Commit | Line | Data |
---|---|---|
c942fddf | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
8cdea7c0 BS |
2 | /* memcontrol.c - Memory Controller |
3 | * | |
4 | * Copyright IBM Corporation, 2007 | |
5 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
6 | * | |
78fb7466 PE |
7 | * Copyright 2007 OpenVZ SWsoft Inc |
8 | * Author: Pavel Emelianov <xemul@openvz.org> | |
9 | * | |
2e72b634 KS |
10 | * Memory thresholds |
11 | * Copyright (C) 2009 Nokia Corporation | |
12 | * Author: Kirill A. Shutemov | |
13 | * | |
7ae1e1d0 GC |
14 | * Kernel Memory Controller |
15 | * Copyright (C) 2012 Parallels Inc. and Google Inc. | |
16 | * Authors: Glauber Costa and Suleiman Souhlal | |
17 | * | |
1575e68b JW |
18 | * Native page reclaim |
19 | * Charge lifetime sanitation | |
20 | * Lockless page tracking & accounting | |
21 | * Unified hierarchy configuration model | |
22 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner | |
6168d0da AS |
23 | * |
24 | * Per memcg lru locking | |
25 | * Copyright (C) 2020 Alibaba, Inc, Alex Shi | |
8cdea7c0 BS |
26 | */ |
27 | ||
3e32cb2e | 28 | #include <linux/page_counter.h> |
8cdea7c0 BS |
29 | #include <linux/memcontrol.h> |
30 | #include <linux/cgroup.h> | |
a520110e | 31 | #include <linux/pagewalk.h> |
6e84f315 | 32 | #include <linux/sched/mm.h> |
3a4f8a0b | 33 | #include <linux/shmem_fs.h> |
4ffef5fe | 34 | #include <linux/hugetlb.h> |
d13d1443 | 35 | #include <linux/pagemap.h> |
1ff9e6e1 | 36 | #include <linux/vm_event_item.h> |
d52aa412 | 37 | #include <linux/smp.h> |
8a9f3ccd | 38 | #include <linux/page-flags.h> |
66e1707b | 39 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
40 | #include <linux/bit_spinlock.h> |
41 | #include <linux/rcupdate.h> | |
e222432b | 42 | #include <linux/limits.h> |
b9e15baf | 43 | #include <linux/export.h> |
8c7c6e34 | 44 | #include <linux/mutex.h> |
bb4cc1a8 | 45 | #include <linux/rbtree.h> |
b6ac57d5 | 46 | #include <linux/slab.h> |
66e1707b | 47 | #include <linux/swap.h> |
02491447 | 48 | #include <linux/swapops.h> |
66e1707b | 49 | #include <linux/spinlock.h> |
2e72b634 | 50 | #include <linux/eventfd.h> |
79bd9814 | 51 | #include <linux/poll.h> |
2e72b634 | 52 | #include <linux/sort.h> |
66e1707b | 53 | #include <linux/fs.h> |
d2ceb9b7 | 54 | #include <linux/seq_file.h> |
70ddf637 | 55 | #include <linux/vmpressure.h> |
b69408e8 | 56 | #include <linux/mm_inline.h> |
5d1ea48b | 57 | #include <linux/swap_cgroup.h> |
cdec2e42 | 58 | #include <linux/cpu.h> |
158e0a2d | 59 | #include <linux/oom.h> |
0056f4e6 | 60 | #include <linux/lockdep.h> |
79bd9814 | 61 | #include <linux/file.h> |
b23afb93 | 62 | #include <linux/tracehook.h> |
0e4b01df | 63 | #include <linux/psi.h> |
c8713d0b | 64 | #include <linux/seq_buf.h> |
08e552c6 | 65 | #include "internal.h" |
d1a4c0b3 | 66 | #include <net/sock.h> |
4bd2c1ee | 67 | #include <net/ip.h> |
f35c3a8e | 68 | #include "slab.h" |
8cdea7c0 | 69 | |
7c0f6ba6 | 70 | #include <linux/uaccess.h> |
8697d331 | 71 | |
cc8e970c KM |
72 | #include <trace/events/vmscan.h> |
73 | ||
073219e9 TH |
74 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
75 | EXPORT_SYMBOL(memory_cgrp_subsys); | |
68ae564b | 76 | |
7d828602 JW |
77 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
78 | ||
37d5985c RG |
79 | /* Active memory cgroup to use from an interrupt context */ |
80 | DEFINE_PER_CPU(struct mem_cgroup *, int_active_memcg); | |
81 | ||
f7e1cb6e JW |
82 | /* Socket memory accounting disabled? */ |
83 | static bool cgroup_memory_nosocket; | |
84 | ||
04823c83 VD |
85 | /* Kernel memory accounting disabled? */ |
86 | static bool cgroup_memory_nokmem; | |
87 | ||
21afa38e | 88 | /* Whether the swap controller is active */ |
c255a458 | 89 | #ifdef CONFIG_MEMCG_SWAP |
eccb52e7 | 90 | bool cgroup_memory_noswap __read_mostly; |
c077719b | 91 | #else |
eccb52e7 | 92 | #define cgroup_memory_noswap 1 |
2d1c4980 | 93 | #endif |
c077719b | 94 | |
97b27821 TH |
95 | #ifdef CONFIG_CGROUP_WRITEBACK |
96 | static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq); | |
97 | #endif | |
98 | ||
7941d214 JW |
99 | /* Whether legacy memory+swap accounting is active */ |
100 | static bool do_memsw_account(void) | |
101 | { | |
eccb52e7 | 102 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_noswap; |
7941d214 JW |
103 | } |
104 | ||
a0db00fc KS |
105 | #define THRESHOLDS_EVENTS_TARGET 128 |
106 | #define SOFTLIMIT_EVENTS_TARGET 1024 | |
e9f8974f | 107 | |
bb4cc1a8 AM |
108 | /* |
109 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
110 | * their hierarchy representation | |
111 | */ | |
112 | ||
ef8f2327 | 113 | struct mem_cgroup_tree_per_node { |
bb4cc1a8 | 114 | struct rb_root rb_root; |
fa90b2fd | 115 | struct rb_node *rb_rightmost; |
bb4cc1a8 AM |
116 | spinlock_t lock; |
117 | }; | |
118 | ||
bb4cc1a8 AM |
119 | struct mem_cgroup_tree { |
120 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
121 | }; | |
122 | ||
123 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
124 | ||
9490ff27 KH |
125 | /* for OOM */ |
126 | struct mem_cgroup_eventfd_list { | |
127 | struct list_head list; | |
128 | struct eventfd_ctx *eventfd; | |
129 | }; | |
2e72b634 | 130 | |
79bd9814 TH |
131 | /* |
132 | * cgroup_event represents events which userspace want to receive. | |
133 | */ | |
3bc942f3 | 134 | struct mem_cgroup_event { |
79bd9814 | 135 | /* |
59b6f873 | 136 | * memcg which the event belongs to. |
79bd9814 | 137 | */ |
59b6f873 | 138 | struct mem_cgroup *memcg; |
79bd9814 TH |
139 | /* |
140 | * eventfd to signal userspace about the event. | |
141 | */ | |
142 | struct eventfd_ctx *eventfd; | |
143 | /* | |
144 | * Each of these stored in a list by the cgroup. | |
145 | */ | |
146 | struct list_head list; | |
fba94807 TH |
147 | /* |
148 | * register_event() callback will be used to add new userspace | |
149 | * waiter for changes related to this event. Use eventfd_signal() | |
150 | * on eventfd to send notification to userspace. | |
151 | */ | |
59b6f873 | 152 | int (*register_event)(struct mem_cgroup *memcg, |
347c4a87 | 153 | struct eventfd_ctx *eventfd, const char *args); |
fba94807 TH |
154 | /* |
155 | * unregister_event() callback will be called when userspace closes | |
156 | * the eventfd or on cgroup removing. This callback must be set, | |
157 | * if you want provide notification functionality. | |
158 | */ | |
59b6f873 | 159 | void (*unregister_event)(struct mem_cgroup *memcg, |
fba94807 | 160 | struct eventfd_ctx *eventfd); |
79bd9814 TH |
161 | /* |
162 | * All fields below needed to unregister event when | |
163 | * userspace closes eventfd. | |
164 | */ | |
165 | poll_table pt; | |
166 | wait_queue_head_t *wqh; | |
ac6424b9 | 167 | wait_queue_entry_t wait; |
79bd9814 TH |
168 | struct work_struct remove; |
169 | }; | |
170 | ||
c0ff4b85 R |
171 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
172 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); | |
2e72b634 | 173 | |
7dc74be0 DN |
174 | /* Stuffs for move charges at task migration. */ |
175 | /* | |
1dfab5ab | 176 | * Types of charges to be moved. |
7dc74be0 | 177 | */ |
1dfab5ab JW |
178 | #define MOVE_ANON 0x1U |
179 | #define MOVE_FILE 0x2U | |
180 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) | |
7dc74be0 | 181 | |
4ffef5fe DN |
182 | /* "mc" and its members are protected by cgroup_mutex */ |
183 | static struct move_charge_struct { | |
b1dd693e | 184 | spinlock_t lock; /* for from, to */ |
264a0ae1 | 185 | struct mm_struct *mm; |
4ffef5fe DN |
186 | struct mem_cgroup *from; |
187 | struct mem_cgroup *to; | |
1dfab5ab | 188 | unsigned long flags; |
4ffef5fe | 189 | unsigned long precharge; |
854ffa8d | 190 | unsigned long moved_charge; |
483c30b5 | 191 | unsigned long moved_swap; |
8033b97c DN |
192 | struct task_struct *moving_task; /* a task moving charges */ |
193 | wait_queue_head_t waitq; /* a waitq for other context */ | |
194 | } mc = { | |
2bd9bb20 | 195 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
196 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
197 | }; | |
4ffef5fe | 198 | |
4e416953 BS |
199 | /* |
200 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
201 | * limit reclaim to prevent infinite loops, if they ever occur. | |
202 | */ | |
a0db00fc | 203 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
bb4cc1a8 | 204 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
4e416953 | 205 | |
8c7c6e34 | 206 | /* for encoding cft->private value on file */ |
86ae53e1 GC |
207 | enum res_type { |
208 | _MEM, | |
209 | _MEMSWAP, | |
210 | _OOM_TYPE, | |
510fc4e1 | 211 | _KMEM, |
d55f90bf | 212 | _TCP, |
86ae53e1 GC |
213 | }; |
214 | ||
a0db00fc KS |
215 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
216 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) | |
8c7c6e34 | 217 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
9490ff27 KH |
218 | /* Used for OOM nofiier */ |
219 | #define OOM_CONTROL (0) | |
8c7c6e34 | 220 | |
b05706f1 KT |
221 | /* |
222 | * Iteration constructs for visiting all cgroups (under a tree). If | |
223 | * loops are exited prematurely (break), mem_cgroup_iter_break() must | |
224 | * be used for reference counting. | |
225 | */ | |
226 | #define for_each_mem_cgroup_tree(iter, root) \ | |
227 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ | |
228 | iter != NULL; \ | |
229 | iter = mem_cgroup_iter(root, iter, NULL)) | |
230 | ||
231 | #define for_each_mem_cgroup(iter) \ | |
232 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ | |
233 | iter != NULL; \ | |
234 | iter = mem_cgroup_iter(NULL, iter, NULL)) | |
235 | ||
7775face TH |
236 | static inline bool should_force_charge(void) |
237 | { | |
238 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || | |
239 | (current->flags & PF_EXITING); | |
240 | } | |
241 | ||
70ddf637 AV |
242 | /* Some nice accessors for the vmpressure. */ |
243 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) | |
244 | { | |
245 | if (!memcg) | |
246 | memcg = root_mem_cgroup; | |
247 | return &memcg->vmpressure; | |
248 | } | |
249 | ||
250 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) | |
251 | { | |
252 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; | |
253 | } | |
254 | ||
84c07d11 | 255 | #ifdef CONFIG_MEMCG_KMEM |
bf4f0599 RG |
256 | extern spinlock_t css_set_lock; |
257 | ||
f1286fae MS |
258 | static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, |
259 | unsigned int nr_pages); | |
c1a660de | 260 | |
bf4f0599 RG |
261 | static void obj_cgroup_release(struct percpu_ref *ref) |
262 | { | |
263 | struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt); | |
264 | struct mem_cgroup *memcg; | |
265 | unsigned int nr_bytes; | |
266 | unsigned int nr_pages; | |
267 | unsigned long flags; | |
268 | ||
269 | /* | |
270 | * At this point all allocated objects are freed, and | |
271 | * objcg->nr_charged_bytes can't have an arbitrary byte value. | |
272 | * However, it can be PAGE_SIZE or (x * PAGE_SIZE). | |
273 | * | |
274 | * The following sequence can lead to it: | |
275 | * 1) CPU0: objcg == stock->cached_objcg | |
276 | * 2) CPU1: we do a small allocation (e.g. 92 bytes), | |
277 | * PAGE_SIZE bytes are charged | |
278 | * 3) CPU1: a process from another memcg is allocating something, | |
279 | * the stock if flushed, | |
280 | * objcg->nr_charged_bytes = PAGE_SIZE - 92 | |
281 | * 5) CPU0: we do release this object, | |
282 | * 92 bytes are added to stock->nr_bytes | |
283 | * 6) CPU0: stock is flushed, | |
284 | * 92 bytes are added to objcg->nr_charged_bytes | |
285 | * | |
286 | * In the result, nr_charged_bytes == PAGE_SIZE. | |
287 | * This page will be uncharged in obj_cgroup_release(). | |
288 | */ | |
289 | nr_bytes = atomic_read(&objcg->nr_charged_bytes); | |
290 | WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); | |
291 | nr_pages = nr_bytes >> PAGE_SHIFT; | |
292 | ||
293 | spin_lock_irqsave(&css_set_lock, flags); | |
294 | memcg = obj_cgroup_memcg(objcg); | |
295 | if (nr_pages) | |
f1286fae | 296 | obj_cgroup_uncharge_pages(objcg, nr_pages); |
bf4f0599 RG |
297 | list_del(&objcg->list); |
298 | mem_cgroup_put(memcg); | |
299 | spin_unlock_irqrestore(&css_set_lock, flags); | |
300 | ||
301 | percpu_ref_exit(ref); | |
302 | kfree_rcu(objcg, rcu); | |
303 | } | |
304 | ||
305 | static struct obj_cgroup *obj_cgroup_alloc(void) | |
306 | { | |
307 | struct obj_cgroup *objcg; | |
308 | int ret; | |
309 | ||
310 | objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL); | |
311 | if (!objcg) | |
312 | return NULL; | |
313 | ||
314 | ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, | |
315 | GFP_KERNEL); | |
316 | if (ret) { | |
317 | kfree(objcg); | |
318 | return NULL; | |
319 | } | |
320 | INIT_LIST_HEAD(&objcg->list); | |
321 | return objcg; | |
322 | } | |
323 | ||
324 | static void memcg_reparent_objcgs(struct mem_cgroup *memcg, | |
325 | struct mem_cgroup *parent) | |
326 | { | |
327 | struct obj_cgroup *objcg, *iter; | |
328 | ||
329 | objcg = rcu_replace_pointer(memcg->objcg, NULL, true); | |
330 | ||
331 | spin_lock_irq(&css_set_lock); | |
332 | ||
333 | /* Move active objcg to the parent's list */ | |
334 | xchg(&objcg->memcg, parent); | |
335 | css_get(&parent->css); | |
336 | list_add(&objcg->list, &parent->objcg_list); | |
337 | ||
338 | /* Move already reparented objcgs to the parent's list */ | |
339 | list_for_each_entry(iter, &memcg->objcg_list, list) { | |
340 | css_get(&parent->css); | |
341 | xchg(&iter->memcg, parent); | |
342 | css_put(&memcg->css); | |
343 | } | |
344 | list_splice(&memcg->objcg_list, &parent->objcg_list); | |
345 | ||
346 | spin_unlock_irq(&css_set_lock); | |
347 | ||
348 | percpu_ref_kill(&objcg->refcnt); | |
349 | } | |
350 | ||
55007d84 | 351 | /* |
9855609b | 352 | * This will be used as a shrinker list's index. |
b8627835 LZ |
353 | * The main reason for not using cgroup id for this: |
354 | * this works better in sparse environments, where we have a lot of memcgs, | |
355 | * but only a few kmem-limited. Or also, if we have, for instance, 200 | |
356 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a | |
357 | * 200 entry array for that. | |
55007d84 | 358 | * |
dbcf73e2 VD |
359 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
360 | * will double each time we have to increase it. | |
55007d84 | 361 | */ |
dbcf73e2 VD |
362 | static DEFINE_IDA(memcg_cache_ida); |
363 | int memcg_nr_cache_ids; | |
749c5415 | 364 | |
05257a1a VD |
365 | /* Protects memcg_nr_cache_ids */ |
366 | static DECLARE_RWSEM(memcg_cache_ids_sem); | |
367 | ||
368 | void memcg_get_cache_ids(void) | |
369 | { | |
370 | down_read(&memcg_cache_ids_sem); | |
371 | } | |
372 | ||
373 | void memcg_put_cache_ids(void) | |
374 | { | |
375 | up_read(&memcg_cache_ids_sem); | |
376 | } | |
377 | ||
55007d84 GC |
378 | /* |
379 | * MIN_SIZE is different than 1, because we would like to avoid going through | |
380 | * the alloc/free process all the time. In a small machine, 4 kmem-limited | |
381 | * cgroups is a reasonable guess. In the future, it could be a parameter or | |
382 | * tunable, but that is strictly not necessary. | |
383 | * | |
b8627835 | 384 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
55007d84 GC |
385 | * this constant directly from cgroup, but it is understandable that this is |
386 | * better kept as an internal representation in cgroup.c. In any case, the | |
b8627835 | 387 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
55007d84 GC |
388 | * increase ours as well if it increases. |
389 | */ | |
390 | #define MEMCG_CACHES_MIN_SIZE 4 | |
b8627835 | 391 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
55007d84 | 392 | |
d7f25f8a GC |
393 | /* |
394 | * A lot of the calls to the cache allocation functions are expected to be | |
272911a4 | 395 | * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are |
d7f25f8a GC |
396 | * conditional to this static branch, we'll have to allow modules that does |
397 | * kmem_cache_alloc and the such to see this symbol as well | |
398 | */ | |
ef12947c | 399 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
d7f25f8a | 400 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
0a432dcb | 401 | #endif |
17cc4dfe | 402 | |
ad7fa852 TH |
403 | /** |
404 | * mem_cgroup_css_from_page - css of the memcg associated with a page | |
405 | * @page: page of interest | |
406 | * | |
407 | * If memcg is bound to the default hierarchy, css of the memcg associated | |
408 | * with @page is returned. The returned css remains associated with @page | |
409 | * until it is released. | |
410 | * | |
411 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup | |
412 | * is returned. | |
ad7fa852 TH |
413 | */ |
414 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) | |
415 | { | |
416 | struct mem_cgroup *memcg; | |
417 | ||
bcfe06bf | 418 | memcg = page_memcg(page); |
ad7fa852 | 419 | |
9e10a130 | 420 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
ad7fa852 TH |
421 | memcg = root_mem_cgroup; |
422 | ||
ad7fa852 TH |
423 | return &memcg->css; |
424 | } | |
425 | ||
2fc04524 VD |
426 | /** |
427 | * page_cgroup_ino - return inode number of the memcg a page is charged to | |
428 | * @page: the page | |
429 | * | |
430 | * Look up the closest online ancestor of the memory cgroup @page is charged to | |
431 | * and return its inode number or 0 if @page is not charged to any cgroup. It | |
432 | * is safe to call this function without holding a reference to @page. | |
433 | * | |
434 | * Note, this function is inherently racy, because there is nothing to prevent | |
435 | * the cgroup inode from getting torn down and potentially reallocated a moment | |
436 | * after page_cgroup_ino() returns, so it only should be used by callers that | |
437 | * do not care (such as procfs interfaces). | |
438 | */ | |
439 | ino_t page_cgroup_ino(struct page *page) | |
440 | { | |
441 | struct mem_cgroup *memcg; | |
442 | unsigned long ino = 0; | |
443 | ||
444 | rcu_read_lock(); | |
bcfe06bf | 445 | memcg = page_memcg_check(page); |
286e04b8 | 446 | |
2fc04524 VD |
447 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
448 | memcg = parent_mem_cgroup(memcg); | |
449 | if (memcg) | |
450 | ino = cgroup_ino(memcg->css.cgroup); | |
451 | rcu_read_unlock(); | |
452 | return ino; | |
453 | } | |
454 | ||
ef8f2327 MG |
455 | static struct mem_cgroup_per_node * |
456 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) | |
f64c3f54 | 457 | { |
97a6c37b | 458 | int nid = page_to_nid(page); |
f64c3f54 | 459 | |
ef8f2327 | 460 | return memcg->nodeinfo[nid]; |
f64c3f54 BS |
461 | } |
462 | ||
ef8f2327 MG |
463 | static struct mem_cgroup_tree_per_node * |
464 | soft_limit_tree_node(int nid) | |
bb4cc1a8 | 465 | { |
ef8f2327 | 466 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
467 | } |
468 | ||
ef8f2327 | 469 | static struct mem_cgroup_tree_per_node * |
bb4cc1a8 AM |
470 | soft_limit_tree_from_page(struct page *page) |
471 | { | |
472 | int nid = page_to_nid(page); | |
bb4cc1a8 | 473 | |
ef8f2327 | 474 | return soft_limit_tree.rb_tree_per_node[nid]; |
bb4cc1a8 AM |
475 | } |
476 | ||
ef8f2327 MG |
477 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
478 | struct mem_cgroup_tree_per_node *mctz, | |
3e32cb2e | 479 | unsigned long new_usage_in_excess) |
bb4cc1a8 AM |
480 | { |
481 | struct rb_node **p = &mctz->rb_root.rb_node; | |
482 | struct rb_node *parent = NULL; | |
ef8f2327 | 483 | struct mem_cgroup_per_node *mz_node; |
fa90b2fd | 484 | bool rightmost = true; |
bb4cc1a8 AM |
485 | |
486 | if (mz->on_tree) | |
487 | return; | |
488 | ||
489 | mz->usage_in_excess = new_usage_in_excess; | |
490 | if (!mz->usage_in_excess) | |
491 | return; | |
492 | while (*p) { | |
493 | parent = *p; | |
ef8f2327 | 494 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
bb4cc1a8 | 495 | tree_node); |
fa90b2fd | 496 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
bb4cc1a8 | 497 | p = &(*p)->rb_left; |
fa90b2fd | 498 | rightmost = false; |
378876b0 | 499 | } else { |
bb4cc1a8 | 500 | p = &(*p)->rb_right; |
378876b0 | 501 | } |
bb4cc1a8 | 502 | } |
fa90b2fd DB |
503 | |
504 | if (rightmost) | |
505 | mctz->rb_rightmost = &mz->tree_node; | |
506 | ||
bb4cc1a8 AM |
507 | rb_link_node(&mz->tree_node, parent, p); |
508 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
509 | mz->on_tree = true; | |
510 | } | |
511 | ||
ef8f2327 MG |
512 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
513 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 AM |
514 | { |
515 | if (!mz->on_tree) | |
516 | return; | |
fa90b2fd DB |
517 | |
518 | if (&mz->tree_node == mctz->rb_rightmost) | |
519 | mctz->rb_rightmost = rb_prev(&mz->tree_node); | |
520 | ||
bb4cc1a8 AM |
521 | rb_erase(&mz->tree_node, &mctz->rb_root); |
522 | mz->on_tree = false; | |
523 | } | |
524 | ||
ef8f2327 MG |
525 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
526 | struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 527 | { |
0a31bc97 JW |
528 | unsigned long flags; |
529 | ||
530 | spin_lock_irqsave(&mctz->lock, flags); | |
cf2c8127 | 531 | __mem_cgroup_remove_exceeded(mz, mctz); |
0a31bc97 | 532 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
533 | } |
534 | ||
3e32cb2e JW |
535 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
536 | { | |
537 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
4db0c3c2 | 538 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
3e32cb2e JW |
539 | unsigned long excess = 0; |
540 | ||
541 | if (nr_pages > soft_limit) | |
542 | excess = nr_pages - soft_limit; | |
543 | ||
544 | return excess; | |
545 | } | |
bb4cc1a8 AM |
546 | |
547 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) | |
548 | { | |
3e32cb2e | 549 | unsigned long excess; |
ef8f2327 MG |
550 | struct mem_cgroup_per_node *mz; |
551 | struct mem_cgroup_tree_per_node *mctz; | |
bb4cc1a8 | 552 | |
e231875b | 553 | mctz = soft_limit_tree_from_page(page); |
bfc7228b LD |
554 | if (!mctz) |
555 | return; | |
bb4cc1a8 AM |
556 | /* |
557 | * Necessary to update all ancestors when hierarchy is used. | |
558 | * because their event counter is not touched. | |
559 | */ | |
560 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
ef8f2327 | 561 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
3e32cb2e | 562 | excess = soft_limit_excess(memcg); |
bb4cc1a8 AM |
563 | /* |
564 | * We have to update the tree if mz is on RB-tree or | |
565 | * mem is over its softlimit. | |
566 | */ | |
567 | if (excess || mz->on_tree) { | |
0a31bc97 JW |
568 | unsigned long flags; |
569 | ||
570 | spin_lock_irqsave(&mctz->lock, flags); | |
bb4cc1a8 AM |
571 | /* if on-tree, remove it */ |
572 | if (mz->on_tree) | |
cf2c8127 | 573 | __mem_cgroup_remove_exceeded(mz, mctz); |
bb4cc1a8 AM |
574 | /* |
575 | * Insert again. mz->usage_in_excess will be updated. | |
576 | * If excess is 0, no tree ops. | |
577 | */ | |
cf2c8127 | 578 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 579 | spin_unlock_irqrestore(&mctz->lock, flags); |
bb4cc1a8 AM |
580 | } |
581 | } | |
582 | } | |
583 | ||
584 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) | |
585 | { | |
ef8f2327 MG |
586 | struct mem_cgroup_tree_per_node *mctz; |
587 | struct mem_cgroup_per_node *mz; | |
588 | int nid; | |
bb4cc1a8 | 589 | |
e231875b | 590 | for_each_node(nid) { |
a3747b53 | 591 | mz = memcg->nodeinfo[nid]; |
ef8f2327 | 592 | mctz = soft_limit_tree_node(nid); |
bfc7228b LD |
593 | if (mctz) |
594 | mem_cgroup_remove_exceeded(mz, mctz); | |
bb4cc1a8 AM |
595 | } |
596 | } | |
597 | ||
ef8f2327 MG |
598 | static struct mem_cgroup_per_node * |
599 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 600 | { |
ef8f2327 | 601 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 AM |
602 | |
603 | retry: | |
604 | mz = NULL; | |
fa90b2fd | 605 | if (!mctz->rb_rightmost) |
bb4cc1a8 AM |
606 | goto done; /* Nothing to reclaim from */ |
607 | ||
fa90b2fd DB |
608 | mz = rb_entry(mctz->rb_rightmost, |
609 | struct mem_cgroup_per_node, tree_node); | |
bb4cc1a8 AM |
610 | /* |
611 | * Remove the node now but someone else can add it back, | |
612 | * we will to add it back at the end of reclaim to its correct | |
613 | * position in the tree. | |
614 | */ | |
cf2c8127 | 615 | __mem_cgroup_remove_exceeded(mz, mctz); |
3e32cb2e | 616 | if (!soft_limit_excess(mz->memcg) || |
8965aa28 | 617 | !css_tryget(&mz->memcg->css)) |
bb4cc1a8 AM |
618 | goto retry; |
619 | done: | |
620 | return mz; | |
621 | } | |
622 | ||
ef8f2327 MG |
623 | static struct mem_cgroup_per_node * |
624 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) | |
bb4cc1a8 | 625 | { |
ef8f2327 | 626 | struct mem_cgroup_per_node *mz; |
bb4cc1a8 | 627 | |
0a31bc97 | 628 | spin_lock_irq(&mctz->lock); |
bb4cc1a8 | 629 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
0a31bc97 | 630 | spin_unlock_irq(&mctz->lock); |
bb4cc1a8 AM |
631 | return mz; |
632 | } | |
633 | ||
db9adbcb JW |
634 | /** |
635 | * __mod_memcg_state - update cgroup memory statistics | |
636 | * @memcg: the memory cgroup | |
637 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item | |
638 | * @val: delta to add to the counter, can be negative | |
639 | */ | |
640 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) | |
641 | { | |
db9adbcb JW |
642 | if (mem_cgroup_disabled()) |
643 | return; | |
644 | ||
2d146aa3 JW |
645 | __this_cpu_add(memcg->vmstats_percpu->state[idx], val); |
646 | cgroup_rstat_updated(memcg->css.cgroup, smp_processor_id()); | |
db9adbcb JW |
647 | } |
648 | ||
2d146aa3 | 649 | /* idx can be of type enum memcg_stat_item or node_stat_item. */ |
a18e6e6e JW |
650 | static unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) |
651 | { | |
2d146aa3 | 652 | long x = READ_ONCE(memcg->vmstats.state[idx]); |
a18e6e6e JW |
653 | #ifdef CONFIG_SMP |
654 | if (x < 0) | |
655 | x = 0; | |
656 | #endif | |
657 | return x; | |
658 | } | |
659 | ||
2d146aa3 | 660 | /* idx can be of type enum memcg_stat_item or node_stat_item. */ |
a18e6e6e JW |
661 | static unsigned long memcg_page_state_local(struct mem_cgroup *memcg, int idx) |
662 | { | |
663 | long x = 0; | |
664 | int cpu; | |
665 | ||
666 | for_each_possible_cpu(cpu) | |
2d146aa3 | 667 | x += per_cpu(memcg->vmstats_percpu->state[idx], cpu); |
a18e6e6e JW |
668 | #ifdef CONFIG_SMP |
669 | if (x < 0) | |
670 | x = 0; | |
671 | #endif | |
672 | return x; | |
673 | } | |
674 | ||
42a30035 JW |
675 | static struct mem_cgroup_per_node * |
676 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) | |
677 | { | |
678 | struct mem_cgroup *parent; | |
679 | ||
680 | parent = parent_mem_cgroup(pn->memcg); | |
681 | if (!parent) | |
682 | return NULL; | |
a3747b53 | 683 | return parent->nodeinfo[nid]; |
42a30035 JW |
684 | } |
685 | ||
eedc4e5a RG |
686 | void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, |
687 | int val) | |
db9adbcb JW |
688 | { |
689 | struct mem_cgroup_per_node *pn; | |
42a30035 | 690 | struct mem_cgroup *memcg; |
ea426c2a | 691 | long x, threshold = MEMCG_CHARGE_BATCH; |
db9adbcb | 692 | |
db9adbcb | 693 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
42a30035 | 694 | memcg = pn->memcg; |
db9adbcb JW |
695 | |
696 | /* Update memcg */ | |
42a30035 | 697 | __mod_memcg_state(memcg, idx, val); |
db9adbcb | 698 | |
b4c46484 RG |
699 | /* Update lruvec */ |
700 | __this_cpu_add(pn->lruvec_stat_local->count[idx], val); | |
701 | ||
ea426c2a RG |
702 | if (vmstat_item_in_bytes(idx)) |
703 | threshold <<= PAGE_SHIFT; | |
704 | ||
db9adbcb | 705 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); |
ea426c2a | 706 | if (unlikely(abs(x) > threshold)) { |
eedc4e5a | 707 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
42a30035 JW |
708 | struct mem_cgroup_per_node *pi; |
709 | ||
42a30035 JW |
710 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) |
711 | atomic_long_add(x, &pi->lruvec_stat[idx]); | |
db9adbcb JW |
712 | x = 0; |
713 | } | |
714 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); | |
715 | } | |
716 | ||
eedc4e5a RG |
717 | /** |
718 | * __mod_lruvec_state - update lruvec memory statistics | |
719 | * @lruvec: the lruvec | |
720 | * @idx: the stat item | |
721 | * @val: delta to add to the counter, can be negative | |
722 | * | |
723 | * The lruvec is the intersection of the NUMA node and a cgroup. This | |
