
2907 | * | |
2d758073 | 2908 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
2909 | * because when this value change, the code to process it is not |
2910 | * patched in yet. | |
2911 | */ | |
2912 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 2913 | memcg->tcpmem_active = true; |
d55f90bf VD |
2914 | } |
2915 | out: | |
2916 | mutex_unlock(&memcg_limit_mutex); | |
2917 | return ret; | |
2918 | } | |
d55f90bf | 2919 | |
628f4235 KH |
2920 | /* |
2921 | * The user of this function is... | |
2922 | * RES_LIMIT. | |
2923 | */ | |
451af504 TH |
2924 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
2925 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 2926 | { |
451af504 | 2927 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 2928 | unsigned long nr_pages; |
628f4235 KH |
2929 | int ret; |
2930 | ||
451af504 | 2931 | buf = strstrip(buf); |
650c5e56 | 2932 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
2933 | if (ret) |
2934 | return ret; | |
af36f906 | 2935 | |
3e32cb2e | 2936 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 2937 | case RES_LIMIT: |
4b3bde4c BS |
2938 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
2939 | ret = -EINVAL; | |
2940 | break; | |
2941 | } | |
3e32cb2e JW |
2942 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
2943 | case _MEM: | |
c054a78c | 2944 | ret = mem_cgroup_resize_limit(memcg, nr_pages, false); |
8c7c6e34 | 2945 | break; |
3e32cb2e | 2946 | case _MEMSWAP: |
c054a78c | 2947 | ret = mem_cgroup_resize_limit(memcg, nr_pages, true); |
296c81d8 | 2948 | break; |
3e32cb2e JW |
2949 | case _KMEM: |
2950 | ret = memcg_update_kmem_limit(memcg, nr_pages); | |
2951 | break; | |
d55f90bf VD |
2952 | case _TCP: |
2953 | ret = memcg_update_tcp_limit(memcg, nr_pages); | |
2954 | break; | |
3e32cb2e | 2955 | } |
296c81d8 | 2956 | break; |
3e32cb2e JW |
2957 | case RES_SOFT_LIMIT: |
2958 | memcg->soft_limit = nr_pages; | |
2959 | ret = 0; | |
628f4235 KH |
2960 | break; |
2961 | } | |
451af504 | 2962 | return ret ?: nbytes; |
8cdea7c0 BS |
2963 | } |
2964 | ||
6770c64e TH |
2965 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
2966 | size_t nbytes, loff_t off) | |
c84872e1 | 2967 | { |
6770c64e | 2968 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 2969 | struct page_counter *counter; |
c84872e1 | 2970 | |
3e32cb2e JW |
2971 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
2972 | case _MEM: | |
2973 | counter = &memcg->memory; | |
2974 | break; | |
2975 | case _MEMSWAP: | |
2976 | counter = &memcg->memsw; | |
2977 | break; | |
2978 | case _KMEM: | |
2979 | counter = &memcg->kmem; | |
2980 | break; | |
d55f90bf | 2981 | case _TCP: |
0db15298 | 2982 | counter = &memcg->tcpmem; |
d55f90bf | 2983 | break; |
3e32cb2e JW |
2984 | default: |
2985 | BUG(); | |
2986 | } | |
af36f906 | 2987 | |
3e32cb2e | 2988 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 2989 | case RES_MAX_USAGE: |
3e32cb2e | 2990 | page_counter_reset_watermark(counter); |
29f2a4da PE |
2991 | break; |
2992 | case RES_FAILCNT: | |
3e32cb2e | 2993 | counter->failcnt = 0; |
29f2a4da | 2994 | break; |
3e32cb2e JW |
2995 | default: |
2996 | BUG(); | |
29f2a4da | 2997 | } |
f64c3f54 | 2998 | |
6770c64e | 2999 | return nbytes; |
c84872e1 PE |
3000 | } |
3001 | ||
182446d0 | 3002 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3003 | struct cftype *cft) |
3004 | { | |
182446d0 | 3005 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3006 | } |
3007 | ||
02491447 | 3008 | #ifdef CONFIG_MMU |
182446d0 | 3009 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3010 | struct cftype *cft, u64 val) |
3011 | { | |
182446d0 | 3012 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3013 | |
1dfab5ab | 3014 | if (val & ~MOVE_MASK) |
7dc74be0 | 3015 | return -EINVAL; |
ee5e8472 | 3016 | |
7dc74be0 | 3017 | /* |
ee5e8472 GC |
3018 | * No kind of locking is needed in here, because ->can_attach() will |
3019 | * check this value once in the beginning of the process, and then carry | |
3020 | * on with stale data. This means that changes to this value will only | |
3021 | * affect task migrations starting after the change. | |
7dc74be0 | 3022 | */ |
c0ff4b85 | 3023 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3024 | return 0; |
3025 | } | |
02491447 | 3026 | #else |
182446d0 | 3027 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3028 | struct cftype *cft, u64 val) |
3029 | { | |
3030 | return -ENOSYS; | |
3031 | } | |
3032 | #endif | |
7dc74be0 | 3033 | |
406eb0c9 | 3034 | #ifdef CONFIG_NUMA |
2da8ca82 | 3035 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3036 | { |
25485de6 GT |
3037 | struct numa_stat { |
3038 | const char *name; | |
3039 | unsigned int lru_mask; | |
3040 | }; | |
3041 | ||
3042 | static const struct numa_stat stats[] = { | |
3043 | { "total", LRU_ALL }, | |
3044 | { "file", LRU_ALL_FILE }, | |
3045 | { "anon", LRU_ALL_ANON }, | |
3046 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3047 | }; | |
3048 | const struct numa_stat *stat; | |
406eb0c9 | 3049 | int nid; |
25485de6 | 3050 | unsigned long nr; |
2da8ca82 | 3051 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
406eb0c9 | 3052 | |
25485de6 GT |
3053 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3054 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); | |
3055 | seq_printf(m, "%s=%lu", stat->name, nr); | |
3056 | for_each_node_state(nid, N_MEMORY) { | |
3057 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3058 | stat->lru_mask); | |
3059 | seq_printf(m, " N%d=%lu", nid, nr); | |
3060 | } | |
3061 | seq_putc(m, '\n'); | |
406eb0c9 | 3062 | } |
406eb0c9 | 3063 | |
071aee13 YH |
3064 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
3065 | struct mem_cgroup *iter; | |
3066 | ||
3067 | nr = 0; | |
3068 | for_each_mem_cgroup_tree(iter, memcg) | |
3069 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); | |
3070 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); | |
3071 | for_each_node_state(nid, N_MEMORY) { | |
3072 | nr = 0; | |
3073 | for_each_mem_cgroup_tree(iter, memcg) | |
3074 | nr += mem_cgroup_node_nr_lru_pages( | |
3075 | iter, nid, stat->lru_mask); | |
3076 | seq_printf(m, " N%d=%lu", nid, nr); | |
3077 | } | |
3078 | seq_putc(m, '\n'); | |
406eb0c9 | 3079 | } |
406eb0c9 | 3080 | |
406eb0c9 YH |
3081 | return 0; |
3082 | } | |
3083 | #endif /* CONFIG_NUMA */ | |
3084 | ||
df0e53d0 JW |
3085 | /* Universal VM events cgroup1 shows, original sort order */ |
3086 | unsigned int memcg1_events[] = { | |
3087 | PGPGIN, | |
3088 | PGPGOUT, | |
3089 | PGFAULT, | |
3090 | PGMAJFAULT, | |
3091 | }; | |
3092 | ||
3093 | static const char *const memcg1_event_names[] = { | |
3094 | "pgpgin", | |
3095 | "pgpgout", | |
3096 | "pgfault", | |
3097 | "pgmajfault", | |
3098 | }; | |
3099 | ||
2da8ca82 | 3100 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3101 | { |
2da8ca82 | 3102 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
3e32cb2e | 3103 | unsigned long memory, memsw; |
af7c4b0e JW |
3104 | struct mem_cgroup *mi; |
3105 | unsigned int i; | |
406eb0c9 | 3106 | |
71cd3113 | 3107 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c RS |
3108 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
3109 | ||
71cd3113 JW |
3110 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
3111 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) | |
1dd3a273 | 3112 | continue; |
71cd3113 | 3113 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], |
ccda7f43 | 3114 | memcg_page_state(memcg, memcg1_stats[i]) * |
71cd3113 | 3115 | PAGE_SIZE); |
1dd3a273 | 3116 | } |
7b854121 | 3117 | |
df0e53d0 JW |
3118 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
3119 | seq_printf(m, "%s %lu\n", memcg1_event_names[i], | |
ccda7f43 | 3120 | memcg_sum_events(memcg, memcg1_events[i])); |
af7c4b0e JW |
3121 | |
3122 | for (i = 0; i < NR_LRU_LISTS; i++) | |
3123 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], | |
3124 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); | |
3125 | ||
14067bb3 | 3126 | /* Hierarchical information */ |
3e32cb2e JW |
3127 | memory = memsw = PAGE_COUNTER_MAX; |
3128 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
3129 | memory = min(memory, mi->memory.limit); | |
3130 | memsw = min(memsw, mi->memsw.limit); | |
fee7b548 | 3131 | } |
3e32cb2e JW |
3132 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3133 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3134 | if (do_memsw_account()) |
3e32cb2e JW |
3135 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3136 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3137 | |
71cd3113 | 3138 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
484ebb3b | 3139 | unsigned long long val = 0; |
af7c4b0e | 3140 | |
71cd3113 | 3141 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3142 | continue; |
af7c4b0e | 3143 | for_each_mem_cgroup_tree(mi, memcg) |
ccda7f43 | 3144 | val += memcg_page_state(mi, memcg1_stats[i]) * |
71cd3113 JW |
3145 | PAGE_SIZE; |
3146 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], val); | |
af7c4b0e JW |
3147 | } |
3148 | ||
df0e53d0 | 3149 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) { |
af7c4b0e JW |
3150 | unsigned long long val = 0; |
3151 | ||
3152 | for_each_mem_cgroup_tree(mi, memcg) | |
ccda7f43 | 3153 | val += memcg_sum_events(mi, memcg1_events[i]); |
df0e53d0 | 3154 | seq_printf(m, "total_%s %llu\n", memcg1_event_names[i], val); |
af7c4b0e JW |
3155 | } |
3156 | ||
3157 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
3158 | unsigned long long val = 0; | |
3159 | ||
3160 | for_each_mem_cgroup_tree(mi, memcg) | |
3161 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; | |
3162 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); | |
1dd3a273 | 3163 | } |
14067bb3 | 3164 | |
7f016ee8 | 3165 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 3166 | { |
ef8f2327 MG |
3167 | pg_data_t *pgdat; |
3168 | struct mem_cgroup_per_node *mz; | |
89abfab1 | 3169 | struct zone_reclaim_stat *rstat; |
7f016ee8 KM |
3170 | unsigned long recent_rotated[2] = {0, 0}; |
3171 | unsigned long recent_scanned[2] = {0, 0}; | |
3172 | ||
ef8f2327 MG |
3173 | for_each_online_pgdat(pgdat) { |
3174 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); | |
3175 | rstat = &mz->lruvec.reclaim_stat; | |
7f016ee8 | 3176 | |
ef8f2327 MG |
3177 | recent_rotated[0] += rstat->recent_rotated[0]; |
3178 | recent_rotated[1] += rstat->recent_rotated[1]; | |
3179 | recent_scanned[0] += rstat->recent_scanned[0]; | |
3180 | recent_scanned[1] += rstat->recent_scanned[1]; | |
3181 | } | |
78ccf5b5 JW |
3182 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
3183 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); | |
3184 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); | |
3185 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); | |
7f016ee8 KM |
3186 | } |
3187 | #endif | |
3188 | ||
d2ceb9b7 KH |
3189 | return 0; |
3190 | } | |
3191 | ||
182446d0 TH |
3192 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3193 | struct cftype *cft) | |
a7885eb8 | 3194 | { |
182446d0 | 3195 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3196 | |
1f4c025b | 3197 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3198 | } |
3199 | ||
182446d0 TH |
3200 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3201 | struct cftype *cft, u64 val) | |
a7885eb8 | 3202 | { |
182446d0 | 3203 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3204 | |
3dae7fec | 3205 | if (val > 100) |
a7885eb8 KM |
3206 | return -EINVAL; |
3207 | ||
14208b0e | 3208 | if (css->parent) |
3dae7fec JW |
3209 | memcg->swappiness = val; |
3210 | else | |
3211 | vm_swappiness = val; | |
068b38c1 | 3212 | |
a7885eb8 KM |
3213 | return 0; |
3214 | } | |
3215 | ||
2e72b634 KS |
3216 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3217 | { | |
3218 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3219 | unsigned long usage; |
2e72b634 KS |
3220 | int i; |
3221 | ||
3222 | rcu_read_lock(); | |
3223 | if (!swap) | |
2c488db2 | 3224 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3225 | else |
2c488db2 | 3226 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3227 | |
3228 | if (!t) | |
3229 | goto unlock; | |
3230 | ||
ce00a967 | 3231 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3232 | |
3233 | /* | |
748dad36 | 3234 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3235 | * If it's not true, a threshold was crossed after last |
3236 | * call of __mem_cgroup_threshold(). | |
3237 | */ | |
5407a562 | 3238 | i = t->current_threshold; |
2e72b634 KS |
3239 | |
3240 | /* | |
3241 | * Iterate backward over array of thresholds starting from | |
3242 | * current_threshold and check if a threshold is crossed. | |
3243 | * If none of thresholds below usage is crossed, we read | |
3244 | * only one element of the array here. | |
3245 | */ | |
3246 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3247 | eventfd_signal(t->entries[i].eventfd, 1); | |
3248 | ||
3249 | /* i = current_threshold + 1 */ | |
3250 | i++; | |
3251 | ||
3252 | /* | |
3253 | * Iterate forward over array of thresholds starting from | |
3254 | * current_threshold+1 and check if a threshold is crossed. | |
3255 | * If none of thresholds above usage is crossed, we read | |
3256 | * only one element of the array here. | |
3257 | */ | |
3258 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3259 | eventfd_signal(t->entries[i].eventfd, 1); | |
3260 | ||
3261 | /* Update current_threshold */ | |
5407a562 | 3262 | t->current_threshold = i - 1; |
2e72b634 KS |
3263 | unlock: |
3264 | rcu_read_unlock(); | |
3265 | } | |
3266 | ||
3267 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3268 | { | |
ad4ca5f4 KS |
3269 | while (memcg) { |
3270 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3271 | if (do_memsw_account()) |
ad4ca5f4 KS |
3272 | __mem_cgroup_threshold(memcg, true); |
3273 | ||
3274 | memcg = parent_mem_cgroup(memcg); | |
3275 | } | |
2e72b634 KS |
3276 | } |
3277 | ||
3278 | static int compare_thresholds(const void *a, const void *b) | |
3279 | { | |
3280 | const struct mem_cgroup_threshold *_a = a; | |
3281 | const struct mem_cgroup_threshold *_b = b; | |
3282 | ||
2bff24a3 GT |
3283 | if (_a->threshold > _b->threshold) |
3284 | return 1; | |
3285 | ||
3286 | if (_a->threshold < _b->threshold) | |
3287 | return -1; | |
3288 | ||
3289 | return 0; | |
2e72b634 KS |
3290 | } |
3291 | ||
c0ff4b85 | 3292 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
3293 | { |
3294 | struct mem_cgroup_eventfd_list *ev; | |
3295 | ||
2bcf2e92 MH |
3296 | spin_lock(&memcg_oom_lock); |
3297 | ||
c0ff4b85 | 3298 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 3299 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
3300 | |
3301 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
3302 | return 0; |
3303 | } | |
3304 | ||
c0ff4b85 | 3305 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 3306 | { |
7d74b06f KH |
3307 | struct mem_cgroup *iter; |
3308 | ||
c0ff4b85 | 3309 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 3310 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
3311 | } |
3312 | ||
59b6f873 | 3313 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3314 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 3315 | { |
2c488db2 KS |
3316 | struct mem_cgroup_thresholds *thresholds; |
3317 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
3318 | unsigned long threshold; |
3319 | unsigned long usage; | |
2c488db2 | 3320 | int i, size, ret; |
2e72b634 | 3321 | |
650c5e56 | 3322 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
3323 | if (ret) |
3324 | return ret; | |
3325 | ||