724 | * function updates the all three counters that are affected by a | |
725 | * change of state at this level: per-node, per-cgroup, per-lruvec. | |
726 | */ | |
727 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, | |
728 | int val) | |
729 | { | |
730 | /* Update node */ | |
731 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); | |
732 | ||
733 | /* Update memcg and lruvec */ | |
734 | if (!mem_cgroup_disabled()) | |
735 | __mod_memcg_lruvec_state(lruvec, idx, val); | |
736 | } | |
737 | ||
c47d5032 SB |
738 | void __mod_lruvec_page_state(struct page *page, enum node_stat_item idx, |
739 | int val) | |
740 | { | |
741 | struct page *head = compound_head(page); /* rmap on tail pages */ | |
b4e0b68f | 742 | struct mem_cgroup *memcg; |
c47d5032 SB |
743 | pg_data_t *pgdat = page_pgdat(page); |
744 | struct lruvec *lruvec; | |
745 | ||
b4e0b68f MS |
746 | rcu_read_lock(); |
747 | memcg = page_memcg(head); | |
c47d5032 | 748 | /* Untracked pages have no memcg, no lruvec. Update only the node */ |
d635a69d | 749 | if (!memcg) { |
b4e0b68f | 750 | rcu_read_unlock(); |
c47d5032 SB |
751 | __mod_node_page_state(pgdat, idx, val); |
752 | return; | |
753 | } | |
754 | ||
d635a69d | 755 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
c47d5032 | 756 | __mod_lruvec_state(lruvec, idx, val); |
b4e0b68f | 757 | rcu_read_unlock(); |
c47d5032 | 758 | } |
f0c0c115 | 759 | EXPORT_SYMBOL(__mod_lruvec_page_state); |
c47d5032 | 760 | |
da3ceeff | 761 | void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val) |
ec9f0238 | 762 | { |
4f103c63 | 763 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
ec9f0238 RG |
764 | struct mem_cgroup *memcg; |
765 | struct lruvec *lruvec; | |
766 | ||
767 | rcu_read_lock(); | |
4f103c63 | 768 | memcg = mem_cgroup_from_obj(p); |
ec9f0238 | 769 | |
8faeb1ff MS |
770 | /* |
771 | * Untracked pages have no memcg, no lruvec. Update only the | |
772 | * node. If we reparent the slab objects to the root memcg, | |
773 | * when we free the slab object, we need to update the per-memcg | |
774 | * vmstats to keep it correct for the root memcg. | |
775 | */ | |
776 | if (!memcg) { | |
ec9f0238 RG |
777 | __mod_node_page_state(pgdat, idx, val); |
778 | } else { | |
867e5e1d | 779 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
ec9f0238 RG |
780 | __mod_lruvec_state(lruvec, idx, val); |
781 | } | |
782 | rcu_read_unlock(); | |
783 | } | |
784 | ||
db9adbcb JW |
785 | /** |
786 | * __count_memcg_events - account VM events in a cgroup | |
787 | * @memcg: the memory cgroup | |
788 | * @idx: the event item | |
789 | * @count: the number of events that occured | |
790 | */ | |
791 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, | |
792 | unsigned long count) | |
793 | { | |
db9adbcb JW |
794 | if (mem_cgroup_disabled()) |
795 | return; | |
796 | ||
2d146aa3 JW |
797 | __this_cpu_add(memcg->vmstats_percpu->events[idx], count); |
798 | cgroup_rstat_updated(memcg->css.cgroup, smp_processor_id()); | |
db9adbcb JW |
799 | } |
800 | ||
42a30035 | 801 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
e9f8974f | 802 | { |
2d146aa3 | 803 | return READ_ONCE(memcg->vmstats.events[event]); |
e9f8974f JW |
804 | } |
805 | ||
42a30035 JW |
806 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
807 | { | |
815744d7 JW |
808 | long x = 0; |
809 | int cpu; | |
810 | ||
811 | for_each_possible_cpu(cpu) | |
2d146aa3 | 812 | x += per_cpu(memcg->vmstats_percpu->events[event], cpu); |
815744d7 | 813 | return x; |
42a30035 JW |
814 | } |
815 | ||
c0ff4b85 | 816 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
b070e65c | 817 | struct page *page, |
3fba69a5 | 818 | int nr_pages) |
d52aa412 | 819 | { |
e401f176 KH |
820 | /* pagein of a big page is an event. So, ignore page size */ |
821 | if (nr_pages > 0) | |
c9019e9b | 822 | __count_memcg_events(memcg, PGPGIN, 1); |
3751d604 | 823 | else { |
c9019e9b | 824 | __count_memcg_events(memcg, PGPGOUT, 1); |
3751d604 KH |
825 | nr_pages = -nr_pages; /* for event */ |
826 | } | |
e401f176 | 827 | |
871789d4 | 828 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
6d12e2d8 KH |
829 | } |
830 | ||
f53d7ce3 JW |
831 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
832 | enum mem_cgroup_events_target target) | |
7a159cc9 JW |
833 | { |
834 | unsigned long val, next; | |
835 | ||
871789d4 CD |
836 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
837 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); | |
7a159cc9 | 838 | /* from time_after() in jiffies.h */ |
6a1a8b80 | 839 | if ((long)(next - val) < 0) { |
f53d7ce3 JW |
840 | switch (target) { |
841 | case MEM_CGROUP_TARGET_THRESH: | |
842 | next = val + THRESHOLDS_EVENTS_TARGET; | |
843 | break; | |
bb4cc1a8 AM |
844 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
845 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
846 | break; | |
f53d7ce3 JW |
847 | default: |
848 | break; | |
849 | } | |
871789d4 | 850 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
f53d7ce3 | 851 | return true; |
7a159cc9 | 852 | } |
f53d7ce3 | 853 | return false; |
d2265e6f KH |
854 | } |
855 | ||
856 | /* | |
857 | * Check events in order. | |
858 | * | |
859 | */ | |
c0ff4b85 | 860 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
d2265e6f KH |
861 | { |
862 | /* threshold event is triggered in finer grain than soft limit */ | |
f53d7ce3 JW |
863 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
864 | MEM_CGROUP_TARGET_THRESH))) { | |
bb4cc1a8 | 865 | bool do_softlimit; |
f53d7ce3 | 866 | |
bb4cc1a8 AM |
867 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
868 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
c0ff4b85 | 869 | mem_cgroup_threshold(memcg); |
bb4cc1a8 AM |
870 | if (unlikely(do_softlimit)) |
871 | mem_cgroup_update_tree(memcg, page); | |
0a31bc97 | 872 | } |
d2265e6f KH |
873 | } |
874 | ||
cf475ad2 | 875 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 876 | { |
31a78f23 BS |
877 | /* |
878 | * mm_update_next_owner() may clear mm->owner to NULL | |
879 | * if it races with swapoff, page migration, etc. | |
880 | * So this can be called with p == NULL. | |
881 | */ | |
882 | if (unlikely(!p)) | |
883 | return NULL; | |
884 | ||
073219e9 | 885 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
78fb7466 | 886 | } |
33398cf2 | 887 | EXPORT_SYMBOL(mem_cgroup_from_task); |
78fb7466 | 888 | |
d46eb14b SB |
889 | /** |
890 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. | |
891 | * @mm: mm from which memcg should be extracted. It can be NULL. | |
892 | * | |
893 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise | |
894 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is | |
895 | * returned. | |
896 | */ | |
897 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) | |
54595fe2 | 898 | { |
d46eb14b SB |
899 | struct mem_cgroup *memcg; |
900 | ||
901 | if (mem_cgroup_disabled()) | |
902 | return NULL; | |
0b7f569e | 903 | |
54595fe2 KH |
904 | rcu_read_lock(); |
905 | do { | |
6f6acb00 MH |
906 | /* |
907 | * Page cache insertions can happen withou an | |
908 | * actual mm context, e.g. during disk probing | |
909 | * on boot, loopback IO, acct() writes etc. | |
910 | */ | |
911 | if (unlikely(!mm)) | |
df381975 | 912 | memcg = root_mem_cgroup; |
6f6acb00 MH |
913 | else { |
914 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
915 | if (unlikely(!memcg)) | |
916 | memcg = root_mem_cgroup; | |
917 | } | |
00d484f3 | 918 | } while (!css_tryget(&memcg->css)); |
54595fe2 | 919 | rcu_read_unlock(); |
c0ff4b85 | 920 | return memcg; |
54595fe2 | 921 | } |
d46eb14b SB |
922 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
923 | ||
37d5985c | 924 | static __always_inline struct mem_cgroup *active_memcg(void) |
d46eb14b | 925 | { |
37d5985c RG |
926 | if (in_interrupt()) |
927 | return this_cpu_read(int_active_memcg); | |
928 | else | |
929 | return current->active_memcg; | |
930 | } | |
279c3393 | 931 | |
4127c650 RG |
932 | static __always_inline bool memcg_kmem_bypass(void) |
933 | { | |
934 | /* Allow remote memcg charging from any context. */ | |
935 | if (unlikely(active_memcg())) | |
936 | return false; | |
937 | ||
938 | /* Memcg to charge can't be determined. */ | |
939 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) | |
940 | return true; | |
941 | ||
942 | return false; | |
943 | } | |
944 | ||
5660048c JW |
945 | /** |
946 | * mem_cgroup_iter - iterate over memory cgroup hierarchy | |
947 | * @root: hierarchy root | |
948 | * @prev: previously returned memcg, NULL on first invocation | |
949 | * @reclaim: cookie for shared reclaim walks, NULL for full walks | |
950 | * | |
951 | * Returns references to children of the hierarchy below @root, or | |
952 | * @root itself, or %NULL after a full round-trip. | |
953 | * | |
954 | * Caller must pass the return value in @prev on subsequent | |
955 | * invocations for reference counting, or use mem_cgroup_iter_break() | |
956 | * to cancel a hierarchy walk before the round-trip is complete. | |
957 | * | |
05bdc520 ML |
958 | * Reclaimers can specify a node in @reclaim to divide up the memcgs |
959 | * in the hierarchy among all concurrent reclaimers operating on the | |
960 | * same node. | |
5660048c | 961 | */ |
694fbc0f | 962 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
5660048c | 963 | struct mem_cgroup *prev, |
694fbc0f | 964 | struct mem_cgroup_reclaim_cookie *reclaim) |
14067bb3 | 965 | { |
3f649ab7 | 966 | struct mem_cgroup_reclaim_iter *iter; |
5ac8fb31 | 967 | struct cgroup_subsys_state *css = NULL; |
9f3a0d09 | 968 | struct mem_cgroup *memcg = NULL; |
5ac8fb31 | 969 | struct mem_cgroup *pos = NULL; |
711d3d2c | 970 | |
694fbc0f AM |
971 | if (mem_cgroup_disabled()) |
972 | return NULL; | |
5660048c | 973 | |
9f3a0d09 JW |
974 | if (!root) |
975 | root = root_mem_cgroup; | |
7d74b06f | 976 | |
9f3a0d09 | 977 | if (prev && !reclaim) |
5ac8fb31 | 978 | pos = prev; |
14067bb3 | 979 | |
542f85f9 | 980 | rcu_read_lock(); |
5f578161 | 981 | |
5ac8fb31 | 982 | if (reclaim) { |
ef8f2327 | 983 | struct mem_cgroup_per_node *mz; |
5ac8fb31 | 984 | |
a3747b53 | 985 | mz = root->nodeinfo[reclaim->pgdat->node_id]; |
9da83f3f | 986 | iter = &mz->iter; |
5ac8fb31 JW |
987 | |
988 | if (prev && reclaim->generation != iter->generation) | |
989 | goto out_unlock; | |
990 | ||
6df38689 | 991 | while (1) { |
4db0c3c2 | 992 | pos = READ_ONCE(iter->position); |
6df38689 VD |
993 | if (!pos || css_tryget(&pos->css)) |
994 | break; | |
5ac8fb31 | 995 | /* |
6df38689 VD |
996 | * css reference reached zero, so iter->position will |
997 | * be cleared by ->css_released. However, we should not | |
998 | * rely on this happening soon, because ->css_released | |
999 | * is called from a work queue, and by busy-waiting we | |
1000 | * might block it. So we clear iter->position right | |
1001 | * away. | |
5ac8fb31 | 1002 | */ |
6df38689 VD |
1003 | (void)cmpxchg(&iter->position, pos, NULL); |
1004 | } | |
5ac8fb31 JW |
1005 | } |
1006 | ||
1007 | if (pos) | |
1008 | css = &pos->css; | |
1009 | ||
1010 | for (;;) { | |
1011 | css = css_next_descendant_pre(css, &root->css); | |
1012 | if (!css) { | |
1013 | /* | |
1014 | * Reclaimers share the hierarchy walk, and a | |
1015 | * new one might jump in right at the end of | |
1016 | * the hierarchy - make sure they see at least | |
1017 | * one group and restart from the beginning. | |
1018 | */ | |
1019 | if (!prev) | |
1020 | continue; | |
1021 | break; | |
527a5ec9 | 1022 | } |
7d74b06f | 1023 | |
5ac8fb31 JW |
1024 | /* |
1025 | * Verify the css and acquire a reference. The root | |
1026 | * is provided by the caller, so we know it's alive | |
1027 | * and kicking, and don't take an extra reference. | |
1028 | */ | |
1029 | memcg = mem_cgroup_from_css(css); | |
14067bb3 | 1030 | |
5ac8fb31 JW |
1031 | if (css == &root->css) |
1032 | break; | |
14067bb3 | 1033 | |
0b8f73e1 JW |
1034 | if (css_tryget(css)) |
1035 | break; | |
9f3a0d09 | 1036 | |
5ac8fb31 | 1037 | memcg = NULL; |
9f3a0d09 | 1038 | } |
5ac8fb31 JW |
1039 | |
1040 | if (reclaim) { | |
5ac8fb31 | 1041 | /* |
6df38689 VD |
1042 | * The position could have already been updated by a competing |
1043 | * thread, so check that the value hasn't changed since we read | |
1044 | * it to avoid reclaiming from the same cgroup twice. | |
5ac8fb31 | 1045 | */ |
6df38689 VD |
1046 | (void)cmpxchg(&iter->position, pos, memcg); |
1047 | ||
5ac8fb31 JW |
1048 | if (pos) |
1049 | css_put(&pos->css); | |
1050 | ||
1051 | if (!memcg) | |
1052 | iter->generation++; | |
1053 | else if (!prev) | |
1054 | reclaim->generation = iter->generation; | |
9f3a0d09 | 1055 | } |
5ac8fb31 | 1056 | |
542f85f9 MH |
1057 | out_unlock: |
1058 | rcu_read_unlock(); | |
c40046f3 MH |
1059 | if (prev && prev != root) |
1060 | css_put(&prev->css); | |
1061 | ||
9f3a0d09 | 1062 | return memcg; |
14067bb3 | 1063 | } |
7d74b06f | 1064 | |
5660048c JW |
1065 | /** |
1066 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely | |
1067 | * @root: hierarchy root | |
1068 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() | |
1069 | */ | |
1070 | void mem_cgroup_iter_break(struct mem_cgroup *root, | |
1071 | struct mem_cgroup *prev) | |
9f3a0d09 JW |
1072 | { |
1073 | if (!root) | |
1074 | root = root_mem_cgroup; | |
1075 | if (prev && prev != root) | |
1076 | css_put(&prev->css); | |
1077 | } | |
7d74b06f | 1078 | |
54a83d6b MC |
1079 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
1080 | struct mem_cgroup *dead_memcg) | |
6df38689 | 1081 | { |
6df38689 | 1082 | struct mem_cgroup_reclaim_iter *iter; |
ef8f2327 MG |
1083 | struct mem_cgroup_per_node *mz; |
1084 | int nid; | |
6df38689 | 1085 | |
54a83d6b | 1086 | for_each_node(nid) { |
a3747b53 | 1087 | mz = from->nodeinfo[nid]; |
9da83f3f YS |
1088 | iter = &mz->iter; |
1089 | cmpxchg(&iter->position, dead_memcg, NULL); | |
6df38689 VD |
1090 | } |
1091 | } | |
1092 | ||
54a83d6b MC |
1093 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
1094 | { | |
1095 | struct mem_cgroup *memcg = dead_memcg; | |
1096 | struct mem_cgroup *last; | |
1097 | ||
1098 | do { | |
1099 | __invalidate_reclaim_iterators(memcg, dead_memcg); | |
1100 | last = memcg; | |
1101 | } while ((memcg = parent_mem_cgroup(memcg))); | |
1102 | ||
1103 | /* | |
1104 | * When cgruop1 non-hierarchy mode is used, | |
1105 | * parent_mem_cgroup() does not walk all the way up to the | |
1106 | * cgroup root (root_mem_cgroup). So we have to handle | |
1107 | * dead_memcg from cgroup root separately. | |
1108 | */ | |
1109 | if (last != root_mem_cgroup) | |
1110 | __invalidate_reclaim_iterators(root_mem_cgroup, | |
1111 | dead_memcg); | |
1112 | } | |
1113 | ||
7c5f64f8 VD |
1114 | /** |
1115 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy | |
1116 | * @memcg: hierarchy root | |
1117 | * @fn: function to call for each task | |
1118 | * @arg: argument passed to @fn | |
1119 | * | |
1120 | * This function iterates over tasks attached to @memcg or to any of its | |
1121 | * descendants and calls @fn for each task. If @fn returns a non-zero | |
1122 | * value, the function breaks the iteration loop and returns the value. | |
1123 | * Otherwise, it will iterate over all tasks and return 0. | |
1124 | * | |
1125 | * This function must not be called for the root memory cgroup. | |
1126 | */ | |
1127 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, | |
1128 | int (*fn)(struct task_struct *, void *), void *arg) | |
1129 | { | |
1130 | struct mem_cgroup *iter; | |
1131 | int ret = 0; | |
1132 | ||
1133 | BUG_ON(memcg == root_mem_cgroup); | |
1134 | ||
1135 | for_each_mem_cgroup_tree(iter, memcg) { | |
1136 | struct css_task_iter it; | |
1137 | struct task_struct *task; | |
1138 | ||
f168a9a5 | 1139 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
7c5f64f8 VD |
1140 | while (!ret && (task = css_task_iter_next(&it))) |
1141 | ret = fn(task, arg); | |
1142 | css_task_iter_end(&it); | |
1143 | if (ret) { | |
1144 | mem_cgroup_iter_break(memcg, iter); | |
1145 | break; | |
1146 | } | |
1147 | } | |
1148 | return ret; | |
1149 | } | |
1150 | ||
6168d0da AS |
1151 | #ifdef CONFIG_DEBUG_VM |
1152 | void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page) | |
1153 | { | |
1154 | struct mem_cgroup *memcg; | |
1155 | ||
1156 | if (mem_cgroup_disabled()) | |
1157 | return; | |
1158 | ||
1159 | memcg = page_memcg(page); | |
1160 | ||
1161 | if (!memcg) | |
1162 | VM_BUG_ON_PAGE(lruvec_memcg(lruvec) != root_mem_cgroup, page); | |
1163 | else | |
1164 | VM_BUG_ON_PAGE(lruvec_memcg(lruvec) != memcg, page); | |
1165 | } | |
1166 | #endif | |
1167 | ||
6168d0da AS |
1168 | /** |
1169 | * lock_page_lruvec - lock and return lruvec for a given page. | |
1170 | * @page: the page | |
1171 | * | |
d7e3aba5 AS |
1172 | * These functions are safe to use under any of the following conditions: |
1173 | * - page locked | |
1174 | * - PageLRU cleared | |
1175 | * - lock_page_memcg() | |
1176 | * - page->_refcount is zero | |
6168d0da AS |
1177 | */ |
1178 | struct lruvec *lock_page_lruvec(struct page *page) | |
1179 | { | |
1180 | struct lruvec *lruvec; | |
1181 | struct pglist_data *pgdat = page_pgdat(page); | |
1182 | ||
6168d0da AS |
1183 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
1184 | spin_lock(&lruvec->lru_lock); | |
6168d0da AS |
1185 | |
1186 | lruvec_memcg_debug(lruvec, page); | |
1187 | ||
1188 | return lruvec; | |
1189 | } | |
1190 | ||
1191 | struct lruvec *lock_page_lruvec_irq(struct page *page) | |
1192 | { | |
1193 | struct lruvec *lruvec; | |
1194 | struct pglist_data *pgdat = page_pgdat(page); | |
1195 | ||
6168d0da AS |
1196 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
1197 | spin_lock_irq(&lruvec->lru_lock); | |
6168d0da AS |
1198 | |
1199 | lruvec_memcg_debug(lruvec, page); | |
1200 | ||
1201 | return lruvec; | |
1202 | } | |
1203 | ||
1204 | struct lruvec *lock_page_lruvec_irqsave(struct page *page, unsigned long *flags) | |
1205 | { | |
1206 | struct lruvec *lruvec; | |
1207 | struct pglist_data *pgdat = page_pgdat(page); | |
1208 | ||
6168d0da AS |
1209 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
1210 | spin_lock_irqsave(&lruvec->lru_lock, *flags); | |
6168d0da AS |
1211 | |
1212 | lruvec_memcg_debug(lruvec, page); | |
1213 | ||
1214 | return lruvec; | |
1215 | } | |
1216 | ||
925b7673 | 1217 | /** |
fa9add64 HD |
1218 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
1219 | * @lruvec: mem_cgroup per zone lru vector | |
1220 | * @lru: index of lru list the page is sitting on | |
b4536f0c | 1221 | * @zid: zone id of the accounted pages |
fa9add64 | 1222 | * @nr_pages: positive when adding or negative when removing |
925b7673 | 1223 | * |
ca707239 HD |
1224 | * This function must be called under lru_lock, just before a page is added |
1225 | * to or just after a page is removed from an lru list (that ordering being | |
1226 | * so as to allow it to check that lru_size 0 is consistent with list_empty). | |
3f58a829 | 1227 | */ |
fa9add64 | 1228 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
b4536f0c | 1229 | int zid, int nr_pages) |
3f58a829 | 1230 | { |
ef8f2327 | 1231 | struct mem_cgroup_per_node *mz; |
fa9add64 | 1232 | unsigned long *lru_size; |
ca707239 | 1233 | long size; |
3f58a829 MK |
1234 | |
1235 | if (mem_cgroup_disabled()) | |
1236 | return; | |
1237 | ||
ef8f2327 | 1238 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
b4536f0c | 1239 | lru_size = &mz->lru_zone_size[zid][lru]; |
ca707239 HD |
1240 | |
1241 | if (nr_pages < 0) | |
1242 | *lru_size += nr_pages; | |
1243 | ||
1244 | size = *lru_size; | |
b4536f0c MH |
1245 | if (WARN_ONCE(size < 0, |
1246 | "%s(%p, %d, %d): lru_size %ld\n", | |
1247 | __func__, lruvec, lru, nr_pages, size)) { | |
ca707239 HD |
1248 | VM_BUG_ON(1); |
1249 | *lru_size = 0; | |
1250 | } | |
1251 | ||
1252 | if (nr_pages > 0) | |
1253 | *lru_size += nr_pages; | |
08e552c6 | 1254 | } |
544122e5 | 1255 | |
19942822 | 1256 | /** |
9d11ea9f | 1257 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
dad7557e | 1258 | * @memcg: the memory cgroup |
19942822 | 1259 | * |
9d11ea9f | 1260 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1261 | * pages. |
19942822 | 1262 | */ |
c0ff4b85 | 1263 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
19942822 | 1264 | { |
3e32cb2e JW |
1265 | unsigned long margin = 0; |
1266 | unsigned long count; | |
1267 | unsigned long limit; | |
9d11ea9f | 1268 | |
3e32cb2e | 1269 | count = page_counter_read(&memcg->memory); |
bbec2e15 | 1270 | limit = READ_ONCE(memcg->memory.max); |
3e32cb2e JW |
1271 | if (count < limit) |
1272 | margin = limit - count; | |
1273 | ||
7941d214 | 1274 | if (do_memsw_account()) { |
3e32cb2e | 1275 | count = page_counter_read(&memcg->memsw); |
bbec2e15 | 1276 | limit = READ_ONCE(memcg->memsw.max); |
1c4448ed | 1277 | if (count < limit) |
3e32cb2e | 1278 | margin = min(margin, limit - count); |
cbedbac3 LR |
1279 | else |
1280 | margin = 0; | |
3e32cb2e JW |
1281 | } |
1282 | ||
1283 | return margin; | |
19942822 JW |
1284 | } |
1285 | ||
32047e2a | 1286 | /* |
bdcbb659 | 1287 | * A routine for checking "mem" is under move_account() or not. |
32047e2a | 1288 | * |
bdcbb659 QH |
1289 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
1290 | * moving cgroups. This is for waiting at high-memory pressure | |
1291 | * caused by "move". | |
32047e2a | 1292 | */ |
c0ff4b85 | 1293 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
4b534334 | 1294 | { |
2bd9bb20 KH |
1295 | struct mem_cgroup *from; |
1296 | struct mem_cgroup *to; | |
4b534334 | 1297 | bool ret = false; |
2bd9bb20 KH |
1298 | /* |
1299 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1300 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1301 | */ | |
1302 | spin_lock(&mc.lock); | |
1303 | from = mc.from; | |
1304 | to = mc.to; | |
1305 | if (!from) | |
1306 | goto unlock; | |
3e92041d | 1307 | |
2314b42d JW |
1308 | ret = mem_cgroup_is_descendant(from, memcg) || |
1309 | mem_cgroup_is_descendant(to, memcg); | |
2bd9bb20 KH |
1310 | unlock: |
1311 | spin_unlock(&mc.lock); | |
4b534334 KH |
1312 | return ret; |
1313 | } | |
1314 | ||
c0ff4b85 | 1315 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
4b534334 KH |
1316 | { |
1317 | if (mc.moving_task && current != mc.moving_task) { | |
c0ff4b85 | 1318 | if (mem_cgroup_under_move(memcg)) { |
4b534334 KH |
1319 | DEFINE_WAIT(wait); |
1320 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1321 | /* moving charge context might have finished. */ | |
1322 | if (mc.moving_task) | |
1323 | schedule(); | |
1324 | finish_wait(&mc.waitq, &wait); | |
1325 | return true; | |
1326 | } | |
1327 | } | |
1328 | return false; | |
1329 | } | |
1330 | ||
5f9a4f4a MS |
1331 | struct memory_stat { |
1332 | const char *name; | |
5f9a4f4a MS |
1333 | unsigned int idx; |
1334 | }; | |
1335 | ||
57b2847d | 1336 | static const struct memory_stat memory_stats[] = { |
fff66b79 MS |
1337 | { "anon", NR_ANON_MAPPED }, |
1338 | { "file", NR_FILE_PAGES }, | |
1339 | { "kernel_stack", NR_KERNEL_STACK_KB }, | |
1340 | { "pagetables", NR_PAGETABLE }, | |
1341 | { "percpu", MEMCG_PERCPU_B }, | |
1342 | { "sock", MEMCG_SOCK }, | |
1343 | { "shmem", NR_SHMEM }, | |
1344 | { "file_mapped", NR_FILE_MAPPED }, | |
1345 | { "file_dirty", NR_FILE_DIRTY }, | |
1346 | { "file_writeback", NR_WRITEBACK }, | |
b6038942 SB |
1347 | #ifdef CONFIG_SWAP |
1348 | { "swapcached", NR_SWAPCACHE }, | |
1349 | #endif | |
5f9a4f4a | 1350 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
fff66b79 MS |
1351 | { "anon_thp", NR_ANON_THPS }, |
1352 | { "file_thp", NR_FILE_THPS }, | |
1353 | { "shmem_thp", NR_SHMEM_THPS }, | |
5f9a4f4a | 1354 | #endif |
fff66b79 MS |
1355 | { "inactive_anon", NR_INACTIVE_ANON }, |
1356 | { "active_anon", NR_ACTIVE_ANON }, | |
1357 | { "inactive_file", NR_INACTIVE_FILE }, | |
1358 | { "active_file", NR_ACTIVE_FILE }, | |
1359 | { "unevictable", NR_UNEVICTABLE }, | |
1360 | { "slab_reclaimable", NR_SLAB_RECLAIMABLE_B }, | |
1361 | { "slab_unreclaimable", NR_SLAB_UNRECLAIMABLE_B }, | |
5f9a4f4a MS |
1362 | |
1363 | /* The memory events */ | |
fff66b79 MS |
1364 | { "workingset_refault_anon", WORKINGSET_REFAULT_ANON }, |
1365 | { "workingset_refault_file", WORKINGSET_REFAULT_FILE }, | |
1366 | { "workingset_activate_anon", WORKINGSET_ACTIVATE_ANON }, | |
1367 | { "workingset_activate_file", WORKINGSET_ACTIVATE_FILE }, | |
1368 | { "workingset_restore_anon", WORKINGSET_RESTORE_ANON }, | |
1369 | { "workingset_restore_file", WORKINGSET_RESTORE_FILE }, | |
1370 | { "workingset_nodereclaim", WORKINGSET_NODERECLAIM }, | |
5f9a4f4a MS |
1371 | }; |
1372 | ||
fff66b79 MS |
1373 | /* Translate stat items to the correct unit for memory.stat output */ |
1374 | static int memcg_page_state_unit(int item) | |
1375 | { | |
1376 | switch (item) { | |
1377 | case MEMCG_PERCPU_B: | |
1378 | case NR_SLAB_RECLAIMABLE_B: | |
1379 | case NR_SLAB_UNRECLAIMABLE_B: | |
1380 | case WORKINGSET_REFAULT_ANON: | |
1381 | case WORKINGSET_REFAULT_FILE: | |
1382 | case WORKINGSET_ACTIVATE_ANON: | |
1383 | case WORKINGSET_ACTIVATE_FILE: | |
1384 | case WORKINGSET_RESTORE_ANON: | |
1385 | case WORKINGSET_RESTORE_FILE: | |
1386 | case WORKINGSET_NODERECLAIM: | |
1387 | return 1; | |
1388 | case NR_KERNEL_STACK_KB: | |
1389 | return SZ_1K; | |
1390 | default: | |
1391 | return PAGE_SIZE; | |
1392 | } | |
1393 | } | |
1394 | ||
1395 | static inline unsigned long memcg_page_state_output(struct mem_cgroup *memcg, | |
1396 | int item) | |
1397 | { | |
1398 | return memcg_page_state(memcg, item) * memcg_page_state_unit(item); | |
1399 | } | |
1400 | ||
c8713d0b JW |
1401 | static char *memory_stat_format(struct mem_cgroup *memcg) |
1402 | { | |
1403 | struct seq_buf s; | |
1404 | int i; | |
71cd3113 | 1405 | |
c8713d0b JW |
1406 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
1407 | if (!s.buffer) | |
1408 | return NULL; | |
1409 | ||
1410 | /* | |
1411 | * Provide statistics on the state of the memory subsystem as | |
1412 | * well as cumulative event counters that show past behavior. | |
1413 | * | |
1414 | * This list is ordered following a combination of these gradients: | |
1415 | * 1) generic big picture -> specifics and details | |
1416 | * 2) reflecting userspace activity -> reflecting kernel heuristics | |
1417 | * | |
1418 | * Current memory state: | |
1419 | */ | |
2d146aa3 | 1420 | cgroup_rstat_flush(memcg->css.cgroup); |
c8713d0b | 1421 | |
5f9a4f4a MS |
1422 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
1423 | u64 size; | |
c8713d0b | 1424 | |
fff66b79 | 1425 | size = memcg_page_state_output(memcg, memory_stats[i].idx); |
5f9a4f4a | 1426 | seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size); |
c8713d0b | 1427 | |
5f9a4f4a | 1428 | if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) { |
fff66b79 MS |
1429 | size += memcg_page_state_output(memcg, |
1430 | NR_SLAB_RECLAIMABLE_B); | |
5f9a4f4a MS |
1431 | seq_buf_printf(&s, "slab %llu\n", size); |
1432 | } | |
1433 | } | |
c8713d0b JW |
1434 | |
1435 | /* Accumulated memory events */ | |
1436 | ||
ebc5d83d KK |
1437 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT), |
1438 | memcg_events(memcg, PGFAULT)); | |
1439 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT), | |
1440 | memcg_events(memcg, PGMAJFAULT)); | |
ebc5d83d KK |
1441 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL), |
1442 | memcg_events(memcg, PGREFILL)); | |
c8713d0b JW |
1443 | seq_buf_printf(&s, "pgscan %lu\n", |
1444 | memcg_events(memcg, PGSCAN_KSWAPD) + | |
1445 | memcg_events(memcg, PGSCAN_DIRECT)); | |
1446 | seq_buf_printf(&s, "pgsteal %lu\n", | |
1447 | memcg_events(memcg, PGSTEAL_KSWAPD) + | |
1448 | memcg_events(memcg, PGSTEAL_DIRECT)); | |
ebc5d83d KK |
1449 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE), |
1450 | memcg_events(memcg, PGACTIVATE)); | |
1451 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE), | |
1452 | memcg_events(memcg, PGDEACTIVATE)); | |
1453 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE), | |
1454 | memcg_events(memcg, PGLAZYFREE)); | |
1455 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED), | |
1456 | memcg_events(memcg, PGLAZYFREED)); | |
c8713d0b JW |
1457 | |
1458 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
ebc5d83d | 1459 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC), |
c8713d0b | 1460 | memcg_events(memcg, THP_FAULT_ALLOC)); |
ebc5d83d | 1461 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC), |
c8713d0b JW |
1462 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
1463 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ | |
1464 | ||
1465 | /* The above should easily fit into one page */ | |
1466 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); | |
1467 | ||
1468 | return s.buffer; | |
1469 | } | |
71cd3113 | 1470 | |
58cf188e | 1471 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
e222432b | 1472 | /** |
f0c867d9 | 1473 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
1474 | * memory controller. | |
e222432b BS |
1475 | * @memcg: The memory cgroup that went over limit |
1476 | * @p: Task that is going to be killed | |
1477 | * | |
1478 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1479 | * enabled | |
1480 | */ | |
f0c867d9 | 1481 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
e222432b | 1482 | { |
e222432b BS |
1483 | rcu_read_lock(); |
1484 | ||
f0c867d9 | 1485 | if (memcg) { |
1486 | pr_cont(",oom_memcg="); | |
1487 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1488 | } else | |
1489 | pr_cont(",global_oom"); | |
2415b9f5 | 1490 | if (p) { |
f0c867d9 | 1491 | pr_cont(",task_memcg="); |
2415b9f5 | 1492 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
2415b9f5 | 1493 | } |
e222432b | 1494 | rcu_read_unlock(); |
f0c867d9 | 1495 | } |
1496 | ||
1497 | /** | |
1498 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to | |
1499 | * memory controller. | |
1500 | * @memcg: The memory cgroup that went over limit | |
1501 | */ | |
1502 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) | |
1503 | { | |
c8713d0b | 1504 | char *buf; |
e222432b | 1505 | |
3e32cb2e JW |
1506 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
1507 | K((u64)page_counter_read(&memcg->memory)), | |
15b42562 | 1508 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
c8713d0b JW |
1509 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
1510 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1511 | K((u64)page_counter_read(&memcg->swap)), | |
32d087cd | 1512 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
c8713d0b JW |
1513 | else { |
1514 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", | |
1515 | K((u64)page_counter_read(&memcg->memsw)), | |
1516 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); | |
1517 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", | |
1518 | K((u64)page_counter_read(&memcg->kmem)), | |
1519 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); | |
58cf188e | 1520 | } |
c8713d0b JW |
1521 | |
1522 | pr_info("Memory cgroup stats for "); | |
1523 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1524 | pr_cont(":"); | |
1525 | buf = memory_stat_format(memcg); | |
1526 | if (!buf) | |
1527 | return; | |
1528 | pr_info("%s", buf); | |
1529 | kfree(buf); | |
e222432b BS |
1530 | } |
1531 | ||
a63d83f4 DR |
1532 | /* |
1533 | * Return the memory (and swap, if configured) limit for a memcg. | |
1534 | */ | |
bbec2e15 | 1535 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
a63d83f4 | 1536 | { |
8d387a5f WL |
1537 | unsigned long max = READ_ONCE(memcg->memory.max); |
1538 | ||
1539 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) { | |
1540 | if (mem_cgroup_swappiness(memcg)) | |
1541 | max += min(READ_ONCE(memcg->swap.max), | |
1542 | (unsigned long)total_swap_pages); | |
1543 | } else { /* v1 */ | |
1544 | if (mem_cgroup_swappiness(memcg)) { | |
1545 | /* Calculate swap excess capacity from memsw limit */ | |
1546 | unsigned long swap = READ_ONCE(memcg->memsw.max) - max; | |
1547 | ||
1548 | max += min(swap, (unsigned long)total_swap_pages); | |
1549 | } | |
9a5a8f19 | 1550 | } |
bbec2e15 | 1551 | return max; |
a63d83f4 DR |
1552 | } |
1553 | ||
9783aa99 CD |
1554 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
1555 | { | |
1556 | return page_counter_read(&memcg->memory); | |
1557 | } | |
1558 | ||
b6e6edcf | 1559 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
19965460 | 1560 | int order) |
9cbb78bb | 1561 | { |
6e0fc46d DR |
1562 | struct oom_control oc = { |
1563 | .zonelist = NULL, | |
1564 | .nodemask = NULL, | |
2a966b77 | 1565 | .memcg = memcg, |
6e0fc46d DR |
1566 | .gfp_mask = gfp_mask, |
1567 | .order = order, | |
6e0fc46d | 1568 | }; |
1378b37d | 1569 | bool ret = true; |
9cbb78bb | 1570 | |
7775face TH |
1571 | if (mutex_lock_killable(&oom_lock)) |
1572 | return true; | |
1378b37d YS |
1573 | |
1574 | if (mem_cgroup_margin(memcg) >= (1 << order)) | |
1575 | goto unlock; | |
1576 | ||
7775face TH |
1577 | /* |
1578 | * A few threads which were not waiting at mutex_lock_killable() can | |
1579 | * fail to bail out. Therefore, check again after holding oom_lock. | |
1580 | */ | |
1581 | ret = should_force_charge() || out_of_memory(&oc); | |
1378b37d YS |
1582 | |
1583 | unlock: | |
dc56401f | 1584 | mutex_unlock(&oom_lock); |
7c5f64f8 | 1585 | return ret; |
9cbb78bb DR |
1586 | } |
1587 | ||
0608f43d | 1588 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
ef8f2327 | 1589 | pg_data_t *pgdat, |
0608f43d AM |
1590 | gfp_t gfp_mask, |
1591 | unsigned long *total_scanned) | |
1592 | { | |
1593 | struct mem_cgroup *victim = NULL; | |
1594 | int total = 0; | |
1595 | int loop = 0; | |
1596 | unsigned long excess; | |
1597 | unsigned long nr_scanned; | |
1598 | struct mem_cgroup_reclaim_cookie reclaim = { | |
ef8f2327 | 1599 | .pgdat = pgdat, |
0608f43d AM |
1600 | }; |
1601 | ||
3e32cb2e | 1602 | excess = soft_limit_excess(root_memcg); |
0608f43d AM |
1603 | |
1604 | while (1) { | |
1605 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); | |
1606 | if (!victim) { | |
1607 | loop++; | |
1608 | if (loop >= 2) { | |
1609 | /* | |
1610 | * If we have not been able to reclaim | |
1611 | * anything, it might because there are | |
1612 | * no reclaimable pages under this hierarchy | |
1613 | */ | |
1614 | if (!total) | |
1615 | break; | |
1616 | /* | |
1617 | * We want to do more targeted reclaim. | |
1618 | * excess >> 2 is not to excessive so as to | |
1619 | * reclaim too much, nor too less that we keep | |
1620 | * coming back to reclaim from this cgroup | |
1621 | */ | |
1622 | if (total >= (excess >> 2) || | |
1623 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) | |
1624 | break; | |
1625 | } | |
1626 | continue; | |
1627 | } | |
a9dd0a83 | 1628 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
ef8f2327 | 1629 | pgdat, &nr_scanned); |
0608f43d | 1630 | *total_scanned += nr_scanned; |
3e32cb2e | 1631 | if (!soft_limit_excess(root_memcg)) |
0608f43d | 1632 | break; |
6d61ef40 | 1633 | } |
0608f43d AM |
1634 | mem_cgroup_iter_break(root_memcg, victim); |
1635 | return total; | |
6d61ef40 BS |
1636 | } |
1637 | ||
0056f4e6 JW |
1638 | #ifdef CONFIG_LOCKDEP |
1639 | static struct lockdep_map memcg_oom_lock_dep_map = { | |
1640 | .name = "memcg_oom_lock", | |
1641 | }; | |
1642 | #endif | |
1643 | ||
fb2a6fc5 JW |
1644 | static DEFINE_SPINLOCK(memcg_oom_lock); |
1645 | ||
867578cb KH |
1646 | /* |
1647 | * Check OOM-Killer is already running under our hierarchy. | |
1648 | * If someone is running, return false. | |
1649 | */ | |
fb2a6fc5 | 1650 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
867578cb | 1651 | { |
79dfdacc | 1652 | struct mem_cgroup *iter, *failed = NULL; |
a636b327 | 1653 | |
fb2a6fc5 JW |
1654 | spin_lock(&memcg_oom_lock); |
1655 | ||
9f3a0d09 | 1656 | for_each_mem_cgroup_tree(iter, memcg) { |
23751be0 | 1657 | if (iter->oom_lock) { |
79dfdacc MH |
1658 | /* |
1659 | * this subtree of our hierarchy is already locked | |
1660 | * so we cannot give a lock. | |
1661 | */ | |
79dfdacc | 1662 | failed = iter; |
9f3a0d09 JW |
1663 | mem_cgroup_iter_break(memcg, iter); |
1664 | break; | |
23751be0 JW |
1665 | } else |
1666 | iter->oom_lock = true; | |
7d74b06f | 1667 | } |
867578cb | 1668 | |
fb2a6fc5 JW |
1669 | if (failed) { |
1670 | /* | |
1671 | * OK, we failed to lock the whole subtree so we have | |
1672 | * to clean up what we set up to the failing subtree | |
1673 | */ | |
1674 | for_each_mem_cgroup_tree(iter, memcg) { | |
1675 | if (iter == failed) { | |
1676 | mem_cgroup_iter_break(memcg, iter); | |
1677 | break; | |
1678 | } | |
1679 | iter->oom_lock = false; | |
79dfdacc | 1680 | } |
0056f4e6 JW |
1681 | } else |
1682 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); | |
fb2a6fc5 JW |
1683 | |
1684 | spin_unlock(&memcg_oom_lock); | |
1685 | ||
1686 | return !failed; | |
a636b327 | 1687 | } |
0b7f569e | 1688 | |
fb2a6fc5 | 1689 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
0b7f569e | 1690 | { |
7d74b06f KH |
1691 | struct mem_cgroup *iter; |
1692 | ||
fb2a6fc5 | 1693 | spin_lock(&memcg_oom_lock); |
5facae4f | 1694 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
c0ff4b85 | 1695 | for_each_mem_cgroup_tree(iter, memcg) |
79dfdacc | 1696 | iter->oom_lock = false; |
fb2a6fc5 | 1697 | spin_unlock(&memcg_oom_lock); |
79dfdacc MH |
1698 | } |
1699 | ||
c0ff4b85 | 1700 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1701 | { |
1702 | struct mem_cgroup *iter; | |
1703 | ||
c2b42d3c | 1704 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1705 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1706 | iter->under_oom++; |
1707 | spin_unlock(&memcg_oom_lock); | |
79dfdacc MH |
1708 | } |
1709 | ||
c0ff4b85 | 1710 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
79dfdacc MH |
1711 | { |
1712 | struct mem_cgroup *iter; | |
1713 | ||
867578cb | 1714 | /* |
7a52d4d8 ML |
1715 | * Be careful about under_oom underflows becase a child memcg |
1716 | * could have been added after mem_cgroup_mark_under_oom. | |
867578cb | 1717 | */ |
c2b42d3c | 1718 | spin_lock(&memcg_oom_lock); |
c0ff4b85 | 1719 | for_each_mem_cgroup_tree(iter, memcg) |
c2b42d3c TH |
1720 | if (iter->under_oom > 0) |
1721 | iter->under_oom--; | |
1722 | spin_unlock(&memcg_oom_lock); | |
0b7f569e KH |
1723 | } |
1724 | ||
867578cb KH |
1725 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1726 | ||
dc98df5a | 1727 | struct oom_wait_info { |
d79154bb | 1728 | struct mem_cgroup *memcg; |
ac6424b9 | 1729 | wait_queue_entry_t wait; |
dc98df5a KH |
1730 | }; |
1731 | ||
ac6424b9 | 1732 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
dc98df5a KH |
1733 | unsigned mode, int sync, void *arg) |
1734 | { | |
d79154bb HD |
1735 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
1736 | struct mem_cgroup *oom_wait_memcg; | |
dc98df5a KH |
1737 | struct oom_wait_info *oom_wait_info; |
1738 | ||
1739 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
d79154bb | 1740 | oom_wait_memcg = oom_wait_info->memcg; |
dc98df5a | 1741 | |
2314b42d JW |
1742 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
1743 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) | |
dc98df5a | 1744 | return 0; |
dc98df5a KH |
1745 | return autoremove_wake_function(wait, mode, sync, arg); |
1746 | } | |
1747 | ||
c0ff4b85 | 1748 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
3c11ecf4 | 1749 | { |
c2b42d3c TH |
1750 | /* |
1751 | * For the following lockless ->under_oom test, the only required | |
1752 | * guarantee is that it must see the state asserted by an OOM when | |
1753 | * this function is called as a result of userland actions | |
1754 | * triggered by the notification of the OOM. This is trivially | |
1755 | * achieved by invoking mem_cgroup_mark_under_oom() before | |
1756 | * triggering notification. | |
1757 | */ | |
1758 | if (memcg && memcg->under_oom) | |
f4b90b70 | 1759 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
3c11ecf4 KH |
1760 | } |
1761 | ||
29ef680a MH |
1762 | enum oom_status { |
1763 | OOM_SUCCESS, | |
1764 | OOM_FAILED, | |
1765 | OOM_ASYNC, | |
1766 | OOM_SKIPPED | |
1767 | }; | |
1768 | ||
1769 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) | |
0b7f569e | 1770 | { |
7056d3a3 MH |
1771 | enum oom_status ret; |
1772 | bool locked; | |
1773 | ||
29ef680a MH |
1774 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
1775 | return OOM_SKIPPED; | |
1776 | ||
7a1adfdd RG |
1777 | memcg_memory_event(memcg, MEMCG_OOM); |
1778 | ||
867578cb | 1779 | /* |
49426420 JW |
1780 | * We are in the middle of the charge context here, so we |
1781 | * don't want to block when potentially sitting on a callstack | |
1782 | * that holds all kinds of filesystem and mm locks. | |
1783 | * | |
29ef680a MH |
1784 | * cgroup1 allows disabling the OOM killer and waiting for outside |
1785 | * handling until the charge can succeed; remember the context and put | |
1786 | * the task to sleep at the end of the page fault when all locks are | |
1787 | * released. | |
49426420 | 1788 | * |
29ef680a MH |
1789 | * On the other hand, in-kernel OOM killer allows for an async victim |
1790 | * memory reclaim (oom_reaper) and that means that we are not solely | |
1791 | * relying on the oom victim to make a forward progress and we can | |
1792 | * invoke the oom killer here. | |
1793 | * | |
1794 | * Please note that mem_cgroup_out_of_memory might fail to find a | |
1795 | * victim and then we have to bail out from the charge path. | |
867578cb | 1796 | */ |
29ef680a MH |
1797 | if (memcg->oom_kill_disable) { |
1798 | if (!current->in_user_fault) | |
1799 | return OOM_SKIPPED; | |
1800 | css_get(&memcg->css); | |
1801 | current->memcg_in_oom = memcg; | |
1802 | current->memcg_oom_gfp_mask = mask; | |
1803 | current->memcg_oom_order = order; | |
1804 | ||
1805 | return OOM_ASYNC; | |
1806 | } | |
1807 | ||
7056d3a3 MH |
1808 | mem_cgroup_mark_under_oom(memcg); |
1809 | ||
1810 | locked = mem_cgroup_oom_trylock(memcg); | |
1811 | ||
1812 | if (locked) | |
1813 | mem_cgroup_oom_notify(memcg); | |
1814 | ||
1815 | mem_cgroup_unmark_under_oom(memcg); | |
29ef680a | 1816 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
7056d3a3 MH |
1817 | ret = OOM_SUCCESS; |
1818 | else | |
1819 | ret = OOM_FAILED; | |
1820 | ||
1821 | if (locked) | |
1822 | mem_cgroup_oom_unlock(memcg); | |
29ef680a | 1823 | |
7056d3a3 | 1824 | return ret; |
3812c8c8 JW |
1825 | } |
1826 | ||
1827 | /** | |
1828 | * mem_cgroup_oom_synchronize - complete memcg OOM handling | |
49426420 | 1829 | * @handle: actually kill/wait or just clean up the OOM state |
3812c8c8 | 1830 | * |
49426420 JW |
1831 | * This has to be called at the end of a page fault if the memcg OOM |
1832 | * handler was enabled. | |
3812c8c8 | 1833 | * |
49426420 | 1834 | * Memcg supports userspace OOM handling where failed allocations must |
3812c8c8 JW |
1835 | * sleep on a waitqueue until the userspace task resolves the |
1836 | * situation. Sleeping directly in the charge context with all kinds | |
1837 | * of locks held is not a good idea, instead we remember an OOM state | |
1838 | * in the task and mem_cgroup_oom_synchronize() has to be called at | |
49426420 | 1839 | * the end of the page fault to complete the OOM handling. |
3812c8c8 JW |
1840 | * |
1841 | * Returns %true if an ongoing memcg OOM situation was detected and | |
49426420 | 1842 | * completed, %false otherwise. |
3812c8c8 | 1843 | */ |
49426420 | 1844 | bool mem_cgroup_oom_synchronize(bool handle) |
3812c8c8 | 1845 | { |
626ebc41 | 1846 | struct mem_cgroup *memcg = current->memcg_in_oom; |
3812c8c8 | 1847 | struct oom_wait_info owait; |
49426420 | 1848 | bool locked; |
3812c8c8 JW |
1849 | |
1850 | /* OOM is global, do not handle */ | |
3812c8c8 | 1851 | if (!memcg) |
49426420 | 1852 | return false; |
3812c8c8 | 1853 | |
7c5f64f8 | 1854 | if (!handle) |
49426420 | 1855 | goto cleanup; |
3812c8c8 JW |
1856 | |
1857 | owait.memcg = memcg; | |
1858 | owait.wait.flags = 0; | |
1859 | owait.wait.func = memcg_oom_wake_function; | |
1860 | owait.wait.private = current; | |
2055da97 | 1861 | INIT_LIST_HEAD(&owait.wait.entry); |
867578cb | 1862 | |
3812c8c8 | 1863 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
49426420 JW |
1864 | mem_cgroup_mark_under_oom(memcg); |
1865 | ||
1866 | locked = mem_cgroup_oom_trylock(memcg); | |
1867 | ||
1868 | if (locked) | |
1869 | mem_cgroup_oom_notify(memcg); | |
1870 | ||
1871 | if (locked && !memcg->oom_kill_disable) { | |
1872 | mem_cgroup_unmark_under_oom(memcg); | |
1873 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
626ebc41 TH |
1874 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
1875 | current->memcg_oom_order); | |
49426420 | 1876 | } else { |
3812c8c8 | 1877 | schedule(); |
49426420 JW |
1878 | mem_cgroup_unmark_under_oom(memcg); |
1879 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
1880 | } | |
1881 | ||
1882 | if (locked) { | |
fb2a6fc5 JW |
1883 | mem_cgroup_oom_unlock(memcg); |
1884 | /* | |
1885 | * There is no guarantee that an OOM-lock contender | |
1886 | * sees the wakeups triggered by the OOM kill | |
1887 | * uncharges. Wake any sleepers explicitely. | |
1888 | */ | |
1889 | memcg_oom_recover(memcg); | |
1890 | } | |
49426420 | 1891 | cleanup: |
626ebc41 | 1892 | current->memcg_in_oom = NULL; |
3812c8c8 | 1893 | css_put(&memcg->css); |
867578cb | 1894 | return true; |
0b7f569e KH |
1895 | } |
1896 | ||
3d8b38eb RG |
1897 | /** |
1898 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM | |
1899 | * @victim: task to be killed by the OOM killer | |
1900 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM | |
1901 | * | |
1902 | * Returns a pointer to a memory cgroup, which has to be cleaned up | |
1903 | * by killing all belonging OOM-killable tasks. | |
1904 | * | |
1905 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. | |
1906 | */ | |
1907 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, | |
1908 | struct mem_cgroup *oom_domain) | |
1909 | { | |
1910 | struct mem_cgroup *oom_group = NULL; | |
1911 | struct mem_cgroup *memcg; | |
1912 | ||
1913 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
1914 | return NULL; | |
1915 | ||
1916 | if (!oom_domain) | |
1917 | oom_domain = root_mem_cgroup; | |
1918 | ||
1919 | rcu_read_lock(); | |
1920 | ||
1921 | memcg = mem_cgroup_from_task(victim); | |
1922 | if (memcg == root_mem_cgroup) | |
1923 | goto out; | |
1924 | ||
48fe267c RG |
1925 | /* |
1926 | * If the victim task has been asynchronously moved to a different | |
1927 | * memory cgroup, we might end up killing tasks outside oom_domain. | |
1928 | * In this case it's better to ignore memory.group.oom. | |
1929 | */ | |
1930 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) | |
1931 | goto out; | |
1932 | ||
3d8b38eb RG |
1933 | /* |
1934 | * Traverse the memory cgroup hierarchy from the victim task's | |
1935 | * cgroup up to the OOMing cgroup (or root) to find the | |
1936 | * highest-level memory cgroup with oom.group set. | |
1937 | */ | |
1938 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { | |
1939 | if (memcg->oom_group) | |
1940 | oom_group = memcg; | |
1941 | ||
1942 | if (memcg == oom_domain) | |
1943 | break; | |
1944 | } | |
1945 | ||
1946 | if (oom_group) | |
1947 | css_get(&oom_group->css); | |
1948 | out: | |
1949 | rcu_read_unlock(); | |
1950 | ||
1951 | return oom_group; | |
1952 | } | |
1953 | ||
1954 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) | |
1955 | { | |
1956 | pr_info("Tasks in "); | |
1957 | pr_cont_cgroup_path(memcg->css.cgroup); | |
1958 | pr_cont(" are going to be killed due to memory.oom.group set\n"); | |
1959 | } | |
1960 | ||
d7365e78 | 1961 | /** |
bcfe06bf | 1962 | * lock_page_memcg - lock a page and memcg binding |
81f8c3a4 | 1963 | * @page: the page |
32047e2a | 1964 | * |
81f8c3a4 | 1965 | * This function protects unlocked LRU pages from being moved to |
739f79fc JW |
1966 | * another cgroup. |
1967 | * | |
1c824a68 JW |
1968 | * It ensures lifetime of the locked memcg. Caller is responsible |
1969 | * for the lifetime of the page. | |
d69b042f | 1970 | */ |
1c824a68 | 1971 | void lock_page_memcg(struct page *page) |
89c06bd5 | 1972 | { |
9da7b521 | 1973 | struct page *head = compound_head(page); /* rmap on tail pages */ |
89c06bd5 | 1974 | struct mem_cgroup *memcg; |
6de22619 | 1975 | unsigned long flags; |
89c06bd5 | 1976 | |
6de22619 JW |
1977 | /* |
1978 | * The RCU lock is held throughout the transaction. The fast | |
1979 | * path can get away without acquiring the memcg->move_lock | |
1980 | * because page moving starts with an RCU grace period. | |
739f79fc | 1981 | */ |
d7365e78 JW |
1982 | rcu_read_lock(); |
1983 | ||
1984 | if (mem_cgroup_disabled()) | |
1c824a68 | 1985 | return; |
89c06bd5 | 1986 | again: |
bcfe06bf | 1987 | memcg = page_memcg(head); |
29833315 | 1988 | if (unlikely(!memcg)) |
1c824a68 | 1989 | return; |
d7365e78 | 1990 | |
20ad50d6 AS |
1991 | #ifdef CONFIG_PROVE_LOCKING |
1992 | local_irq_save(flags); | |
1993 | might_lock(&memcg->move_lock); | |
1994 | local_irq_restore(flags); | |
1995 | #endif | |
1996 | ||
bdcbb659 | 1997 | if (atomic_read(&memcg->moving_account) <= 0) |
1c824a68 | 1998 | return; |
89c06bd5 | 1999 | |
6de22619 | 2000 | spin_lock_irqsave(&memcg->move_lock, flags); |
bcfe06bf | 2001 | if (memcg != page_memcg(head)) { |
6de22619 | 2002 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
89c06bd5 KH |
2003 | goto again; |
2004 | } | |
6de22619 JW |
2005 | |
2006 | /* | |
1c824a68 JW |
2007 | * When charge migration first begins, we can have multiple |
2008 | * critical sections holding the fast-path RCU lock and one | |
2009 | * holding the slowpath move_lock. Track the task who has the | |
2010 | * move_lock for unlock_page_memcg(). | |
6de22619 JW |
2011 | */ |
2012 | memcg->move_lock_task = current; | |
2013 | memcg->move_lock_flags = flags; | |
89c06bd5 | 2014 | } |
81f8c3a4 | 2015 | EXPORT_SYMBOL(lock_page_memcg); |
89c06bd5 | 2016 | |
1c824a68 | 2017 | static void __unlock_page_memcg(struct mem_cgroup *memcg) |
89c06bd5 | 2018 | { |
6de22619 JW |
2019 | if (memcg && memcg->move_lock_task == current) { |
2020 | unsigned long flags = memcg->move_lock_flags; | |
2021 | ||
2022 | memcg->move_lock_task = NULL; | |
2023 | memcg->move_lock_flags = 0; | |
2024 | ||
2025 | spin_unlock_irqrestore(&memcg->move_lock, flags); | |
2026 | } | |
89c06bd5 | 2027 | |
d7365e78 | 2028 | rcu_read_unlock(); |
89c06bd5 | 2029 | } |
739f79fc JW |
2030 | |
2031 | /** | |
bcfe06bf | 2032 | * unlock_page_memcg - unlock a page and memcg binding |
739f79fc JW |
2033 | * @page: the page |
2034 | */ | |
2035 | void unlock_page_memcg(struct page *page) | |
2036 | { | |
9da7b521 JW |
2037 | struct page *head = compound_head(page); |
2038 | ||
bcfe06bf | 2039 | __unlock_page_memcg(page_memcg(head)); |
739f79fc | 2040 | } |
81f8c3a4 | 2041 | EXPORT_SYMBOL(unlock_page_memcg); |
89c06bd5 | 2042 | |
cdec2e42 KH |
2043 | struct memcg_stock_pcp { |
2044 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 2045 | unsigned int nr_pages; |
bf4f0599 RG |
2046 | |
2047 | #ifdef CONFIG_MEMCG_KMEM | |
2048 | struct obj_cgroup *cached_objcg; | |
2049 | unsigned int nr_bytes; | |
2050 | #endif | |
2051 | ||
cdec2e42 | 2052 | struct work_struct work; |
26fe6168 | 2053 | unsigned long flags; |
a0db00fc | 2054 | #define FLUSHING_CACHED_CHARGE 0 |
cdec2e42 KH |
2055 | }; |
2056 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 2057 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 | 2058 | |
bf4f0599 RG |
2059 | #ifdef CONFIG_MEMCG_KMEM |
2060 | static void drain_obj_stock(struct memcg_stock_pcp *stock); | |
2061 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2062 | struct mem_cgroup *root_memcg); | |
2063 | ||
2064 | #else | |
2065 | static inline void drain_obj_stock(struct memcg_stock_pcp *stock) | |
2066 | { | |
2067 | } | |
2068 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
2069 | struct mem_cgroup *root_memcg) | |
2070 | { | |
2071 | return false; | |
2072 | } | |
2073 | #endif | |
2074 | ||
a0956d54 SS |
2075 | /** |
2076 | * consume_stock: Try to consume stocked charge on this cpu. | |
2077 | * @memcg: memcg to consume from. | |
2078 | * @nr_pages: how many pages to charge. | |
2079 | * | |
2080 | * The charges will only happen if @memcg matches the current cpu's memcg | |
2081 | * stock, and at least @nr_pages are available in that stock. Failure to | |
2082 | * service an allocation will refill the stock. | |
2083 | * | |
2084 | * returns true if successful, false otherwise. | |
cdec2e42 | 2085 | */ |
a0956d54 | 2086 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 KH |
2087 | { |
2088 | struct memcg_stock_pcp *stock; | |
db2ba40c | 2089 | unsigned long flags; |
3e32cb2e | 2090 | bool ret = false; |
cdec2e42 | 2091 | |
a983b5eb | 2092 | if (nr_pages > MEMCG_CHARGE_BATCH) |
3e32cb2e | 2093 | return ret; |
a0956d54 | 2094 | |
db2ba40c JW |
2095 | local_irq_save(flags); |
2096 | ||
2097 | stock = this_cpu_ptr(&memcg_stock); | |
3e32cb2e | 2098 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
a0956d54 | 2099 | stock->nr_pages -= nr_pages; |
3e32cb2e JW |
2100 | ret = true; |
2101 | } | |
db2ba40c JW |
2102 | |
2103 | local_irq_restore(flags); | |
2104 | ||
cdec2e42 KH |
2105 | return ret; |
2106 | } | |
2107 | ||
2108 | /* | |
3e32cb2e | 2109 | * Returns stocks cached in percpu and reset cached information. |
cdec2e42 KH |
2110 | */ |
2111 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2112 | { | |
2113 | struct mem_cgroup *old = stock->cached; | |
2114 | ||
1a3e1f40 JW |
2115 | if (!old) |
2116 | return; | |
2117 | ||
11c9ea4e | 2118 | if (stock->nr_pages) { |
3e32cb2e | 2119 | page_counter_uncharge(&old->memory, stock->nr_pages); |
7941d214 | 2120 | if (do_memsw_account()) |
3e32cb2e | 2121 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
11c9ea4e | 2122 | stock->nr_pages = 0; |
cdec2e42 | 2123 | } |
1a3e1f40 JW |
2124 | |
2125 | css_put(&old->css); | |
cdec2e42 | 2126 | stock->cached = NULL; |
cdec2e42 KH |
2127 | } |
2128 | ||
cdec2e42 KH |
2129 | static void drain_local_stock(struct work_struct *dummy) |
2130 | { | |
db2ba40c JW |
2131 | struct memcg_stock_pcp *stock; |
2132 | unsigned long flags; | |
2133 | ||
72f0184c MH |
2134 | /* |
2135 | * The only protection from memory hotplug vs. drain_stock races is | |
2136 | * that we always operate on local CPU stock here with IRQ disabled | |
2137 | */ | |
db2ba40c JW |
2138 | local_irq_save(flags); |
2139 | ||
2140 | stock = this_cpu_ptr(&memcg_stock); | |
bf4f0599 | 2141 | drain_obj_stock(stock); |
cdec2e42 | 2142 | drain_stock(stock); |
26fe6168 | 2143 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
db2ba40c JW |
2144 | |
2145 | local_irq_restore(flags); | |
cdec2e42 KH |
2146 | } |
2147 | ||
2148 | /* | |
3e32cb2e | 2149 | * Cache charges(val) to local per_cpu area. |
320cc51d | 2150 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2151 | */ |
c0ff4b85 | 2152 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
cdec2e42 | 2153 | { |
db2ba40c JW |
2154 | struct memcg_stock_pcp *stock; |
2155 | unsigned long flags; | |
2156 | ||
2157 | local_irq_save(flags); | |
cdec2e42 | 2158 | |
db2ba40c | 2159 | stock = this_cpu_ptr(&memcg_stock); |
c0ff4b85 | 2160 | if (stock->cached != memcg) { /* reset if necessary */ |
cdec2e42 | 2161 | drain_stock(stock); |
1a3e1f40 | 2162 | css_get(&memcg->css); |
c0ff4b85 | 2163 | stock->cached = memcg; |
cdec2e42 | 2164 | } |
11c9ea4e | 2165 | stock->nr_pages += nr_pages; |
db2ba40c | 2166 | |
a983b5eb | 2167 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
475d0487 RG |
2168 | drain_stock(stock); |
2169 | ||
db2ba40c | 2170 | local_irq_restore(flags); |
cdec2e42 KH |
2171 | } |
2172 | ||
2173 | /* | |
c0ff4b85 | 2174 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
6d3d6aa2 | 2175 | * of the hierarchy under it. |
cdec2e42 | 2176 | */ |
6d3d6aa2 | 2177 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
cdec2e42 | 2178 | { |
26fe6168 | 2179 | int cpu, curcpu; |
d38144b7 | 2180 | |
6d3d6aa2 JW |
2181 | /* If someone's already draining, avoid adding running more workers. */ |
2182 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2183 | return; | |
72f0184c MH |
2184 | /* |
2185 | * Notify other cpus that system-wide "drain" is running | |
2186 | * We do not care about races with the cpu hotplug because cpu down | |
2187 | * as well as workers from this path always operate on the local | |
2188 | * per-cpu data. CPU up doesn't touch memcg_stock at all. | |