3326 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3327 | |
05b84301 | 3328 | if (type == _MEM) { |
2c488db2 | 3329 | thresholds = &memcg->thresholds; |
ce00a967 | 3330 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3331 | } else if (type == _MEMSWAP) { |
2c488db2 | 3332 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3333 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3334 | } else |
2e72b634 KS |
3335 | BUG(); |
3336 | ||
2e72b634 | 3337 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 3338 | if (thresholds->primary) |
2e72b634 KS |
3339 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3340 | ||
2c488db2 | 3341 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3342 | |
3343 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3344 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3345 | GFP_KERNEL); |
2c488db2 | 3346 | if (!new) { |
2e72b634 KS |
3347 | ret = -ENOMEM; |
3348 | goto unlock; | |
3349 | } | |
2c488db2 | 3350 | new->size = size; |
2e72b634 KS |
3351 | |
3352 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3353 | if (thresholds->primary) { |
3354 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3355 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3356 | } |
3357 | ||
2e72b634 | 3358 | /* Add new threshold */ |
2c488db2 KS |
3359 | new->entries[size - 1].eventfd = eventfd; |
3360 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3361 | |
3362 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3363 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3364 | compare_thresholds, NULL); |
3365 | ||
3366 | /* Find current threshold */ | |
2c488db2 | 3367 | new->current_threshold = -1; |
2e72b634 | 3368 | for (i = 0; i < size; i++) { |
748dad36 | 3369 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 3370 | /* |
2c488db2 KS |
3371 | * new->current_threshold will not be used until |
3372 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3373 | * it here. |
3374 | */ | |
2c488db2 | 3375 | ++new->current_threshold; |
748dad36 SZ |
3376 | } else |
3377 | break; | |
2e72b634 KS |
3378 | } |
3379 | ||
2c488db2 KS |
3380 | /* Free old spare buffer and save old primary buffer as spare */ |
3381 | kfree(thresholds->spare); | |
3382 | thresholds->spare = thresholds->primary; | |
3383 | ||
3384 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3385 | |
907860ed | 3386 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3387 | synchronize_rcu(); |
3388 | ||
2e72b634 KS |
3389 | unlock: |
3390 | mutex_unlock(&memcg->thresholds_lock); | |
3391 | ||
3392 | return ret; | |
3393 | } | |
3394 | ||
59b6f873 | 3395 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3396 | struct eventfd_ctx *eventfd, const char *args) |
3397 | { | |
59b6f873 | 3398 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
3399 | } |
3400 | ||
59b6f873 | 3401 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3402 | struct eventfd_ctx *eventfd, const char *args) |
3403 | { | |
59b6f873 | 3404 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
3405 | } |
3406 | ||
59b6f873 | 3407 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3408 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 3409 | { |
2c488db2 KS |
3410 | struct mem_cgroup_thresholds *thresholds; |
3411 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 3412 | unsigned long usage; |
2c488db2 | 3413 | int i, j, size; |
2e72b634 KS |
3414 | |
3415 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
3416 | |
3417 | if (type == _MEM) { | |
2c488db2 | 3418 | thresholds = &memcg->thresholds; |
ce00a967 | 3419 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 3420 | } else if (type == _MEMSWAP) { |
2c488db2 | 3421 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 3422 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 3423 | } else |
2e72b634 KS |
3424 | BUG(); |
3425 | ||
371528ca AV |
3426 | if (!thresholds->primary) |
3427 | goto unlock; | |
3428 | ||
2e72b634 KS |
3429 | /* Check if a threshold crossed before removing */ |
3430 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3431 | ||
3432 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3433 | size = 0; |
3434 | for (i = 0; i < thresholds->primary->size; i++) { | |
3435 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3436 | size++; |
3437 | } | |
3438 | ||
2c488db2 | 3439 | new = thresholds->spare; |
907860ed | 3440 | |
2e72b634 KS |
3441 | /* Set thresholds array to NULL if we don't have thresholds */ |
3442 | if (!size) { | |
2c488db2 KS |
3443 | kfree(new); |
3444 | new = NULL; | |
907860ed | 3445 | goto swap_buffers; |
2e72b634 KS |
3446 | } |
3447 | ||
2c488db2 | 3448 | new->size = size; |
2e72b634 KS |
3449 | |
3450 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3451 | new->current_threshold = -1; |
3452 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3453 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3454 | continue; |
3455 | ||
2c488db2 | 3456 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 3457 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 3458 | /* |
2c488db2 | 3459 | * new->current_threshold will not be used |
2e72b634 KS |
3460 | * until rcu_assign_pointer(), so it's safe to increment |
3461 | * it here. | |
3462 | */ | |
2c488db2 | 3463 | ++new->current_threshold; |
2e72b634 KS |
3464 | } |
3465 | j++; | |
3466 | } | |
3467 | ||
907860ed | 3468 | swap_buffers: |
2c488db2 KS |
3469 | /* Swap primary and spare array */ |
3470 | thresholds->spare = thresholds->primary; | |
8c757763 | 3471 | |
2c488db2 | 3472 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 3473 | |
907860ed | 3474 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 3475 | synchronize_rcu(); |
6611d8d7 MC |
3476 | |
3477 | /* If all events are unregistered, free the spare array */ | |
3478 | if (!new) { | |
3479 | kfree(thresholds->spare); | |
3480 | thresholds->spare = NULL; | |
3481 | } | |
371528ca | 3482 | unlock: |
2e72b634 | 3483 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3484 | } |
c1e862c1 | 3485 | |
59b6f873 | 3486 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3487 | struct eventfd_ctx *eventfd) |
3488 | { | |
59b6f873 | 3489 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
3490 | } |
3491 | ||
59b6f873 | 3492 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
3493 | struct eventfd_ctx *eventfd) |
3494 | { | |
59b6f873 | 3495 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
3496 | } |
3497 | ||
59b6f873 | 3498 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 3499 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 3500 | { |
9490ff27 | 3501 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 3502 | |
9490ff27 KH |
3503 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
3504 | if (!event) | |
3505 | return -ENOMEM; | |
3506 | ||
1af8efe9 | 3507 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
3508 | |
3509 | event->eventfd = eventfd; | |
3510 | list_add(&event->list, &memcg->oom_notify); | |
3511 | ||
3512 | /* already in OOM ? */ | |
c2b42d3c | 3513 | if (memcg->under_oom) |
9490ff27 | 3514 | eventfd_signal(eventfd, 1); |
1af8efe9 | 3515 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3516 | |
3517 | return 0; | |
3518 | } | |
3519 | ||
59b6f873 | 3520 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 3521 | struct eventfd_ctx *eventfd) |
9490ff27 | 3522 | { |
9490ff27 | 3523 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 3524 | |
1af8efe9 | 3525 | spin_lock(&memcg_oom_lock); |
9490ff27 | 3526 | |
c0ff4b85 | 3527 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
3528 | if (ev->eventfd == eventfd) { |
3529 | list_del(&ev->list); | |
3530 | kfree(ev); | |
3531 | } | |
3532 | } | |
3533 | ||
1af8efe9 | 3534 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
3535 | } |
3536 | ||
2da8ca82 | 3537 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 3538 | { |
2da8ca82 | 3539 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
3c11ecf4 | 3540 | |
791badbd | 3541 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 3542 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
8e675f7a | 3543 | seq_printf(sf, "oom_kill %lu\n", memcg_sum_events(memcg, OOM_KILL)); |
3c11ecf4 KH |
3544 | return 0; |
3545 | } | |
3546 | ||
182446d0 | 3547 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
3548 | struct cftype *cft, u64 val) |
3549 | { | |
182446d0 | 3550 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
3551 | |
3552 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
14208b0e | 3553 | if (!css->parent || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
3554 | return -EINVAL; |
3555 | ||
c0ff4b85 | 3556 | memcg->oom_kill_disable = val; |
4d845ebf | 3557 | if (!val) |
c0ff4b85 | 3558 | memcg_oom_recover(memcg); |
3dae7fec | 3559 | |
3c11ecf4 KH |
3560 | return 0; |
3561 | } | |
3562 | ||
52ebea74 TH |
3563 | #ifdef CONFIG_CGROUP_WRITEBACK |
3564 | ||
3565 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) | |
3566 | { | |
3567 | return &memcg->cgwb_list; | |
3568 | } | |
3569 | ||
841710aa TH |
3570 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
3571 | { | |
3572 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
3573 | } | |
3574 | ||
3575 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3576 | { | |
3577 | wb_domain_exit(&memcg->cgwb_domain); | |
3578 | } | |
3579 | ||
2529bb3a TH |
3580 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3581 | { | |
3582 | wb_domain_size_changed(&memcg->cgwb_domain); | |
3583 | } | |
3584 | ||
841710aa TH |
3585 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
3586 | { | |
3587 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3588 | ||
3589 | if (!memcg->css.parent) | |
3590 | return NULL; | |
3591 | ||
3592 | return &memcg->cgwb_domain; | |
3593 | } | |
3594 | ||
c2aa723a TH |
3595 | /** |
3596 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
3597 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
3598 | * @pfilepages: out parameter for number of file pages |
3599 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
3600 | * @pdirty: out parameter for number of dirty pages |
3601 | * @pwriteback: out parameter for number of pages under writeback | |
3602 | * | |
c5edf9cd TH |
3603 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
3604 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
3605 | * is a bit more involved. | |
c2aa723a | 3606 | * |
c5edf9cd TH |
3607 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
3608 | * headroom is calculated as the lowest headroom of itself and the | |
3609 | * ancestors. Note that this doesn't consider the actual amount of | |
3610 | * available memory in the system. The caller should further cap | |
3611 | * *@pheadroom accordingly. | |
c2aa723a | 3612 | */ |
c5edf9cd TH |
3613 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
3614 | unsigned long *pheadroom, unsigned long *pdirty, | |
3615 | unsigned long *pwriteback) | |
c2aa723a TH |
3616 | { |
3617 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
3618 | struct mem_cgroup *parent; | |
c2aa723a | 3619 | |
ccda7f43 | 3620 | *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY); |
c2aa723a TH |
3621 | |
3622 | /* this should eventually include NR_UNSTABLE_NFS */ | |
ccda7f43 | 3623 | *pwriteback = memcg_page_state(memcg, NR_WRITEBACK); |
c5edf9cd TH |
3624 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
3625 | (1 << LRU_ACTIVE_FILE)); | |
3626 | *pheadroom = PAGE_COUNTER_MAX; | |
c2aa723a | 3627 | |
c2aa723a TH |
3628 | while ((parent = parent_mem_cgroup(memcg))) { |
3629 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); | |
3630 | unsigned long used = page_counter_read(&memcg->memory); | |
3631 | ||
c5edf9cd | 3632 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
3633 | memcg = parent; |
3634 | } | |
c2aa723a TH |
3635 | } |
3636 | ||
841710aa TH |
3637 | #else /* CONFIG_CGROUP_WRITEBACK */ |
3638 | ||
3639 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
3640 | { | |
3641 | return 0; | |
3642 | } | |
3643 | ||
3644 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
3645 | { | |
3646 | } | |
3647 | ||
2529bb3a TH |
3648 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
3649 | { | |
3650 | } | |
3651 | ||
52ebea74 TH |
3652 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
3653 | ||
3bc942f3 TH |
3654 | /* |
3655 | * DO NOT USE IN NEW FILES. | |
3656 | * | |
3657 | * "cgroup.event_control" implementation. | |
3658 | * | |
3659 | * This is way over-engineered. It tries to support fully configurable | |
3660 | * events for each user. Such level of flexibility is completely | |
3661 | * unnecessary especially in the light of the planned unified hierarchy. | |
3662 | * | |
3663 | * Please deprecate this and replace with something simpler if at all | |
3664 | * possible. | |
3665 | */ | |
3666 | ||
79bd9814 TH |
3667 | /* |
3668 | * Unregister event and free resources. | |
3669 | * | |
3670 | * Gets called from workqueue. | |
3671 | */ | |
3bc942f3 | 3672 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 3673 | { |
3bc942f3 TH |
3674 | struct mem_cgroup_event *event = |
3675 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 3676 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
3677 | |
3678 | remove_wait_queue(event->wqh, &event->wait); | |
3679 | ||
59b6f873 | 3680 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
3681 | |
3682 | /* Notify userspace the event is going away. */ | |
3683 | eventfd_signal(event->eventfd, 1); | |
3684 | ||
3685 | eventfd_ctx_put(event->eventfd); | |
3686 | kfree(event); | |
59b6f873 | 3687 | css_put(&memcg->css); |
79bd9814 TH |
3688 | } |
3689 | ||
3690 | /* | |
3691 | * Gets called on POLLHUP on eventfd when user closes it. | |
3692 | * | |
3693 | * Called with wqh->lock held and interrupts disabled. | |
3694 | */ | |
ac6424b9 | 3695 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 3696 | int sync, void *key) |
79bd9814 | 3697 | { |
3bc942f3 TH |
3698 | struct mem_cgroup_event *event = |
3699 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 3700 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 3701 | __poll_t flags = key_to_poll(key); |
79bd9814 TH |
3702 | |
3703 | if (flags & POLLHUP) { | |
3704 | /* | |
3705 | * If the event has been detached at cgroup removal, we | |
3706 | * can simply return knowing the other side will cleanup | |
3707 | * for us. | |
3708 | * | |
3709 | * We can't race against event freeing since the other | |
3710 | * side will require wqh->lock via remove_wait_queue(), | |
3711 | * which we hold. | |
3712 | */ | |
fba94807 | 3713 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
3714 | if (!list_empty(&event->list)) { |
3715 | list_del_init(&event->list); | |
3716 | /* | |
3717 | * We are in atomic context, but cgroup_event_remove() | |
3718 | * may sleep, so we have to call it in workqueue. | |
3719 | */ | |
3720 | schedule_work(&event->remove); | |
3721 | } | |
fba94807 | 3722 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
3723 | } |
3724 | ||
3725 | return 0; | |
3726 | } | |
3727 | ||
3bc942f3 | 3728 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
3729 | wait_queue_head_t *wqh, poll_table *pt) |
3730 | { | |
3bc942f3 TH |
3731 | struct mem_cgroup_event *event = |
3732 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
3733 | |
3734 | event->wqh = wqh; | |
3735 | add_wait_queue(wqh, &event->wait); | |
3736 | } | |
3737 | ||
3738 | /* | |
3bc942f3 TH |
3739 | * DO NOT USE IN NEW FILES. |
3740 | * | |
79bd9814 TH |
3741 | * Parse input and register new cgroup event handler. |
3742 | * | |
3743 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