2189 | */ | |
5af12d0e | 2190 | curcpu = get_cpu(); |
cdec2e42 KH |
2191 | for_each_online_cpu(cpu) { |
2192 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
c0ff4b85 | 2193 | struct mem_cgroup *memcg; |
e1a366be | 2194 | bool flush = false; |
26fe6168 | 2195 | |
e1a366be | 2196 | rcu_read_lock(); |
c0ff4b85 | 2197 | memcg = stock->cached; |
e1a366be RG |
2198 | if (memcg && stock->nr_pages && |
2199 | mem_cgroup_is_descendant(memcg, root_memcg)) | |
2200 | flush = true; | |
bf4f0599 RG |
2201 | if (obj_stock_flush_required(stock, root_memcg)) |
2202 | flush = true; | |
e1a366be RG |
2203 | rcu_read_unlock(); |
2204 | ||
2205 | if (flush && | |
2206 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { | |
d1a05b69 MH |
2207 | if (cpu == curcpu) |
2208 | drain_local_stock(&stock->work); | |
2209 | else | |
2210 | schedule_work_on(cpu, &stock->work); | |
2211 | } | |
cdec2e42 | 2212 | } |
5af12d0e | 2213 | put_cpu(); |
9f50fad6 | 2214 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2215 | } |
2216 | ||
2cd21c89 | 2217 | static void memcg_flush_lruvec_page_state(struct mem_cgroup *memcg, int cpu) |
cdec2e42 | 2218 | { |
2cd21c89 | 2219 | int nid; |
a983b5eb | 2220 | |
2cd21c89 JW |
2221 | for_each_node(nid) { |
2222 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[nid]; | |
2223 | unsigned long stat[NR_VM_NODE_STAT_ITEMS]; | |
2224 | struct batched_lruvec_stat *lstatc; | |
a983b5eb JW |
2225 | int i; |
2226 | ||
2cd21c89 | 2227 | lstatc = per_cpu_ptr(pn->lruvec_stat_cpu, cpu); |
2d146aa3 | 2228 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) { |
2cd21c89 JW |
2229 | stat[i] = lstatc->count[i]; |
2230 | lstatc->count[i] = 0; | |
2231 | } | |
a983b5eb | 2232 | |
2cd21c89 JW |
2233 | do { |
2234 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) | |
2235 | atomic_long_add(stat[i], &pn->lruvec_stat[i]); | |
2236 | } while ((pn = parent_nodeinfo(pn, nid))); | |
2237 | } | |
2238 | } | |
a983b5eb | 2239 | |
2cd21c89 JW |
2240 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
2241 | { | |
2242 | struct memcg_stock_pcp *stock; | |
2243 | struct mem_cgroup *memcg; | |
a3d4c05a | 2244 | |
2cd21c89 JW |
2245 | stock = &per_cpu(memcg_stock, cpu); |
2246 | drain_stock(stock); | |
a3d4c05a | 2247 | |
2cd21c89 JW |
2248 | for_each_mem_cgroup(memcg) |
2249 | memcg_flush_lruvec_page_state(memcg, cpu); | |
a983b5eb | 2250 | |
308167fc | 2251 | return 0; |
cdec2e42 KH |
2252 | } |
2253 | ||
b3ff9291 CD |
2254 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
2255 | unsigned int nr_pages, | |
2256 | gfp_t gfp_mask) | |
f7e1cb6e | 2257 | { |
b3ff9291 CD |
2258 | unsigned long nr_reclaimed = 0; |
2259 | ||
f7e1cb6e | 2260 | do { |
e22c6ed9 JW |
2261 | unsigned long pflags; |
2262 | ||
d1663a90 JK |
2263 | if (page_counter_read(&memcg->memory) <= |
2264 | READ_ONCE(memcg->memory.high)) | |
f7e1cb6e | 2265 | continue; |
e22c6ed9 | 2266 | |
e27be240 | 2267 | memcg_memory_event(memcg, MEMCG_HIGH); |
e22c6ed9 JW |
2268 | |
2269 | psi_memstall_enter(&pflags); | |
b3ff9291 CD |
2270 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
2271 | gfp_mask, true); | |
e22c6ed9 | 2272 | psi_memstall_leave(&pflags); |
4bf17307 CD |
2273 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2274 | !mem_cgroup_is_root(memcg)); | |
b3ff9291 CD |
2275 | |
2276 | return nr_reclaimed; | |
f7e1cb6e JW |
2277 | } |
2278 | ||
2279 | static void high_work_func(struct work_struct *work) | |
2280 | { | |
2281 | struct mem_cgroup *memcg; | |
2282 | ||
2283 | memcg = container_of(work, struct mem_cgroup, high_work); | |
a983b5eb | 2284 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
f7e1cb6e JW |
2285 | } |
2286 | ||
0e4b01df CD |
2287 | /* |
2288 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is | |
2289 | * enough to still cause a significant slowdown in most cases, while still | |
2290 | * allowing diagnostics and tracing to proceed without becoming stuck. | |
2291 | */ | |
2292 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) | |
2293 | ||
2294 | /* | |
2295 | * When calculating the delay, we use these either side of the exponentiation to | |
2296 | * maintain precision and scale to a reasonable number of jiffies (see the table | |
2297 | * below. | |
2298 | * | |
2299 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the | |
2300 | * overage ratio to a delay. | |
ac5ddd0f | 2301 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
0e4b01df CD |
2302 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
2303 | * to produce a reasonable delay curve. | |
2304 | * | |
2305 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a | |
2306 | * reasonable delay curve compared to precision-adjusted overage, not | |
2307 | * penalising heavily at first, but still making sure that growth beyond the | |
2308 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For | |
2309 | * example, with a high of 100 megabytes: | |
2310 | * | |
2311 | * +-------+------------------------+ | |
2312 | * | usage | time to allocate in ms | | |
2313 | * +-------+------------------------+ | |
2314 | * | 100M | 0 | | |
2315 | * | 101M | 6 | | |
2316 | * | 102M | 25 | | |
2317 | * | 103M | 57 | | |
2318 | * | 104M | 102 | | |
2319 | * | 105M | 159 | | |
2320 | * | 106M | 230 | | |
2321 | * | 107M | 313 | | |
2322 | * | 108M | 409 | | |
2323 | * | 109M | 518 | | |
2324 | * | 110M | 639 | | |
2325 | * | 111M | 774 | | |
2326 | * | 112M | 921 | | |
2327 | * | 113M | 1081 | | |
2328 | * | 114M | 1254 | | |
2329 | * | 115M | 1439 | | |
2330 | * | 116M | 1638 | | |
2331 | * | 117M | 1849 | | |
2332 | * | 118M | 2000 | | |
2333 | * | 119M | 2000 | | |
2334 | * | 120M | 2000 | | |
2335 | * +-------+------------------------+ | |
2336 | */ | |
2337 | #define MEMCG_DELAY_PRECISION_SHIFT 20 | |
2338 | #define MEMCG_DELAY_SCALING_SHIFT 14 | |
2339 | ||
8a5dbc65 | 2340 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
b23afb93 | 2341 | { |
8a5dbc65 | 2342 | u64 overage; |
b23afb93 | 2343 | |
8a5dbc65 JK |
2344 | if (usage <= high) |
2345 | return 0; | |
e26733e0 | 2346 | |
8a5dbc65 JK |
2347 | /* |
2348 | * Prevent division by 0 in overage calculation by acting as if | |
2349 | * it was a threshold of 1 page | |
2350 | */ | |
2351 | high = max(high, 1UL); | |
9b8b1754 | 2352 | |
8a5dbc65 JK |
2353 | overage = usage - high; |
2354 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; | |
2355 | return div64_u64(overage, high); | |
2356 | } | |
e26733e0 | 2357 | |
8a5dbc65 JK |
2358 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
2359 | { | |
2360 | u64 overage, max_overage = 0; | |
e26733e0 | 2361 | |
8a5dbc65 JK |
2362 | do { |
2363 | overage = calculate_overage(page_counter_read(&memcg->memory), | |
d1663a90 | 2364 | READ_ONCE(memcg->memory.high)); |
8a5dbc65 | 2365 | max_overage = max(overage, max_overage); |
e26733e0 CD |
2366 | } while ((memcg = parent_mem_cgroup(memcg)) && |
2367 | !mem_cgroup_is_root(memcg)); | |
2368 | ||
8a5dbc65 JK |
2369 | return max_overage; |
2370 | } | |
2371 | ||
4b82ab4f JK |
2372 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
2373 | { | |
2374 | u64 overage, max_overage = 0; | |
2375 | ||
2376 | do { | |
2377 | overage = calculate_overage(page_counter_read(&memcg->swap), | |
2378 | READ_ONCE(memcg->swap.high)); | |
2379 | if (overage) | |
2380 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); | |
2381 | max_overage = max(overage, max_overage); | |
2382 | } while ((memcg = parent_mem_cgroup(memcg)) && | |
2383 | !mem_cgroup_is_root(memcg)); | |
2384 | ||
2385 | return max_overage; | |
2386 | } | |
2387 | ||
8a5dbc65 JK |
2388 | /* |
2389 | * Get the number of jiffies that we should penalise a mischievous cgroup which | |
2390 | * is exceeding its memory.high by checking both it and its ancestors. | |
2391 | */ | |
2392 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, | |
2393 | unsigned int nr_pages, | |
2394 | u64 max_overage) | |
2395 | { | |
2396 | unsigned long penalty_jiffies; | |
2397 | ||
e26733e0 CD |
2398 | if (!max_overage) |
2399 | return 0; | |
0e4b01df CD |
2400 | |
2401 | /* | |
0e4b01df CD |
2402 | * We use overage compared to memory.high to calculate the number of |
2403 | * jiffies to sleep (penalty_jiffies). Ideally this value should be | |
2404 | * fairly lenient on small overages, and increasingly harsh when the | |
2405 | * memcg in question makes it clear that it has no intention of stopping | |
2406 | * its crazy behaviour, so we exponentially increase the delay based on | |
2407 | * overage amount. | |
2408 | */ | |
e26733e0 CD |
2409 | penalty_jiffies = max_overage * max_overage * HZ; |
2410 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; | |
2411 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; | |
0e4b01df CD |
2412 | |
2413 | /* | |
2414 | * Factor in the task's own contribution to the overage, such that four | |
2415 | * N-sized allocations are throttled approximately the same as one | |
2416 | * 4N-sized allocation. | |
2417 | * | |
2418 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or | |
2419 | * larger the current charge patch is than that. | |
2420 | */ | |
ff144e69 | 2421 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
e26733e0 CD |
2422 | } |
2423 | ||
2424 | /* | |
2425 | * Scheduled by try_charge() to be executed from the userland return path | |
2426 | * and reclaims memory over the high limit. | |
2427 | */ | |
2428 | void mem_cgroup_handle_over_high(void) | |
2429 | { | |
2430 | unsigned long penalty_jiffies; | |
2431 | unsigned long pflags; | |
b3ff9291 | 2432 | unsigned long nr_reclaimed; |
e26733e0 | 2433 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
d977aa93 | 2434 | int nr_retries = MAX_RECLAIM_RETRIES; |
e26733e0 | 2435 | struct mem_cgroup *memcg; |
b3ff9291 | 2436 | bool in_retry = false; |
e26733e0 CD |
2437 | |
2438 | if (likely(!nr_pages)) | |
2439 | return; | |
2440 | ||
2441 | memcg = get_mem_cgroup_from_mm(current->mm); | |
e26733e0 CD |
2442 | current->memcg_nr_pages_over_high = 0; |
2443 | ||
b3ff9291 CD |
2444 | retry_reclaim: |
2445 | /* | |
2446 | * The allocating task should reclaim at least the batch size, but for | |
2447 | * subsequent retries we only want to do what's necessary to prevent oom | |
2448 | * or breaching resource isolation. | |
2449 | * | |
2450 | * This is distinct from memory.max or page allocator behaviour because | |
2451 | * memory.high is currently batched, whereas memory.max and the page | |
2452 | * allocator run every time an allocation is made. | |
2453 | */ | |
2454 | nr_reclaimed = reclaim_high(memcg, | |
2455 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, | |
2456 | GFP_KERNEL); | |
2457 | ||
e26733e0 CD |
2458 | /* |
2459 | * memory.high is breached and reclaim is unable to keep up. Throttle | |
2460 | * allocators proactively to slow down excessive growth. | |
2461 | */ | |
8a5dbc65 JK |
2462 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
2463 | mem_find_max_overage(memcg)); | |
0e4b01df | 2464 | |
4b82ab4f JK |
2465 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
2466 | swap_find_max_overage(memcg)); | |
2467 | ||
ff144e69 JK |
2468 | /* |
2469 | * Clamp the max delay per usermode return so as to still keep the | |
2470 | * application moving forwards and also permit diagnostics, albeit | |
2471 | * extremely slowly. | |
2472 | */ | |
2473 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); | |
2474 | ||
0e4b01df CD |
2475 | /* |
2476 | * Don't sleep if the amount of jiffies this memcg owes us is so low | |
2477 | * that it's not even worth doing, in an attempt to be nice to those who | |
2478 | * go only a small amount over their memory.high value and maybe haven't | |
2479 | * been aggressively reclaimed enough yet. | |
2480 | */ | |
2481 | if (penalty_jiffies <= HZ / 100) | |
2482 | goto out; | |
2483 | ||
b3ff9291 CD |
2484 | /* |
2485 | * If reclaim is making forward progress but we're still over | |
2486 | * memory.high, we want to encourage that rather than doing allocator | |
2487 | * throttling. | |
2488 | */ | |
2489 | if (nr_reclaimed || nr_retries--) { | |
2490 | in_retry = true; | |
2491 | goto retry_reclaim; | |
2492 | } | |
2493 | ||
0e4b01df CD |
2494 | /* |
2495 | * If we exit early, we're guaranteed to die (since | |
2496 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't | |
2497 | * need to account for any ill-begotten jiffies to pay them off later. | |
2498 | */ | |
2499 | psi_memstall_enter(&pflags); | |
2500 | schedule_timeout_killable(penalty_jiffies); | |
2501 | psi_memstall_leave(&pflags); | |
2502 | ||
2503 | out: | |
2504 | css_put(&memcg->css); | |
b23afb93 TH |
2505 | } |
2506 | ||
00501b53 JW |
2507 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
2508 | unsigned int nr_pages) | |
8a9f3ccd | 2509 | { |
a983b5eb | 2510 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
d977aa93 | 2511 | int nr_retries = MAX_RECLAIM_RETRIES; |
6539cc05 | 2512 | struct mem_cgroup *mem_over_limit; |
3e32cb2e | 2513 | struct page_counter *counter; |
e22c6ed9 | 2514 | enum oom_status oom_status; |
6539cc05 | 2515 | unsigned long nr_reclaimed; |
b70a2a21 JW |
2516 | bool may_swap = true; |
2517 | bool drained = false; | |
e22c6ed9 | 2518 | unsigned long pflags; |
a636b327 | 2519 | |
ce00a967 | 2520 | if (mem_cgroup_is_root(memcg)) |
10d53c74 | 2521 | return 0; |
6539cc05 | 2522 | retry: |
b6b6cc72 | 2523 | if (consume_stock(memcg, nr_pages)) |
10d53c74 | 2524 | return 0; |
8a9f3ccd | 2525 | |
7941d214 | 2526 | if (!do_memsw_account() || |
6071ca52 JW |
2527 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
2528 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) | |
6539cc05 | 2529 | goto done_restock; |
7941d214 | 2530 | if (do_memsw_account()) |
3e32cb2e JW |
2531 | page_counter_uncharge(&memcg->memsw, batch); |
2532 | mem_over_limit = mem_cgroup_from_counter(counter, memory); | |
3fbe7244 | 2533 | } else { |
3e32cb2e | 2534 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
b70a2a21 | 2535 | may_swap = false; |
3fbe7244 | 2536 | } |
7a81b88c | 2537 | |
6539cc05 JW |
2538 | if (batch > nr_pages) { |
2539 | batch = nr_pages; | |
2540 | goto retry; | |
2541 | } | |
6d61ef40 | 2542 | |
869712fd JW |
2543 | /* |
2544 | * Memcg doesn't have a dedicated reserve for atomic | |
2545 | * allocations. But like the global atomic pool, we need to | |
2546 | * put the burden of reclaim on regular allocation requests | |
2547 | * and let these go through as privileged allocations. | |
2548 | */ | |
2549 | if (gfp_mask & __GFP_ATOMIC) | |
2550 | goto force; | |
2551 | ||
06b078fc JW |
2552 | /* |
2553 | * Unlike in global OOM situations, memcg is not in a physical | |
2554 | * memory shortage. Allow dying and OOM-killed tasks to | |
2555 | * bypass the last charges so that they can exit quickly and | |
2556 | * free their memory. | |
2557 | */ | |
7775face | 2558 | if (unlikely(should_force_charge())) |
10d53c74 | 2559 | goto force; |
06b078fc | 2560 | |
89a28483 JW |
2561 | /* |
2562 | * Prevent unbounded recursion when reclaim operations need to | |
2563 | * allocate memory. This might exceed the limits temporarily, | |
2564 | * but we prefer facilitating memory reclaim and getting back | |
2565 | * under the limit over triggering OOM kills in these cases. | |
2566 | */ | |
2567 | if (unlikely(current->flags & PF_MEMALLOC)) | |
2568 | goto force; | |
2569 | ||
06b078fc JW |
2570 | if (unlikely(task_in_memcg_oom(current))) |
2571 | goto nomem; | |
2572 | ||
d0164adc | 2573 | if (!gfpflags_allow_blocking(gfp_mask)) |
6539cc05 | 2574 | goto nomem; |
4b534334 | 2575 | |
e27be240 | 2576 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
241994ed | 2577 | |
e22c6ed9 | 2578 | psi_memstall_enter(&pflags); |
b70a2a21 JW |
2579 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
2580 | gfp_mask, may_swap); | |
e22c6ed9 | 2581 | psi_memstall_leave(&pflags); |
6539cc05 | 2582 | |
61e02c74 | 2583 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
6539cc05 | 2584 | goto retry; |
28c34c29 | 2585 | |
b70a2a21 | 2586 | if (!drained) { |
6d3d6aa2 | 2587 | drain_all_stock(mem_over_limit); |
b70a2a21 JW |
2588 | drained = true; |
2589 | goto retry; | |
2590 | } | |
2591 | ||
28c34c29 JW |
2592 | if (gfp_mask & __GFP_NORETRY) |
2593 | goto nomem; | |
6539cc05 JW |
2594 | /* |
2595 | * Even though the limit is exceeded at this point, reclaim | |
2596 | * may have been able to free some pages. Retry the charge | |
2597 | * before killing the task. | |
2598 | * | |
2599 | * Only for regular pages, though: huge pages are rather | |
2600 | * unlikely to succeed so close to the limit, and we fall back | |
2601 | * to regular pages anyway in case of failure. | |
2602 | */ | |
61e02c74 | 2603 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
6539cc05 JW |
2604 | goto retry; |
2605 | /* | |
2606 | * At task move, charge accounts can be doubly counted. So, it's | |
2607 | * better to wait until the end of task_move if something is going on. | |
2608 | */ | |
2609 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2610 | goto retry; | |
2611 | ||
9b130619 JW |
2612 | if (nr_retries--) |
2613 | goto retry; | |
2614 | ||
38d38493 | 2615 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
29ef680a MH |
2616 | goto nomem; |
2617 | ||
6539cc05 | 2618 | if (fatal_signal_pending(current)) |
10d53c74 | 2619 | goto force; |
6539cc05 | 2620 | |
29ef680a MH |
2621 | /* |
2622 | * keep retrying as long as the memcg oom killer is able to make | |
2623 | * a forward progress or bypass the charge if the oom killer | |
2624 | * couldn't make any progress. | |
2625 | */ | |
2626 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, | |
3608de07 | 2627 | get_order(nr_pages * PAGE_SIZE)); |
29ef680a MH |
2628 | switch (oom_status) { |
2629 | case OOM_SUCCESS: | |
d977aa93 | 2630 | nr_retries = MAX_RECLAIM_RETRIES; |
29ef680a MH |
2631 | goto retry; |
2632 | case OOM_FAILED: | |
2633 | goto force; | |
2634 | default: | |
2635 | goto nomem; | |
2636 | } | |
7a81b88c | 2637 | nomem: |
6d1fdc48 | 2638 | if (!(gfp_mask & __GFP_NOFAIL)) |
3168ecbe | 2639 | return -ENOMEM; |
10d53c74 TH |
2640 | force: |
2641 | /* | |
2642 | * The allocation either can't fail or will lead to more memory | |
2643 | * being freed very soon. Allow memory usage go over the limit | |
2644 | * temporarily by force charging it. | |
2645 | */ | |
2646 | page_counter_charge(&memcg->memory, nr_pages); | |
7941d214 | 2647 | if (do_memsw_account()) |
10d53c74 | 2648 | page_counter_charge(&memcg->memsw, nr_pages); |
10d53c74 TH |
2649 | |
2650 | return 0; | |
6539cc05 JW |
2651 | |
2652 | done_restock: | |
2653 | if (batch > nr_pages) | |
2654 | refill_stock(memcg, batch - nr_pages); | |
b23afb93 | 2655 | |
241994ed | 2656 | /* |
b23afb93 TH |
2657 | * If the hierarchy is above the normal consumption range, schedule |
2658 | * reclaim on returning to userland. We can perform reclaim here | |
71baba4b | 2659 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
b23afb93 TH |
2660 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
2661 | * not recorded as it most likely matches current's and won't | |
2662 | * change in the meantime. As high limit is checked again before | |
2663 | * reclaim, the cost of mismatch is negligible. | |
241994ed JW |
2664 | */ |
2665 | do { | |
4b82ab4f JK |
2666 | bool mem_high, swap_high; |
2667 | ||
2668 | mem_high = page_counter_read(&memcg->memory) > | |
2669 | READ_ONCE(memcg->memory.high); | |
2670 | swap_high = page_counter_read(&memcg->swap) > | |
2671 | READ_ONCE(memcg->swap.high); | |
2672 | ||
2673 | /* Don't bother a random interrupted task */ | |
2674 | if (in_interrupt()) { | |
2675 | if (mem_high) { | |
f7e1cb6e JW |
2676 | schedule_work(&memcg->high_work); |
2677 | break; | |
2678 | } | |
4b82ab4f JK |
2679 | continue; |
2680 | } | |
2681 | ||
2682 | if (mem_high || swap_high) { | |
2683 | /* | |
2684 | * The allocating tasks in this cgroup will need to do | |
2685 | * reclaim or be throttled to prevent further growth | |
2686 | * of the memory or swap footprints. | |
2687 | * | |
2688 | * Target some best-effort fairness between the tasks, | |
2689 | * and distribute reclaim work and delay penalties | |
2690 | * based on how much each task is actually allocating. | |
2691 | */ | |
9516a18a | 2692 | current->memcg_nr_pages_over_high += batch; |
b23afb93 TH |
2693 | set_notify_resume(current); |
2694 | break; | |
2695 | } | |
241994ed | 2696 | } while ((memcg = parent_mem_cgroup(memcg))); |
10d53c74 TH |
2697 | |
2698 | return 0; | |
7a81b88c | 2699 | } |
8a9f3ccd | 2700 | |
f0e45fb4 | 2701 | #if defined(CONFIG_MEMCG_KMEM) || defined(CONFIG_MMU) |
00501b53 | 2702 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
a3032a2c | 2703 | { |
ce00a967 JW |
2704 | if (mem_cgroup_is_root(memcg)) |
2705 | return; | |
2706 | ||
3e32cb2e | 2707 | page_counter_uncharge(&memcg->memory, nr_pages); |
7941d214 | 2708 | if (do_memsw_account()) |
3e32cb2e | 2709 | page_counter_uncharge(&memcg->memsw, nr_pages); |
d01dd17f | 2710 | } |
f0e45fb4 | 2711 | #endif |
d01dd17f | 2712 | |
d9eb1ea2 | 2713 | static void commit_charge(struct page *page, struct mem_cgroup *memcg) |
0a31bc97 | 2714 | { |
bcfe06bf | 2715 | VM_BUG_ON_PAGE(page_memcg(page), page); |
0a31bc97 | 2716 | /* |
a5eb011a | 2717 | * Any of the following ensures page's memcg stability: |
0a31bc97 | 2718 | * |
a0b5b414 JW |
2719 | * - the page lock |
2720 | * - LRU isolation | |
2721 | * - lock_page_memcg() | |
2722 | * - exclusive reference | |
0a31bc97 | 2723 | */ |
bcfe06bf | 2724 | page->memcg_data = (unsigned long)memcg; |
7a81b88c | 2725 | } |
66e1707b | 2726 | |
e74d2259 MS |
2727 | static struct mem_cgroup *get_mem_cgroup_from_objcg(struct obj_cgroup *objcg) |
2728 | { | |
2729 | struct mem_cgroup *memcg; | |
2730 | ||
2731 | rcu_read_lock(); | |
2732 | retry: | |
2733 | memcg = obj_cgroup_memcg(objcg); | |
2734 | if (unlikely(!css_tryget(&memcg->css))) | |
2735 | goto retry; | |
2736 | rcu_read_unlock(); | |
2737 | ||
2738 | return memcg; | |
2739 | } | |
2740 | ||
84c07d11 | 2741 | #ifdef CONFIG_MEMCG_KMEM |
10befea9 | 2742 | int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s, |
2e9bd483 | 2743 | gfp_t gfp, bool new_page) |
10befea9 RG |
2744 | { |
2745 | unsigned int objects = objs_per_slab_page(s, page); | |
2e9bd483 | 2746 | unsigned long memcg_data; |
10befea9 RG |
2747 | void *vec; |
2748 | ||
2749 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, | |
2750 | page_to_nid(page)); | |
2751 | if (!vec) | |
2752 | return -ENOMEM; | |
2753 | ||
2e9bd483 RG |
2754 | memcg_data = (unsigned long) vec | MEMCG_DATA_OBJCGS; |
2755 | if (new_page) { | |
2756 | /* | |
2757 | * If the slab page is brand new and nobody can yet access | |
2758 | * it's memcg_data, no synchronization is required and | |
2759 | * memcg_data can be simply assigned. | |
2760 | */ | |
2761 | page->memcg_data = memcg_data; | |
2762 | } else if (cmpxchg(&page->memcg_data, 0, memcg_data)) { | |
2763 | /* | |
2764 | * If the slab page is already in use, somebody can allocate | |
2765 | * and assign obj_cgroups in parallel. In this case the existing | |
2766 | * objcg vector should be reused. | |
2767 | */ | |
10befea9 | 2768 | kfree(vec); |
2e9bd483 RG |
2769 | return 0; |
2770 | } | |
10befea9 | 2771 | |
2e9bd483 | 2772 | kmemleak_not_leak(vec); |
10befea9 RG |
2773 | return 0; |
2774 | } | |
2775 | ||
8380ce47 RG |
2776 | /* |
2777 | * Returns a pointer to the memory cgroup to which the kernel object is charged. | |
2778 | * | |
bcfe06bf RG |
2779 | * A passed kernel object can be a slab object or a generic kernel page, so |
2780 | * different mechanisms for getting the memory cgroup pointer should be used. | |
2781 | * In certain cases (e.g. kernel stacks or large kmallocs with SLUB) the caller | |
2782 | * can not know for sure how the kernel object is implemented. | |
2783 | * mem_cgroup_from_obj() can be safely used in such cases. | |
2784 | * | |
8380ce47 RG |
2785 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), |
2786 | * cgroup_mutex, etc. | |
2787 | */ | |
2788 | struct mem_cgroup *mem_cgroup_from_obj(void *p) | |
2789 | { | |
2790 | struct page *page; | |
2791 | ||
2792 | if (mem_cgroup_disabled()) | |
2793 | return NULL; | |
2794 | ||
2795 | page = virt_to_head_page(p); | |
2796 | ||
2797 | /* | |
9855609b RG |
2798 | * Slab objects are accounted individually, not per-page. |
2799 | * Memcg membership data for each individual object is saved in | |
2800 | * the page->obj_cgroups. | |
8380ce47 | 2801 | */ |
270c6a71 | 2802 | if (page_objcgs_check(page)) { |
9855609b RG |
2803 | struct obj_cgroup *objcg; |
2804 | unsigned int off; | |
2805 | ||
2806 | off = obj_to_index(page->slab_cache, page, p); | |
270c6a71 | 2807 | objcg = page_objcgs(page)[off]; |
10befea9 RG |
2808 | if (objcg) |
2809 | return obj_cgroup_memcg(objcg); | |
2810 | ||
2811 | return NULL; | |
9855609b | 2812 | } |
8380ce47 | 2813 | |
bcfe06bf RG |
2814 | /* |
2815 | * page_memcg_check() is used here, because page_has_obj_cgroups() | |
2816 | * check above could fail because the object cgroups vector wasn't set | |
2817 | * at that moment, but it can be set concurrently. | |
2818 | * page_memcg_check(page) will guarantee that a proper memory | |
2819 | * cgroup pointer or NULL will be returned. | |
2820 | */ | |
2821 | return page_memcg_check(page); | |
8380ce47 RG |
2822 | } |
2823 | ||
bf4f0599 RG |
2824 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
2825 | { | |
2826 | struct obj_cgroup *objcg = NULL; | |
2827 | struct mem_cgroup *memcg; | |
2828 | ||
279c3393 RG |
2829 | if (memcg_kmem_bypass()) |
2830 | return NULL; | |
2831 | ||
bf4f0599 | 2832 | rcu_read_lock(); |
37d5985c RG |
2833 | if (unlikely(active_memcg())) |
2834 | memcg = active_memcg(); | |
bf4f0599 RG |
2835 | else |
2836 | memcg = mem_cgroup_from_task(current); | |
2837 | ||
2838 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { | |
2839 | objcg = rcu_dereference(memcg->objcg); | |
2840 | if (objcg && obj_cgroup_tryget(objcg)) | |
2841 | break; | |
2f7659a3 | 2842 | objcg = NULL; |
bf4f0599 RG |
2843 | } |
2844 | rcu_read_unlock(); | |
2845 | ||
2846 | return objcg; | |
2847 | } | |
2848 | ||
f3bb3043 | 2849 | static int memcg_alloc_cache_id(void) |
55007d84 | 2850 | { |
f3bb3043 VD |
2851 | int id, size; |
2852 | int err; | |
2853 | ||
dbcf73e2 | 2854 | id = ida_simple_get(&memcg_cache_ida, |
f3bb3043 VD |
2855 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
2856 | if (id < 0) | |
2857 | return id; | |
55007d84 | 2858 | |
dbcf73e2 | 2859 | if (id < memcg_nr_cache_ids) |
f3bb3043 VD |
2860 | return id; |
2861 | ||
2862 | /* | |
2863 | * There's no space for the new id in memcg_caches arrays, | |
2864 | * so we have to grow them. | |
2865 | */ | |
05257a1a | 2866 | down_write(&memcg_cache_ids_sem); |
f3bb3043 VD |
2867 | |
2868 | size = 2 * (id + 1); | |
55007d84 GC |
2869 | if (size < MEMCG_CACHES_MIN_SIZE) |
2870 | size = MEMCG_CACHES_MIN_SIZE; | |
2871 | else if (size > MEMCG_CACHES_MAX_SIZE) | |
2872 | size = MEMCG_CACHES_MAX_SIZE; | |
2873 | ||
9855609b | 2874 | err = memcg_update_all_list_lrus(size); |
05257a1a VD |
2875 | if (!err) |
2876 | memcg_nr_cache_ids = size; | |
2877 | ||
2878 | up_write(&memcg_cache_ids_sem); | |
2879 | ||
f3bb3043 | 2880 | if (err) { |
dbcf73e2 | 2881 | ida_simple_remove(&memcg_cache_ida, id); |
f3bb3043 VD |
2882 | return err; |
2883 | } | |
2884 | return id; | |
2885 | } | |
2886 | ||
2887 | static void memcg_free_cache_id(int id) | |
2888 | { | |
dbcf73e2 | 2889 | ida_simple_remove(&memcg_cache_ida, id); |
55007d84 GC |
2890 | } |
2891 | ||
f1286fae MS |
2892 | /* |
2893 | * obj_cgroup_uncharge_pages: uncharge a number of kernel pages from a objcg | |
2894 | * @objcg: object cgroup to uncharge | |
2895 | * @nr_pages: number of pages to uncharge | |
2896 | */ | |
e74d2259 MS |
2897 | static void obj_cgroup_uncharge_pages(struct obj_cgroup *objcg, |
2898 | unsigned int nr_pages) | |
2899 | { | |
2900 | struct mem_cgroup *memcg; | |
2901 | ||
2902 | memcg = get_mem_cgroup_from_objcg(objcg); | |
e74d2259 | 2903 | |
f1286fae MS |
2904 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
2905 | page_counter_uncharge(&memcg->kmem, nr_pages); | |
2906 | refill_stock(memcg, nr_pages); | |