3744 | * Interpretation of args is defined by control file implementation. | |
3745 | */ | |
451af504 TH |
3746 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
3747 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 3748 | { |
451af504 | 3749 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 3750 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 3751 | struct mem_cgroup_event *event; |
79bd9814 TH |
3752 | struct cgroup_subsys_state *cfile_css; |
3753 | unsigned int efd, cfd; | |
3754 | struct fd efile; | |
3755 | struct fd cfile; | |
fba94807 | 3756 | const char *name; |
79bd9814 TH |
3757 | char *endp; |
3758 | int ret; | |
3759 | ||
451af504 TH |
3760 | buf = strstrip(buf); |
3761 | ||
3762 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
3763 | if (*endp != ' ') |
3764 | return -EINVAL; | |
451af504 | 3765 | buf = endp + 1; |
79bd9814 | 3766 | |
451af504 | 3767 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
3768 | if ((*endp != ' ') && (*endp != '\0')) |
3769 | return -EINVAL; | |
451af504 | 3770 | buf = endp + 1; |
79bd9814 TH |
3771 | |
3772 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
3773 | if (!event) | |
3774 | return -ENOMEM; | |
3775 | ||
59b6f873 | 3776 | event->memcg = memcg; |
79bd9814 | 3777 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
3778 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
3779 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
3780 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
3781 | |
3782 | efile = fdget(efd); | |
3783 | if (!efile.file) { | |
3784 | ret = -EBADF; | |
3785 | goto out_kfree; | |
3786 | } | |
3787 | ||
3788 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
3789 | if (IS_ERR(event->eventfd)) { | |
3790 | ret = PTR_ERR(event->eventfd); | |
3791 | goto out_put_efile; | |
3792 | } | |
3793 | ||
3794 | cfile = fdget(cfd); | |
3795 | if (!cfile.file) { | |
3796 | ret = -EBADF; | |
3797 | goto out_put_eventfd; | |
3798 | } | |
3799 | ||
3800 | /* the process need read permission on control file */ | |
3801 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
3802 | ret = inode_permission(file_inode(cfile.file), MAY_READ); | |
3803 | if (ret < 0) | |
3804 | goto out_put_cfile; | |
3805 | ||
fba94807 TH |
3806 | /* |
3807 | * Determine the event callbacks and set them in @event. This used | |
3808 | * to be done via struct cftype but cgroup core no longer knows | |
3809 | * about these events. The following is crude but the whole thing | |
3810 | * is for compatibility anyway. | |
3bc942f3 TH |
3811 | * |
3812 | * DO NOT ADD NEW FILES. | |
fba94807 | 3813 | */ |
b583043e | 3814 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
3815 | |
3816 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
3817 | event->register_event = mem_cgroup_usage_register_event; | |
3818 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
3819 | } else if (!strcmp(name, "memory.oom_control")) { | |
3820 | event->register_event = mem_cgroup_oom_register_event; | |
3821 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
3822 | } else if (!strcmp(name, "memory.pressure_level")) { | |
3823 | event->register_event = vmpressure_register_event; | |
3824 | event->unregister_event = vmpressure_unregister_event; | |
3825 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
3826 | event->register_event = memsw_cgroup_usage_register_event; |
3827 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
3828 | } else { |
3829 | ret = -EINVAL; | |
3830 | goto out_put_cfile; | |
3831 | } | |
3832 | ||
79bd9814 | 3833 | /* |
b5557c4c TH |
3834 | * Verify @cfile should belong to @css. Also, remaining events are |
3835 | * automatically removed on cgroup destruction but the removal is | |
3836 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 3837 | */ |
b583043e | 3838 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 3839 | &memory_cgrp_subsys); |
79bd9814 | 3840 | ret = -EINVAL; |
5a17f543 | 3841 | if (IS_ERR(cfile_css)) |
79bd9814 | 3842 | goto out_put_cfile; |
5a17f543 TH |
3843 | if (cfile_css != css) { |
3844 | css_put(cfile_css); | |
79bd9814 | 3845 | goto out_put_cfile; |
5a17f543 | 3846 | } |
79bd9814 | 3847 | |
451af504 | 3848 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
3849 | if (ret) |
3850 | goto out_put_css; | |
3851 | ||
3852 | efile.file->f_op->poll(efile.file, &event->pt); | |
3853 | ||
fba94807 TH |
3854 | spin_lock(&memcg->event_list_lock); |
3855 | list_add(&event->list, &memcg->event_list); | |
3856 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
3857 | |
3858 | fdput(cfile); | |
3859 | fdput(efile); | |
3860 | ||
451af504 | 3861 | return nbytes; |
79bd9814 TH |
3862 | |
3863 | out_put_css: | |
b5557c4c | 3864 | css_put(css); |
79bd9814 TH |
3865 | out_put_cfile: |
3866 | fdput(cfile); | |
3867 | out_put_eventfd: | |
3868 | eventfd_ctx_put(event->eventfd); | |
3869 | out_put_efile: | |
3870 | fdput(efile); | |
3871 | out_kfree: | |
3872 | kfree(event); | |
3873 | ||
3874 | return ret; | |
3875 | } | |
3876 | ||
241994ed | 3877 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 3878 | { |
0eea1030 | 3879 | .name = "usage_in_bytes", |
8c7c6e34 | 3880 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 3881 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3882 | }, |
c84872e1 PE |
3883 | { |
3884 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 3885 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 3886 | .write = mem_cgroup_reset, |
791badbd | 3887 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 3888 | }, |
8cdea7c0 | 3889 | { |
0eea1030 | 3890 | .name = "limit_in_bytes", |
8c7c6e34 | 3891 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 3892 | .write = mem_cgroup_write, |
791badbd | 3893 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3894 | }, |
296c81d8 BS |
3895 | { |
3896 | .name = "soft_limit_in_bytes", | |
3897 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 3898 | .write = mem_cgroup_write, |
791badbd | 3899 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 3900 | }, |
8cdea7c0 BS |
3901 | { |
3902 | .name = "failcnt", | |
8c7c6e34 | 3903 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 3904 | .write = mem_cgroup_reset, |
791badbd | 3905 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 3906 | }, |
d2ceb9b7 KH |
3907 | { |
3908 | .name = "stat", | |
2da8ca82 | 3909 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 3910 | }, |
c1e862c1 KH |
3911 | { |
3912 | .name = "force_empty", | |
6770c64e | 3913 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 3914 | }, |
18f59ea7 BS |
3915 | { |
3916 | .name = "use_hierarchy", | |
3917 | .write_u64 = mem_cgroup_hierarchy_write, | |
3918 | .read_u64 = mem_cgroup_hierarchy_read, | |
3919 | }, | |
79bd9814 | 3920 | { |
3bc942f3 | 3921 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 3922 | .write = memcg_write_event_control, |
7dbdb199 | 3923 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 3924 | }, |
a7885eb8 KM |
3925 | { |
3926 | .name = "swappiness", | |
3927 | .read_u64 = mem_cgroup_swappiness_read, | |
3928 | .write_u64 = mem_cgroup_swappiness_write, | |
3929 | }, | |
7dc74be0 DN |
3930 | { |
3931 | .name = "move_charge_at_immigrate", | |
3932 | .read_u64 = mem_cgroup_move_charge_read, | |
3933 | .write_u64 = mem_cgroup_move_charge_write, | |
3934 | }, | |
9490ff27 KH |
3935 | { |
3936 | .name = "oom_control", | |
2da8ca82 | 3937 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 3938 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
3939 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
3940 | }, | |
70ddf637 AV |
3941 | { |
3942 | .name = "pressure_level", | |
70ddf637 | 3943 | }, |
406eb0c9 YH |
3944 | #ifdef CONFIG_NUMA |
3945 | { | |
3946 | .name = "numa_stat", | |
2da8ca82 | 3947 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
3948 | }, |
3949 | #endif | |
510fc4e1 GC |
3950 | { |
3951 | .name = "kmem.limit_in_bytes", | |
3952 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 3953 | .write = mem_cgroup_write, |
791badbd | 3954 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
3955 | }, |
3956 | { | |
3957 | .name = "kmem.usage_in_bytes", | |
3958 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 3959 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
3960 | }, |
3961 | { | |
3962 | .name = "kmem.failcnt", | |
3963 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 3964 | .write = mem_cgroup_reset, |
791badbd | 3965 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
3966 | }, |
3967 | { | |
3968 | .name = "kmem.max_usage_in_bytes", | |
3969 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 3970 | .write = mem_cgroup_reset, |
791badbd | 3971 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 3972 | }, |
5b365771 | 3973 | #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG) |
749c5415 GC |
3974 | { |
3975 | .name = "kmem.slabinfo", | |
bc2791f8 TH |
3976 | .seq_start = memcg_slab_start, |
3977 | .seq_next = memcg_slab_next, | |
3978 | .seq_stop = memcg_slab_stop, | |
b047501c | 3979 | .seq_show = memcg_slab_show, |
749c5415 GC |
3980 | }, |
3981 | #endif | |
d55f90bf VD |
3982 | { |
3983 | .name = "kmem.tcp.limit_in_bytes", | |
3984 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
3985 | .write = mem_cgroup_write, | |
3986 | .read_u64 = mem_cgroup_read_u64, | |
3987 | }, | |
3988 | { | |
3989 | .name = "kmem.tcp.usage_in_bytes", | |
3990 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
3991 | .read_u64 = mem_cgroup_read_u64, | |
3992 | }, | |
3993 | { | |
3994 | .name = "kmem.tcp.failcnt", | |
3995 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
3996 | .write = mem_cgroup_reset, | |
3997 | .read_u64 = mem_cgroup_read_u64, | |
3998 | }, | |
3999 | { | |
4000 | .name = "kmem.tcp.max_usage_in_bytes", | |
4001 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4002 | .write = mem_cgroup_reset, | |
4003 | .read_u64 = mem_cgroup_read_u64, | |
4004 | }, | |
6bc10349 | 4005 | { }, /* terminate */ |
af36f906 | 4006 | }; |
8c7c6e34 | 4007 | |
73f576c0 JW |
4008 | /* |
4009 | * Private memory cgroup IDR | |
4010 | * | |
4011 | * Swap-out records and page cache shadow entries need to store memcg | |
4012 | * references in constrained space, so we maintain an ID space that is | |
4013 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4014 | * memory-controlled cgroups to 64k. | |
4015 | * | |
4016 | * However, there usually are many references to the oflline CSS after | |
4017 | * the cgroup has been destroyed, such as page cache or reclaimable | |
4018 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4019 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4020 | * relatively small ID space and prevent the creation of new cgroups | |
4021 | * even when there are much fewer than 64k cgroups - possibly none. | |
4022 | * | |
4023 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4024 | * be freed and recycled when it's no longer needed, which is usually | |
4025 | * when the CSS is offlined. | |
4026 | * | |
4027 | * The only exception to that are records of swapped out tmpfs/shmem | |
4028 | * pages that need to be attributed to live ancestors on swapin. But | |
4029 | * those references are manageable from userspace. | |
4030 | */ | |
4031 | ||
4032 | static DEFINE_IDR(mem_cgroup_idr); | |
4033 | ||
615d66c3 | 4034 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4035 | { |
58fa2a55 | 4036 | VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0); |
615d66c3 | 4037 | atomic_add(n, &memcg->id.ref); |
73f576c0 JW |
4038 | } |
4039 | ||
615d66c3 | 4040 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4041 | { |
58fa2a55 | 4042 | VM_BUG_ON(atomic_read(&memcg->id.ref) < n); |
615d66c3 | 4043 | if (atomic_sub_and_test(n, &memcg->id.ref)) { |
73f576c0 JW |
4044 | idr_remove(&mem_cgroup_idr, memcg->id.id); |
4045 | memcg->id.id = 0; | |
4046 | ||
4047 | /* Memcg ID pins CSS */ | |
4048 | css_put(&memcg->css); | |
4049 | } | |
4050 | } | |
4051 | ||
615d66c3 VD |
4052 | static inline void mem_cgroup_id_get(struct mem_cgroup *memcg) |
4053 | { | |
4054 | mem_cgroup_id_get_many(memcg, 1); | |
4055 | } | |
4056 | ||
4057 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) | |
4058 | { | |
4059 | mem_cgroup_id_put_many(memcg, 1); | |
4060 | } | |
4061 | ||
73f576c0 JW |
4062 | /** |
4063 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4064 | * @id: the memcg id to look up | |
4065 | * | |
4066 | * Caller must hold rcu_read_lock(). | |
4067 | */ | |
4068 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4069 | { | |
4070 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4071 | return idr_find(&mem_cgroup_idr, id); | |
4072 | } | |
4073 | ||
ef8f2327 | 4074 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4075 | { |
4076 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 4077 | int tmp = node; |
1ecaab2b KH |
4078 | /* |
4079 | * This routine is called against possible nodes. | |
4080 | * But it's BUG to call kmalloc() against offline node. | |
4081 | * | |
4082 | * TODO: this routine can waste much memory for nodes which will | |
4083 | * never be onlined. It's better to use memory hotplug callback | |
4084 | * function. | |
4085 | */ | |
41e3355d KH |
4086 | if (!node_state(node, N_NORMAL_MEMORY)) |
4087 | tmp = -1; | |
17295c88 | 4088 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4089 | if (!pn) |
4090 | return 1; | |
1ecaab2b | 4091 | |
a983b5eb JW |
4092 | pn->lruvec_stat_cpu = alloc_percpu(struct lruvec_stat); |
4093 | if (!pn->lruvec_stat_cpu) { | |
00f3ca2c JW |
4094 | kfree(pn); |
4095 | return 1; | |
4096 | } | |
4097 | ||
ef8f2327 MG |
4098 | lruvec_init(&pn->lruvec); |
4099 | pn->usage_in_excess = 0; | |
4100 | pn->on_tree = false; | |
4101 | pn->memcg = memcg; | |
4102 | ||
54f72fe0 | 4103 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4104 | return 0; |
4105 | } | |
4106 | ||
ef8f2327 | 4107 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4108 | { |
00f3ca2c JW |
4109 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
4110 | ||
a983b5eb | 4111 | free_percpu(pn->lruvec_stat_cpu); |
00f3ca2c | 4112 | kfree(pn); |
1ecaab2b KH |
4113 | } |
4114 | ||
40e952f9 | 4115 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4116 | { |
c8b2a36f | 4117 | int node; |
59927fb9 | 4118 | |
c8b2a36f | 4119 | for_each_node(node) |
ef8f2327 | 4120 | free_mem_cgroup_per_node_info(memcg, node); |
a983b5eb | 4121 | free_percpu(memcg->stat_cpu); |
8ff69e2c | 4122 | kfree(memcg); |
59927fb9 | 4123 | } |
3afe36b1 | 4124 | |
40e952f9 TE |
4125 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
4126 | { | |
4127 | memcg_wb_domain_exit(memcg); | |
4128 | __mem_cgroup_free(memcg); | |
4129 | } | |
4130 | ||
0b8f73e1 | 4131 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 4132 | { |
d142e3e6 | 4133 | struct mem_cgroup *memcg; |
0b8f73e1 | 4134 | size_t size; |
6d12e2d8 | 4135 | int node; |
8cdea7c0 | 4136 | |
0b8f73e1 JW |
4137 | size = sizeof(struct mem_cgroup); |
4138 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
4139 | ||
4140 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 4141 | if (!memcg) |
0b8f73e1 JW |
4142 | return NULL; |
4143 | ||
73f576c0 JW |
4144 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
4145 | 1, MEM_CGROUP_ID_MAX, | |
4146 | GFP_KERNEL); | |
4147 | if (memcg->id.id < 0) | |
4148 | goto fail; | |
4149 | ||
a983b5eb JW |
4150 | memcg->stat_cpu = alloc_percpu(struct mem_cgroup_stat_cpu); |
4151 | if (!memcg->stat_cpu) | |
0b8f73e1 | 4152 | goto fail; |
78fb7466 | 4153 | |
3ed28fa1 | 4154 | for_each_node(node) |