e74d2259 | 2907 | |
e74d2259 | 2908 | css_put(&memcg->css); |
e74d2259 MS |
2909 | } |
2910 | ||
f1286fae MS |
2911 | /* |
2912 | * obj_cgroup_charge_pages: charge a number of kernel pages to a objcg | |
2913 | * @objcg: object cgroup to charge | |
45264778 | 2914 | * @gfp: reclaim mode |
92d0510c | 2915 | * @nr_pages: number of pages to charge |
45264778 VD |
2916 | * |
2917 | * Returns 0 on success, an error code on failure. | |
2918 | */ | |
f1286fae MS |
2919 | static int obj_cgroup_charge_pages(struct obj_cgroup *objcg, gfp_t gfp, |
2920 | unsigned int nr_pages) | |
7ae1e1d0 | 2921 | { |
f3ccb2c4 | 2922 | struct page_counter *counter; |
f1286fae | 2923 | struct mem_cgroup *memcg; |
7ae1e1d0 GC |
2924 | int ret; |
2925 | ||
f1286fae MS |
2926 | memcg = get_mem_cgroup_from_objcg(objcg); |
2927 | ||
f3ccb2c4 | 2928 | ret = try_charge(memcg, gfp, nr_pages); |
52c29b04 | 2929 | if (ret) |
f1286fae | 2930 | goto out; |
52c29b04 JW |
2931 | |
2932 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && | |
2933 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { | |
e55d9d9b MH |
2934 | |
2935 | /* | |
2936 | * Enforce __GFP_NOFAIL allocation because callers are not | |
2937 | * prepared to see failures and likely do not have any failure | |
2938 | * handling code. | |
2939 | */ | |
2940 | if (gfp & __GFP_NOFAIL) { | |
2941 | page_counter_charge(&memcg->kmem, nr_pages); | |
f1286fae | 2942 | goto out; |
e55d9d9b | 2943 | } |
52c29b04 | 2944 | cancel_charge(memcg, nr_pages); |
f1286fae | 2945 | ret = -ENOMEM; |
7ae1e1d0 | 2946 | } |
f1286fae MS |
2947 | out: |
2948 | css_put(&memcg->css); | |
4b13f64d | 2949 | |
f1286fae | 2950 | return ret; |
4b13f64d RG |
2951 | } |
2952 | ||
45264778 | 2953 | /** |
f4b00eab | 2954 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
45264778 VD |
2955 | * @page: page to charge |
2956 | * @gfp: reclaim mode | |
2957 | * @order: allocation order | |
2958 | * | |
2959 | * Returns 0 on success, an error code on failure. | |
2960 | */ | |
f4b00eab | 2961 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
7ae1e1d0 | 2962 | { |
b4e0b68f | 2963 | struct obj_cgroup *objcg; |
fcff7d7e | 2964 | int ret = 0; |
7ae1e1d0 | 2965 | |
b4e0b68f MS |
2966 | objcg = get_obj_cgroup_from_current(); |
2967 | if (objcg) { | |
2968 | ret = obj_cgroup_charge_pages(objcg, gfp, 1 << order); | |
4d96ba35 | 2969 | if (!ret) { |
b4e0b68f | 2970 | page->memcg_data = (unsigned long)objcg | |
18b2db3b | 2971 | MEMCG_DATA_KMEM; |
1a3e1f40 | 2972 | return 0; |
4d96ba35 | 2973 | } |
b4e0b68f | 2974 | obj_cgroup_put(objcg); |
c4159a75 | 2975 | } |
d05e83a6 | 2976 | return ret; |
7ae1e1d0 | 2977 | } |
49a18eae | 2978 | |
45264778 | 2979 | /** |
f4b00eab | 2980 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
45264778 VD |
2981 | * @page: page to uncharge |
2982 | * @order: allocation order | |
2983 | */ | |
f4b00eab | 2984 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
7ae1e1d0 | 2985 | { |
b4e0b68f | 2986 | struct obj_cgroup *objcg; |
f3ccb2c4 | 2987 | unsigned int nr_pages = 1 << order; |
7ae1e1d0 | 2988 | |
b4e0b68f | 2989 | if (!PageMemcgKmem(page)) |
7ae1e1d0 GC |
2990 | return; |
2991 | ||
b4e0b68f MS |
2992 | objcg = __page_objcg(page); |
2993 | obj_cgroup_uncharge_pages(objcg, nr_pages); | |
bcfe06bf | 2994 | page->memcg_data = 0; |
b4e0b68f | 2995 | obj_cgroup_put(objcg); |
60d3fd32 | 2996 | } |
bf4f0599 RG |
2997 | |
2998 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
2999 | { | |
3000 | struct memcg_stock_pcp *stock; | |
3001 | unsigned long flags; | |
3002 | bool ret = false; | |
3003 | ||
3004 | local_irq_save(flags); | |
3005 | ||
3006 | stock = this_cpu_ptr(&memcg_stock); | |
3007 | if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { | |
3008 | stock->nr_bytes -= nr_bytes; | |
3009 | ret = true; | |
3010 | } | |
3011 | ||
3012 | local_irq_restore(flags); | |
3013 | ||
3014 | return ret; | |
3015 | } | |
3016 | ||
3017 | static void drain_obj_stock(struct memcg_stock_pcp *stock) | |
3018 | { | |
3019 | struct obj_cgroup *old = stock->cached_objcg; | |
3020 | ||
3021 | if (!old) | |
3022 | return; | |
3023 | ||
3024 | if (stock->nr_bytes) { | |
3025 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; | |
3026 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); | |
3027 | ||
e74d2259 MS |
3028 | if (nr_pages) |
3029 | obj_cgroup_uncharge_pages(old, nr_pages); | |
bf4f0599 RG |
3030 | |
3031 | /* | |
3032 | * The leftover is flushed to the centralized per-memcg value. | |
3033 | * On the next attempt to refill obj stock it will be moved | |
3034 | * to a per-cpu stock (probably, on an other CPU), see | |
3035 | * refill_obj_stock(). | |
3036 | * | |
3037 | * How often it's flushed is a trade-off between the memory | |
3038 | * limit enforcement accuracy and potential CPU contention, | |
3039 | * so it might be changed in the future. | |
3040 | */ | |
3041 | atomic_add(nr_bytes, &old->nr_charged_bytes); | |
3042 | stock->nr_bytes = 0; | |
3043 | } | |
3044 | ||
3045 | obj_cgroup_put(old); | |
3046 | stock->cached_objcg = NULL; | |
3047 | } | |
3048 | ||
3049 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, | |
3050 | struct mem_cgroup *root_memcg) | |
3051 | { | |
3052 | struct mem_cgroup *memcg; | |
3053 | ||
3054 | if (stock->cached_objcg) { | |
3055 | memcg = obj_cgroup_memcg(stock->cached_objcg); | |
3056 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) | |
3057 | return true; | |
3058 | } | |
3059 | ||
3060 | return false; | |
3061 | } | |
3062 | ||
3063 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) | |
3064 | { | |
3065 | struct memcg_stock_pcp *stock; | |
3066 | unsigned long flags; | |
3067 | ||
3068 | local_irq_save(flags); | |
3069 | ||
3070 | stock = this_cpu_ptr(&memcg_stock); | |
3071 | if (stock->cached_objcg != objcg) { /* reset if necessary */ | |
3072 | drain_obj_stock(stock); | |
3073 | obj_cgroup_get(objcg); | |
3074 | stock->cached_objcg = objcg; | |
3075 | stock->nr_bytes = atomic_xchg(&objcg->nr_charged_bytes, 0); | |
3076 | } | |
3077 | stock->nr_bytes += nr_bytes; | |
3078 | ||
3079 | if (stock->nr_bytes > PAGE_SIZE) | |
3080 | drain_obj_stock(stock); | |
3081 | ||
3082 | local_irq_restore(flags); | |
3083 | } | |
3084 | ||
3085 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) | |
3086 | { | |
bf4f0599 RG |
3087 | unsigned int nr_pages, nr_bytes; |
3088 | int ret; | |
3089 | ||
3090 | if (consume_obj_stock(objcg, size)) | |
3091 | return 0; | |
3092 | ||
3093 | /* | |
3094 | * In theory, memcg->nr_charged_bytes can have enough | |
3095 | * pre-charged bytes to satisfy the allocation. However, | |
3096 | * flushing memcg->nr_charged_bytes requires two atomic | |
3097 | * operations, and memcg->nr_charged_bytes can't be big, | |
3098 | * so it's better to ignore it and try grab some new pages. | |
3099 | * memcg->nr_charged_bytes will be flushed in | |
3100 | * refill_obj_stock(), called from this function or | |
3101 | * independently later. | |
3102 | */ | |
bf4f0599 RG |
3103 | nr_pages = size >> PAGE_SHIFT; |
3104 | nr_bytes = size & (PAGE_SIZE - 1); | |
3105 | ||
3106 | if (nr_bytes) | |
3107 | nr_pages += 1; | |
3108 | ||
e74d2259 | 3109 | ret = obj_cgroup_charge_pages(objcg, gfp, nr_pages); |
bf4f0599 RG |
3110 | if (!ret && nr_bytes) |
3111 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes); | |
3112 | ||
bf4f0599 RG |
3113 | return ret; |
3114 | } | |
3115 | ||
3116 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) | |
3117 | { | |
3118 | refill_obj_stock(objcg, size); | |
3119 | } | |
3120 | ||
84c07d11 | 3121 | #endif /* CONFIG_MEMCG_KMEM */ |
7ae1e1d0 | 3122 | |
ca3e0214 | 3123 | /* |
be6c8982 | 3124 | * Because page_memcg(head) is not set on tails, set it now. |
ca3e0214 | 3125 | */ |
be6c8982 | 3126 | void split_page_memcg(struct page *head, unsigned int nr) |
ca3e0214 | 3127 | { |
bcfe06bf | 3128 | struct mem_cgroup *memcg = page_memcg(head); |
e94c8a9c | 3129 | int i; |
ca3e0214 | 3130 | |
be6c8982 | 3131 | if (mem_cgroup_disabled() || !memcg) |
3d37c4a9 | 3132 | return; |
b070e65c | 3133 | |
be6c8982 ZG |
3134 | for (i = 1; i < nr; i++) |
3135 | head[i].memcg_data = head->memcg_data; | |
b4e0b68f MS |
3136 | |
3137 | if (PageMemcgKmem(head)) | |
3138 | obj_cgroup_get_many(__page_objcg(head), nr - 1); | |
3139 | else | |
3140 | css_get_many(&memcg->css, nr - 1); | |
ca3e0214 | 3141 | } |
ca3e0214 | 3142 | |
c255a458 | 3143 | #ifdef CONFIG_MEMCG_SWAP |
02491447 DN |
3144 | /** |
3145 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3146 | * @entry: swap entry to be moved | |
3147 | * @from: mem_cgroup which the entry is moved from | |
3148 | * @to: mem_cgroup which the entry is moved to | |
3149 | * | |
3150 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3151 | * as the mem_cgroup's id of @from. | |
3152 | * | |
3153 | * Returns 0 on success, -EINVAL on failure. | |
3154 | * | |
3e32cb2e | 3155 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
02491447 DN |
3156 | * both res and memsw, and called css_get(). |
3157 | */ | |
3158 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3159 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3160 | { |
3161 | unsigned short old_id, new_id; | |
3162 | ||
34c00c31 LZ |
3163 | old_id = mem_cgroup_id(from); |
3164 | new_id = mem_cgroup_id(to); | |
02491447 DN |
3165 | |
3166 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
c9019e9b JW |
3167 | mod_memcg_state(from, MEMCG_SWAP, -1); |
3168 | mod_memcg_state(to, MEMCG_SWAP, 1); | |
02491447 DN |
3169 | return 0; |
3170 | } | |
3171 | return -EINVAL; | |
3172 | } | |
3173 | #else | |
3174 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
e91cbb42 | 3175 | struct mem_cgroup *from, struct mem_cgroup *to) |
02491447 DN |
3176 | { |
3177 | return -EINVAL; | |
3178 | } | |
8c7c6e34 | 3179 | #endif |
d13d1443 | 3180 | |
bbec2e15 | 3181 | static DEFINE_MUTEX(memcg_max_mutex); |
f212ad7c | 3182 | |
bbec2e15 RG |
3183 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
3184 | unsigned long max, bool memsw) | |
628f4235 | 3185 | { |
3e32cb2e | 3186 | bool enlarge = false; |
bb4a7ea2 | 3187 | bool drained = false; |
3e32cb2e | 3188 | int ret; |
c054a78c YZ |
3189 | bool limits_invariant; |
3190 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; | |
81d39c20 | 3191 | |
3e32cb2e | 3192 | do { |
628f4235 KH |
3193 | if (signal_pending(current)) { |
3194 | ret = -EINTR; | |
3195 | break; | |
3196 | } | |
3e32cb2e | 3197 | |
bbec2e15 | 3198 | mutex_lock(&memcg_max_mutex); |
c054a78c YZ |
3199 | /* |
3200 | * Make sure that the new limit (memsw or memory limit) doesn't | |
bbec2e15 | 3201 | * break our basic invariant rule memory.max <= memsw.max. |
c054a78c | 3202 | */ |
15b42562 | 3203 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
bbec2e15 | 3204 | max <= memcg->memsw.max; |
c054a78c | 3205 | if (!limits_invariant) { |
bbec2e15 | 3206 | mutex_unlock(&memcg_max_mutex); |
8c7c6e34 | 3207 | ret = -EINVAL; |
8c7c6e34 KH |
3208 | break; |
3209 | } | |
bbec2e15 | 3210 | if (max > counter->max) |
3e32cb2e | 3211 | enlarge = true; |
bbec2e15 RG |
3212 | ret = page_counter_set_max(counter, max); |
3213 | mutex_unlock(&memcg_max_mutex); | |
8c7c6e34 KH |
3214 | |
3215 | if (!ret) | |
3216 | break; | |
3217 | ||
bb4a7ea2 SB |
3218 | if (!drained) { |
3219 | drain_all_stock(memcg); | |
3220 | drained = true; | |
3221 | continue; | |
3222 | } | |
3223 | ||
1ab5c056 AR |
3224 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
3225 | GFP_KERNEL, !memsw)) { | |
3226 | ret = -EBUSY; | |
3227 | break; | |
3228 | } | |
3229 | } while (true); | |
3e32cb2e | 3230 | |
3c11ecf4 KH |
3231 | if (!ret && enlarge) |
3232 | memcg_oom_recover(memcg); | |
3e32cb2e | 3233 | |
628f4235 KH |
3234 | return ret; |
3235 | } | |
3236 | ||
ef8f2327 | 3237 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
0608f43d AM |
3238 | gfp_t gfp_mask, |
3239 | unsigned long *total_scanned) | |
3240 | { | |
3241 | unsigned long nr_reclaimed = 0; | |
ef8f2327 | 3242 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
0608f43d AM |
3243 | unsigned long reclaimed; |
3244 | int loop = 0; | |
ef8f2327 | 3245 | struct mem_cgroup_tree_per_node *mctz; |
3e32cb2e | 3246 | unsigned long excess; |
0608f43d AM |
3247 | unsigned long nr_scanned; |
3248 | ||
3249 | if (order > 0) | |
3250 | return 0; | |
3251 | ||
ef8f2327 | 3252 | mctz = soft_limit_tree_node(pgdat->node_id); |
d6507ff5 MH |
3253 | |
3254 | /* | |
3255 | * Do not even bother to check the largest node if the root | |
3256 | * is empty. Do it lockless to prevent lock bouncing. Races | |
3257 | * are acceptable as soft limit is best effort anyway. | |
3258 | */ | |
bfc7228b | 3259 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
d6507ff5 MH |
3260 | return 0; |
3261 | ||
0608f43d AM |
3262 | /* |
3263 | * This loop can run a while, specially if mem_cgroup's continuously | |
3264 | * keep exceeding their soft limit and putting the system under | |
3265 | * pressure | |
3266 | */ | |
3267 | do { | |
3268 | if (next_mz) | |
3269 | mz = next_mz; | |
3270 | else | |
3271 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3272 | if (!mz) | |
3273 | break; | |
3274 | ||
3275 | nr_scanned = 0; | |
ef8f2327 | 3276 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
0608f43d AM |
3277 | gfp_mask, &nr_scanned); |
3278 | nr_reclaimed += reclaimed; | |
3279 | *total_scanned += nr_scanned; | |
0a31bc97 | 3280 | spin_lock_irq(&mctz->lock); |
bc2f2e7f | 3281 | __mem_cgroup_remove_exceeded(mz, mctz); |
0608f43d AM |
3282 | |
3283 | /* | |
3284 | * If we failed to reclaim anything from this memory cgroup | |
3285 | * it is time to move on to the next cgroup | |
3286 | */ | |
3287 | next_mz = NULL; | |
bc2f2e7f VD |
3288 | if (!reclaimed) |
3289 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
3290 | ||
3e32cb2e | 3291 | excess = soft_limit_excess(mz->memcg); |
0608f43d AM |
3292 | /* |
3293 | * One school of thought says that we should not add | |
3294 | * back the node to the tree if reclaim returns 0. | |
3295 | * But our reclaim could return 0, simply because due | |
3296 | * to priority we are exposing a smaller subset of | |
3297 | * memory to reclaim from. Consider this as a longer | |
3298 | * term TODO. | |
3299 | */ | |
3300 | /* If excess == 0, no tree ops */ | |
cf2c8127 | 3301 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
0a31bc97 | 3302 | spin_unlock_irq(&mctz->lock); |
0608f43d AM |
3303 | css_put(&mz->memcg->css); |
3304 | loop++; | |
3305 | /* | |
3306 | * Could not reclaim anything and there are no more | |
3307 | * mem cgroups to try or we seem to be looping without | |
3308 | * reclaiming anything. | |
3309 | */ | |
3310 | if (!nr_reclaimed && | |
3311 | (next_mz == NULL || | |
3312 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3313 | break; | |
3314 | } while (!nr_reclaimed); | |
3315 | if (next_mz) | |
3316 | css_put(&next_mz->memcg->css); | |
3317 | return nr_reclaimed; | |
3318 | } | |
3319 | ||
c26251f9 | 3320 | /* |
51038171 | 3321 | * Reclaims as many pages from the given memcg as possible. |
c26251f9 MH |
3322 | * |
3323 | * Caller is responsible for holding css reference for memcg. | |
3324 | */ | |
3325 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) | |
3326 | { | |
d977aa93 | 3327 | int nr_retries = MAX_RECLAIM_RETRIES; |
c26251f9 | 3328 | |
c1e862c1 KH |
3329 | /* we call try-to-free pages for make this cgroup empty */ |
3330 | lru_add_drain_all(); | |
d12c60f6 JS |
3331 | |
3332 | drain_all_stock(memcg); | |
3333 | ||
f817ed48 | 3334 | /* try to free all pages in this cgroup */ |
3e32cb2e | 3335 | while (nr_retries && page_counter_read(&memcg->memory)) { |
f817ed48 | 3336 | int progress; |
c1e862c1 | 3337 | |
c26251f9 MH |
3338 | if (signal_pending(current)) |
3339 | return -EINTR; | |
3340 | ||
b70a2a21 JW |
3341 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
3342 | GFP_KERNEL, true); | |
c1e862c1 | 3343 | if (!progress) { |
f817ed48 | 3344 | nr_retries--; |
c1e862c1 | 3345 | /* maybe some writeback is necessary */ |
8aa7e847 | 3346 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3347 | } |
f817ed48 KH |
3348 | |
3349 | } | |
ab5196c2 MH |
3350 | |
3351 | return 0; | |
cc847582 KH |
3352 | } |
3353 | ||
6770c64e TH |
3354 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
3355 | char *buf, size_t nbytes, | |
3356 | loff_t off) | |
c1e862c1 | 3357 | { |
6770c64e | 3358 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
c26251f9 | 3359 | |
d8423011 MH |
3360 | if (mem_cgroup_is_root(memcg)) |
3361 | return -EINVAL; | |
6770c64e | 3362 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
c1e862c1 KH |
3363 | } |
3364 | ||
182446d0 TH |
3365 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
3366 | struct cftype *cft) | |
18f59ea7 | 3367 | { |
bef8620c | 3368 | return 1; |
18f59ea7 BS |
3369 | } |
3370 | ||
182446d0 TH |
3371 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
3372 | struct cftype *cft, u64 val) | |
18f59ea7 | 3373 | { |
bef8620c | 3374 | if (val == 1) |
0b8f73e1 | 3375 | return 0; |
567fb435 | 3376 | |
bef8620c RG |
3377 | pr_warn_once("Non-hierarchical mode is deprecated. " |
3378 | "Please report your usecase to linux-mm@kvack.org if you " | |
3379 | "depend on this functionality.\n"); | |
567fb435 | 3380 | |
bef8620c | 3381 | return -EINVAL; |
18f59ea7 BS |
3382 | } |
3383 | ||
6f646156 | 3384 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
ce00a967 | 3385 | { |
42a30035 | 3386 | unsigned long val; |
ce00a967 | 3387 | |
3e32cb2e | 3388 | if (mem_cgroup_is_root(memcg)) { |
2d146aa3 | 3389 | cgroup_rstat_flush(memcg->css.cgroup); |
0d1c2072 | 3390 | val = memcg_page_state(memcg, NR_FILE_PAGES) + |
be5d0a74 | 3391 | memcg_page_state(memcg, NR_ANON_MAPPED); |
42a30035 JW |
3392 | if (swap) |
3393 | val += memcg_page_state(memcg, MEMCG_SWAP); | |
3e32cb2e | 3394 | } else { |
ce00a967 | 3395 | if (!swap) |
3e32cb2e | 3396 | val = page_counter_read(&memcg->memory); |
ce00a967 | 3397 | else |
3e32cb2e | 3398 | val = page_counter_read(&memcg->memsw); |
ce00a967 | 3399 | } |
c12176d3 | 3400 | return val; |
ce00a967 JW |
3401 | } |
3402 | ||
3e32cb2e JW |
3403 | enum { |
3404 | RES_USAGE, | |
3405 | RES_LIMIT, | |
3406 | RES_MAX_USAGE, | |
3407 | RES_FAILCNT, | |
3408 | RES_SOFT_LIMIT, | |
3409 | }; | |
ce00a967 | 3410 | |
791badbd | 3411 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
05b84301 | 3412 | struct cftype *cft) |
8cdea7c0 | 3413 | { |
182446d0 | 3414 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3e32cb2e | 3415 | struct page_counter *counter; |
af36f906 | 3416 | |
3e32cb2e | 3417 | switch (MEMFILE_TYPE(cft->private)) { |
8c7c6e34 | 3418 | case _MEM: |
3e32cb2e JW |
3419 | counter = &memcg->memory; |
3420 | break; | |
8c7c6e34 | 3421 | case _MEMSWAP: |
3e32cb2e JW |
3422 | counter = &memcg->memsw; |
3423 | break; | |
510fc4e1 | 3424 | case _KMEM: |
3e32cb2e | 3425 | counter = &memcg->kmem; |
510fc4e1 | 3426 | break; |
d55f90bf | 3427 | case _TCP: |
0db15298 | 3428 | counter = &memcg->tcpmem; |
d55f90bf | 3429 | break; |
8c7c6e34 KH |
3430 | default: |
3431 | BUG(); | |
8c7c6e34 | 3432 | } |
3e32cb2e JW |
3433 | |
3434 | switch (MEMFILE_ATTR(cft->private)) { | |
3435 | case RES_USAGE: | |
3436 | if (counter == &memcg->memory) | |
c12176d3 | 3437 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
3e32cb2e | 3438 | if (counter == &memcg->memsw) |
c12176d3 | 3439 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
3e32cb2e JW |
3440 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
3441 | case RES_LIMIT: | |
bbec2e15 | 3442 | return (u64)counter->max * PAGE_SIZE; |
3e32cb2e JW |
3443 | case RES_MAX_USAGE: |
3444 | return (u64)counter->watermark * PAGE_SIZE; | |
3445 | case RES_FAILCNT: | |
3446 | return counter->failcnt; | |
3447 | case RES_SOFT_LIMIT: | |
3448 | return (u64)memcg->soft_limit * PAGE_SIZE; | |
3449 | default: | |
3450 | BUG(); | |
3451 | } | |
8cdea7c0 | 3452 | } |
510fc4e1 | 3453 | |
84c07d11 | 3454 | #ifdef CONFIG_MEMCG_KMEM |
567e9ab2 | 3455 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
d6441637 | 3456 | { |
bf4f0599 | 3457 | struct obj_cgroup *objcg; |
d6441637 VD |
3458 | int memcg_id; |
3459 | ||
b313aeee VD |
3460 | if (cgroup_memory_nokmem) |
3461 | return 0; | |
3462 | ||
2a4db7eb | 3463 | BUG_ON(memcg->kmemcg_id >= 0); |
567e9ab2 | 3464 | BUG_ON(memcg->kmem_state); |
d6441637 | 3465 | |
f3bb3043 | 3466 | memcg_id = memcg_alloc_cache_id(); |
0b8f73e1 JW |
3467 | if (memcg_id < 0) |
3468 | return memcg_id; | |
d6441637 | 3469 | |
bf4f0599 RG |
3470 | objcg = obj_cgroup_alloc(); |
3471 | if (!objcg) { | |
3472 | memcg_free_cache_id(memcg_id); | |
3473 | return -ENOMEM; | |
3474 | } | |
3475 | objcg->memcg = memcg; | |
3476 | rcu_assign_pointer(memcg->objcg, objcg); | |
3477 | ||
d648bcc7 RG |
3478 | static_branch_enable(&memcg_kmem_enabled_key); |
3479 | ||
900a38f0 | 3480 | memcg->kmemcg_id = memcg_id; |
567e9ab2 | 3481 | memcg->kmem_state = KMEM_ONLINE; |
0b8f73e1 JW |
3482 | |
3483 | return 0; | |
d6441637 VD |
3484 | } |
3485 | ||
8e0a8912 JW |
3486 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
3487 | { | |
3488 | struct cgroup_subsys_state *css; | |
3489 | struct mem_cgroup *parent, *child; | |
3490 | int kmemcg_id; | |
3491 | ||
3492 | if (memcg->kmem_state != KMEM_ONLINE) | |
3493 | return; | |
9855609b | 3494 | |
8e0a8912 JW |
3495 | memcg->kmem_state = KMEM_ALLOCATED; |
3496 | ||
8e0a8912 JW |
3497 | parent = parent_mem_cgroup(memcg); |
3498 | if (!parent) | |
3499 | parent = root_mem_cgroup; | |
3500 | ||
bf4f0599 | 3501 | memcg_reparent_objcgs(memcg, parent); |
fb2f2b0a RG |
3502 | |
3503 | kmemcg_id = memcg->kmemcg_id; | |
3504 | BUG_ON(kmemcg_id < 0); | |
3505 | ||
8e0a8912 JW |
3506 | /* |
3507 | * Change kmemcg_id of this cgroup and all its descendants to the | |
3508 | * parent's id, and then move all entries from this cgroup's list_lrus | |
3509 | * to ones of the parent. After we have finished, all list_lrus | |
3510 | * corresponding to this cgroup are guaranteed to remain empty. The | |
3511 | * ordering is imposed by list_lru_node->lock taken by | |
3512 | * memcg_drain_all_list_lrus(). | |
3513 | */ | |
3a06bb78 | 3514 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
8e0a8912 JW |
3515 | css_for_each_descendant_pre(css, &memcg->css) { |
3516 | child = mem_cgroup_from_css(css); | |
3517 | BUG_ON(child->kmemcg_id != kmemcg_id); | |
3518 | child->kmemcg_id = parent->kmemcg_id; | |
8e0a8912 | 3519 | } |
3a06bb78 TH |
3520 | rcu_read_unlock(); |
3521 | ||
9bec5c35 | 3522 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
8e0a8912 JW |
3523 | |
3524 | memcg_free_cache_id(kmemcg_id); | |
3525 | } | |
3526 | ||
3527 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3528 | { | |
0b8f73e1 JW |
3529 | /* css_alloc() failed, offlining didn't happen */ |
3530 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) | |
3531 | memcg_offline_kmem(memcg); | |
8e0a8912 | 3532 | } |
d6441637 | 3533 | #else |
0b8f73e1 | 3534 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
127424c8 JW |
3535 | { |
3536 | return 0; | |
3537 | } | |
3538 | static void memcg_offline_kmem(struct mem_cgroup *memcg) | |
3539 | { | |
3540 | } | |
3541 | static void memcg_free_kmem(struct mem_cgroup *memcg) | |
3542 | { | |
3543 | } | |
84c07d11 | 3544 | #endif /* CONFIG_MEMCG_KMEM */ |
127424c8 | 3545 | |
bbec2e15 RG |
3546 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
3547 | unsigned long max) | |
d6441637 | 3548 | { |
b313aeee | 3549 | int ret; |
127424c8 | 3550 | |
bbec2e15 RG |
3551 | mutex_lock(&memcg_max_mutex); |
3552 | ret = page_counter_set_max(&memcg->kmem, max); | |
3553 | mutex_unlock(&memcg_max_mutex); | |
127424c8 | 3554 | return ret; |
d6441637 | 3555 | } |
510fc4e1 | 3556 | |
bbec2e15 | 3557 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
d55f90bf VD |
3558 | { |
3559 | int ret; | |
3560 | ||
bbec2e15 | 3561 | mutex_lock(&memcg_max_mutex); |
d55f90bf | 3562 | |
bbec2e15 | 3563 | ret = page_counter_set_max(&memcg->tcpmem, max); |
d55f90bf VD |
3564 | if (ret) |
3565 | goto out; | |
3566 | ||
0db15298 | 3567 | if (!memcg->tcpmem_active) { |
d55f90bf VD |
3568 | /* |
3569 | * The active flag needs to be written after the static_key | |
3570 | * update. This is what guarantees that the socket activation | |
2d758073 JW |
3571 | * function is the last one to run. See mem_cgroup_sk_alloc() |
3572 | * for details, and note that we don't mark any socket as | |
3573 | * belonging to this memcg until that flag is up. | |
d55f90bf VD |
3574 | * |
3575 | * We need to do this, because static_keys will span multiple | |
3576 | * sites, but we can't control their order. If we mark a socket | |
3577 | * as accounted, but the accounting functions are not patched in | |
3578 | * yet, we'll lose accounting. | |
3579 | * | |
2d758073 | 3580 | * We never race with the readers in mem_cgroup_sk_alloc(), |
d55f90bf VD |
3581 | * because when this value change, the code to process it is not |
3582 | * patched in yet. | |
3583 | */ | |
3584 | static_branch_inc(&memcg_sockets_enabled_key); | |
0db15298 | 3585 | memcg->tcpmem_active = true; |
d55f90bf VD |
3586 | } |
3587 | out: | |
bbec2e15 | 3588 | mutex_unlock(&memcg_max_mutex); |
d55f90bf VD |
3589 | return ret; |
3590 | } | |
d55f90bf | 3591 | |
628f4235 KH |
3592 | /* |
3593 | * The user of this function is... | |
3594 | * RES_LIMIT. | |
3595 | */ | |
451af504 TH |
3596 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
3597 | char *buf, size_t nbytes, loff_t off) | |
8cdea7c0 | 3598 | { |
451af504 | 3599 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3600 | unsigned long nr_pages; |
628f4235 KH |
3601 | int ret; |
3602 | ||
451af504 | 3603 | buf = strstrip(buf); |
650c5e56 | 3604 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
3e32cb2e JW |
3605 | if (ret) |
3606 | return ret; | |
af36f906 | 3607 | |
3e32cb2e | 3608 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
628f4235 | 3609 | case RES_LIMIT: |
4b3bde4c BS |
3610 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3611 | ret = -EINVAL; | |
3612 | break; | |
3613 | } | |
3e32cb2e JW |
3614 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3615 | case _MEM: | |
bbec2e15 | 3616 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
8c7c6e34 | 3617 | break; |
3e32cb2e | 3618 | case _MEMSWAP: |
bbec2e15 | 3619 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
296c81d8 | 3620 | break; |
3e32cb2e | 3621 | case _KMEM: |
0158115f MH |
3622 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " |
3623 | "Please report your usecase to linux-mm@kvack.org if you " | |
3624 | "depend on this functionality.\n"); | |
bbec2e15 | 3625 | ret = memcg_update_kmem_max(memcg, nr_pages); |
3e32cb2e | 3626 | break; |
d55f90bf | 3627 | case _TCP: |
bbec2e15 | 3628 | ret = memcg_update_tcp_max(memcg, nr_pages); |
d55f90bf | 3629 | break; |
3e32cb2e | 3630 | } |
296c81d8 | 3631 | break; |
3e32cb2e JW |
3632 | case RES_SOFT_LIMIT: |
3633 | memcg->soft_limit = nr_pages; | |
3634 | ret = 0; | |
628f4235 KH |
3635 | break; |
3636 | } | |
451af504 | 3637 | return ret ?: nbytes; |
8cdea7c0 BS |
3638 | } |
3639 | ||
6770c64e TH |
3640 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
3641 | size_t nbytes, loff_t off) | |
c84872e1 | 3642 | { |
6770c64e | 3643 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
3e32cb2e | 3644 | struct page_counter *counter; |
c84872e1 | 3645 | |
3e32cb2e JW |
3646 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
3647 | case _MEM: | |
3648 | counter = &memcg->memory; | |
3649 | break; | |
3650 | case _MEMSWAP: | |
3651 | counter = &memcg->memsw; | |
3652 | break; | |
3653 | case _KMEM: | |
3654 | counter = &memcg->kmem; | |
3655 | break; | |
d55f90bf | 3656 | case _TCP: |
0db15298 | 3657 | counter = &memcg->tcpmem; |
d55f90bf | 3658 | break; |
3e32cb2e JW |
3659 | default: |
3660 | BUG(); | |
3661 | } | |
af36f906 | 3662 | |
3e32cb2e | 3663 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
29f2a4da | 3664 | case RES_MAX_USAGE: |
3e32cb2e | 3665 | page_counter_reset_watermark(counter); |
29f2a4da PE |
3666 | break; |
3667 | case RES_FAILCNT: | |
3e32cb2e | 3668 | counter->failcnt = 0; |
29f2a4da | 3669 | break; |
3e32cb2e JW |
3670 | default: |
3671 | BUG(); | |
29f2a4da | 3672 | } |
f64c3f54 | 3673 | |
6770c64e | 3674 | return nbytes; |
c84872e1 PE |
3675 | } |
3676 | ||
182446d0 | 3677 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3678 | struct cftype *cft) |
3679 | { | |
182446d0 | 3680 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
7dc74be0 DN |
3681 | } |
3682 | ||
02491447 | 3683 | #ifdef CONFIG_MMU |
182446d0 | 3684 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
7dc74be0 DN |
3685 | struct cftype *cft, u64 val) |
3686 | { | |
182446d0 | 3687 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
7dc74be0 | 3688 | |
1dfab5ab | 3689 | if (val & ~MOVE_MASK) |
7dc74be0 | 3690 | return -EINVAL; |
ee5e8472 | 3691 | |
7dc74be0 | 3692 | /* |
ee5e8472 GC |
3693 | * No kind of locking is needed in here, because ->can_attach() will |
3694 | * check this value once in the beginning of the process, and then carry | |
3695 | * on with stale data. This means that changes to this value will only | |
3696 | * affect task migrations starting after the change. | |
7dc74be0 | 3697 | */ |
c0ff4b85 | 3698 | memcg->move_charge_at_immigrate = val; |
7dc74be0 DN |
3699 | return 0; |
3700 | } | |
02491447 | 3701 | #else |
182446d0 | 3702 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
02491447 DN |
3703 | struct cftype *cft, u64 val) |
3704 | { | |
3705 | return -ENOSYS; | |
3706 | } | |
3707 | #endif | |
7dc74be0 | 3708 | |
406eb0c9 | 3709 | #ifdef CONFIG_NUMA |
113b7dfd JW |
3710 | |
3711 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) | |
3712 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) | |
3713 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) | |
3714 | ||
3715 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 | 3716 | int nid, unsigned int lru_mask, bool tree) |
113b7dfd | 3717 | { |
867e5e1d | 3718 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
113b7dfd JW |
3719 | unsigned long nr = 0; |
3720 | enum lru_list lru; | |
3721 | ||
3722 | VM_BUG_ON((unsigned)nid >= nr_node_ids); | |
3723 | ||
3724 | for_each_lru(lru) { | |
3725 | if (!(BIT(lru) & lru_mask)) | |
3726 | continue; | |
dd8657b6 SB |
3727 | if (tree) |
3728 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); | |
3729 | else | |
3730 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); | |
113b7dfd JW |
3731 | } |
3732 | return nr; | |
3733 | } | |
3734 | ||
3735 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, | |
dd8657b6 SB |
3736 | unsigned int lru_mask, |
3737 | bool tree) | |
113b7dfd JW |
3738 | { |
3739 | unsigned long nr = 0; | |
3740 | enum lru_list lru; | |
3741 | ||
3742 | for_each_lru(lru) { | |
3743 | if (!(BIT(lru) & lru_mask)) | |
3744 | continue; | |
dd8657b6 SB |
3745 | if (tree) |
3746 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); | |
3747 | else | |
3748 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); | |
113b7dfd JW |
3749 | } |
3750 | return nr; | |
3751 | } | |
3752 | ||
2da8ca82 | 3753 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
406eb0c9 | 3754 | { |
25485de6 GT |
3755 | struct numa_stat { |
3756 | const char *name; | |
3757 | unsigned int lru_mask; | |
3758 | }; | |
3759 | ||
3760 | static const struct numa_stat stats[] = { | |
3761 | { "total", LRU_ALL }, | |
3762 | { "file", LRU_ALL_FILE }, | |
3763 | { "anon", LRU_ALL_ANON }, | |
3764 | { "unevictable", BIT(LRU_UNEVICTABLE) }, | |
3765 | }; | |
3766 | const struct numa_stat *stat; | |
406eb0c9 | 3767 | int nid; |
aa9694bb | 3768 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
406eb0c9 | 3769 | |
2d146aa3 JW |
3770 | cgroup_rstat_flush(memcg->css.cgroup); |
3771 | ||
25485de6 | 3772 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3773 | seq_printf(m, "%s=%lu", stat->name, |
3774 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3775 | false)); | |
3776 | for_each_node_state(nid, N_MEMORY) | |
3777 | seq_printf(m, " N%d=%lu", nid, | |
3778 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3779 | stat->lru_mask, false)); | |
25485de6 | 3780 | seq_putc(m, '\n'); |
406eb0c9 | 3781 | } |
406eb0c9 | 3782 | |
071aee13 | 3783 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
dd8657b6 SB |
3784 | |
3785 | seq_printf(m, "hierarchical_%s=%lu", stat->name, | |
3786 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, | |
3787 | true)); | |
3788 | for_each_node_state(nid, N_MEMORY) | |
3789 | seq_printf(m, " N%d=%lu", nid, | |
3790 | mem_cgroup_node_nr_lru_pages(memcg, nid, | |
3791 | stat->lru_mask, true)); | |
071aee13 | 3792 | seq_putc(m, '\n'); |
406eb0c9 | 3793 | } |
406eb0c9 | 3794 | |
406eb0c9 YH |
3795 | return 0; |
3796 | } | |
3797 | #endif /* CONFIG_NUMA */ | |
3798 | ||
c8713d0b | 3799 | static const unsigned int memcg1_stats[] = { |
0d1c2072 | 3800 | NR_FILE_PAGES, |
be5d0a74 | 3801 | NR_ANON_MAPPED, |
468c3982 JW |
3802 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
3803 | NR_ANON_THPS, | |
3804 | #endif | |
c8713d0b JW |
3805 | NR_SHMEM, |
3806 | NR_FILE_MAPPED, | |
3807 | NR_FILE_DIRTY, | |
3808 | NR_WRITEBACK, | |
3809 | MEMCG_SWAP, | |
3810 | }; | |
3811 | ||
3812 | static const char *const memcg1_stat_names[] = { | |
3813 | "cache", | |
3814 | "rss", | |
468c3982 | 3815 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
c8713d0b | 3816 | "rss_huge", |
468c3982 | 3817 | #endif |
c8713d0b JW |
3818 | "shmem", |
3819 | "mapped_file", | |
3820 | "dirty", | |
3821 | "writeback", | |
3822 | "swap", | |
3823 | }; | |
3824 | ||
df0e53d0 | 3825 | /* Universal VM events cgroup1 shows, original sort order */ |
8dd53fd3 | 3826 | static const unsigned int memcg1_events[] = { |
df0e53d0 JW |
3827 | PGPGIN, |
3828 | PGPGOUT, | |
3829 | PGFAULT, | |
3830 | PGMAJFAULT, | |
3831 | }; | |
3832 | ||
2da8ca82 | 3833 | static int memcg_stat_show(struct seq_file *m, void *v) |
d2ceb9b7 | 3834 | { |
aa9694bb | 3835 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3e32cb2e | 3836 | unsigned long memory, memsw; |
af7c4b0e JW |
3837 | struct mem_cgroup *mi; |
3838 | unsigned int i; | |
406eb0c9 | 3839 | |
71cd3113 | 3840 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
70bc068c | 3841 | |
2d146aa3 JW |
3842 | cgroup_rstat_flush(memcg->css.cgroup); |
3843 | ||
71cd3113 | 3844 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
468c3982 JW |
3845 | unsigned long nr; |
3846 | ||
71cd3113 | 3847 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3848 | continue; |
468c3982 | 3849 | nr = memcg_page_state_local(memcg, memcg1_stats[i]); |
468c3982 | 3850 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE); |
1dd3a273 | 3851 | } |
7b854121 | 3852 | |
df0e53d0 | 3853 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d | 3854 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
205b20cc | 3855 | memcg_events_local(memcg, memcg1_events[i])); |
af7c4b0e JW |
3856 | |
3857 | for (i = 0; i < NR_LRU_LISTS; i++) | |
ebc5d83d | 3858 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
205b20cc | 3859 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
21d89d15 | 3860 | PAGE_SIZE); |
af7c4b0e | 3861 | |
14067bb3 | 3862 | /* Hierarchical information */ |
3e32cb2e JW |
3863 | memory = memsw = PAGE_COUNTER_MAX; |
3864 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { | |
15b42562 CD |
3865 | memory = min(memory, READ_ONCE(mi->memory.max)); |
3866 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); | |
fee7b548 | 3867 | } |
3e32cb2e JW |
3868 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
3869 | (u64)memory * PAGE_SIZE); | |
7941d214 | 3870 | if (do_memsw_account()) |
3e32cb2e JW |
3871 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
3872 | (u64)memsw * PAGE_SIZE); | |
7f016ee8 | 3873 | |
8de7ecc6 | 3874 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
7de2e9f1 | 3875 | unsigned long nr; |
3876 | ||
71cd3113 | 3877 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
1dd3a273 | 3878 | continue; |
7de2e9f1 | 3879 | nr = memcg_page_state(memcg, memcg1_stats[i]); |
8de7ecc6 | 3880 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
7de2e9f1 | 3881 | (u64)nr * PAGE_SIZE); |
af7c4b0e JW |
3882 | } |
3883 | ||
8de7ecc6 | 3884 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
ebc5d83d KK |
3885 | seq_printf(m, "total_%s %llu\n", |
3886 | vm_event_name(memcg1_events[i]), | |
dd923990 | 3887 | (u64)memcg_events(memcg, memcg1_events[i])); |
af7c4b0e | 3888 | |
8de7ecc6 | 3889 | for (i = 0; i < NR_LRU_LISTS; i++) |
ebc5d83d | 3890 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
42a30035 JW |
3891 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
3892 | PAGE_SIZE); | |
14067bb3 | 3893 | |
7f016ee8 | 3894 | #ifdef CONFIG_DEBUG_VM |
7f016ee8 | 3895 | { |
ef8f2327 MG |
3896 | pg_data_t *pgdat; |
3897 | struct mem_cgroup_per_node *mz; | |
1431d4d1 JW |
3898 | unsigned long anon_cost = 0; |
3899 | unsigned long file_cost = 0; | |
7f016ee8 | 3900 | |
ef8f2327 | 3901 | for_each_online_pgdat(pgdat) { |
a3747b53 | 3902 | mz = memcg->nodeinfo[pgdat->node_id]; |
7f016ee8 | 3903 | |
1431d4d1 JW |
3904 | anon_cost += mz->lruvec.anon_cost; |
3905 | file_cost += mz->lruvec.file_cost; | |
ef8f2327 | 3906 | } |
1431d4d1 JW |
3907 | seq_printf(m, "anon_cost %lu\n", anon_cost); |
3908 | seq_printf(m, "file_cost %lu\n", file_cost); | |
7f016ee8 KM |
3909 | } |
3910 | #endif | |
3911 | ||
d2ceb9b7 KH |
3912 | return 0; |
3913 | } | |
3914 | ||
182446d0 TH |
3915 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
3916 | struct cftype *cft) | |
a7885eb8 | 3917 | { |
182446d0 | 3918 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3919 | |
1f4c025b | 3920 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
3921 | } |
3922 | ||
182446d0 TH |
3923 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
3924 | struct cftype *cft, u64 val) | |
a7885eb8 | 3925 | { |
182446d0 | 3926 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
a7885eb8 | 3927 | |
3dae7fec | 3928 | if (val > 100) |
a7885eb8 KM |
3929 | return -EINVAL; |
3930 | ||
a4792030 | 3931 | if (!mem_cgroup_is_root(memcg)) |
3dae7fec JW |
3932 | memcg->swappiness = val; |
3933 | else | |
3934 | vm_swappiness = val; | |
068b38c1 | 3935 | |
a7885eb8 KM |
3936 | return 0; |
3937 | } | |
3938 | ||
2e72b634 KS |
3939 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3940 | { | |
3941 | struct mem_cgroup_threshold_ary *t; | |
3e32cb2e | 3942 | unsigned long usage; |
2e72b634 KS |
3943 | int i; |
3944 | ||
3945 | rcu_read_lock(); | |
3946 | if (!swap) | |
2c488db2 | 3947 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3948 | else |
2c488db2 | 3949 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3950 | |
3951 | if (!t) | |
3952 | goto unlock; | |
3953 | ||
ce00a967 | 3954 | usage = mem_cgroup_usage(memcg, swap); |
2e72b634 KS |
3955 | |
3956 | /* | |
748dad36 | 3957 | * current_threshold points to threshold just below or equal to usage. |
2e72b634 KS |
3958 | * If it's not true, a threshold was crossed after last |
3959 | * call of __mem_cgroup_threshold(). | |
3960 | */ | |
5407a562 | 3961 | i = t->current_threshold; |
2e72b634 KS |
3962 | |
3963 | /* | |
3964 | * Iterate backward over array of thresholds starting from | |
3965 | * current_threshold and check if a threshold is crossed. | |
3966 | * If none of thresholds below usage is crossed, we read | |
3967 | * only one element of the array here. | |
3968 | */ | |
3969 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3970 | eventfd_signal(t->entries[i].eventfd, 1); | |
3971 | ||
3972 | /* i = current_threshold + 1 */ | |
3973 | i++; | |
3974 | ||
3975 | /* | |
3976 | * Iterate forward over array of thresholds starting from | |
3977 | * current_threshold+1 and check if a threshold is crossed. | |
3978 | * If none of thresholds above usage is crossed, we read | |
3979 | * only one element of the array here. | |
3980 | */ | |
3981 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3982 | eventfd_signal(t->entries[i].eventfd, 1); | |
3983 | ||
3984 | /* Update current_threshold */ | |
5407a562 | 3985 | t->current_threshold = i - 1; |
2e72b634 KS |
3986 | unlock: |
3987 | rcu_read_unlock(); | |
3988 | } | |
3989 | ||
3990 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3991 | { | |
ad4ca5f4 KS |
3992 | while (memcg) { |
3993 | __mem_cgroup_threshold(memcg, false); | |
7941d214 | 3994 | if (do_memsw_account()) |
ad4ca5f4 KS |
3995 | __mem_cgroup_threshold(memcg, true); |
3996 | ||
3997 | memcg = parent_mem_cgroup(memcg); | |
3998 | } | |
2e72b634 KS |
3999 | } |
4000 | ||
4001 | static int compare_thresholds(const void *a, const void *b) | |
4002 | { | |
4003 | const struct mem_cgroup_threshold *_a = a; | |
4004 | const struct mem_cgroup_threshold *_b = b; | |
4005 | ||
2bff24a3 GT |
4006 | if (_a->threshold > _b->threshold) |
4007 | return 1; | |
4008 | ||
4009 | if (_a->threshold < _b->threshold) | |
4010 | return -1; | |
4011 | ||
4012 | return 0; | |
2e72b634 KS |
4013 | } |
4014 | ||
c0ff4b85 | 4015 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
9490ff27 KH |
4016 | { |
4017 | struct mem_cgroup_eventfd_list *ev; | |
4018 | ||
2bcf2e92 MH |
4019 | spin_lock(&memcg_oom_lock); |
4020 | ||
c0ff4b85 | 4021 | list_for_each_entry(ev, &memcg->oom_notify, list) |
9490ff27 | 4022 | eventfd_signal(ev->eventfd, 1); |
2bcf2e92 MH |
4023 | |
4024 | spin_unlock(&memcg_oom_lock); | |
9490ff27 KH |
4025 | return 0; |
4026 | } | |
4027 | ||
c0ff4b85 | 4028 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
9490ff27 | 4029 | { |
7d74b06f KH |
4030 | struct mem_cgroup *iter; |
4031 | ||
c0ff4b85 | 4032 | for_each_mem_cgroup_tree(iter, memcg) |
7d74b06f | 4033 | mem_cgroup_oom_notify_cb(iter); |
9490ff27 KH |
4034 | } |
4035 | ||
59b6f873 | 4036 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4037 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
2e72b634 | 4038 | { |
2c488db2 KS |
4039 | struct mem_cgroup_thresholds *thresholds; |
4040 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e JW |
4041 | unsigned long threshold; |
4042 | unsigned long usage; | |
2c488db2 | 4043 | int i, size, ret; |
2e72b634 | 4044 | |
650c5e56 | 4045 | ret = page_counter_memparse(args, "-1", &threshold); |
2e72b634 KS |
4046 | if (ret) |
4047 | return ret; | |
4048 | ||
4049 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4050 | |
05b84301 | 4051 | if (type == _MEM) { |
2c488db2 | 4052 | thresholds = &memcg->thresholds; |
ce00a967 | 4053 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4054 | } else if (type == _MEMSWAP) { |
2c488db2 | 4055 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4056 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4057 | } else |
2e72b634 KS |
4058 | BUG(); |
4059 | ||
2e72b634 | 4060 | /* Check if a threshold crossed before adding a new one */ |
2c488db2 | 4061 | if (thresholds->primary) |
2e72b634 KS |
4062 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4063 | ||
2c488db2 | 4064 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4065 | |
4066 | /* Allocate memory for new array of thresholds */ | |
67b8046f | 4067 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
2c488db2 | 4068 | if (!new) { |
2e72b634 KS |
4069 | ret = -ENOMEM; |
4070 | goto unlock; | |
4071 | } | |
2c488db2 | 4072 | new->size = size; |
2e72b634 KS |
4073 | |
4074 | /* Copy thresholds (if any) to new array */ | |
e90342e6 GS |
4075 | if (thresholds->primary) |
4076 | memcpy(new->entries, thresholds->primary->entries, | |
4077 | flex_array_size(new, entries, size - 1)); | |
2c488db2 | 4078 | |
2e72b634 | 4079 | /* Add new threshold */ |
2c488db2 KS |
4080 | new->entries[size - 1].eventfd = eventfd; |
4081 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4082 | |
4083 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
61e604e6 | 4084 | sort(new->entries, size, sizeof(*new->entries), |
2e72b634 KS |
4085 | compare_thresholds, NULL); |
4086 | ||
4087 | /* Find current threshold */ | |
2c488db2 | 4088 | new->current_threshold = -1; |
2e72b634 | 4089 | for (i = 0; i < size; i++) { |
748dad36 | 4090 | if (new->entries[i].threshold <= usage) { |
2e72b634 | 4091 | /* |
2c488db2 KS |
4092 | * new->current_threshold will not be used until |
4093 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4094 | * it here. |
4095 | */ | |
2c488db2 | 4096 | ++new->current_threshold; |
748dad36 SZ |
4097 | } else |
4098 | break; | |
2e72b634 KS |
4099 | } |
4100 | ||
2c488db2 KS |
4101 | /* Free old spare buffer and save old primary buffer as spare */ |
4102 | kfree(thresholds->spare); | |
4103 | thresholds->spare = thresholds->primary; | |
4104 | ||
4105 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4106 | |
907860ed | 4107 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4108 | synchronize_rcu(); |
4109 | ||
2e72b634 KS |
4110 | unlock: |
4111 | mutex_unlock(&memcg->thresholds_lock); | |
4112 | ||
4113 | return ret; | |
4114 | } | |
4115 | ||
59b6f873 | 4116 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4117 | struct eventfd_ctx *eventfd, const char *args) |
4118 | { | |
59b6f873 | 4119 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
347c4a87 TH |
4120 | } |
4121 | ||
59b6f873 | 4122 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4123 | struct eventfd_ctx *eventfd, const char *args) |
4124 | { | |
59b6f873 | 4125 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
347c4a87 TH |
4126 | } |
4127 | ||
59b6f873 | 4128 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4129 | struct eventfd_ctx *eventfd, enum res_type type) |
2e72b634 | 4130 | { |
2c488db2 KS |
4131 | struct mem_cgroup_thresholds *thresholds; |
4132 | struct mem_cgroup_threshold_ary *new; | |
3e32cb2e | 4133 | unsigned long usage; |
7d36665a | 4134 | int i, j, size, entries; |
2e72b634 KS |
4135 | |
4136 | mutex_lock(&memcg->thresholds_lock); | |
05b84301 JW |
4137 | |
4138 | if (type == _MEM) { | |
2c488db2 | 4139 | thresholds = &memcg->thresholds; |
ce00a967 | 4140 | usage = mem_cgroup_usage(memcg, false); |
05b84301 | 4141 | } else if (type == _MEMSWAP) { |
2c488db2 | 4142 | thresholds = &memcg->memsw_thresholds; |
ce00a967 | 4143 | usage = mem_cgroup_usage(memcg, true); |
05b84301 | 4144 | } else |
2e72b634 KS |
4145 | BUG(); |
4146 | ||
371528ca AV |
4147 | if (!thresholds->primary) |
4148 | goto unlock; | |
4149 | ||
2e72b634 KS |
4150 | /* Check if a threshold crossed before removing */ |
4151 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4152 | ||
4153 | /* Calculate new number of threshold */ | |
7d36665a | 4154 | size = entries = 0; |
2c488db2 KS |
4155 | for (i = 0; i < thresholds->primary->size; i++) { |
4156 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 | 4157 | size++; |
7d36665a CX |
4158 | else |
4159 | entries++; | |
2e72b634 KS |
4160 | } |
4161 | ||
2c488db2 | 4162 | new = thresholds->spare; |
907860ed | 4163 | |
7d36665a CX |
4164 | /* If no items related to eventfd have been cleared, nothing to do */ |
4165 | if (!entries) | |
4166 | goto unlock; | |
4167 | ||
2e72b634 KS |
4168 | /* Set thresholds array to NULL if we don't have thresholds */ |
4169 | if (!size) { | |
2c488db2 KS |
4170 | kfree(new); |
4171 | new = NULL; | |
907860ed | 4172 | goto swap_buffers; |
2e72b634 KS |
4173 | } |
4174 | ||
2c488db2 | 4175 | new->size = size; |
2e72b634 KS |
4176 | |
4177 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4178 | new->current_threshold = -1; |
4179 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4180 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4181 | continue; |
4182 | ||
2c488db2 | 4183 | new->entries[j] = thresholds->primary->entries[i]; |
748dad36 | 4184 | if (new->entries[j].threshold <= usage) { |
2e72b634 | 4185 | /* |
2c488db2 | 4186 | * new->current_threshold will not be used |
2e72b634 KS |
4187 | * until rcu_assign_pointer(), so it's safe to increment |
4188 | * it here. | |
4189 | */ | |
2c488db2 | 4190 | ++new->current_threshold; |
2e72b634 KS |
4191 | } |
4192 | j++; | |
4193 | } | |
4194 | ||
907860ed | 4195 | swap_buffers: |
2c488db2 KS |
4196 | /* Swap primary and spare array */ |
4197 | thresholds->spare = thresholds->primary; | |
8c757763 | 4198 | |
2c488db2 | 4199 | rcu_assign_pointer(thresholds->primary, new); |
2e72b634 | 4200 | |
907860ed | 4201 | /* To be sure that nobody uses thresholds */ |
2e72b634 | 4202 | synchronize_rcu(); |
6611d8d7 MC |
4203 | |
4204 | /* If all events are unregistered, free the spare array */ | |
4205 | if (!new) { | |
4206 | kfree(thresholds->spare); | |
4207 | thresholds->spare = NULL; | |
4208 | } | |
371528ca | 4209 | unlock: |
2e72b634 | 4210 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4211 | } |
c1e862c1 | 4212 | |
59b6f873 | 4213 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4214 | struct eventfd_ctx *eventfd) |
4215 | { | |
59b6f873 | 4216 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
347c4a87 TH |
4217 | } |
4218 | ||
59b6f873 | 4219 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
347c4a87 TH |
4220 | struct eventfd_ctx *eventfd) |
4221 | { | |
59b6f873 | 4222 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
347c4a87 TH |
4223 | } |
4224 | ||
59b6f873 | 4225 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
347c4a87 | 4226 | struct eventfd_ctx *eventfd, const char *args) |
9490ff27 | 4227 | { |
9490ff27 | 4228 | struct mem_cgroup_eventfd_list *event; |
9490ff27 | 4229 | |
9490ff27 KH |
4230 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
4231 | if (!event) | |
4232 | return -ENOMEM; | |
4233 | ||
1af8efe9 | 4234 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4235 | |
4236 | event->eventfd = eventfd; | |
4237 | list_add(&event->list, &memcg->oom_notify); | |
4238 | ||
4239 | /* already in OOM ? */ | |
c2b42d3c | 4240 | if (memcg->under_oom) |
9490ff27 | 4241 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4242 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4243 | |
4244 | return 0; | |
4245 | } | |
4246 | ||
59b6f873 | 4247 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
347c4a87 | 4248 | struct eventfd_ctx *eventfd) |
9490ff27 | 4249 | { |
9490ff27 | 4250 | struct mem_cgroup_eventfd_list *ev, *tmp; |
9490ff27 | 4251 | |
1af8efe9 | 4252 | spin_lock(&memcg_oom_lock); |
9490ff27 | 4253 | |
c0ff4b85 | 4254 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
9490ff27 KH |
4255 | if (ev->eventfd == eventfd) { |
4256 | list_del(&ev->list); | |
4257 | kfree(ev); | |
4258 | } | |
4259 | } | |
4260 | ||
1af8efe9 | 4261 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4262 | } |
4263 | ||
2da8ca82 | 4264 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
3c11ecf4 | 4265 | { |
aa9694bb | 4266 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
3c11ecf4 | 4267 | |
791badbd | 4268 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
c2b42d3c | 4269 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
fe6bdfc8 RG |
4270 | seq_printf(sf, "oom_kill %lu\n", |
4271 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); | |
3c11ecf4 KH |
4272 | return 0; |
4273 | } | |
4274 | ||
182446d0 | 4275 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
3c11ecf4 KH |
4276 | struct cftype *cft, u64 val) |
4277 | { | |
182446d0 | 4278 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3c11ecf4 KH |
4279 | |
4280 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
a4792030 | 4281 | if (mem_cgroup_is_root(memcg) || !((val == 0) || (val == 1))) |
3c11ecf4 KH |
4282 | return -EINVAL; |
4283 | ||
c0ff4b85 | 4284 | memcg->oom_kill_disable = val; |
4d845ebf | 4285 | if (!val) |
c0ff4b85 | 4286 | memcg_oom_recover(memcg); |
3dae7fec | 4287 | |
3c11ecf4 KH |
4288 | return 0; |
4289 | } | |
4290 | ||
52ebea74 TH |
4291 | #ifdef CONFIG_CGROUP_WRITEBACK |
4292 | ||
3a8e9ac8 TH |
4293 | #include <trace/events/writeback.h> |
4294 | ||
841710aa TH |
4295 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
4296 | { | |
4297 | return wb_domain_init(&memcg->cgwb_domain, gfp); | |
4298 | } | |
4299 | ||
4300 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4301 | { | |
4302 | wb_domain_exit(&memcg->cgwb_domain); | |
4303 | } | |
4304 | ||
2529bb3a TH |
4305 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4306 | { | |
4307 | wb_domain_size_changed(&memcg->cgwb_domain); | |
4308 | } | |
4309 | ||
841710aa TH |
4310 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
4311 | { | |
4312 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4313 | ||
4314 | if (!memcg->css.parent) | |
4315 | return NULL; | |
4316 | ||
4317 | return &memcg->cgwb_domain; | |
4318 | } | |
4319 | ||
c2aa723a TH |
4320 | /** |
4321 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg | |
4322 | * @wb: bdi_writeback in question | |
c5edf9cd TH |
4323 | * @pfilepages: out parameter for number of file pages |
4324 | * @pheadroom: out parameter for number of allocatable pages according to memcg | |
c2aa723a TH |
4325 | * @pdirty: out parameter for number of dirty pages |
4326 | * @pwriteback: out parameter for number of pages under writeback | |
4327 | * | |
c5edf9cd TH |
4328 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
4329 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom | |
4330 | * is a bit more involved. | |
c2aa723a | 4331 | * |
c5edf9cd TH |
4332 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
4333 | * headroom is calculated as the lowest headroom of itself and the | |
4334 | * ancestors. Note that this doesn't consider the actual amount of | |
4335 | * available memory in the system. The caller should further cap | |
4336 | * *@pheadroom accordingly. | |
c2aa723a | 4337 | */ |
c5edf9cd TH |
4338 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
4339 | unsigned long *pheadroom, unsigned long *pdirty, | |
4340 | unsigned long *pwriteback) | |
c2aa723a TH |
4341 | { |
4342 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4343 | struct mem_cgroup *parent; | |
c2aa723a | 4344 | |
2d146aa3 | 4345 | cgroup_rstat_flush_irqsafe(memcg->css.cgroup); |
c2aa723a | 4346 | |
2d146aa3 JW |
4347 | *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY); |
4348 | *pwriteback = memcg_page_state(memcg, NR_WRITEBACK); | |
4349 | *pfilepages = memcg_page_state(memcg, NR_INACTIVE_FILE) + | |
4350 | memcg_page_state(memcg, NR_ACTIVE_FILE); | |
c2aa723a | 4351 | |
2d146aa3 | 4352 | *pheadroom = PAGE_COUNTER_MAX; |
c2aa723a | 4353 | while ((parent = parent_mem_cgroup(memcg))) { |
15b42562 | 4354 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
d1663a90 | 4355 | READ_ONCE(memcg->memory.high)); |
c2aa723a TH |
4356 | unsigned long used = page_counter_read(&memcg->memory); |
4357 | ||
c5edf9cd | 4358 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
c2aa723a TH |
4359 | memcg = parent; |
4360 | } | |
c2aa723a TH |
4361 | } |
4362 | ||
97b27821 TH |
4363 | /* |
4364 | * Foreign dirty flushing | |
4365 | * | |
4366 | * There's an inherent mismatch between memcg and writeback. The former | |
4367 | * trackes ownership per-page while the latter per-inode. This was a | |
4368 | * deliberate design decision because honoring per-page ownership in the | |
4369 | * writeback path is complicated, may lead to higher CPU and IO overheads | |
4370 | * and deemed unnecessary given that write-sharing an inode across | |
4371 | * different cgroups isn't a common use-case. | |
4372 | * | |
4373 | * Combined with inode majority-writer ownership switching, this works well | |
4374 | * enough in most cases but there are some pathological cases. For | |
4375 | * example, let's say there are two cgroups A and B which keep writing to | |
4376 | * different but confined parts of the same inode. B owns the inode and | |
4377 | * A's memory is limited far below B's. A's dirty ratio can rise enough to | |
4378 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid | |
4379 | * triggering background writeback. A will be slowed down without a way to | |
4380 | * make writeback of the dirty pages happen. | |
4381 | * | |
4382 | * Conditions like the above can lead to a cgroup getting repatedly and | |
4383 | * severely throttled after making some progress after each | |
4384 | * dirty_expire_interval while the underyling IO device is almost | |
4385 | * completely idle. | |
4386 | * | |
4387 | * Solving this problem completely requires matching the ownership tracking | |
4388 | * granularities between memcg and writeback in either direction. However, | |