ef8f2327 | 4155 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 4156 | goto fail; |
f64c3f54 | 4157 | |
0b8f73e1 JW |
4158 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
4159 | goto fail; | |
28dbc4b6 | 4160 | |
f7e1cb6e | 4161 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 GC |
4162 | memcg->last_scanned_node = MAX_NUMNODES; |
4163 | INIT_LIST_HEAD(&memcg->oom_notify); | |
d142e3e6 GC |
4164 | mutex_init(&memcg->thresholds_lock); |
4165 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 4166 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
4167 | INIT_LIST_HEAD(&memcg->event_list); |
4168 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 4169 | memcg->socket_pressure = jiffies; |
127424c8 | 4170 | #ifndef CONFIG_SLOB |
900a38f0 | 4171 | memcg->kmemcg_id = -1; |
900a38f0 | 4172 | #endif |
52ebea74 TH |
4173 | #ifdef CONFIG_CGROUP_WRITEBACK |
4174 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
4175 | #endif | |
73f576c0 | 4176 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
4177 | return memcg; |
4178 | fail: | |
73f576c0 JW |
4179 | if (memcg->id.id > 0) |
4180 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
40e952f9 | 4181 | __mem_cgroup_free(memcg); |
0b8f73e1 | 4182 | return NULL; |
d142e3e6 GC |
4183 | } |
4184 | ||
0b8f73e1 JW |
4185 | static struct cgroup_subsys_state * __ref |
4186 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 4187 | { |
0b8f73e1 JW |
4188 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
4189 | struct mem_cgroup *memcg; | |
4190 | long error = -ENOMEM; | |
d142e3e6 | 4191 | |
0b8f73e1 JW |
4192 | memcg = mem_cgroup_alloc(); |
4193 | if (!memcg) | |
4194 | return ERR_PTR(error); | |
d142e3e6 | 4195 | |
0b8f73e1 JW |
4196 | memcg->high = PAGE_COUNTER_MAX; |
4197 | memcg->soft_limit = PAGE_COUNTER_MAX; | |
4198 | if (parent) { | |
4199 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
4200 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
4201 | } | |
4202 | if (parent && parent->use_hierarchy) { | |
4203 | memcg->use_hierarchy = true; | |
3e32cb2e | 4204 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 4205 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e JW |
4206 | page_counter_init(&memcg->memsw, &parent->memsw); |
4207 | page_counter_init(&memcg->kmem, &parent->kmem); | |
0db15298 | 4208 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 4209 | } else { |
3e32cb2e | 4210 | page_counter_init(&memcg->memory, NULL); |
37e84351 | 4211 | page_counter_init(&memcg->swap, NULL); |
3e32cb2e JW |
4212 | page_counter_init(&memcg->memsw, NULL); |
4213 | page_counter_init(&memcg->kmem, NULL); | |
0db15298 | 4214 | page_counter_init(&memcg->tcpmem, NULL); |
8c7f6edb TH |
4215 | /* |
4216 | * Deeper hierachy with use_hierarchy == false doesn't make | |
4217 | * much sense so let cgroup subsystem know about this | |
4218 | * unfortunate state in our controller. | |
4219 | */ | |
d142e3e6 | 4220 | if (parent != root_mem_cgroup) |
073219e9 | 4221 | memory_cgrp_subsys.broken_hierarchy = true; |
18f59ea7 | 4222 | } |
d6441637 | 4223 | |
0b8f73e1 JW |
4224 | /* The following stuff does not apply to the root */ |
4225 | if (!parent) { | |
4226 | root_mem_cgroup = memcg; | |
4227 | return &memcg->css; | |
4228 | } | |
4229 | ||
b313aeee | 4230 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
4231 | if (error) |
4232 | goto fail; | |
127424c8 | 4233 | |
f7e1cb6e | 4234 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4235 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 4236 | |
0b8f73e1 JW |
4237 | return &memcg->css; |
4238 | fail: | |
4239 | mem_cgroup_free(memcg); | |
ea3a9645 | 4240 | return ERR_PTR(-ENOMEM); |
0b8f73e1 JW |
4241 | } |
4242 | ||
73f576c0 | 4243 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 4244 | { |
58fa2a55 VD |
4245 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
4246 | ||
73f576c0 | 4247 | /* Online state pins memcg ID, memcg ID pins CSS */ |
58fa2a55 | 4248 | atomic_set(&memcg->id.ref, 1); |
73f576c0 | 4249 | css_get(css); |
2f7dd7a4 | 4250 | return 0; |
8cdea7c0 BS |
4251 | } |
4252 | ||
eb95419b | 4253 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 4254 | { |
eb95419b | 4255 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4256 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
4257 | |
4258 | /* | |
4259 | * Unregister events and notify userspace. | |
4260 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
4261 | * directory to avoid race between userspace and kernelspace. | |
4262 | */ | |
fba94807 TH |
4263 | spin_lock(&memcg->event_list_lock); |
4264 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
4265 | list_del_init(&event->list); |
4266 | schedule_work(&event->remove); | |
4267 | } | |
fba94807 | 4268 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 4269 | |
63677c74 RG |
4270 | memcg->low = 0; |
4271 | ||
567e9ab2 | 4272 | memcg_offline_kmem(memcg); |
52ebea74 | 4273 | wb_memcg_offline(memcg); |
73f576c0 JW |
4274 | |
4275 | mem_cgroup_id_put(memcg); | |
df878fb0 KH |
4276 | } |
4277 | ||
6df38689 VD |
4278 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
4279 | { | |
4280 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4281 | ||
4282 | invalidate_reclaim_iterators(memcg); | |
4283 | } | |
4284 | ||
eb95419b | 4285 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 4286 | { |
eb95419b | 4287 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
c268e994 | 4288 | |
f7e1cb6e | 4289 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 4290 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 4291 | |
0db15298 | 4292 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 4293 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 4294 | |
0b8f73e1 JW |
4295 | vmpressure_cleanup(&memcg->vmpressure); |
4296 | cancel_work_sync(&memcg->high_work); | |
4297 | mem_cgroup_remove_from_trees(memcg); | |
d886f4e4 | 4298 | memcg_free_kmem(memcg); |
0b8f73e1 | 4299 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
4300 | } |
4301 | ||
1ced953b TH |
4302 | /** |
4303 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
4304 | * @css: the target css | |
4305 | * | |
4306 | * Reset the states of the mem_cgroup associated with @css. This is | |
4307 | * invoked when the userland requests disabling on the default hierarchy | |
4308 | * but the memcg is pinned through dependency. The memcg should stop | |
4309 | * applying policies and should revert to the vanilla state as it may be | |
4310 | * made visible again. | |
4311 | * | |
4312 | * The current implementation only resets the essential configurations. | |
4313 | * This needs to be expanded to cover all the visible parts. | |
4314 | */ | |
4315 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
4316 | { | |
4317 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
4318 | ||
d334c9bc VD |
4319 | page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX); |
4320 | page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX); | |
4321 | page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX); | |
4322 | page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX); | |
4323 | page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
241994ed JW |
4324 | memcg->low = 0; |
4325 | memcg->high = PAGE_COUNTER_MAX; | |
24d404dc | 4326 | memcg->soft_limit = PAGE_COUNTER_MAX; |
2529bb3a | 4327 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
4328 | } |
4329 | ||
02491447 | 4330 | #ifdef CONFIG_MMU |
7dc74be0 | 4331 | /* Handlers for move charge at task migration. */ |
854ffa8d | 4332 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 4333 | { |
05b84301 | 4334 | int ret; |
9476db97 | 4335 | |
d0164adc MG |
4336 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
4337 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 4338 | if (!ret) { |
854ffa8d | 4339 | mc.precharge += count; |
854ffa8d DN |
4340 | return ret; |
4341 | } | |
9476db97 | 4342 | |
3674534b | 4343 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 4344 | while (count--) { |
3674534b | 4345 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 4346 | if (ret) |
38c5d72f | 4347 | return ret; |
854ffa8d | 4348 | mc.precharge++; |
9476db97 | 4349 | cond_resched(); |
854ffa8d | 4350 | } |
9476db97 | 4351 | return 0; |
4ffef5fe DN |
4352 | } |
4353 | ||
4ffef5fe DN |
4354 | union mc_target { |
4355 | struct page *page; | |
02491447 | 4356 | swp_entry_t ent; |
4ffef5fe DN |
4357 | }; |
4358 | ||
4ffef5fe | 4359 | enum mc_target_type { |
8d32ff84 | 4360 | MC_TARGET_NONE = 0, |
4ffef5fe | 4361 | MC_TARGET_PAGE, |
02491447 | 4362 | MC_TARGET_SWAP, |
c733a828 | 4363 | MC_TARGET_DEVICE, |
4ffef5fe DN |
4364 | }; |
4365 | ||
90254a65 DN |
4366 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4367 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4368 | { |
c733a828 | 4369 | struct page *page = _vm_normal_page(vma, addr, ptent, true); |
4ffef5fe | 4370 | |
90254a65 DN |
4371 | if (!page || !page_mapped(page)) |
4372 | return NULL; | |
4373 | if (PageAnon(page)) { | |
1dfab5ab | 4374 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 4375 | return NULL; |
1dfab5ab JW |
4376 | } else { |
4377 | if (!(mc.flags & MOVE_FILE)) | |
4378 | return NULL; | |
4379 | } | |
90254a65 DN |
4380 | if (!get_page_unless_zero(page)) |
4381 | return NULL; | |
4382 | ||
4383 | return page; | |
4384 | } | |
4385 | ||
c733a828 | 4386 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 4387 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 4388 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 4389 | { |
90254a65 DN |
4390 | struct page *page = NULL; |
4391 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4392 | ||
1dfab5ab | 4393 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
90254a65 | 4394 | return NULL; |
c733a828 JG |
4395 | |
4396 | /* | |
4397 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
4398 | * a device and because they are not accessible by CPU they are store | |
4399 | * as special swap entry in the CPU page table. | |
4400 | */ | |
4401 | if (is_device_private_entry(ent)) { | |
4402 | page = device_private_entry_to_page(ent); | |
4403 | /* | |
4404 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
4405 | * a refcount of 1 when free (unlike normal page) | |
4406 | */ | |
4407 | if (!page_ref_add_unless(page, 1, 1)) | |
4408 | return NULL; | |
4409 | return page; | |
4410 | } | |
4411 | ||
4b91355e KH |
4412 | /* |
4413 | * Because lookup_swap_cache() updates some statistics counter, | |
4414 | * we call find_get_page() with swapper_space directly. | |
4415 | */ | |
f6ab1f7f | 4416 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
7941d214 | 4417 | if (do_memsw_account()) |
90254a65 DN |
4418 | entry->val = ent.val; |
4419 | ||
4420 | return page; | |
4421 | } | |
4b91355e KH |
4422 | #else |
4423 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 4424 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
4425 | { |
4426 | return NULL; | |
4427 | } | |
4428 | #endif | |
90254a65 | 4429 | |
87946a72 DN |
4430 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4431 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4432 | { | |
4433 | struct page *page = NULL; | |
87946a72 DN |
4434 | struct address_space *mapping; |
4435 | pgoff_t pgoff; | |
4436 | ||
4437 | if (!vma->vm_file) /* anonymous vma */ | |
4438 | return NULL; | |
1dfab5ab | 4439 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
4440 | return NULL; |
4441 | ||
87946a72 | 4442 | mapping = vma->vm_file->f_mapping; |
0661a336 | 4443 | pgoff = linear_page_index(vma, addr); |
87946a72 DN |
4444 | |
4445 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
4446 | #ifdef CONFIG_SWAP |
4447 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
139b6a6f JW |
4448 | if (shmem_mapping(mapping)) { |
4449 | page = find_get_entry(mapping, pgoff); | |
4450 | if (radix_tree_exceptional_entry(page)) { | |
4451 | swp_entry_t swp = radix_to_swp_entry(page); | |
7941d214 | 4452 | if (do_memsw_account()) |
139b6a6f | 4453 | *entry = swp; |
f6ab1f7f HY |
4454 | page = find_get_page(swap_address_space(swp), |
4455 | swp_offset(swp)); | |
139b6a6f JW |
4456 | } |
4457 | } else | |
4458 | page = find_get_page(mapping, pgoff); | |
4459 | #else | |
4460 | page = find_get_page(mapping, pgoff); | |
aa3b1895 | 4461 | #endif |
87946a72 DN |
4462 | return page; |
4463 | } | |
4464 | ||
b1b0deab CG |
4465 | /** |
4466 | * mem_cgroup_move_account - move account of the page | |
4467 | * @page: the page | |
25843c2b | 4468 | * @compound: charge the page as compound or small page |
b1b0deab CG |
4469 | * @from: mem_cgroup which the page is moved from. |
4470 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
4471 | * | |
3ac808fd | 4472 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
4473 | * |
4474 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
4475 | * from old cgroup. | |
4476 | */ | |
4477 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 4478 | bool compound, |
b1b0deab CG |
4479 | struct mem_cgroup *from, |
4480 | struct mem_cgroup *to) | |
4481 | { | |
4482 | unsigned long flags; | |
f627c2f5 | 4483 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
b1b0deab | 4484 | int ret; |
c4843a75 | 4485 | bool anon; |
b1b0deab CG |
4486 | |
4487 | VM_BUG_ON(from == to); | |
4488 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 4489 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
4490 | |
4491 | /* | |
6a93ca8f | 4492 | * Prevent mem_cgroup_migrate() from looking at |
45637bab | 4493 | * page->mem_cgroup of its source page while we change it. |
b1b0deab | 4494 | */ |
f627c2f5 | 4495 | ret = -EBUSY; |
b1b0deab CG |
4496 | if (!trylock_page(page)) |
4497 | goto out; | |
4498 | ||
4499 | ret = -EINVAL; | |
4500 | if (page->mem_cgroup != from) | |
4501 | goto out_unlock; | |
4502 | ||
c4843a75 GT |
4503 | anon = PageAnon(page); |
4504 | ||
b1b0deab CG |
4505 | spin_lock_irqsave(&from->move_lock, flags); |
4506 | ||
c4843a75 | 4507 | if (!anon && page_mapped(page)) { |
c9019e9b JW |
4508 | __mod_memcg_state(from, NR_FILE_MAPPED, -nr_pages); |
4509 | __mod_memcg_state(to, NR_FILE_MAPPED, nr_pages); | |
b1b0deab CG |
4510 | } |
4511 | ||
c4843a75 GT |
4512 | /* |
4513 | * move_lock grabbed above and caller set from->moving_account, so | |
ccda7f43 | 4514 | * mod_memcg_page_state will serialize updates to PageDirty. |
c4843a75 GT |
4515 | * So mapping should be stable for dirty pages. |
4516 | */ | |
4517 | if (!anon && PageDirty(page)) { | |
4518 | struct address_space *mapping = page_mapping(page); | |
4519 | ||
4520 | if (mapping_cap_account_dirty(mapping)) { | |
c9019e9b JW |
4521 | __mod_memcg_state(from, NR_FILE_DIRTY, -nr_pages); |
4522 | __mod_memcg_state(to, NR_FILE_DIRTY, nr_pages); | |
c4843a75 GT |
4523 | } |
4524 | } | |
4525 | ||
b1b0deab | 4526 | if (PageWriteback(page)) { |
c9019e9b JW |
4527 | __mod_memcg_state(from, NR_WRITEBACK, -nr_pages); |
4528 | __mod_memcg_state(to, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
4529 | } |
4530 | ||
4531 | /* | |
4532 | * It is safe to change page->mem_cgroup here because the page | |
4533 | * is referenced, charged, and isolated - we can't race with | |
4534 | * uncharging, charging, migration, or LRU putback. | |
4535 | */ | |
4536 | ||
4537 | /* caller should have done css_get */ | |
4538 | page->mem_cgroup = to; | |
4539 | spin_unlock_irqrestore(&from->move_lock, flags); | |
4540 | ||
4541 | ret = 0; | |
4542 | ||
4543 | local_irq_disable(); | |
f627c2f5 | 4544 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
b1b0deab | 4545 | memcg_check_events(to, page); |
f627c2f5 | 4546 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
b1b0deab CG |
4547 | memcg_check_events(from, page); |
4548 | local_irq_enable(); | |
4549 | out_unlock: | |
4550 | unlock_page(page); | |
4551 | out: | |
4552 | return ret; | |
4553 | } | |
4554 | ||
7cf7806c LR |
4555 | /** |
4556 | * get_mctgt_type - get target type of moving charge | |
4557 | * @vma: the vma the pte to be checked belongs | |
4558 | * @addr: the address corresponding to the pte to be checked | |
4559 | * @ptent: the pte to be checked | |
4560 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
4561 | * | |
4562 | * Returns | |
4563 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4564 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4565 | * move charge. if @target is not NULL, the page is stored in target->page | |
4566 | * with extra refcnt got(Callers should handle it). | |
4567 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
4568 | * target for charge migration. if @target is not NULL, the entry is stored | |
4569 | * in target->ent. | |
df6ad698 JG |
4570 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PUBLIC |
4571 | * or MEMORY_DEVICE_PRIVATE (so ZONE_DEVICE page and thus not on the lru). | |
4572 | * For now we such page is charge like a regular page would be as for all | |
4573 | * intent and purposes it is just special memory taking the place of a | |
4574 | * regular page. | |
c733a828 JG |
4575 | * |
4576 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
4577 | * |
4578 | * Called with pte lock held. | |
4579 | */ | |
4580 | ||
8d32ff84 | 4581 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
4582 | unsigned long addr, pte_t ptent, union mc_target *target) |
4583 | { | |
4584 | struct page *page = NULL; | |
8d32ff84 | 4585 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
4586 | swp_entry_t ent = { .val = 0 }; |
4587 | ||
4588 | if (pte_present(ptent)) | |
4589 | page = mc_handle_present_pte(vma, addr, ptent); | |
4590 | else if (is_swap_pte(ptent)) | |
48406ef8 | 4591 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 4592 | else if (pte_none(ptent)) |
87946a72 | 4593 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
4594 | |
4595 | if (!page && !ent.val) | |
8d32ff84 | 4596 | return ret; |
02491447 | 4597 | if (page) { |
02491447 | 4598 | /* |
0a31bc97 | 4599 | * Do only loose check w/o serialization. |
1306a85a | 4600 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 4601 | * not under LRU exclusion. |
02491447 | 4602 | */ |
1306a85a | 4603 | if (page->mem_cgroup == mc.from) { |
02491447 | 4604 | ret = MC_TARGET_PAGE; |
df6ad698 JG |
4605 | if (is_device_private_page(page) || |
4606 | is_device_public_page(page)) | |
c733a828 | 4607 | ret = MC_TARGET_DEVICE; |
02491447 DN |
4608 | if (target) |
4609 | target->page = page; | |
4610 | } | |
4611 | if (!ret || !target) | |
4612 | put_page(page); | |
4613 | } | |
3e14a57b HY |
4614 | /* |
4615 | * There is a swap entry and a page doesn't exist or isn't charged. | |
4616 | * But we cannot move a tail-page in a THP. | |
4617 | */ | |
4618 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 4619 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
4620 | ret = MC_TARGET_SWAP; |
4621 | if (target) | |
4622 | target->ent = ent; | |
4ffef5fe | 4623 | } |
4ffef5fe DN |
4624 | return ret; |
4625 | } | |
4626 | ||
12724850 NH |
4627 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
4628 | /* | |
d6810d73 HY |
4629 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
4630 | * not support them for now. | |
12724850 NH |
4631 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
4632 | */ | |
4633 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4634 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4635 | { | |
4636 | struct page *page = NULL; | |
12724850 NH |
4637 | enum mc_target_type ret = MC_TARGET_NONE; |
4638 | ||
84c3fc4e ZY |
4639 | if (unlikely(is_swap_pmd(pmd))) { |
4640 | VM_BUG_ON(thp_migration_supported() && | |
4641 | !is_pmd_migration_entry(pmd)); | |
4642 | return ret; | |
4643 | } | |
12724850 | 4644 | page = pmd_page(pmd); |
309381fe | 4645 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 4646 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 4647 | return ret; |
1306a85a | 4648 | if (page->mem_cgroup == mc.from) { |
12724850 NH |
4649 | ret = MC_TARGET_PAGE; |
4650 | if (target) { | |
4651 | get_page(page); | |
4652 | target->page = page; | |
4653 | } | |
4654 | } | |
4655 | return ret; | |
4656 | } | |
4657 | #else | |
4658 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
4659 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
4660 | { | |
4661 | return MC_TARGET_NONE; | |
4662 | } | |
4663 | #endif | |
4664 | ||
4ffef5fe DN |
4665 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
4666 | unsigned long addr, unsigned long end, | |
4667 | struct mm_walk *walk) | |
4668 | { | |
26bcd64a | 4669 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4670 | pte_t *pte; |
4671 | spinlock_t *ptl; | |
4672 | ||
b6ec57f4 KS |
4673 | ptl = pmd_trans_huge_lock(pmd, vma); |
4674 | if (ptl) { | |
c733a828 JG |
4675 | /* |
4676 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
4677 | * support transparent huge page with MEMORY_DEVICE_PUBLIC or | |
4678 | * MEMORY_DEVICE_PRIVATE but this might change. | |
4679 | */ | |
12724850 NH |
4680 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
4681 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 4682 | spin_unlock(ptl); |
1a5a9906 | 4683 | return 0; |
12724850 | 4684 | } |
03319327 | 4685 | |
45f83cef AA |
4686 | if (pmd_trans_unstable(pmd)) |
4687 | return 0; | |
4ffef5fe DN |
4688 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4689 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 4690 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
4691 | mc.precharge++; /* increment precharge temporarily */ |
4692 | pte_unmap_unlock(pte - 1, ptl); | |
4693 | cond_resched(); | |
4694 | ||
7dc74be0 DN |
4695 | return 0; |
4696 | } | |
4697 | ||
4ffef5fe DN |
4698 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4699 | { | |
4700 | unsigned long precharge; | |
4ffef5fe | 4701 | |
26bcd64a NH |
4702 | struct mm_walk mem_cgroup_count_precharge_walk = { |
4703 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4704 | .mm = mm, | |
4705 | }; | |
dfe076b0 | 4706 | down_read(&mm->mmap_sem); |
0247f3f4 JM |
4707 | walk_page_range(0, mm->highest_vm_end, |
4708 | &mem_cgroup_count_precharge_walk); | |
dfe076b0 | 4709 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4710 | |
4711 | precharge = mc.precharge; | |
4712 | mc.precharge = 0; | |
4713 | ||
4714 | return precharge; | |
4715 | } | |
4716 | ||
4ffef5fe DN |
4717 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4718 | { | |
dfe076b0 DN |
4719 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4720 | ||
4721 | VM_BUG_ON(mc.moving_task); | |
4722 | mc.moving_task = current; | |
4723 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4724 | } |
4725 | ||
dfe076b0 DN |
4726 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4727 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4728 | { |
2bd9bb20 KH |
4729 | struct mem_cgroup *from = mc.from; |
4730 | struct mem_cgroup *to = mc.to; | |
4731 | ||
4ffef5fe | 4732 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 4733 | if (mc.precharge) { |
00501b53 | 4734 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
4735 | mc.precharge = 0; |
4736 | } | |
4737 | /* | |
4738 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4739 | * we must uncharge here. | |
4740 | */ | |
4741 | if (mc.moved_charge) { | |
00501b53 | 4742 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 4743 | mc.moved_charge = 0; |
4ffef5fe | 4744 | } |
483c30b5 DN |
4745 | /* we must fixup refcnts and charges */ |
4746 | if (mc.moved_swap) { | |
483c30b5 | 4747 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 4748 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 4749 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 4750 | |
615d66c3 VD |
4751 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
4752 | ||
05b84301 | 4753 | /* |
3e32cb2e JW |
4754 | * we charged both to->memory and to->memsw, so we |
4755 | * should uncharge to->memory. | |
05b84301 | 4756 | */ |
ce00a967 | 4757 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
4758 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
4759 | ||
615d66c3 VD |
4760 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
4761 | css_put_many(&mc.to->css, mc.moved_swap); | |
3e32cb2e | 4762 | |
483c30b5 DN |
4763 | mc.moved_swap = 0; |
4764 | } | |
dfe076b0 DN |
4765 | memcg_oom_recover(from); |
4766 | memcg_oom_recover(to); | |
4767 | wake_up_all(&mc.waitq); | |
4768 | } | |
4769 | ||
4770 | static void mem_cgroup_clear_mc(void) | |
4771 | { | |
264a0ae1 TH |
4772 | struct mm_struct *mm = mc.mm; |
4773 | ||
dfe076b0 DN |
4774 | /* |
4775 | * we must clear moving_task before waking up waiters at the end of | |
4776 | * task migration. | |
4777 | */ | |
4778 | mc.moving_task = NULL; | |
4779 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4780 | spin_lock(&mc.lock); |
4ffef5fe DN |
4781 | mc.from = NULL; |
4782 | mc.to = NULL; | |
264a0ae1 | 4783 | mc.mm = NULL; |
2bd9bb20 | 4784 | spin_unlock(&mc.lock); |
264a0ae1 TH |
4785 | |
4786 | mmput(mm); | |
4ffef5fe DN |
4787 | } |
4788 | ||
1f7dd3e5 | 4789 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4790 | { |
1f7dd3e5 | 4791 | struct cgroup_subsys_state *css; |
eed67d75 | 4792 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 4793 | struct mem_cgroup *from; |
4530eddb | 4794 | struct task_struct *leader, *p; |
9f2115f9 | 4795 | struct mm_struct *mm; |
1dfab5ab | 4796 | unsigned long move_flags; |
9f2115f9 | 4797 | int ret = 0; |
7dc74be0 | 4798 | |
1f7dd3e5 TH |
4799 | /* charge immigration isn't supported on the default hierarchy */ |
4800 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
4801 | return 0; |
4802 | ||
4530eddb TH |
4803 | /* |
4804 | * Multi-process migrations only happen on the default hierarchy | |
4805 | * where charge immigration is not used. Perform charge | |
4806 | * immigration if @tset contains a leader and whine if there are | |
4807 | * multiple. | |
4808 | */ | |
4809 | p = NULL; | |
1f7dd3e5 | 4810 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
4811 | WARN_ON_ONCE(p); |
4812 | p = leader; | |
1f7dd3e5 | 4813 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
4814 | } |
4815 | if (!p) | |
4816 | return 0; | |
4817 | ||
1f7dd3e5 TH |
4818 | /* |
4819 | * We are now commited to this value whatever it is. Changes in this | |
4820 | * tunable will only affect upcoming migrations, not the current one. | |
4821 | * So we need to save it, and keep it going. | |
4822 | */ | |
4823 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
4824 | if (!move_flags) | |
4825 | return 0; | |
4826 | ||
9f2115f9 TH |
4827 | from = mem_cgroup_from_task(p); |
4828 | ||
4829 | VM_BUG_ON(from == memcg); | |
4830 | ||
4831 | mm = get_task_mm(p); | |
4832 | if (!mm) | |
4833 | return 0; | |
4834 | /* We move charges only when we move a owner of the mm */ | |
4835 | if (mm->owner == p) { | |
4836 | VM_BUG_ON(mc.from); | |
4837 | VM_BUG_ON(mc.to); | |
4838 | VM_BUG_ON(mc.precharge); | |
4839 | VM_BUG_ON(mc.moved_charge); | |
4840 | VM_BUG_ON(mc.moved_swap); | |
4841 | ||
4842 | spin_lock(&mc.lock); | |
264a0ae1 | 4843 | mc.mm = mm; |
9f2115f9 TH |
4844 | mc.from = from; |
4845 | mc.to = memcg; | |
4846 | mc.flags = move_flags; | |
4847 | spin_unlock(&mc.lock); | |
4848 | /* We set mc.moving_task later */ | |
4849 | ||
4850 | ret = mem_cgroup_precharge_mc(mm); | |
4851 | if (ret) | |
4852 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
4853 | } else { |
4854 | mmput(mm); | |
7dc74be0 DN |
4855 | } |
4856 | return ret; | |
4857 | } | |
4858 | ||
1f7dd3e5 | 4859 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 4860 | { |
4e2f245d JW |
4861 | if (mc.to) |
4862 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
4863 | } |
4864 | ||
4ffef5fe DN |
4865 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4866 | unsigned long addr, unsigned long end, | |
4867 | struct mm_walk *walk) | |
7dc74be0 | 4868 | { |
4ffef5fe | 4869 | int ret = 0; |
26bcd64a | 4870 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
4871 | pte_t *pte; |
4872 | spinlock_t *ptl; | |
12724850 NH |
4873 | enum mc_target_type target_type; |
4874 | union mc_target target; | |
4875 | struct page *page; | |
4ffef5fe | 4876 | |
b6ec57f4 KS |
4877 | ptl = pmd_trans_huge_lock(pmd, vma); |
4878 | if (ptl) { | |
62ade86a | 4879 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 4880 | spin_unlock(ptl); |
12724850 NH |
4881 | return 0; |
4882 | } | |
4883 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
4884 | if (target_type == MC_TARGET_PAGE) { | |
4885 | page = target.page; | |
4886 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 4887 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 4888 | mc.from, mc.to)) { |
12724850 NH |
4889 | mc.precharge -= HPAGE_PMD_NR; |
4890 | mc.moved_charge += HPAGE_PMD_NR; | |
4891 | } | |
4892 | putback_lru_page(page); | |
4893 | } | |
4894 | put_page(page); | |
c733a828 JG |
4895 | } else if (target_type == MC_TARGET_DEVICE) { |
4896 | page = target.page; | |
4897 | if (!mem_cgroup_move_account(page, true, | |
4898 | mc.from, mc.to)) { | |
4899 | mc.precharge -= HPAGE_PMD_NR; | |
4900 | mc.moved_charge += HPAGE_PMD_NR; | |
4901 | } | |
4902 | put_page(page); | |
12724850 | 4903 | } |
bf929152 | 4904 | spin_unlock(ptl); |
1a5a9906 | 4905 | return 0; |
12724850 NH |
4906 | } |
4907 | ||
45f83cef AA |
4908 | if (pmd_trans_unstable(pmd)) |
4909 | return 0; | |
4ffef5fe DN |
4910 | retry: |
4911 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4912 | for (; addr != end; addr += PAGE_SIZE) { | |
4913 | pte_t ptent = *(pte++); | |
c733a828 | 4914 | bool device = false; |
02491447 | 4915 | swp_entry_t ent; |
4ffef5fe DN |
4916 | |
4917 | if (!mc.precharge) | |
4918 | break; | |
4919 | ||
8d32ff84 | 4920 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
4921 | case MC_TARGET_DEVICE: |
4922 | device = true; | |
4923 | /* fall through */ | |
4ffef5fe DN |
4924 | case MC_TARGET_PAGE: |
4925 | page = target.page; | |
53f9263b KS |
4926 | /* |
4927 | * We can have a part of the split pmd here. Moving it | |
4928 | * can be done but it would be too convoluted so simply | |
4929 | * ignore such a partial THP and keep it in original | |
4930 | * memcg. There should be somebody mapping the head. | |
4931 | */ | |
4932 | if (PageTransCompound(page)) | |
4933 | goto put; | |
c733a828 | 4934 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 4935 | goto put; |
f627c2f5 KS |
4936 | if (!mem_cgroup_move_account(page, false, |
4937 | mc.from, mc.to)) { | |
4ffef5fe | 4938 | mc.precharge--; |
854ffa8d DN |
4939 | /* we uncharge from mc.from later. */ |
4940 | mc.moved_charge++; | |
4ffef5fe | 4941 | } |
c733a828 JG |
4942 | if (!device) |
4943 | putback_lru_page(page); | |
8d32ff84 | 4944 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
4945 | put_page(page); |
4946 | break; | |
02491447 DN |
4947 | case MC_TARGET_SWAP: |
4948 | ent = target.ent; | |
e91cbb42 | 4949 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 4950 | mc.precharge--; |
483c30b5 DN |
4951 | /* we fixup refcnts and charges later. */ |
4952 | mc.moved_swap++; | |
4953 | } | |
02491447 | 4954 | break; |
4ffef5fe DN |
4955 | default: |
4956 | break; | |
4957 | } | |
4958 | } | |
4959 | pte_unmap_unlock(pte - 1, ptl); | |
4960 | cond_resched(); | |
4961 | ||
4962 | if (addr != end) { | |
4963 | /* | |