4389 | * the more egregious behaviors can be avoided by simply remembering the | |
4390 | * most recent foreign dirtying events and initiating remote flushes on | |
4391 | * them when local writeback isn't enough to keep the memory clean enough. | |
4392 | * | |
4393 | * The following two functions implement such mechanism. When a foreign | |
4394 | * page - a page whose memcg and writeback ownerships don't match - is | |
4395 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning | |
4396 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() | |
4397 | * decides that the memcg needs to sleep due to high dirty ratio, it calls | |
4398 | * mem_cgroup_flush_foreign() which queues writeback on the recorded | |
4399 | * foreign bdi_writebacks which haven't expired. Both the numbers of | |
4400 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are | |
4401 | * limited to MEMCG_CGWB_FRN_CNT. | |
4402 | * | |
4403 | * The mechanism only remembers IDs and doesn't hold any object references. | |
4404 | * As being wrong occasionally doesn't matter, updates and accesses to the | |
4405 | * records are lockless and racy. | |
4406 | */ | |
4407 | void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, | |
4408 | struct bdi_writeback *wb) | |
4409 | { | |
bcfe06bf | 4410 | struct mem_cgroup *memcg = page_memcg(page); |
97b27821 TH |
4411 | struct memcg_cgwb_frn *frn; |
4412 | u64 now = get_jiffies_64(); | |
4413 | u64 oldest_at = now; | |
4414 | int oldest = -1; | |
4415 | int i; | |
4416 | ||
3a8e9ac8 TH |
4417 | trace_track_foreign_dirty(page, wb); |
4418 | ||
97b27821 TH |
4419 | /* |
4420 | * Pick the slot to use. If there is already a slot for @wb, keep | |
4421 | * using it. If not replace the oldest one which isn't being | |
4422 | * written out. | |
4423 | */ | |
4424 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4425 | frn = &memcg->cgwb_frn[i]; | |
4426 | if (frn->bdi_id == wb->bdi->id && | |
4427 | frn->memcg_id == wb->memcg_css->id) | |
4428 | break; | |
4429 | if (time_before64(frn->at, oldest_at) && | |
4430 | atomic_read(&frn->done.cnt) == 1) { | |
4431 | oldest = i; | |
4432 | oldest_at = frn->at; | |
4433 | } | |
4434 | } | |
4435 | ||
4436 | if (i < MEMCG_CGWB_FRN_CNT) { | |
4437 | /* | |
4438 | * Re-using an existing one. Update timestamp lazily to | |
4439 | * avoid making the cacheline hot. We want them to be | |
4440 | * reasonably up-to-date and significantly shorter than | |
4441 | * dirty_expire_interval as that's what expires the record. | |
4442 | * Use the shorter of 1s and dirty_expire_interval / 8. | |
4443 | */ | |
4444 | unsigned long update_intv = | |
4445 | min_t(unsigned long, HZ, | |
4446 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); | |
4447 | ||
4448 | if (time_before64(frn->at, now - update_intv)) | |
4449 | frn->at = now; | |
4450 | } else if (oldest >= 0) { | |
4451 | /* replace the oldest free one */ | |
4452 | frn = &memcg->cgwb_frn[oldest]; | |
4453 | frn->bdi_id = wb->bdi->id; | |
4454 | frn->memcg_id = wb->memcg_css->id; | |
4455 | frn->at = now; | |
4456 | } | |
4457 | } | |
4458 | ||
4459 | /* issue foreign writeback flushes for recorded foreign dirtying events */ | |
4460 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) | |
4461 | { | |
4462 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); | |
4463 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); | |
4464 | u64 now = jiffies_64; | |
4465 | int i; | |
4466 | ||
4467 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { | |
4468 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; | |
4469 | ||
4470 | /* | |
4471 | * If the record is older than dirty_expire_interval, | |
4472 | * writeback on it has already started. No need to kick it | |
4473 | * off again. Also, don't start a new one if there's | |
4474 | * already one in flight. | |
4475 | */ | |
4476 | if (time_after64(frn->at, now - intv) && | |
4477 | atomic_read(&frn->done.cnt) == 1) { | |
4478 | frn->at = 0; | |
3a8e9ac8 | 4479 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
97b27821 TH |
4480 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0, |
4481 | WB_REASON_FOREIGN_FLUSH, | |
4482 | &frn->done); | |
4483 | } | |
4484 | } | |
4485 | } | |
4486 | ||
841710aa TH |
4487 | #else /* CONFIG_CGROUP_WRITEBACK */ |
4488 | ||
4489 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) | |
4490 | { | |
4491 | return 0; | |
4492 | } | |
4493 | ||
4494 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) | |
4495 | { | |
4496 | } | |
4497 | ||
2529bb3a TH |
4498 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
4499 | { | |
4500 | } | |
4501 | ||
52ebea74 TH |
4502 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
4503 | ||
3bc942f3 TH |
4504 | /* |
4505 | * DO NOT USE IN NEW FILES. | |
4506 | * | |
4507 | * "cgroup.event_control" implementation. | |
4508 | * | |
4509 | * This is way over-engineered. It tries to support fully configurable | |
4510 | * events for each user. Such level of flexibility is completely | |
4511 | * unnecessary especially in the light of the planned unified hierarchy. | |
4512 | * | |
4513 | * Please deprecate this and replace with something simpler if at all | |
4514 | * possible. | |
4515 | */ | |
4516 | ||
79bd9814 TH |
4517 | /* |
4518 | * Unregister event and free resources. | |
4519 | * | |
4520 | * Gets called from workqueue. | |
4521 | */ | |
3bc942f3 | 4522 | static void memcg_event_remove(struct work_struct *work) |
79bd9814 | 4523 | { |
3bc942f3 TH |
4524 | struct mem_cgroup_event *event = |
4525 | container_of(work, struct mem_cgroup_event, remove); | |
59b6f873 | 4526 | struct mem_cgroup *memcg = event->memcg; |
79bd9814 TH |
4527 | |
4528 | remove_wait_queue(event->wqh, &event->wait); | |
4529 | ||
59b6f873 | 4530 | event->unregister_event(memcg, event->eventfd); |
79bd9814 TH |
4531 | |
4532 | /* Notify userspace the event is going away. */ | |
4533 | eventfd_signal(event->eventfd, 1); | |
4534 | ||
4535 | eventfd_ctx_put(event->eventfd); | |
4536 | kfree(event); | |
59b6f873 | 4537 | css_put(&memcg->css); |
79bd9814 TH |
4538 | } |
4539 | ||
4540 | /* | |
a9a08845 | 4541 | * Gets called on EPOLLHUP on eventfd when user closes it. |
79bd9814 TH |
4542 | * |
4543 | * Called with wqh->lock held and interrupts disabled. | |
4544 | */ | |
ac6424b9 | 4545 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
3bc942f3 | 4546 | int sync, void *key) |
79bd9814 | 4547 | { |
3bc942f3 TH |
4548 | struct mem_cgroup_event *event = |
4549 | container_of(wait, struct mem_cgroup_event, wait); | |
59b6f873 | 4550 | struct mem_cgroup *memcg = event->memcg; |
3ad6f93e | 4551 | __poll_t flags = key_to_poll(key); |
79bd9814 | 4552 | |
a9a08845 | 4553 | if (flags & EPOLLHUP) { |
79bd9814 TH |
4554 | /* |
4555 | * If the event has been detached at cgroup removal, we | |
4556 | * can simply return knowing the other side will cleanup | |
4557 | * for us. | |
4558 | * | |
4559 | * We can't race against event freeing since the other | |
4560 | * side will require wqh->lock via remove_wait_queue(), | |
4561 | * which we hold. | |
4562 | */ | |
fba94807 | 4563 | spin_lock(&memcg->event_list_lock); |
79bd9814 TH |
4564 | if (!list_empty(&event->list)) { |
4565 | list_del_init(&event->list); | |
4566 | /* | |
4567 | * We are in atomic context, but cgroup_event_remove() | |
4568 | * may sleep, so we have to call it in workqueue. | |
4569 | */ | |
4570 | schedule_work(&event->remove); | |
4571 | } | |
fba94807 | 4572 | spin_unlock(&memcg->event_list_lock); |
79bd9814 TH |
4573 | } |
4574 | ||
4575 | return 0; | |
4576 | } | |
4577 | ||
3bc942f3 | 4578 | static void memcg_event_ptable_queue_proc(struct file *file, |
79bd9814 TH |
4579 | wait_queue_head_t *wqh, poll_table *pt) |
4580 | { | |
3bc942f3 TH |
4581 | struct mem_cgroup_event *event = |
4582 | container_of(pt, struct mem_cgroup_event, pt); | |
79bd9814 TH |
4583 | |
4584 | event->wqh = wqh; | |
4585 | add_wait_queue(wqh, &event->wait); | |
4586 | } | |
4587 | ||
4588 | /* | |
3bc942f3 TH |
4589 | * DO NOT USE IN NEW FILES. |
4590 | * | |
79bd9814 TH |
4591 | * Parse input and register new cgroup event handler. |
4592 | * | |
4593 | * Input must be in format '<event_fd> <control_fd> <args>'. | |
4594 | * Interpretation of args is defined by control file implementation. | |
4595 | */ | |
451af504 TH |
4596 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
4597 | char *buf, size_t nbytes, loff_t off) | |
79bd9814 | 4598 | { |
451af504 | 4599 | struct cgroup_subsys_state *css = of_css(of); |
fba94807 | 4600 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 4601 | struct mem_cgroup_event *event; |
79bd9814 TH |
4602 | struct cgroup_subsys_state *cfile_css; |
4603 | unsigned int efd, cfd; | |
4604 | struct fd efile; | |
4605 | struct fd cfile; | |
fba94807 | 4606 | const char *name; |
79bd9814 TH |
4607 | char *endp; |
4608 | int ret; | |
4609 | ||
451af504 TH |
4610 | buf = strstrip(buf); |
4611 | ||
4612 | efd = simple_strtoul(buf, &endp, 10); | |
79bd9814 TH |
4613 | if (*endp != ' ') |
4614 | return -EINVAL; | |
451af504 | 4615 | buf = endp + 1; |
79bd9814 | 4616 | |
451af504 | 4617 | cfd = simple_strtoul(buf, &endp, 10); |
79bd9814 TH |
4618 | if ((*endp != ' ') && (*endp != '\0')) |
4619 | return -EINVAL; | |
451af504 | 4620 | buf = endp + 1; |
79bd9814 TH |
4621 | |
4622 | event = kzalloc(sizeof(*event), GFP_KERNEL); | |
4623 | if (!event) | |
4624 | return -ENOMEM; | |
4625 | ||
59b6f873 | 4626 | event->memcg = memcg; |
79bd9814 | 4627 | INIT_LIST_HEAD(&event->list); |
3bc942f3 TH |
4628 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
4629 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); | |
4630 | INIT_WORK(&event->remove, memcg_event_remove); | |
79bd9814 TH |
4631 | |
4632 | efile = fdget(efd); | |
4633 | if (!efile.file) { | |
4634 | ret = -EBADF; | |
4635 | goto out_kfree; | |
4636 | } | |
4637 | ||
4638 | event->eventfd = eventfd_ctx_fileget(efile.file); | |
4639 | if (IS_ERR(event->eventfd)) { | |
4640 | ret = PTR_ERR(event->eventfd); | |
4641 | goto out_put_efile; | |
4642 | } | |
4643 | ||
4644 | cfile = fdget(cfd); | |
4645 | if (!cfile.file) { | |
4646 | ret = -EBADF; | |
4647 | goto out_put_eventfd; | |
4648 | } | |
4649 | ||
4650 | /* the process need read permission on control file */ | |
4651 | /* AV: shouldn't we check that it's been opened for read instead? */ | |
02f92b38 | 4652 | ret = file_permission(cfile.file, MAY_READ); |
79bd9814 TH |
4653 | if (ret < 0) |
4654 | goto out_put_cfile; | |
4655 | ||
fba94807 TH |
4656 | /* |
4657 | * Determine the event callbacks and set them in @event. This used | |
4658 | * to be done via struct cftype but cgroup core no longer knows | |
4659 | * about these events. The following is crude but the whole thing | |
4660 | * is for compatibility anyway. | |
3bc942f3 TH |
4661 | * |
4662 | * DO NOT ADD NEW FILES. | |
fba94807 | 4663 | */ |
b583043e | 4664 | name = cfile.file->f_path.dentry->d_name.name; |
fba94807 TH |
4665 | |
4666 | if (!strcmp(name, "memory.usage_in_bytes")) { | |
4667 | event->register_event = mem_cgroup_usage_register_event; | |
4668 | event->unregister_event = mem_cgroup_usage_unregister_event; | |
4669 | } else if (!strcmp(name, "memory.oom_control")) { | |
4670 | event->register_event = mem_cgroup_oom_register_event; | |
4671 | event->unregister_event = mem_cgroup_oom_unregister_event; | |
4672 | } else if (!strcmp(name, "memory.pressure_level")) { | |
4673 | event->register_event = vmpressure_register_event; | |
4674 | event->unregister_event = vmpressure_unregister_event; | |
4675 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { | |
347c4a87 TH |
4676 | event->register_event = memsw_cgroup_usage_register_event; |
4677 | event->unregister_event = memsw_cgroup_usage_unregister_event; | |
fba94807 TH |
4678 | } else { |
4679 | ret = -EINVAL; | |
4680 | goto out_put_cfile; | |
4681 | } | |
4682 | ||
79bd9814 | 4683 | /* |
b5557c4c TH |
4684 | * Verify @cfile should belong to @css. Also, remaining events are |
4685 | * automatically removed on cgroup destruction but the removal is | |
4686 | * asynchronous, so take an extra ref on @css. | |
79bd9814 | 4687 | */ |
b583043e | 4688 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
ec903c0c | 4689 | &memory_cgrp_subsys); |
79bd9814 | 4690 | ret = -EINVAL; |
5a17f543 | 4691 | if (IS_ERR(cfile_css)) |
79bd9814 | 4692 | goto out_put_cfile; |
5a17f543 TH |
4693 | if (cfile_css != css) { |
4694 | css_put(cfile_css); | |
79bd9814 | 4695 | goto out_put_cfile; |
5a17f543 | 4696 | } |
79bd9814 | 4697 | |
451af504 | 4698 | ret = event->register_event(memcg, event->eventfd, buf); |
79bd9814 TH |
4699 | if (ret) |
4700 | goto out_put_css; | |
4701 | ||
9965ed17 | 4702 | vfs_poll(efile.file, &event->pt); |
79bd9814 | 4703 | |
fba94807 TH |
4704 | spin_lock(&memcg->event_list_lock); |
4705 | list_add(&event->list, &memcg->event_list); | |
4706 | spin_unlock(&memcg->event_list_lock); | |
79bd9814 TH |
4707 | |
4708 | fdput(cfile); | |
4709 | fdput(efile); | |
4710 | ||
451af504 | 4711 | return nbytes; |
79bd9814 TH |
4712 | |
4713 | out_put_css: | |
b5557c4c | 4714 | css_put(css); |
79bd9814 TH |
4715 | out_put_cfile: |
4716 | fdput(cfile); | |
4717 | out_put_eventfd: | |
4718 | eventfd_ctx_put(event->eventfd); | |
4719 | out_put_efile: | |
4720 | fdput(efile); | |
4721 | out_kfree: | |
4722 | kfree(event); | |
4723 | ||
4724 | return ret; | |
4725 | } | |
4726 | ||
241994ed | 4727 | static struct cftype mem_cgroup_legacy_files[] = { |
8cdea7c0 | 4728 | { |
0eea1030 | 4729 | .name = "usage_in_bytes", |
8c7c6e34 | 4730 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
791badbd | 4731 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4732 | }, |
c84872e1 PE |
4733 | { |
4734 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4735 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
6770c64e | 4736 | .write = mem_cgroup_reset, |
791badbd | 4737 | .read_u64 = mem_cgroup_read_u64, |
c84872e1 | 4738 | }, |
8cdea7c0 | 4739 | { |
0eea1030 | 4740 | .name = "limit_in_bytes", |
8c7c6e34 | 4741 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
451af504 | 4742 | .write = mem_cgroup_write, |
791badbd | 4743 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4744 | }, |
296c81d8 BS |
4745 | { |
4746 | .name = "soft_limit_in_bytes", | |
4747 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
451af504 | 4748 | .write = mem_cgroup_write, |
791badbd | 4749 | .read_u64 = mem_cgroup_read_u64, |
296c81d8 | 4750 | }, |
8cdea7c0 BS |
4751 | { |
4752 | .name = "failcnt", | |
8c7c6e34 | 4753 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
6770c64e | 4754 | .write = mem_cgroup_reset, |
791badbd | 4755 | .read_u64 = mem_cgroup_read_u64, |
8cdea7c0 | 4756 | }, |
d2ceb9b7 KH |
4757 | { |
4758 | .name = "stat", | |
2da8ca82 | 4759 | .seq_show = memcg_stat_show, |
d2ceb9b7 | 4760 | }, |
c1e862c1 KH |
4761 | { |
4762 | .name = "force_empty", | |
6770c64e | 4763 | .write = mem_cgroup_force_empty_write, |
c1e862c1 | 4764 | }, |
18f59ea7 BS |
4765 | { |
4766 | .name = "use_hierarchy", | |
4767 | .write_u64 = mem_cgroup_hierarchy_write, | |
4768 | .read_u64 = mem_cgroup_hierarchy_read, | |
4769 | }, | |
79bd9814 | 4770 | { |
3bc942f3 | 4771 | .name = "cgroup.event_control", /* XXX: for compat */ |
451af504 | 4772 | .write = memcg_write_event_control, |
7dbdb199 | 4773 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
79bd9814 | 4774 | }, |
a7885eb8 KM |
4775 | { |
4776 | .name = "swappiness", | |
4777 | .read_u64 = mem_cgroup_swappiness_read, | |
4778 | .write_u64 = mem_cgroup_swappiness_write, | |
4779 | }, | |
7dc74be0 DN |
4780 | { |
4781 | .name = "move_charge_at_immigrate", | |
4782 | .read_u64 = mem_cgroup_move_charge_read, | |
4783 | .write_u64 = mem_cgroup_move_charge_write, | |
4784 | }, | |
9490ff27 KH |
4785 | { |
4786 | .name = "oom_control", | |
2da8ca82 | 4787 | .seq_show = mem_cgroup_oom_control_read, |
3c11ecf4 | 4788 | .write_u64 = mem_cgroup_oom_control_write, |
9490ff27 KH |
4789 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
4790 | }, | |
70ddf637 AV |
4791 | { |
4792 | .name = "pressure_level", | |
70ddf637 | 4793 | }, |
406eb0c9 YH |
4794 | #ifdef CONFIG_NUMA |
4795 | { | |
4796 | .name = "numa_stat", | |
2da8ca82 | 4797 | .seq_show = memcg_numa_stat_show, |
406eb0c9 YH |
4798 | }, |
4799 | #endif | |
510fc4e1 GC |
4800 | { |
4801 | .name = "kmem.limit_in_bytes", | |
4802 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), | |
451af504 | 4803 | .write = mem_cgroup_write, |
791badbd | 4804 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4805 | }, |
4806 | { | |
4807 | .name = "kmem.usage_in_bytes", | |
4808 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), | |
791badbd | 4809 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4810 | }, |
4811 | { | |
4812 | .name = "kmem.failcnt", | |
4813 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), | |
6770c64e | 4814 | .write = mem_cgroup_reset, |
791badbd | 4815 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 GC |
4816 | }, |
4817 | { | |
4818 | .name = "kmem.max_usage_in_bytes", | |
4819 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), | |
6770c64e | 4820 | .write = mem_cgroup_reset, |
791badbd | 4821 | .read_u64 = mem_cgroup_read_u64, |
510fc4e1 | 4822 | }, |
a87425a3 YS |
4823 | #if defined(CONFIG_MEMCG_KMEM) && \ |
4824 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) | |
749c5415 GC |
4825 | { |
4826 | .name = "kmem.slabinfo", | |
b047501c | 4827 | .seq_show = memcg_slab_show, |
749c5415 GC |
4828 | }, |
4829 | #endif | |
d55f90bf VD |
4830 | { |
4831 | .name = "kmem.tcp.limit_in_bytes", | |
4832 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), | |
4833 | .write = mem_cgroup_write, | |
4834 | .read_u64 = mem_cgroup_read_u64, | |
4835 | }, | |
4836 | { | |
4837 | .name = "kmem.tcp.usage_in_bytes", | |
4838 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), | |
4839 | .read_u64 = mem_cgroup_read_u64, | |
4840 | }, | |
4841 | { | |
4842 | .name = "kmem.tcp.failcnt", | |
4843 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), | |
4844 | .write = mem_cgroup_reset, | |
4845 | .read_u64 = mem_cgroup_read_u64, | |
4846 | }, | |
4847 | { | |
4848 | .name = "kmem.tcp.max_usage_in_bytes", | |
4849 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), | |
4850 | .write = mem_cgroup_reset, | |
4851 | .read_u64 = mem_cgroup_read_u64, | |
4852 | }, | |
6bc10349 | 4853 | { }, /* terminate */ |
af36f906 | 4854 | }; |
8c7c6e34 | 4855 | |
73f576c0 JW |
4856 | /* |
4857 | * Private memory cgroup IDR | |
4858 | * | |
4859 | * Swap-out records and page cache shadow entries need to store memcg | |
4860 | * references in constrained space, so we maintain an ID space that is | |
4861 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of | |
4862 | * memory-controlled cgroups to 64k. | |
4863 | * | |
b8f2935f | 4864 | * However, there usually are many references to the offline CSS after |
73f576c0 JW |
4865 | * the cgroup has been destroyed, such as page cache or reclaimable |
4866 | * slab objects, that don't need to hang on to the ID. We want to keep | |
4867 | * those dead CSS from occupying IDs, or we might quickly exhaust the | |
4868 | * relatively small ID space and prevent the creation of new cgroups | |
4869 | * even when there are much fewer than 64k cgroups - possibly none. | |
4870 | * | |
4871 | * Maintain a private 16-bit ID space for memcg, and allow the ID to | |
4872 | * be freed and recycled when it's no longer needed, which is usually | |
4873 | * when the CSS is offlined. | |
4874 | * | |
4875 | * The only exception to that are records of swapped out tmpfs/shmem | |
4876 | * pages that need to be attributed to live ancestors on swapin. But | |
4877 | * those references are manageable from userspace. | |
4878 | */ | |
4879 | ||
4880 | static DEFINE_IDR(mem_cgroup_idr); | |
4881 | ||
7e97de0b KT |
4882 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
4883 | { | |
4884 | if (memcg->id.id > 0) { | |
4885 | idr_remove(&mem_cgroup_idr, memcg->id.id); | |
4886 | memcg->id.id = 0; | |
4887 | } | |
4888 | } | |
4889 | ||
c1514c0a VF |
4890 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
4891 | unsigned int n) | |
73f576c0 | 4892 | { |
1c2d479a | 4893 | refcount_add(n, &memcg->id.ref); |
73f576c0 JW |
4894 | } |
4895 | ||
615d66c3 | 4896 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
73f576c0 | 4897 | { |
1c2d479a | 4898 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
7e97de0b | 4899 | mem_cgroup_id_remove(memcg); |
73f576c0 JW |
4900 | |
4901 | /* Memcg ID pins CSS */ | |
4902 | css_put(&memcg->css); | |
4903 | } | |
4904 | } | |
4905 | ||
615d66c3 VD |
4906 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
4907 | { | |
4908 | mem_cgroup_id_put_many(memcg, 1); | |
4909 | } | |
4910 | ||
73f576c0 JW |
4911 | /** |
4912 | * mem_cgroup_from_id - look up a memcg from a memcg id | |
4913 | * @id: the memcg id to look up | |
4914 | * | |
4915 | * Caller must hold rcu_read_lock(). | |
4916 | */ | |
4917 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) | |
4918 | { | |
4919 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
4920 | return idr_find(&mem_cgroup_idr, id); | |
4921 | } | |
4922 | ||
ef8f2327 | 4923 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
6d12e2d8 KH |
4924 | { |
4925 | struct mem_cgroup_per_node *pn; | |
ef8f2327 | 4926 | int tmp = node; |
1ecaab2b KH |
4927 | /* |
4928 | * This routine is called against possible nodes. | |
4929 | * But it's BUG to call kmalloc() against offline node. | |
4930 | * | |
4931 | * TODO: this routine can waste much memory for nodes which will | |
4932 | * never be onlined. It's better to use memory hotplug callback | |
4933 | * function. | |
4934 | */ | |
41e3355d KH |
4935 | if (!node_state(node, N_NORMAL_MEMORY)) |
4936 | tmp = -1; | |
17295c88 | 4937 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4938 | if (!pn) |
4939 | return 1; | |
1ecaab2b | 4940 | |
3e38e0aa RG |
4941 | pn->lruvec_stat_local = alloc_percpu_gfp(struct lruvec_stat, |
4942 | GFP_KERNEL_ACCOUNT); | |
815744d7 JW |
4943 | if (!pn->lruvec_stat_local) { |
4944 | kfree(pn); | |
4945 | return 1; | |
4946 | } | |
4947 | ||
f3344adf | 4948 | pn->lruvec_stat_cpu = alloc_percpu_gfp(struct batched_lruvec_stat, |
3e38e0aa | 4949 | GFP_KERNEL_ACCOUNT); |
a983b5eb | 4950 | if (!pn->lruvec_stat_cpu) { |
815744d7 | 4951 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c JW |
4952 | kfree(pn); |
4953 | return 1; | |
4954 | } | |
4955 | ||
ef8f2327 MG |
4956 | lruvec_init(&pn->lruvec); |
4957 | pn->usage_in_excess = 0; | |
4958 | pn->on_tree = false; | |
4959 | pn->memcg = memcg; | |
4960 | ||
54f72fe0 | 4961 | memcg->nodeinfo[node] = pn; |
6d12e2d8 KH |
4962 | return 0; |
4963 | } | |
4964 | ||
ef8f2327 | 4965 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
1ecaab2b | 4966 | { |
00f3ca2c JW |
4967 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
4968 | ||
4eaf431f MH |
4969 | if (!pn) |
4970 | return; | |
4971 | ||
a983b5eb | 4972 | free_percpu(pn->lruvec_stat_cpu); |
815744d7 | 4973 | free_percpu(pn->lruvec_stat_local); |
00f3ca2c | 4974 | kfree(pn); |
1ecaab2b KH |
4975 | } |
4976 | ||
40e952f9 | 4977 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
59927fb9 | 4978 | { |
c8b2a36f | 4979 | int node; |
59927fb9 | 4980 | |
c8b2a36f | 4981 | for_each_node(node) |
ef8f2327 | 4982 | free_mem_cgroup_per_node_info(memcg, node); |
871789d4 | 4983 | free_percpu(memcg->vmstats_percpu); |
8ff69e2c | 4984 | kfree(memcg); |
59927fb9 | 4985 | } |
3afe36b1 | 4986 | |
40e952f9 TE |
4987 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
4988 | { | |
2cd21c89 JW |
4989 | int cpu; |
4990 | ||
40e952f9 | 4991 | memcg_wb_domain_exit(memcg); |
7961eee3 | 4992 | /* |
2d146aa3 JW |
4993 | * Flush percpu lruvec stats to guarantee the value |
4994 | * correctness on parent's and all ancestor levels. | |
7961eee3 | 4995 | */ |
2cd21c89 JW |
4996 | for_each_online_cpu(cpu) |
4997 | memcg_flush_lruvec_page_state(memcg, cpu); | |
40e952f9 TE |
4998 | __mem_cgroup_free(memcg); |
4999 | } | |
5000 | ||
0b8f73e1 | 5001 | static struct mem_cgroup *mem_cgroup_alloc(void) |
8cdea7c0 | 5002 | { |
d142e3e6 | 5003 | struct mem_cgroup *memcg; |
b9726c26 | 5004 | unsigned int size; |
6d12e2d8 | 5005 | int node; |
97b27821 | 5006 | int __maybe_unused i; |
11d67612 | 5007 | long error = -ENOMEM; |
8cdea7c0 | 5008 | |
0b8f73e1 JW |
5009 | size = sizeof(struct mem_cgroup); |
5010 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); | |
5011 | ||
5012 | memcg = kzalloc(size, GFP_KERNEL); | |
c0ff4b85 | 5013 | if (!memcg) |
11d67612 | 5014 | return ERR_PTR(error); |
0b8f73e1 | 5015 | |
73f576c0 JW |
5016 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
5017 | 1, MEM_CGROUP_ID_MAX, | |
5018 | GFP_KERNEL); | |
11d67612 YS |
5019 | if (memcg->id.id < 0) { |
5020 | error = memcg->id.id; | |
73f576c0 | 5021 | goto fail; |
11d67612 | 5022 | } |
73f576c0 | 5023 | |
3e38e0aa RG |
5024 | memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
5025 | GFP_KERNEL_ACCOUNT); | |
871789d4 | 5026 | if (!memcg->vmstats_percpu) |
0b8f73e1 | 5027 | goto fail; |
78fb7466 | 5028 | |
3ed28fa1 | 5029 | for_each_node(node) |
ef8f2327 | 5030 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
0b8f73e1 | 5031 | goto fail; |
f64c3f54 | 5032 | |
0b8f73e1 JW |
5033 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
5034 | goto fail; | |
28dbc4b6 | 5035 | |
f7e1cb6e | 5036 | INIT_WORK(&memcg->high_work, high_work_func); |
d142e3e6 | 5037 | INIT_LIST_HEAD(&memcg->oom_notify); |
d142e3e6 GC |
5038 | mutex_init(&memcg->thresholds_lock); |
5039 | spin_lock_init(&memcg->move_lock); | |
70ddf637 | 5040 | vmpressure_init(&memcg->vmpressure); |
fba94807 TH |
5041 | INIT_LIST_HEAD(&memcg->event_list); |
5042 | spin_lock_init(&memcg->event_list_lock); | |
d886f4e4 | 5043 | memcg->socket_pressure = jiffies; |
84c07d11 | 5044 | #ifdef CONFIG_MEMCG_KMEM |
900a38f0 | 5045 | memcg->kmemcg_id = -1; |
bf4f0599 | 5046 | INIT_LIST_HEAD(&memcg->objcg_list); |
900a38f0 | 5047 | #endif |
52ebea74 TH |
5048 | #ifdef CONFIG_CGROUP_WRITEBACK |
5049 | INIT_LIST_HEAD(&memcg->cgwb_list); | |
97b27821 TH |
5050 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
5051 | memcg->cgwb_frn[i].done = | |
5052 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); | |
87eaceb3 YS |
5053 | #endif |
5054 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE | |
5055 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); | |
5056 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); | |
5057 | memcg->deferred_split_queue.split_queue_len = 0; | |
52ebea74 | 5058 | #endif |
73f576c0 | 5059 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
0b8f73e1 JW |
5060 | return memcg; |
5061 | fail: | |
7e97de0b | 5062 | mem_cgroup_id_remove(memcg); |
40e952f9 | 5063 | __mem_cgroup_free(memcg); |
11d67612 | 5064 | return ERR_PTR(error); |
d142e3e6 GC |
5065 | } |
5066 | ||
0b8f73e1 JW |
5067 | static struct cgroup_subsys_state * __ref |
5068 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) | |
d142e3e6 | 5069 | { |
0b8f73e1 | 5070 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
b87d8cef | 5071 | struct mem_cgroup *memcg, *old_memcg; |
0b8f73e1 | 5072 | long error = -ENOMEM; |
d142e3e6 | 5073 | |
b87d8cef | 5074 | old_memcg = set_active_memcg(parent); |
0b8f73e1 | 5075 | memcg = mem_cgroup_alloc(); |
b87d8cef | 5076 | set_active_memcg(old_memcg); |
11d67612 YS |
5077 | if (IS_ERR(memcg)) |
5078 | return ERR_CAST(memcg); | |
d142e3e6 | 5079 | |
d1663a90 | 5080 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
0b8f73e1 | 5081 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5082 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
0b8f73e1 JW |
5083 | if (parent) { |
5084 | memcg->swappiness = mem_cgroup_swappiness(parent); | |
5085 | memcg->oom_kill_disable = parent->oom_kill_disable; | |
bef8620c | 5086 | |
3e32cb2e | 5087 | page_counter_init(&memcg->memory, &parent->memory); |
37e84351 | 5088 | page_counter_init(&memcg->swap, &parent->swap); |
3e32cb2e | 5089 | page_counter_init(&memcg->kmem, &parent->kmem); |
0db15298 | 5090 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