4964 | * We have consumed all precharges we got in can_attach(). | |
4965 | * We try charge one by one, but don't do any additional | |
4966 | * charges to mc.to if we have failed in charge once in attach() | |
4967 | * phase. | |
4968 | */ | |
854ffa8d | 4969 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4970 | if (!ret) |
4971 | goto retry; | |
4972 | } | |
4973 | ||
4974 | return ret; | |
4975 | } | |
4976 | ||
264a0ae1 | 4977 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 4978 | { |
26bcd64a NH |
4979 | struct mm_walk mem_cgroup_move_charge_walk = { |
4980 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
264a0ae1 | 4981 | .mm = mc.mm, |
26bcd64a | 4982 | }; |
4ffef5fe DN |
4983 | |
4984 | lru_add_drain_all(); | |
312722cb | 4985 | /* |
81f8c3a4 JW |
4986 | * Signal lock_page_memcg() to take the memcg's move_lock |
4987 | * while we're moving its pages to another memcg. Then wait | |
4988 | * for already started RCU-only updates to finish. | |
312722cb JW |
4989 | */ |
4990 | atomic_inc(&mc.from->moving_account); | |
4991 | synchronize_rcu(); | |
dfe076b0 | 4992 | retry: |
264a0ae1 | 4993 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
dfe076b0 DN |
4994 | /* |
4995 | * Someone who are holding the mmap_sem might be waiting in | |
4996 | * waitq. So we cancel all extra charges, wake up all waiters, | |
4997 | * and retry. Because we cancel precharges, we might not be able | |
4998 | * to move enough charges, but moving charge is a best-effort | |
4999 | * feature anyway, so it wouldn't be a big problem. | |
5000 | */ | |
5001 | __mem_cgroup_clear_mc(); | |
5002 | cond_resched(); | |
5003 | goto retry; | |
5004 | } | |
26bcd64a NH |
5005 | /* |
5006 | * When we have consumed all precharges and failed in doing | |
5007 | * additional charge, the page walk just aborts. | |
5008 | */ | |
0247f3f4 JM |
5009 | walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk); |
5010 | ||
264a0ae1 | 5011 | up_read(&mc.mm->mmap_sem); |
312722cb | 5012 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5013 | } |
5014 | ||
264a0ae1 | 5015 | static void mem_cgroup_move_task(void) |
67e465a7 | 5016 | { |
264a0ae1 TH |
5017 | if (mc.to) { |
5018 | mem_cgroup_move_charge(); | |
a433658c | 5019 | mem_cgroup_clear_mc(); |
264a0ae1 | 5020 | } |
67e465a7 | 5021 | } |
5cfb80a7 | 5022 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5023 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5024 | { |
5025 | return 0; | |
5026 | } | |
1f7dd3e5 | 5027 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5028 | { |
5029 | } | |
264a0ae1 | 5030 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
5031 | { |
5032 | } | |
5033 | #endif | |
67e465a7 | 5034 | |
f00baae7 TH |
5035 | /* |
5036 | * Cgroup retains root cgroups across [un]mount cycles making it necessary | |
aa6ec29b TH |
5037 | * to verify whether we're attached to the default hierarchy on each mount |
5038 | * attempt. | |
f00baae7 | 5039 | */ |
eb95419b | 5040 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
f00baae7 TH |
5041 | { |
5042 | /* | |
aa6ec29b | 5043 | * use_hierarchy is forced on the default hierarchy. cgroup core |
f00baae7 TH |
5044 | * guarantees that @root doesn't have any children, so turning it |
5045 | * on for the root memcg is enough. | |
5046 | */ | |
9e10a130 | 5047 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
7feee590 VD |
5048 | root_mem_cgroup->use_hierarchy = true; |
5049 | else | |
5050 | root_mem_cgroup->use_hierarchy = false; | |
f00baae7 TH |
5051 | } |
5052 | ||
241994ed JW |
5053 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
5054 | struct cftype *cft) | |
5055 | { | |
f5fc3c5d JW |
5056 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5057 | ||
5058 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
5059 | } |
5060 | ||
5061 | static int memory_low_show(struct seq_file *m, void *v) | |
5062 | { | |
5063 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5064 | unsigned long low = READ_ONCE(memcg->low); |
241994ed JW |
5065 | |
5066 | if (low == PAGE_COUNTER_MAX) | |
d2973697 | 5067 | seq_puts(m, "max\n"); |
241994ed JW |
5068 | else |
5069 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); | |
5070 | ||
5071 | return 0; | |
5072 | } | |
5073 | ||
5074 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
5075 | char *buf, size_t nbytes, loff_t off) | |
5076 | { | |
5077 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
5078 | unsigned long low; | |
5079 | int err; | |
5080 | ||
5081 | buf = strstrip(buf); | |
d2973697 | 5082 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
5083 | if (err) |
5084 | return err; | |
5085 | ||
5086 | memcg->low = low; | |
5087 | ||
5088 | return nbytes; | |
5089 | } | |
5090 | ||
5091 | static int memory_high_show(struct seq_file *m, void *v) | |
5092 | { | |
5093 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5094 | unsigned long high = READ_ONCE(memcg->high); |
241994ed JW |
5095 | |
5096 | if (high == PAGE_COUNTER_MAX) | |
d2973697 | 5097 | seq_puts(m, "max\n"); |
241994ed JW |
5098 | else |
5099 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); | |
5100 | ||
5101 | return 0; | |
5102 | } | |
5103 | ||
5104 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
5105 | char *buf, size_t nbytes, loff_t off) | |
5106 | { | |
5107 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
588083bb | 5108 | unsigned long nr_pages; |
241994ed JW |
5109 | unsigned long high; |
5110 | int err; | |
5111 | ||
5112 | buf = strstrip(buf); | |
d2973697 | 5113 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
5114 | if (err) |
5115 | return err; | |
5116 | ||
5117 | memcg->high = high; | |
5118 | ||
588083bb JW |
5119 | nr_pages = page_counter_read(&memcg->memory); |
5120 | if (nr_pages > high) | |
5121 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
5122 | GFP_KERNEL, true); | |
5123 | ||
2529bb3a | 5124 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5125 | return nbytes; |
5126 | } | |
5127 | ||
5128 | static int memory_max_show(struct seq_file *m, void *v) | |
5129 | { | |
5130 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
4db0c3c2 | 5131 | unsigned long max = READ_ONCE(memcg->memory.limit); |
241994ed JW |
5132 | |
5133 | if (max == PAGE_COUNTER_MAX) | |
d2973697 | 5134 | seq_puts(m, "max\n"); |
241994ed JW |
5135 | else |
5136 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
5137 | ||
5138 | return 0; | |
5139 | } | |
5140 | ||
5141 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
5142 | char *buf, size_t nbytes, loff_t off) | |
5143 | { | |
5144 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
b6e6edcf JW |
5145 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
5146 | bool drained = false; | |
241994ed JW |
5147 | unsigned long max; |
5148 | int err; | |
5149 | ||
5150 | buf = strstrip(buf); | |
d2973697 | 5151 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
5152 | if (err) |
5153 | return err; | |
5154 | ||
b6e6edcf JW |
5155 | xchg(&memcg->memory.limit, max); |
5156 | ||
5157 | for (;;) { | |
5158 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
5159 | ||
5160 | if (nr_pages <= max) | |
5161 | break; | |
5162 | ||
5163 | if (signal_pending(current)) { | |
5164 | err = -EINTR; | |
5165 | break; | |
5166 | } | |
5167 | ||
5168 | if (!drained) { | |
5169 | drain_all_stock(memcg); | |
5170 | drained = true; | |
5171 | continue; | |
5172 | } | |
5173 | ||
5174 | if (nr_reclaims) { | |
5175 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
5176 | GFP_KERNEL, true)) | |
5177 | nr_reclaims--; | |
5178 | continue; | |
5179 | } | |
5180 | ||
31176c78 | 5181 | mem_cgroup_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
5182 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
5183 | break; | |
5184 | } | |
241994ed | 5185 | |
2529bb3a | 5186 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
5187 | return nbytes; |
5188 | } | |
5189 | ||
5190 | static int memory_events_show(struct seq_file *m, void *v) | |
5191 | { | |
5192 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
5193 | ||
ccda7f43 JW |
5194 | seq_printf(m, "low %lu\n", memcg_sum_events(memcg, MEMCG_LOW)); |
5195 | seq_printf(m, "high %lu\n", memcg_sum_events(memcg, MEMCG_HIGH)); | |
5196 | seq_printf(m, "max %lu\n", memcg_sum_events(memcg, MEMCG_MAX)); | |
5197 | seq_printf(m, "oom %lu\n", memcg_sum_events(memcg, MEMCG_OOM)); | |
8e675f7a | 5198 | seq_printf(m, "oom_kill %lu\n", memcg_sum_events(memcg, OOM_KILL)); |
241994ed JW |
5199 | |
5200 | return 0; | |
5201 | } | |
5202 | ||
587d9f72 JW |
5203 | static int memory_stat_show(struct seq_file *m, void *v) |
5204 | { | |
5205 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
72b54e73 VD |
5206 | unsigned long stat[MEMCG_NR_STAT]; |
5207 | unsigned long events[MEMCG_NR_EVENTS]; | |
587d9f72 JW |
5208 | int i; |
5209 | ||
5210 | /* | |
5211 | * Provide statistics on the state of the memory subsystem as | |
5212 | * well as cumulative event counters that show past behavior. | |
5213 | * | |
5214 | * This list is ordered following a combination of these gradients: | |
5215 | * 1) generic big picture -> specifics and details | |
5216 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
5217 | * | |
5218 | * Current memory state: | |
5219 | */ | |
5220 | ||
72b54e73 VD |
5221 | tree_stat(memcg, stat); |
5222 | tree_events(memcg, events); | |
5223 | ||
587d9f72 | 5224 | seq_printf(m, "anon %llu\n", |
71cd3113 | 5225 | (u64)stat[MEMCG_RSS] * PAGE_SIZE); |
587d9f72 | 5226 | seq_printf(m, "file %llu\n", |
71cd3113 | 5227 | (u64)stat[MEMCG_CACHE] * PAGE_SIZE); |
12580e4b | 5228 | seq_printf(m, "kernel_stack %llu\n", |
efdc9490 | 5229 | (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024); |
27ee57c9 | 5230 | seq_printf(m, "slab %llu\n", |
32049296 JW |
5231 | (u64)(stat[NR_SLAB_RECLAIMABLE] + |
5232 | stat[NR_SLAB_UNRECLAIMABLE]) * PAGE_SIZE); | |
b2807f07 | 5233 | seq_printf(m, "sock %llu\n", |
72b54e73 | 5234 | (u64)stat[MEMCG_SOCK] * PAGE_SIZE); |
587d9f72 | 5235 | |
9a4caf1e | 5236 | seq_printf(m, "shmem %llu\n", |
71cd3113 | 5237 | (u64)stat[NR_SHMEM] * PAGE_SIZE); |
587d9f72 | 5238 | seq_printf(m, "file_mapped %llu\n", |
71cd3113 | 5239 | (u64)stat[NR_FILE_MAPPED] * PAGE_SIZE); |
587d9f72 | 5240 | seq_printf(m, "file_dirty %llu\n", |
71cd3113 | 5241 | (u64)stat[NR_FILE_DIRTY] * PAGE_SIZE); |
587d9f72 | 5242 | seq_printf(m, "file_writeback %llu\n", |
71cd3113 | 5243 | (u64)stat[NR_WRITEBACK] * PAGE_SIZE); |
587d9f72 JW |
5244 | |
5245 | for (i = 0; i < NR_LRU_LISTS; i++) { | |
5246 | struct mem_cgroup *mi; | |
5247 | unsigned long val = 0; | |
5248 | ||
5249 | for_each_mem_cgroup_tree(mi, memcg) | |
5250 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)); | |
5251 | seq_printf(m, "%s %llu\n", | |
5252 | mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE); | |
5253 | } | |
5254 | ||
27ee57c9 | 5255 | seq_printf(m, "slab_reclaimable %llu\n", |
32049296 | 5256 | (u64)stat[NR_SLAB_RECLAIMABLE] * PAGE_SIZE); |
27ee57c9 | 5257 | seq_printf(m, "slab_unreclaimable %llu\n", |
32049296 | 5258 | (u64)stat[NR_SLAB_UNRECLAIMABLE] * PAGE_SIZE); |
27ee57c9 | 5259 | |
587d9f72 JW |
5260 | /* Accumulated memory events */ |
5261 | ||
df0e53d0 JW |
5262 | seq_printf(m, "pgfault %lu\n", events[PGFAULT]); |
5263 | seq_printf(m, "pgmajfault %lu\n", events[PGMAJFAULT]); | |
587d9f72 | 5264 | |
2262185c RG |
5265 | seq_printf(m, "pgrefill %lu\n", events[PGREFILL]); |
5266 | seq_printf(m, "pgscan %lu\n", events[PGSCAN_KSWAPD] + | |
5267 | events[PGSCAN_DIRECT]); | |
5268 | seq_printf(m, "pgsteal %lu\n", events[PGSTEAL_KSWAPD] + | |
5269 | events[PGSTEAL_DIRECT]); | |
5270 | seq_printf(m, "pgactivate %lu\n", events[PGACTIVATE]); | |
5271 | seq_printf(m, "pgdeactivate %lu\n", events[PGDEACTIVATE]); | |
5272 | seq_printf(m, "pglazyfree %lu\n", events[PGLAZYFREE]); | |
5273 | seq_printf(m, "pglazyfreed %lu\n", events[PGLAZYFREED]); | |
5274 | ||
2a2e4885 | 5275 | seq_printf(m, "workingset_refault %lu\n", |
71cd3113 | 5276 | stat[WORKINGSET_REFAULT]); |
2a2e4885 | 5277 | seq_printf(m, "workingset_activate %lu\n", |
71cd3113 | 5278 | stat[WORKINGSET_ACTIVATE]); |
2a2e4885 | 5279 | seq_printf(m, "workingset_nodereclaim %lu\n", |
71cd3113 | 5280 | stat[WORKINGSET_NODERECLAIM]); |
2a2e4885 | 5281 | |
587d9f72 JW |
5282 | return 0; |
5283 | } | |
5284 | ||
241994ed JW |
5285 | static struct cftype memory_files[] = { |
5286 | { | |
5287 | .name = "current", | |
f5fc3c5d | 5288 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
5289 | .read_u64 = memory_current_read, |
5290 | }, | |
5291 | { | |
5292 | .name = "low", | |
5293 | .flags = CFTYPE_NOT_ON_ROOT, | |
5294 | .seq_show = memory_low_show, | |
5295 | .write = memory_low_write, | |
5296 | }, | |
5297 | { | |
5298 | .name = "high", | |
5299 | .flags = CFTYPE_NOT_ON_ROOT, | |
5300 | .seq_show = memory_high_show, | |
5301 | .write = memory_high_write, | |
5302 | }, | |
5303 | { | |
5304 | .name = "max", | |
5305 | .flags = CFTYPE_NOT_ON_ROOT, | |
5306 | .seq_show = memory_max_show, | |
5307 | .write = memory_max_write, | |
5308 | }, | |
5309 | { | |
5310 | .name = "events", | |
5311 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 5312 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
5313 | .seq_show = memory_events_show, |
5314 | }, | |
587d9f72 JW |
5315 | { |
5316 | .name = "stat", | |
5317 | .flags = CFTYPE_NOT_ON_ROOT, | |
5318 | .seq_show = memory_stat_show, | |
5319 | }, | |
241994ed JW |
5320 | { } /* terminate */ |
5321 | }; | |
5322 | ||
073219e9 | 5323 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 5324 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 5325 | .css_online = mem_cgroup_css_online, |
92fb9748 | 5326 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 5327 | .css_released = mem_cgroup_css_released, |
92fb9748 | 5328 | .css_free = mem_cgroup_css_free, |
1ced953b | 5329 | .css_reset = mem_cgroup_css_reset, |
7dc74be0 DN |
5330 | .can_attach = mem_cgroup_can_attach, |
5331 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 5332 | .post_attach = mem_cgroup_move_task, |
f00baae7 | 5333 | .bind = mem_cgroup_bind, |
241994ed JW |
5334 | .dfl_cftypes = memory_files, |
5335 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 5336 | .early_init = 0, |
8cdea7c0 | 5337 | }; |
c077719b | 5338 | |
241994ed JW |
5339 | /** |
5340 | * mem_cgroup_low - check if memory consumption is below the normal range | |
34c81057 | 5341 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
5342 | * @memcg: the memory cgroup to check |
5343 | * | |
5344 | * Returns %true if memory consumption of @memcg, and that of all | |
34c81057 SC |
5345 | * ancestors up to (but not including) @root, is below the normal range. |
5346 | * | |
5347 | * @root is exclusive; it is never low when looked at directly and isn't | |
5348 | * checked when traversing the hierarchy. | |
5349 | * | |
5350 | * Excluding @root enables using memory.low to prioritize memory usage | |
5351 | * between cgroups within a subtree of the hierarchy that is limited by | |
5352 | * memory.high or memory.max. | |
5353 | * | |
5354 | * For example, given cgroup A with children B and C: | |
5355 | * | |
5356 | * A | |
5357 | * / \ | |
5358 | * B C | |
5359 | * | |
5360 | * and | |
5361 | * | |
5362 | * 1. A/memory.current > A/memory.high | |
5363 | * 2. A/B/memory.current < A/B/memory.low | |
5364 | * 3. A/C/memory.current >= A/C/memory.low | |
5365 | * | |
5366 | * As 'A' is high, i.e. triggers reclaim from 'A', and 'B' is low, we | |
5367 | * should reclaim from 'C' until 'A' is no longer high or until we can | |
5368 | * no longer reclaim from 'C'. If 'A', i.e. @root, isn't excluded by | |
5369 | * mem_cgroup_low when reclaming from 'A', then 'B' won't be considered | |
5370 | * low and we will reclaim indiscriminately from both 'B' and 'C'. | |
241994ed JW |
5371 | */ |
5372 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) | |
5373 | { | |
5374 | if (mem_cgroup_disabled()) | |
5375 | return false; | |
5376 | ||
34c81057 SC |
5377 | if (!root) |
5378 | root = root_mem_cgroup; | |
5379 | if (memcg == root) | |
241994ed JW |
5380 | return false; |
5381 | ||
34c81057 | 5382 | for (; memcg != root; memcg = parent_mem_cgroup(memcg)) { |
4e54dede | 5383 | if (page_counter_read(&memcg->memory) >= memcg->low) |
241994ed JW |
5384 | return false; |
5385 | } | |
34c81057 | 5386 | |
241994ed JW |
5387 | return true; |
5388 | } | |
5389 | ||
00501b53 JW |
5390 | /** |
5391 | * mem_cgroup_try_charge - try charging a page | |
5392 | * @page: page to charge | |
5393 | * @mm: mm context of the victim | |
5394 | * @gfp_mask: reclaim mode | |
5395 | * @memcgp: charged memcg return | |
25843c2b | 5396 | * @compound: charge the page as compound or small page |
00501b53 JW |
5397 | * |
5398 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
5399 | * pages according to @gfp_mask if necessary. | |
5400 | * | |
5401 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. | |
5402 | * Otherwise, an error code is returned. | |
5403 | * | |
5404 | * After page->mapping has been set up, the caller must finalize the | |
5405 | * charge with mem_cgroup_commit_charge(). Or abort the transaction | |
5406 | * with mem_cgroup_cancel_charge() in case page instantiation fails. | |
5407 | */ | |
5408 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, | |
f627c2f5 KS |
5409 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
5410 | bool compound) | |
00501b53 JW |
5411 | { |
5412 | struct mem_cgroup *memcg = NULL; | |
f627c2f5 | 5413 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5414 | int ret = 0; |
5415 | ||
5416 | if (mem_cgroup_disabled()) | |
5417 | goto out; | |
5418 | ||
5419 | if (PageSwapCache(page)) { | |
00501b53 JW |
5420 | /* |
5421 | * Every swap fault against a single page tries to charge the | |
5422 | * page, bail as early as possible. shmem_unuse() encounters | |
5423 | * already charged pages, too. The USED bit is protected by | |
5424 | * the page lock, which serializes swap cache removal, which | |
5425 | * in turn serializes uncharging. | |
5426 | */ | |
e993d905 | 5427 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
abe2895b | 5428 | if (compound_head(page)->mem_cgroup) |
00501b53 | 5429 | goto out; |
e993d905 | 5430 | |
37e84351 | 5431 | if (do_swap_account) { |
e993d905 VD |
5432 | swp_entry_t ent = { .val = page_private(page), }; |
5433 | unsigned short id = lookup_swap_cgroup_id(ent); | |
5434 | ||
5435 | rcu_read_lock(); | |
5436 | memcg = mem_cgroup_from_id(id); | |
5437 | if (memcg && !css_tryget_online(&memcg->css)) | |
5438 | memcg = NULL; | |
5439 | rcu_read_unlock(); | |
5440 | } | |
00501b53 JW |
5441 | } |
5442 | ||
00501b53 JW |
5443 | if (!memcg) |
5444 | memcg = get_mem_cgroup_from_mm(mm); | |
5445 | ||
5446 | ret = try_charge(memcg, gfp_mask, nr_pages); | |
5447 | ||
5448 | css_put(&memcg->css); | |
00501b53 JW |
5449 | out: |
5450 | *memcgp = memcg; | |
5451 | return ret; | |
5452 | } | |
5453 | ||
5454 | /** | |
5455 | * mem_cgroup_commit_charge - commit a page charge | |
5456 | * @page: page to charge | |
5457 | * @memcg: memcg to charge the page to | |
5458 | * @lrucare: page might be on LRU already | |
25843c2b | 5459 | * @compound: charge the page as compound or small page |
00501b53 JW |
5460 | * |
5461 | * Finalize a charge transaction started by mem_cgroup_try_charge(), | |
5462 | * after page->mapping has been set up. This must happen atomically | |
5463 | * as part of the page instantiation, i.e. under the page table lock | |
5464 | * for anonymous pages, under the page lock for page and swap cache. | |
5465 | * | |
5466 | * In addition, the page must not be on the LRU during the commit, to | |
5467 | * prevent racing with task migration. If it might be, use @lrucare. | |
5468 | * | |
5469 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. | |
5470 | */ | |
5471 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, | |
f627c2f5 | 5472 | bool lrucare, bool compound) |
00501b53 | 5473 | { |
f627c2f5 | 5474 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5475 | |
5476 | VM_BUG_ON_PAGE(!page->mapping, page); | |
5477 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); | |
5478 | ||
5479 | if (mem_cgroup_disabled()) | |
5480 | return; | |
5481 | /* | |
5482 | * Swap faults will attempt to charge the same page multiple | |
5483 | * times. But reuse_swap_page() might have removed the page | |
5484 | * from swapcache already, so we can't check PageSwapCache(). | |
5485 | */ | |
5486 | if (!memcg) | |
5487 | return; | |
5488 | ||
6abb5a86 JW |
5489 | commit_charge(page, memcg, lrucare); |
5490 | ||
6abb5a86 | 5491 | local_irq_disable(); |
f627c2f5 | 5492 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
6abb5a86 JW |
5493 | memcg_check_events(memcg, page); |
5494 | local_irq_enable(); | |
00501b53 | 5495 | |
7941d214 | 5496 | if (do_memsw_account() && PageSwapCache(page)) { |
00501b53 JW |
5497 | swp_entry_t entry = { .val = page_private(page) }; |
5498 | /* | |
5499 | * The swap entry might not get freed for a long time, | |
5500 | * let's not wait for it. The page already received a | |
5501 | * memory+swap charge, drop the swap entry duplicate. | |
5502 | */ | |
38d8b4e6 | 5503 | mem_cgroup_uncharge_swap(entry, nr_pages); |
00501b53 JW |
5504 | } |
5505 | } | |
5506 | ||
5507 | /** | |
5508 | * mem_cgroup_cancel_charge - cancel a page charge | |
5509 | * @page: page to charge | |
5510 | * @memcg: memcg to charge the page to | |
25843c2b | 5511 | * @compound: charge the page as compound or small page |
00501b53 JW |
5512 | * |
5513 | * Cancel a charge transaction started by mem_cgroup_try_charge(). | |
5514 | */ | |
f627c2f5 KS |
5515 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
5516 | bool compound) | |
00501b53 | 5517 | { |
f627c2f5 | 5518 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
00501b53 JW |
5519 | |
5520 | if (mem_cgroup_disabled()) | |
5521 | return; | |
5522 | /* | |
5523 | * Swap faults will attempt to charge the same page multiple | |
5524 | * times. But reuse_swap_page() might have removed the page | |
5525 | * from swapcache already, so we can't check PageSwapCache(). | |
5526 | */ | |
5527 | if (!memcg) | |
5528 | return; | |
5529 | ||
00501b53 JW |
5530 | cancel_charge(memcg, nr_pages); |
5531 | } | |
5532 | ||
a9d5adee JG |
5533 | struct uncharge_gather { |
5534 | struct mem_cgroup *memcg; | |
5535 | unsigned long pgpgout; | |
5536 | unsigned long nr_anon; | |
5537 | unsigned long nr_file; | |
5538 | unsigned long nr_kmem; | |
5539 | unsigned long nr_huge; | |
5540 | unsigned long nr_shmem; | |
5541 | struct page *dummy_page; | |
5542 | }; | |
5543 | ||
5544 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 5545 | { |
a9d5adee JG |
5546 | memset(ug, 0, sizeof(*ug)); |
5547 | } | |
5548 | ||
5549 | static void uncharge_batch(const struct uncharge_gather *ug) | |
5550 | { | |
5551 | unsigned long nr_pages = ug->nr_anon + ug->nr_file + ug->nr_kmem; | |
747db954 JW |
5552 | unsigned long flags; |
5553 | ||
a9d5adee JG |
5554 | if (!mem_cgroup_is_root(ug->memcg)) { |
5555 | page_counter_uncharge(&ug->memcg->memory, nr_pages); | |
7941d214 | 5556 | if (do_memsw_account()) |
a9d5adee JG |
5557 | page_counter_uncharge(&ug->memcg->memsw, nr_pages); |
5558 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) | |
5559 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
5560 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 5561 | } |
747db954 JW |
5562 | |
5563 | local_irq_save(flags); | |
c9019e9b JW |
5564 | __mod_memcg_state(ug->memcg, MEMCG_RSS, -ug->nr_anon); |
5565 | __mod_memcg_state(ug->memcg, MEMCG_CACHE, -ug->nr_file); | |
5566 | __mod_memcg_state(ug->memcg, MEMCG_RSS_HUGE, -ug->nr_huge); | |
5567 | __mod_memcg_state(ug->memcg, NR_SHMEM, -ug->nr_shmem); | |
5568 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); | |
a983b5eb | 5569 | __this_cpu_add(ug->memcg->stat_cpu->nr_page_events, nr_pages); |
a9d5adee | 5570 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 5571 | local_irq_restore(flags); |
e8ea14cc | 5572 | |
a9d5adee JG |
5573 | if (!mem_cgroup_is_root(ug->memcg)) |
5574 | css_put_many(&ug->memcg->css, nr_pages); | |
5575 | } | |
5576 | ||
5577 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
5578 | { | |
5579 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
3f2eb028 JG |
5580 | VM_BUG_ON_PAGE(page_count(page) && !is_zone_device_page(page) && |
5581 | !PageHWPoison(page) , page); | |
a9d5adee JG |
5582 | |
5583 | if (!page->mem_cgroup) | |
5584 | return; | |
5585 | ||
5586 | /* | |
5587 | * Nobody should be changing or seriously looking at | |
5588 | * page->mem_cgroup at this point, we have fully | |
5589 | * exclusive access to the page. | |
5590 | */ | |
5591 | ||
5592 | if (ug->memcg != page->mem_cgroup) { | |
5593 | if (ug->memcg) { | |
5594 | uncharge_batch(ug); | |
5595 | uncharge_gather_clear(ug); | |
5596 | } | |
5597 | ug->memcg = page->mem_cgroup; | |
5598 | } | |
5599 | ||
5600 | if (!PageKmemcg(page)) { | |
5601 | unsigned int nr_pages = 1; | |
5602 | ||
5603 | if (PageTransHuge(page)) { | |
5604 | nr_pages <<= compound_order(page); | |
5605 | ug->nr_huge += nr_pages; | |
5606 | } | |
5607 | if (PageAnon(page)) | |
5608 | ug->nr_anon += nr_pages; | |
5609 | else { | |
5610 | ug->nr_file += nr_pages; | |
5611 | if (PageSwapBacked(page)) | |
5612 | ug->nr_shmem += nr_pages; | |
5613 | } | |
5614 | ug->pgpgout++; | |
5615 | } else { | |
5616 | ug->nr_kmem += 1 << compound_order(page); | |
5617 | __ClearPageKmemcg(page); | |
5618 | } | |
5619 | ||
5620 | ug->dummy_page = page; | |
5621 | page->mem_cgroup = NULL; | |
747db954 JW |
5622 | } |
5623 | ||
5624 | static void uncharge_list(struct list_head *page_list) | |
5625 | { | |
a9d5adee | 5626 | struct uncharge_gather ug; |
747db954 | 5627 | struct list_head *next; |
a9d5adee JG |
5628 | |
5629 | uncharge_gather_clear(&ug); | |
747db954 | 5630 | |
8b592656 JW |
5631 | /* |
5632 | * Note that the list can be a single page->lru; hence the | |
5633 | * do-while loop instead of a simple list_for_each_entry(). | |
5634 | */ | |
747db954 JW |
5635 | next = page_list->next; |
5636 | do { | |
a9d5adee JG |
5637 | struct page *page; |
5638 | ||
747db954 JW |
5639 | page = list_entry(next, struct page, lru); |
5640 | next = page->lru.next; | |
5641 | ||
a9d5adee | 5642 | uncharge_page(page, &ug); |
747db954 JW |
5643 | } while (next != page_list); |
5644 | ||
a9d5adee JG |
5645 | if (ug.memcg) |
5646 | uncharge_batch(&ug); | |
747db954 JW |
5647 | } |
5648 | ||
0a31bc97 JW |
5649 | /** |
5650 | * mem_cgroup_uncharge - uncharge a page | |
5651 | * @page: page to uncharge | |
5652 | * | |
5653 | * Uncharge a page previously charged with mem_cgroup_try_charge() and | |
5654 | * mem_cgroup_commit_charge(). | |
5655 | */ | |
5656 | void mem_cgroup_uncharge(struct page *page) | |
5657 | { | |
a9d5adee JG |
5658 | struct uncharge_gather ug; |
5659 | ||
0a31bc97 JW |
5660 | if (mem_cgroup_disabled()) |
5661 | return; | |
5662 | ||
747db954 | 5663 | /* Don't touch page->lru of any random page, pre-check: */ |
1306a85a | 5664 | if (!page->mem_cgroup) |
0a31bc97 JW |
5665 | return; |
5666 | ||
a9d5adee JG |
5667 | uncharge_gather_clear(&ug); |
5668 | uncharge_page(page, &ug); | |
5669 | uncharge_batch(&ug); | |
747db954 | 5670 | } |
0a31bc97 | 5671 | |
747db954 JW |
5672 | /** |
5673 | * mem_cgroup_uncharge_list - uncharge a list of page | |
5674 | * @page_list: list of pages to uncharge | |
5675 | * | |
5676 | * Uncharge a list of pages previously charged with | |
5677 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). | |
5678 | */ | |
5679 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
5680 | { | |
5681 | if (mem_cgroup_disabled()) | |
5682 | return; | |
0a31bc97 | 5683 | |
747db954 JW |
5684 | if (!list_empty(page_list)) |
5685 | uncharge_list(page_list); | |
0a31bc97 JW |
5686 | } |
5687 | ||
5688 | /** | |
6a93ca8f JW |
5689 | * mem_cgroup_migrate - charge a page's replacement |
5690 | * @oldpage: currently circulating page | |
5691 | * @newpage: replacement page | |
0a31bc97 | 5692 | * |
6a93ca8f JW |
5693 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
5694 | * be uncharged upon free. | |
0a31bc97 JW |
5695 | * |
5696 | * Both pages must be locked, @newpage->mapping must be set up. | |
5697 | */ | |
6a93ca8f | 5698 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 5699 | { |
29833315 | 5700 | struct mem_cgroup *memcg; |
44b7a8d3 JW |
5701 | unsigned int nr_pages; |
5702 | bool compound; | |
d93c4130 | 5703 | unsigned long flags; |
0a31bc97 JW |
5704 | |
5705 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
5706 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 5707 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
5708 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
5709 | newpage); | |
0a31bc97 JW |
5710 | |
5711 | if (mem_cgroup_disabled()) | |
5712 | return; | |
5713 | ||
5714 | /* Page cache replacement: new page already charged? */ | |
1306a85a | 5715 | if (newpage->mem_cgroup) |
0a31bc97 JW |
5716 | return; |
5717 | ||
45637bab | 5718 | /* Swapcache readahead pages can get replaced before being charged */ |
1306a85a | 5719 | memcg = oldpage->mem_cgroup; |
29833315 | 5720 | if (!memcg) |
0a31bc97 JW |
5721 | return; |
5722 | ||
44b7a8d3 JW |
5723 | /* Force-charge the new page. The old one will be freed soon */ |
5724 | compound = PageTransHuge(newpage); | |
5725 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; | |
5726 | ||
5727 | page_counter_charge(&memcg->memory, nr_pages); | |
5728 | if (do_memsw_account()) | |
5729 | page_counter_charge(&memcg->memsw, nr_pages); | |
5730 | css_get_many(&memcg->css, nr_pages); | |
0a31bc97 | 5731 | |
9cf7666a | 5732 | commit_charge(newpage, memcg, false); |
44b7a8d3 | 5733 | |
d93c4130 | 5734 | local_irq_save(flags); |
44b7a8d3 JW |
5735 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
5736 | memcg_check_events(memcg, newpage); | |
d93c4130 | 5737 | local_irq_restore(flags); |
0a31bc97 JW |
5738 | } |
5739 | ||
ef12947c | 5740 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
5741 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
5742 | ||
2d758073 | 5743 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
5744 | { |
5745 | struct mem_cgroup *memcg; | |
5746 | ||
2d758073 JW |
5747 | if (!mem_cgroup_sockets_enabled) |
5748 | return; | |
5749 | ||
edbe69ef RG |
5750 | /* |
5751 | * Socket cloning can throw us here with sk_memcg already | |
5752 | * filled. It won't however, necessarily happen from | |
5753 | * process context. So the test for root memcg given | |
5754 | * the current task's memcg won't help us in this case. | |
5755 | * | |
5756 | * Respecting the original socket's memcg is a better | |
5757 | * decision in this case. | |
5758 | */ | |
5759 | if (sk->sk_memcg) { | |
5760 | css_get(&sk->sk_memcg->css); | |
5761 | return; | |
5762 | } | |
5763 | ||
11092087 JW |
5764 | rcu_read_lock(); |
5765 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
5766 | if (memcg == root_mem_cgroup) |
5767 | goto out; | |
0db15298 | 5768 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 5769 | goto out; |
f7e1cb6e | 5770 | if (css_tryget_online(&memcg->css)) |
11092087 | 5771 | sk->sk_memcg = memcg; |
f7e1cb6e | 5772 | out: |
11092087 JW |
5773 | rcu_read_unlock(); |
5774 | } | |
11092087 | 5775 | |
2d758073 | 5776 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 5777 | { |
2d758073 JW |
5778 | if (sk->sk_memcg) |
5779 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
5780 | } |
5781 | ||
5782 | /** | |
5783 | * mem_cgroup_charge_skmem - charge socket memory | |
5784 | * @memcg: memcg to charge | |
5785 | * @nr_pages: number of pages to charge | |
5786 | * | |
5787 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
5788 | * @memcg's configured limit, %false if the charge had to be forced. | |
5789 | */ | |
5790 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5791 | { | |
f7e1cb6e | 5792 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 5793 | |
f7e1cb6e | 5794 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5795 | struct page_counter *fail; |
f7e1cb6e | 5796 | |
0db15298 JW |
5797 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
5798 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
5799 | return true; |
5800 | } | |
0db15298 JW |
5801 | page_counter_charge(&memcg->tcpmem, nr_pages); |
5802 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 5803 | return false; |
11092087 | 5804 | } |
d886f4e4 | 5805 | |
f7e1cb6e JW |
5806 | /* Don't block in the packet receive path */ |
5807 | if (in_softirq()) | |
5808 | gfp_mask = GFP_NOWAIT; | |
5809 | ||
c9019e9b | 5810 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 5811 | |
f7e1cb6e JW |
5812 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
5813 | return true; | |
5814 | ||
5815 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
5816 | return false; |
5817 | } | |
5818 | ||
5819 | /** | |
5820 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
5821 | * @memcg - memcg to uncharge | |
5822 | * @nr_pages - number of pages to uncharge | |
5823 | */ | |
5824 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
5825 | { | |
f7e1cb6e | 5826 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 5827 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
5828 | return; |
5829 | } | |
d886f4e4 | 5830 | |
c9019e9b | 5831 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 5832 | |
475d0487 | 5833 | refill_stock(memcg, nr_pages); |
11092087 JW |
5834 | } |
5835 | ||
f7e1cb6e JW |
5836 | static int __init cgroup_memory(char *s) |
5837 | { | |
5838 | char *token; | |
5839 | ||
5840 | while ((token = strsep(&s, ",")) != NULL) { | |
5841 | if (!*token) | |
5842 | continue; | |
5843 | if (!strcmp(token, "nosocket")) | |
5844 | cgroup_memory_nosocket = true; | |
04823c83 VD |
5845 | if (!strcmp(token, "nokmem")) |
5846 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
5847 | } |
5848 | return 0; | |
5849 | } | |
5850 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 5851 | |
2d11085e | 5852 | /* |
1081312f MH |
5853 | * subsys_initcall() for memory controller. |
5854 | * | |
308167fc SAS |
5855 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
5856 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
5857 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
5858 | * should be initialized from here. | |
2d11085e MH |
5859 | */ |
5860 | static int __init mem_cgroup_init(void) | |
5861 | { | |
95a045f6 JW |
5862 | int cpu, node; |
5863 | ||
13583c3d VD |
5864 | #ifndef CONFIG_SLOB |
5865 | /* | |
5866 | * Kmem cache creation is mostly done with the slab_mutex held, | |
17cc4dfe TH |
5867 | * so use a workqueue with limited concurrency to avoid stalling |
5868 | * all worker threads in case lots of cgroups are created and | |
5869 | * destroyed simultaneously. | |
13583c3d | 5870 | */ |
17cc4dfe TH |
5871 | memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1); |
5872 | BUG_ON(!memcg_kmem_cache_wq); | |
13583c3d VD |
5873 | #endif |
5874 | ||
308167fc SAS |
5875 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
5876 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
5877 | |
5878 | for_each_possible_cpu(cpu) | |
5879 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
5880 | drain_local_stock); | |
5881 | ||
5882 | for_each_node(node) { | |
5883 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
5884 | |
5885 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
5886 | node_online(node) ? node : NUMA_NO_NODE); | |
5887 | ||
ef8f2327 | 5888 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 5889 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 5890 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
5891 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
5892 | } | |
5893 | ||
2d11085e MH |
5894 | return 0; |
5895 | } | |
5896 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
5897 | |
5898 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
5899 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
5900 | { | |
5901 | while (!atomic_inc_not_zero(&memcg->id.ref)) { | |
5902 | /* | |
5903 | * The root cgroup cannot be destroyed, so it's refcount must | |
5904 | * always be >= 1. | |
5905 | */ | |
5906 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
5907 | VM_BUG_ON(1); | |
5908 | break; | |
5909 | } | |
5910 | memcg = parent_mem_cgroup(memcg); | |
5911 | if (!memcg) | |
5912 | memcg = root_mem_cgroup; | |
5913 | } | |
5914 | return memcg; | |
5915 | } | |
5916 | ||
21afa38e JW |
5917 | /** |
5918 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
5919 | * @page: page whose memsw charge to transfer | |
5920 | * @entry: swap entry to move the charge to | |
5921 | * | |
5922 | * Transfer the memsw charge of @page to @entry. | |
5923 | */ | |
5924 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
5925 | { | |
1f47b61f | 5926 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 5927 | unsigned int nr_entries; |
21afa38e JW |
5928 | unsigned short oldid; |
5929 | ||
5930 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
5931 | VM_BUG_ON_PAGE(page_count(page), page); | |
5932 | ||
7941d214 | 5933 | if (!do_memsw_account()) |
21afa38e JW |
5934 | return; |
5935 | ||
5936 | memcg = page->mem_cgroup; | |
5937 | ||
5938 | /* Readahead page, never charged */ | |
5939 | if (!memcg) | |
5940 | return; | |
5941 | ||
1f47b61f VD |
5942 | /* |
5943 | * In case the memcg owning these pages has been offlined and doesn't | |
5944 | * have an ID allocated to it anymore, charge the closest online | |
5945 | * ancestor for the swap instead and transfer the memory+swap charge. | |
5946 | */ | |
5947 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
d6810d73 HY |
5948 | nr_entries = hpage_nr_pages(page); |
5949 | /* Get references for the tail pages, too */ | |
5950 | if (nr_entries > 1) | |
5951 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
5952 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
5953 | nr_entries); | |
21afa38e | 5954 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 5955 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e JW |
5956 | |
5957 | page->mem_cgroup = NULL; | |
5958 | ||
5959 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 5960 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 5961 | |
1f47b61f VD |
5962 | if (memcg != swap_memcg) { |
5963 | if (!mem_cgroup_is_root(swap_memcg)) | |
d6810d73 HY |
5964 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
5965 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
5966 | } |
5967 | ||
ce9ce665 SAS |
5968 | /* |
5969 | * Interrupts should be disabled here because the caller holds the | |
5970 | * mapping->tree_lock lock which is taken with interrupts-off. It is | |
5971 | * important here to have the interrupts disabled because it is the | |
5972 | * only synchronisation we have for udpating the per-CPU variables. | |
5973 | */ | |
5974 | VM_BUG_ON(!irqs_disabled()); | |
d6810d73 HY |
5975 | mem_cgroup_charge_statistics(memcg, page, PageTransHuge(page), |
5976 | -nr_entries); | |
21afa38e | 5977 | memcg_check_events(memcg, page); |
73f576c0 JW |
5978 | |
5979 | if (!mem_cgroup_is_root(memcg)) | |
d08afa14 | 5980 | css_put_many(&memcg->css, nr_entries); |
21afa38e JW |
5981 | } |
5982 | ||
38d8b4e6 HY |
5983 | /** |
5984 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
5985 | * @page: page being added to swap |
5986 | * @entry: swap entry to charge | |
5987 | * | |
38d8b4e6 | 5988 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
5989 | * |
5990 | * Returns 0 on success, -ENOMEM on failure. | |
5991 | */ | |
5992 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
5993 | { | |
38d8b4e6 | 5994 | unsigned int nr_pages = hpage_nr_pages(page); |
37e84351 | 5995 | struct page_counter *counter; |
38d8b4e6 | 5996 | struct mem_cgroup *memcg; |
37e84351 VD |
5997 | unsigned short oldid; |
5998 | ||
5999 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) | |
6000 | return 0; | |
6001 | ||
6002 | memcg = page->mem_cgroup; | |
6003 | ||
6004 | /* Readahead page, never charged */ | |
6005 | if (!memcg) | |
6006 | return 0; | |
6007 | ||
1f47b61f VD |
6008 | memcg = mem_cgroup_id_get_online(memcg); |
6009 | ||
37e84351 | 6010 | if (!mem_cgroup_is_root(memcg) && |
38d8b4e6 | 6011 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
1f47b61f | 6012 | mem_cgroup_id_put(memcg); |
37e84351 | 6013 | return -ENOMEM; |
1f47b61f | 6014 | } |
37e84351 | 6015 | |
38d8b4e6 HY |
6016 | /* Get references for the tail pages, too */ |
6017 | if (nr_pages > 1) | |
6018 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
6019 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 6020 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 6021 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 6022 | |
37e84351 VD |
6023 | return 0; |
6024 | } | |
6025 | ||
21afa38e | 6026 | /** |
38d8b4e6 | 6027 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 6028 | * @entry: swap entry to uncharge |
38d8b4e6 | 6029 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 6030 | */ |
38d8b4e6 | 6031 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
6032 | { |
6033 | struct mem_cgroup *memcg; | |
6034 | unsigned short id; | |
6035 | ||
37e84351 | 6036 | if (!do_swap_account) |
21afa38e JW |
6037 | return; |
6038 | ||
38d8b4e6 | 6039 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 6040 | rcu_read_lock(); |
adbe427b | 6041 | memcg = mem_cgroup_from_id(id); |
21afa38e | 6042 | if (memcg) { |
37e84351 VD |
6043 | if (!mem_cgroup_is_root(memcg)) { |
6044 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
38d8b4e6 | 6045 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 6046 | else |
38d8b4e6 | 6047 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 6048 | } |
c9019e9b | 6049 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 6050 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
6051 | } |
6052 | rcu_read_unlock(); | |
6053 | } | |
6054 | ||
d8b38438 VD |
6055 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
6056 | { | |
6057 | long nr_swap_pages = get_nr_swap_pages(); | |
6058 | ||
6059 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6060 | return nr_swap_pages; | |
6061 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
6062 | nr_swap_pages = min_t(long, nr_swap_pages, | |
6063 | READ_ONCE(memcg->swap.limit) - | |
6064 | page_counter_read(&memcg->swap)); | |
6065 | return nr_swap_pages; | |
6066 | } | |
6067 | ||
5ccc5aba VD |
6068 | bool mem_cgroup_swap_full(struct page *page) |
6069 | { | |
6070 | struct mem_cgroup *memcg; | |
6071 | ||
6072 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
6073 | ||
6074 | if (vm_swap_full()) | |
6075 | return true; | |
6076 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
6077 | return false; | |
6078 | ||
6079 | memcg = page->mem_cgroup; | |
6080 | if (!memcg) | |
6081 | return false; | |
6082 | ||
6083 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
6084 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit) | |
6085 | return true; | |
6086 | ||
6087 | return false; | |
6088 | } | |
6089 | ||
21afa38e JW |
6090 | /* for remember boot option*/ |
6091 | #ifdef CONFIG_MEMCG_SWAP_ENABLED | |
6092 | static int really_do_swap_account __initdata = 1; | |
6093 | #else | |
6094 | static int really_do_swap_account __initdata; | |
6095 | #endif | |
6096 | ||
6097 | static int __init enable_swap_account(char *s) | |
6098 | { | |
6099 | if (!strcmp(s, "1")) | |
6100 | really_do_swap_account = 1; | |
6101 | else if (!strcmp(s, "0")) | |
6102 | really_do_swap_account = 0; | |
6103 | return 1; | |
6104 | } | |
6105 | __setup("swapaccount=", enable_swap_account); | |
6106 | ||
37e84351 VD |
6107 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
6108 | struct cftype *cft) | |
6109 | { | |
6110 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
6111 | ||
6112 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
6113 | } | |
6114 | ||
6115 | static int swap_max_show(struct seq_file *m, void *v) | |
6116 | { | |
6117 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); | |
6118 | unsigned long max = READ_ONCE(memcg->swap.limit); | |
6119 | ||
6120 | if (max == PAGE_COUNTER_MAX) | |
6121 | seq_puts(m, "max\n"); | |
6122 | else | |
6123 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); | |
6124 | ||
6125 | return 0; | |
6126 | } | |
6127 | ||
6128 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
6129 | char *buf, size_t nbytes, loff_t off) | |
6130 | { | |
6131 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6132 | unsigned long max; | |
6133 | int err; | |
6134 | ||
6135 | buf = strstrip(buf); | |
6136 | err = page_counter_memparse(buf, "max", &max); | |
6137 | if (err) | |
6138 | return err; | |
6139 | ||
6140 | mutex_lock(&memcg_limit_mutex); | |
6141 | err = page_counter_limit(&memcg->swap, max); | |
6142 | mutex_unlock(&memcg_limit_mutex); | |
6143 | if (err) | |
6144 | return err; | |
6145 | ||
6146 | return nbytes; | |
6147 | } | |
6148 | ||
6149 | static struct cftype swap_files[] = { | |
6150 | { | |
6151 | .name = "swap.current", | |
6152 | .flags = CFTYPE_NOT_ON_ROOT, | |
6153 | .read_u64 = swap_current_read, | |
6154 | }, | |
6155 | { | |
6156 | .name = "swap.max", | |
6157 | .flags = CFTYPE_NOT_ON_ROOT, | |
6158 | .seq_show = swap_max_show, | |
6159 | .write = swap_max_write, | |
6160 | }, | |
6161 | { } /* terminate */ | |
6162 | }; | |
6163 | ||
21afa38e JW |
6164 | static struct cftype memsw_cgroup_files[] = { |
6165 | { | |
6166 | .name = "memsw.usage_in_bytes", | |
6167 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
6168 | .read_u64 = mem_cgroup_read_u64, | |
6169 | }, | |
6170 | { | |
6171 | .name = "memsw.max_usage_in_bytes", | |
6172 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
6173 | .write = mem_cgroup_reset, | |
6174 | .read_u64 = mem_cgroup_read_u64, | |
6175 | }, | |
6176 | { | |
6177 | .name = "memsw.limit_in_bytes", | |
6178 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
6179 | .write = mem_cgroup_write, | |
6180 | .read_u64 = mem_cgroup_read_u64, | |
6181 | }, | |
6182 | { | |
6183 | .name = "memsw.failcnt", | |
6184 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
6185 | .write = mem_cgroup_reset, | |
6186 | .read_u64 = mem_cgroup_read_u64, | |
6187 | }, | |
6188 | { }, /* terminate */ | |
6189 | }; | |
6190 | ||
6191 | static int __init mem_cgroup_swap_init(void) | |
6192 | { | |
6193 | if (!mem_cgroup_disabled() && really_do_swap_account) { | |
6194 | do_swap_account = 1; | |
37e84351 VD |
6195 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
6196 | swap_files)); | |
21afa38e JW |
6197 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
6198 | memsw_cgroup_files)); | |
6199 | } | |
6200 | return 0; | |
6201 | } | |
6202 | subsys_initcall(mem_cgroup_swap_init); | |
6203 | ||
6204 | #endif /* CONFIG_MEMCG_SWAP */ |