18f59ea7 | 5091 | } else { |
bef8620c RG |
5092 | page_counter_init(&memcg->memory, NULL); |
5093 | page_counter_init(&memcg->swap, NULL); | |
5094 | page_counter_init(&memcg->kmem, NULL); | |
5095 | page_counter_init(&memcg->tcpmem, NULL); | |
d6441637 | 5096 | |
0b8f73e1 JW |
5097 | root_mem_cgroup = memcg; |
5098 | return &memcg->css; | |
5099 | } | |
5100 | ||
bef8620c | 5101 | /* The following stuff does not apply to the root */ |
b313aeee | 5102 | error = memcg_online_kmem(memcg); |
0b8f73e1 JW |
5103 | if (error) |
5104 | goto fail; | |
127424c8 | 5105 | |
f7e1cb6e | 5106 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5107 | static_branch_inc(&memcg_sockets_enabled_key); |
f7e1cb6e | 5108 | |
0b8f73e1 JW |
5109 | return &memcg->css; |
5110 | fail: | |
7e97de0b | 5111 | mem_cgroup_id_remove(memcg); |
0b8f73e1 | 5112 | mem_cgroup_free(memcg); |
11d67612 | 5113 | return ERR_PTR(error); |
0b8f73e1 JW |
5114 | } |
5115 | ||
73f576c0 | 5116 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
0b8f73e1 | 5117 | { |
58fa2a55 VD |
5118 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
5119 | ||
0a4465d3 | 5120 | /* |
e4262c4f | 5121 | * A memcg must be visible for expand_shrinker_info() |
0a4465d3 KT |
5122 | * by the time the maps are allocated. So, we allocate maps |
5123 | * here, when for_each_mem_cgroup() can't skip it. | |
5124 | */ | |
e4262c4f | 5125 | if (alloc_shrinker_info(memcg)) { |
0a4465d3 KT |
5126 | mem_cgroup_id_remove(memcg); |
5127 | return -ENOMEM; | |
5128 | } | |
5129 | ||
73f576c0 | 5130 | /* Online state pins memcg ID, memcg ID pins CSS */ |
1c2d479a | 5131 | refcount_set(&memcg->id.ref, 1); |
73f576c0 | 5132 | css_get(css); |
2f7dd7a4 | 5133 | return 0; |
8cdea7c0 BS |
5134 | } |
5135 | ||
eb95419b | 5136 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
df878fb0 | 5137 | { |
eb95419b | 5138 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
3bc942f3 | 5139 | struct mem_cgroup_event *event, *tmp; |
79bd9814 TH |
5140 | |
5141 | /* | |
5142 | * Unregister events and notify userspace. | |
5143 | * Notify userspace about cgroup removing only after rmdir of cgroup | |
5144 | * directory to avoid race between userspace and kernelspace. | |
5145 | */ | |
fba94807 TH |
5146 | spin_lock(&memcg->event_list_lock); |
5147 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { | |
79bd9814 TH |
5148 | list_del_init(&event->list); |
5149 | schedule_work(&event->remove); | |
5150 | } | |
fba94807 | 5151 | spin_unlock(&memcg->event_list_lock); |
ec64f515 | 5152 | |
bf8d5d52 | 5153 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5154 | page_counter_set_low(&memcg->memory, 0); |
63677c74 | 5155 | |
567e9ab2 | 5156 | memcg_offline_kmem(memcg); |
a178015c | 5157 | reparent_shrinker_deferred(memcg); |
52ebea74 | 5158 | wb_memcg_offline(memcg); |
73f576c0 | 5159 | |
591edfb1 RG |
5160 | drain_all_stock(memcg); |
5161 | ||
73f576c0 | 5162 | mem_cgroup_id_put(memcg); |
df878fb0 KH |
5163 | } |
5164 | ||
6df38689 VD |
5165 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
5166 | { | |
5167 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5168 | ||
5169 | invalidate_reclaim_iterators(memcg); | |
5170 | } | |
5171 | ||
eb95419b | 5172 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
8cdea7c0 | 5173 | { |
eb95419b | 5174 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
97b27821 | 5175 | int __maybe_unused i; |
c268e994 | 5176 | |
97b27821 TH |
5177 | #ifdef CONFIG_CGROUP_WRITEBACK |
5178 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) | |
5179 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); | |
5180 | #endif | |
f7e1cb6e | 5181 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
ef12947c | 5182 | static_branch_dec(&memcg_sockets_enabled_key); |
127424c8 | 5183 | |
0db15298 | 5184 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
d55f90bf | 5185 | static_branch_dec(&memcg_sockets_enabled_key); |
3893e302 | 5186 | |
0b8f73e1 JW |
5187 | vmpressure_cleanup(&memcg->vmpressure); |
5188 | cancel_work_sync(&memcg->high_work); | |
5189 | mem_cgroup_remove_from_trees(memcg); | |
e4262c4f | 5190 | free_shrinker_info(memcg); |
d886f4e4 | 5191 | memcg_free_kmem(memcg); |
0b8f73e1 | 5192 | mem_cgroup_free(memcg); |
8cdea7c0 BS |
5193 | } |
5194 | ||
1ced953b TH |
5195 | /** |
5196 | * mem_cgroup_css_reset - reset the states of a mem_cgroup | |
5197 | * @css: the target css | |
5198 | * | |
5199 | * Reset the states of the mem_cgroup associated with @css. This is | |
5200 | * invoked when the userland requests disabling on the default hierarchy | |
5201 | * but the memcg is pinned through dependency. The memcg should stop | |
5202 | * applying policies and should revert to the vanilla state as it may be | |
5203 | * made visible again. | |
5204 | * | |
5205 | * The current implementation only resets the essential configurations. | |
5206 | * This needs to be expanded to cover all the visible parts. | |
5207 | */ | |
5208 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) | |
5209 | { | |
5210 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5211 | ||
bbec2e15 RG |
5212 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
5213 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); | |
bbec2e15 RG |
5214 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); |
5215 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); | |
bf8d5d52 | 5216 | page_counter_set_min(&memcg->memory, 0); |
23067153 | 5217 | page_counter_set_low(&memcg->memory, 0); |
d1663a90 | 5218 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
24d404dc | 5219 | memcg->soft_limit = PAGE_COUNTER_MAX; |
4b82ab4f | 5220 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
2529bb3a | 5221 | memcg_wb_domain_size_changed(memcg); |
1ced953b TH |
5222 | } |
5223 | ||
2d146aa3 JW |
5224 | static void mem_cgroup_css_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
5225 | { | |
5226 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
5227 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); | |
5228 | struct memcg_vmstats_percpu *statc; | |
5229 | long delta, v; | |
5230 | int i; | |
5231 | ||
5232 | statc = per_cpu_ptr(memcg->vmstats_percpu, cpu); | |
5233 | ||
5234 | for (i = 0; i < MEMCG_NR_STAT; i++) { | |
5235 | /* | |
5236 | * Collect the aggregated propagation counts of groups | |
5237 | * below us. We're in a per-cpu loop here and this is | |
5238 | * a global counter, so the first cycle will get them. | |
5239 | */ | |
5240 | delta = memcg->vmstats.state_pending[i]; | |
5241 | if (delta) | |
5242 | memcg->vmstats.state_pending[i] = 0; | |
5243 | ||
5244 | /* Add CPU changes on this level since the last flush */ | |
5245 | v = READ_ONCE(statc->state[i]); | |
5246 | if (v != statc->state_prev[i]) { | |
5247 | delta += v - statc->state_prev[i]; | |
5248 | statc->state_prev[i] = v; | |
5249 | } | |
5250 | ||
5251 | if (!delta) | |
5252 | continue; | |
5253 | ||
5254 | /* Aggregate counts on this level and propagate upwards */ | |
5255 | memcg->vmstats.state[i] += delta; | |
5256 | if (parent) | |
5257 | parent->vmstats.state_pending[i] += delta; | |
5258 | } | |
5259 | ||
5260 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { | |
5261 | delta = memcg->vmstats.events_pending[i]; | |
5262 | if (delta) | |
5263 | memcg->vmstats.events_pending[i] = 0; | |
5264 | ||
5265 | v = READ_ONCE(statc->events[i]); | |
5266 | if (v != statc->events_prev[i]) { | |
5267 | delta += v - statc->events_prev[i]; | |
5268 | statc->events_prev[i] = v; | |
5269 | } | |
5270 | ||
5271 | if (!delta) | |
5272 | continue; | |
5273 | ||
5274 | memcg->vmstats.events[i] += delta; | |
5275 | if (parent) | |
5276 | parent->vmstats.events_pending[i] += delta; | |
5277 | } | |
5278 | } | |
5279 | ||
02491447 | 5280 | #ifdef CONFIG_MMU |
7dc74be0 | 5281 | /* Handlers for move charge at task migration. */ |
854ffa8d | 5282 | static int mem_cgroup_do_precharge(unsigned long count) |
7dc74be0 | 5283 | { |
05b84301 | 5284 | int ret; |
9476db97 | 5285 | |
d0164adc MG |
5286 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
5287 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); | |
9476db97 | 5288 | if (!ret) { |
854ffa8d | 5289 | mc.precharge += count; |
854ffa8d DN |
5290 | return ret; |
5291 | } | |
9476db97 | 5292 | |
3674534b | 5293 | /* Try charges one by one with reclaim, but do not retry */ |
854ffa8d | 5294 | while (count--) { |
3674534b | 5295 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
38c5d72f | 5296 | if (ret) |
38c5d72f | 5297 | return ret; |
854ffa8d | 5298 | mc.precharge++; |
9476db97 | 5299 | cond_resched(); |
854ffa8d | 5300 | } |
9476db97 | 5301 | return 0; |
4ffef5fe DN |
5302 | } |
5303 | ||
4ffef5fe DN |
5304 | union mc_target { |
5305 | struct page *page; | |
02491447 | 5306 | swp_entry_t ent; |
4ffef5fe DN |
5307 | }; |
5308 | ||
4ffef5fe | 5309 | enum mc_target_type { |
8d32ff84 | 5310 | MC_TARGET_NONE = 0, |
4ffef5fe | 5311 | MC_TARGET_PAGE, |
02491447 | 5312 | MC_TARGET_SWAP, |
c733a828 | 5313 | MC_TARGET_DEVICE, |
4ffef5fe DN |
5314 | }; |
5315 | ||
90254a65 DN |
5316 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5317 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5318 | { |
25b2995a | 5319 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5320 | |
90254a65 DN |
5321 | if (!page || !page_mapped(page)) |
5322 | return NULL; | |
5323 | if (PageAnon(page)) { | |
1dfab5ab | 5324 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5325 | return NULL; |
1dfab5ab JW |
5326 | } else { |
5327 | if (!(mc.flags & MOVE_FILE)) | |
5328 | return NULL; | |
5329 | } | |
90254a65 DN |
5330 | if (!get_page_unless_zero(page)) |
5331 | return NULL; | |
5332 | ||
5333 | return page; | |
5334 | } | |
5335 | ||
c733a828 | 5336 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
90254a65 | 5337 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
48406ef8 | 5338 | pte_t ptent, swp_entry_t *entry) |
90254a65 | 5339 | { |
90254a65 DN |
5340 | struct page *page = NULL; |
5341 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5342 | ||
9a137153 | 5343 | if (!(mc.flags & MOVE_ANON)) |
90254a65 | 5344 | return NULL; |
c733a828 JG |
5345 | |
5346 | /* | |
5347 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to | |
5348 | * a device and because they are not accessible by CPU they are store | |
5349 | * as special swap entry in the CPU page table. | |
5350 | */ | |
5351 | if (is_device_private_entry(ent)) { | |
5352 | page = device_private_entry_to_page(ent); | |
5353 | /* | |
5354 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have | |
5355 | * a refcount of 1 when free (unlike normal page) | |
5356 | */ | |
5357 | if (!page_ref_add_unless(page, 1, 1)) | |
5358 | return NULL; | |
5359 | return page; | |
5360 | } | |
5361 | ||
9a137153 RC |
5362 | if (non_swap_entry(ent)) |
5363 | return NULL; | |
5364 | ||
4b91355e KH |
5365 | /* |
5366 | * Because lookup_swap_cache() updates some statistics counter, | |
5367 | * we call find_get_page() with swapper_space directly. | |
5368 | */ | |
f6ab1f7f | 5369 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
2d1c4980 | 5370 | entry->val = ent.val; |
90254a65 DN |
5371 | |
5372 | return page; | |
5373 | } | |
4b91355e KH |
5374 | #else |
5375 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
48406ef8 | 5376 | pte_t ptent, swp_entry_t *entry) |
4b91355e KH |
5377 | { |
5378 | return NULL; | |
5379 | } | |
5380 | #endif | |
90254a65 | 5381 | |
87946a72 DN |
5382 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5383 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5384 | { | |
87946a72 DN |
5385 | if (!vma->vm_file) /* anonymous vma */ |
5386 | return NULL; | |
1dfab5ab | 5387 | if (!(mc.flags & MOVE_FILE)) |
87946a72 DN |
5388 | return NULL; |
5389 | ||
87946a72 | 5390 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
aa3b1895 | 5391 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
f5df8635 MWO |
5392 | return find_get_incore_page(vma->vm_file->f_mapping, |
5393 | linear_page_index(vma, addr)); | |
87946a72 DN |
5394 | } |
5395 | ||
b1b0deab CG |
5396 | /** |
5397 | * mem_cgroup_move_account - move account of the page | |
5398 | * @page: the page | |
25843c2b | 5399 | * @compound: charge the page as compound or small page |
b1b0deab CG |
5400 | * @from: mem_cgroup which the page is moved from. |
5401 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
5402 | * | |
3ac808fd | 5403 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
b1b0deab CG |
5404 | * |
5405 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" | |
5406 | * from old cgroup. | |
5407 | */ | |
5408 | static int mem_cgroup_move_account(struct page *page, | |
f627c2f5 | 5409 | bool compound, |
b1b0deab CG |
5410 | struct mem_cgroup *from, |
5411 | struct mem_cgroup *to) | |
5412 | { | |
ae8af438 KK |
5413 | struct lruvec *from_vec, *to_vec; |
5414 | struct pglist_data *pgdat; | |
6c357848 | 5415 | unsigned int nr_pages = compound ? thp_nr_pages(page) : 1; |
b1b0deab CG |
5416 | int ret; |
5417 | ||
5418 | VM_BUG_ON(from == to); | |
5419 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
f627c2f5 | 5420 | VM_BUG_ON(compound && !PageTransHuge(page)); |
b1b0deab CG |
5421 | |
5422 | /* | |
6a93ca8f | 5423 | * Prevent mem_cgroup_migrate() from looking at |
bcfe06bf | 5424 | * page's memory cgroup of its source page while we change it. |
b1b0deab | 5425 | */ |
f627c2f5 | 5426 | ret = -EBUSY; |
b1b0deab CG |
5427 | if (!trylock_page(page)) |
5428 | goto out; | |
5429 | ||
5430 | ret = -EINVAL; | |
bcfe06bf | 5431 | if (page_memcg(page) != from) |
b1b0deab CG |
5432 | goto out_unlock; |
5433 | ||
ae8af438 | 5434 | pgdat = page_pgdat(page); |
867e5e1d JW |
5435 | from_vec = mem_cgroup_lruvec(from, pgdat); |
5436 | to_vec = mem_cgroup_lruvec(to, pgdat); | |
ae8af438 | 5437 | |
abb242f5 | 5438 | lock_page_memcg(page); |
b1b0deab | 5439 | |
be5d0a74 JW |
5440 | if (PageAnon(page)) { |
5441 | if (page_mapped(page)) { | |
5442 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); | |
5443 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); | |
468c3982 | 5444 | if (PageTransHuge(page)) { |
69473e5d MS |
5445 | __mod_lruvec_state(from_vec, NR_ANON_THPS, |
5446 | -nr_pages); | |
5447 | __mod_lruvec_state(to_vec, NR_ANON_THPS, | |
5448 | nr_pages); | |
468c3982 | 5449 | } |
be5d0a74 JW |
5450 | } |
5451 | } else { | |
0d1c2072 JW |
5452 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
5453 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); | |
5454 | ||
5455 | if (PageSwapBacked(page)) { | |
5456 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); | |
5457 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); | |
5458 | } | |
5459 | ||
49e50d27 JW |
5460 | if (page_mapped(page)) { |
5461 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); | |
5462 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); | |
5463 | } | |
b1b0deab | 5464 | |
49e50d27 JW |
5465 | if (PageDirty(page)) { |
5466 | struct address_space *mapping = page_mapping(page); | |
c4843a75 | 5467 | |
f56753ac | 5468 | if (mapping_can_writeback(mapping)) { |
49e50d27 JW |
5469 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
5470 | -nr_pages); | |
5471 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, | |
5472 | nr_pages); | |
5473 | } | |
c4843a75 GT |
5474 | } |
5475 | } | |
5476 | ||
b1b0deab | 5477 | if (PageWriteback(page)) { |
ae8af438 KK |
5478 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
5479 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); | |
b1b0deab CG |
5480 | } |
5481 | ||
5482 | /* | |
abb242f5 JW |
5483 | * All state has been migrated, let's switch to the new memcg. |
5484 | * | |
bcfe06bf | 5485 | * It is safe to change page's memcg here because the page |
abb242f5 JW |
5486 | * is referenced, charged, isolated, and locked: we can't race |
5487 | * with (un)charging, migration, LRU putback, or anything else | |
bcfe06bf | 5488 | * that would rely on a stable page's memory cgroup. |
abb242f5 JW |
5489 | * |
5490 | * Note that lock_page_memcg is a memcg lock, not a page lock, | |
bcfe06bf | 5491 | * to save space. As soon as we switch page's memory cgroup to a |
abb242f5 JW |
5492 | * new memcg that isn't locked, the above state can change |
5493 | * concurrently again. Make sure we're truly done with it. | |
b1b0deab | 5494 | */ |
abb242f5 | 5495 | smp_mb(); |
b1b0deab | 5496 | |
1a3e1f40 JW |
5497 | css_get(&to->css); |
5498 | css_put(&from->css); | |
5499 | ||
bcfe06bf | 5500 | page->memcg_data = (unsigned long)to; |
87eaceb3 | 5501 | |
abb242f5 | 5502 | __unlock_page_memcg(from); |
b1b0deab CG |
5503 | |
5504 | ret = 0; | |
5505 | ||
5506 | local_irq_disable(); | |
3fba69a5 | 5507 | mem_cgroup_charge_statistics(to, page, nr_pages); |
b1b0deab | 5508 | memcg_check_events(to, page); |
3fba69a5 | 5509 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
b1b0deab CG |
5510 | memcg_check_events(from, page); |
5511 | local_irq_enable(); | |
5512 | out_unlock: | |
5513 | unlock_page(page); | |
5514 | out: | |
5515 | return ret; | |
5516 | } | |
5517 | ||
7cf7806c LR |
5518 | /** |
5519 | * get_mctgt_type - get target type of moving charge | |
5520 | * @vma: the vma the pte to be checked belongs | |
5521 | * @addr: the address corresponding to the pte to be checked | |
5522 | * @ptent: the pte to be checked | |
5523 | * @target: the pointer the target page or swap ent will be stored(can be NULL) | |
5524 | * | |
5525 | * Returns | |
5526 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5527 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5528 | * move charge. if @target is not NULL, the page is stored in target->page | |
5529 | * with extra refcnt got(Callers should handle it). | |
5530 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a | |
5531 | * target for charge migration. if @target is not NULL, the entry is stored | |
5532 | * in target->ent. | |
25b2995a CH |
5533 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
5534 | * (so ZONE_DEVICE page and thus not on the lru). | |
df6ad698 JG |
5535 | * For now we such page is charge like a regular page would be as for all |
5536 | * intent and purposes it is just special memory taking the place of a | |
5537 | * regular page. | |
c733a828 JG |
5538 | * |
5539 | * See Documentations/vm/hmm.txt and include/linux/hmm.h | |
7cf7806c LR |
5540 | * |
5541 | * Called with pte lock held. | |
5542 | */ | |
5543 | ||
8d32ff84 | 5544 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
90254a65 DN |
5545 | unsigned long addr, pte_t ptent, union mc_target *target) |
5546 | { | |
5547 | struct page *page = NULL; | |
8d32ff84 | 5548 | enum mc_target_type ret = MC_TARGET_NONE; |
90254a65 DN |
5549 | swp_entry_t ent = { .val = 0 }; |
5550 | ||
5551 | if (pte_present(ptent)) | |
5552 | page = mc_handle_present_pte(vma, addr, ptent); | |
5553 | else if (is_swap_pte(ptent)) | |
48406ef8 | 5554 | page = mc_handle_swap_pte(vma, ptent, &ent); |
0661a336 | 5555 | else if (pte_none(ptent)) |
87946a72 | 5556 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
90254a65 DN |
5557 | |
5558 | if (!page && !ent.val) | |
8d32ff84 | 5559 | return ret; |
02491447 | 5560 | if (page) { |
02491447 | 5561 | /* |
0a31bc97 | 5562 | * Do only loose check w/o serialization. |
1306a85a | 5563 | * mem_cgroup_move_account() checks the page is valid or |
0a31bc97 | 5564 | * not under LRU exclusion. |
02491447 | 5565 | */ |
bcfe06bf | 5566 | if (page_memcg(page) == mc.from) { |
02491447 | 5567 | ret = MC_TARGET_PAGE; |
25b2995a | 5568 | if (is_device_private_page(page)) |
c733a828 | 5569 | ret = MC_TARGET_DEVICE; |
02491447 DN |
5570 | if (target) |
5571 | target->page = page; | |
5572 | } | |
5573 | if (!ret || !target) | |
5574 | put_page(page); | |
5575 | } | |
3e14a57b HY |
5576 | /* |
5577 | * There is a swap entry and a page doesn't exist or isn't charged. | |
5578 | * But we cannot move a tail-page in a THP. | |
5579 | */ | |
5580 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && | |
34c00c31 | 5581 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
7f0f1546 KH |
5582 | ret = MC_TARGET_SWAP; |
5583 | if (target) | |
5584 | target->ent = ent; | |
4ffef5fe | 5585 | } |
4ffef5fe DN |
5586 | return ret; |
5587 | } | |
5588 | ||
12724850 NH |
5589 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
5590 | /* | |
d6810d73 HY |
5591 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
5592 | * not support them for now. | |
12724850 NH |
5593 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
5594 | */ | |
5595 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5596 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5597 | { | |
5598 | struct page *page = NULL; | |
12724850 NH |
5599 | enum mc_target_type ret = MC_TARGET_NONE; |
5600 | ||
84c3fc4e ZY |
5601 | if (unlikely(is_swap_pmd(pmd))) { |
5602 | VM_BUG_ON(thp_migration_supported() && | |
5603 | !is_pmd_migration_entry(pmd)); | |
5604 | return ret; | |
5605 | } | |
12724850 | 5606 | page = pmd_page(pmd); |
309381fe | 5607 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
1dfab5ab | 5608 | if (!(mc.flags & MOVE_ANON)) |
12724850 | 5609 | return ret; |
bcfe06bf | 5610 | if (page_memcg(page) == mc.from) { |
12724850 NH |
5611 | ret = MC_TARGET_PAGE; |
5612 | if (target) { | |
5613 | get_page(page); | |
5614 | target->page = page; | |
5615 | } | |
5616 | } | |
5617 | return ret; | |
5618 | } | |
5619 | #else | |
5620 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, | |
5621 | unsigned long addr, pmd_t pmd, union mc_target *target) | |
5622 | { | |
5623 | return MC_TARGET_NONE; | |
5624 | } | |
5625 | #endif | |
5626 | ||
4ffef5fe DN |
5627 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
5628 | unsigned long addr, unsigned long end, | |
5629 | struct mm_walk *walk) | |
5630 | { | |
26bcd64a | 5631 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5632 | pte_t *pte; |
5633 | spinlock_t *ptl; | |
5634 | ||
b6ec57f4 KS |
5635 | ptl = pmd_trans_huge_lock(pmd, vma); |
5636 | if (ptl) { | |
c733a828 JG |
5637 | /* |
5638 | * Note their can not be MC_TARGET_DEVICE for now as we do not | |
25b2995a CH |
5639 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
5640 | * this might change. | |
c733a828 | 5641 | */ |
12724850 NH |
5642 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
5643 | mc.precharge += HPAGE_PMD_NR; | |
bf929152 | 5644 | spin_unlock(ptl); |
1a5a9906 | 5645 | return 0; |
12724850 | 5646 | } |
03319327 | 5647 | |
45f83cef AA |
5648 | if (pmd_trans_unstable(pmd)) |
5649 | return 0; | |
4ffef5fe DN |
5650 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5651 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
8d32ff84 | 5652 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
4ffef5fe DN |
5653 | mc.precharge++; /* increment precharge temporarily */ |
5654 | pte_unmap_unlock(pte - 1, ptl); | |
5655 | cond_resched(); | |
5656 | ||
7dc74be0 DN |
5657 | return 0; |
5658 | } | |
5659 | ||
7b86ac33 CH |
5660 | static const struct mm_walk_ops precharge_walk_ops = { |
5661 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5662 | }; | |
5663 | ||
4ffef5fe DN |
5664 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5665 | { | |
5666 | unsigned long precharge; | |
4ffef5fe | 5667 | |
d8ed45c5 | 5668 | mmap_read_lock(mm); |
7b86ac33 | 5669 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
d8ed45c5 | 5670 | mmap_read_unlock(mm); |
4ffef5fe DN |
5671 | |
5672 | precharge = mc.precharge; | |
5673 | mc.precharge = 0; | |
5674 | ||
5675 | return precharge; | |
5676 | } | |
5677 | ||
4ffef5fe DN |
5678 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5679 | { | |
dfe076b0 DN |
5680 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5681 | ||
5682 | VM_BUG_ON(mc.moving_task); | |
5683 | mc.moving_task = current; | |
5684 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5685 | } |
5686 | ||
dfe076b0 DN |
5687 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5688 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5689 | { |
2bd9bb20 KH |
5690 | struct mem_cgroup *from = mc.from; |
5691 | struct mem_cgroup *to = mc.to; | |
5692 | ||
4ffef5fe | 5693 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d | 5694 | if (mc.precharge) { |
00501b53 | 5695 | cancel_charge(mc.to, mc.precharge); |
854ffa8d DN |
5696 | mc.precharge = 0; |
5697 | } | |
5698 | /* | |
5699 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5700 | * we must uncharge here. | |
5701 | */ | |
5702 | if (mc.moved_charge) { | |
00501b53 | 5703 | cancel_charge(mc.from, mc.moved_charge); |
854ffa8d | 5704 | mc.moved_charge = 0; |
4ffef5fe | 5705 | } |
483c30b5 DN |
5706 | /* we must fixup refcnts and charges */ |
5707 | if (mc.moved_swap) { | |
483c30b5 | 5708 | /* uncharge swap account from the old cgroup */ |
ce00a967 | 5709 | if (!mem_cgroup_is_root(mc.from)) |
3e32cb2e | 5710 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
483c30b5 | 5711 | |
615d66c3 VD |
5712 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
5713 | ||
05b84301 | 5714 | /* |
3e32cb2e JW |
5715 | * we charged both to->memory and to->memsw, so we |
5716 | * should uncharge to->memory. | |
05b84301 | 5717 | */ |
ce00a967 | 5718 | if (!mem_cgroup_is_root(mc.to)) |
3e32cb2e JW |
5719 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
5720 | ||
483c30b5 DN |
5721 | mc.moved_swap = 0; |
5722 | } | |
dfe076b0 DN |
5723 | memcg_oom_recover(from); |
5724 | memcg_oom_recover(to); | |
5725 | wake_up_all(&mc.waitq); | |
5726 | } | |
5727 | ||
5728 | static void mem_cgroup_clear_mc(void) | |
5729 | { | |
264a0ae1 TH |
5730 | struct mm_struct *mm = mc.mm; |
5731 | ||
dfe076b0 DN |
5732 | /* |
5733 | * we must clear moving_task before waking up waiters at the end of | |
5734 | * task migration. | |
5735 | */ | |
5736 | mc.moving_task = NULL; | |
5737 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5738 | spin_lock(&mc.lock); |
4ffef5fe DN |
5739 | mc.from = NULL; |
5740 | mc.to = NULL; | |
264a0ae1 | 5741 | mc.mm = NULL; |
2bd9bb20 | 5742 | spin_unlock(&mc.lock); |
264a0ae1 TH |
5743 | |
5744 | mmput(mm); | |
4ffef5fe DN |
5745 | } |
5746 | ||
1f7dd3e5 | 5747 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5748 | { |
1f7dd3e5 | 5749 | struct cgroup_subsys_state *css; |
eed67d75 | 5750 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
9f2115f9 | 5751 | struct mem_cgroup *from; |
4530eddb | 5752 | struct task_struct *leader, *p; |
9f2115f9 | 5753 | struct mm_struct *mm; |
1dfab5ab | 5754 | unsigned long move_flags; |
9f2115f9 | 5755 | int ret = 0; |
7dc74be0 | 5756 | |
1f7dd3e5 TH |
5757 | /* charge immigration isn't supported on the default hierarchy */ |
5758 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) | |
9f2115f9 TH |
5759 | return 0; |
5760 | ||
4530eddb TH |
5761 | /* |
5762 | * Multi-process migrations only happen on the default hierarchy | |
5763 | * where charge immigration is not used. Perform charge | |
5764 | * immigration if @tset contains a leader and whine if there are | |
5765 | * multiple. | |
5766 | */ | |
5767 | p = NULL; | |
1f7dd3e5 | 5768 | cgroup_taskset_for_each_leader(leader, css, tset) { |
4530eddb TH |
5769 | WARN_ON_ONCE(p); |
5770 | p = leader; | |
1f7dd3e5 | 5771 | memcg = mem_cgroup_from_css(css); |
4530eddb TH |
5772 | } |
5773 | if (!p) | |
5774 | return 0; | |
5775 | ||
1f7dd3e5 TH |
5776 | /* |
5777 | * We are now commited to this value whatever it is. Changes in this | |
5778 | * tunable will only affect upcoming migrations, not the current one. | |
5779 | * So we need to save it, and keep it going. | |
5780 | */ | |
5781 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); | |
5782 | if (!move_flags) | |
5783 | return 0; | |
5784 | ||
9f2115f9 TH |
5785 | from = mem_cgroup_from_task(p); |
5786 | ||
5787 | VM_BUG_ON(from == memcg); | |
5788 | ||
5789 | mm = get_task_mm(p); | |
5790 | if (!mm) | |
5791 | return 0; | |
5792 | /* We move charges only when we move a owner of the mm */ | |
5793 | if (mm->owner == p) { | |
5794 | VM_BUG_ON(mc.from); | |
5795 | VM_BUG_ON(mc.to); | |
5796 | VM_BUG_ON(mc.precharge); | |
5797 | VM_BUG_ON(mc.moved_charge); | |
5798 | VM_BUG_ON(mc.moved_swap); | |
5799 | ||
5800 | spin_lock(&mc.lock); | |
264a0ae1 | 5801 | mc.mm = mm; |
9f2115f9 TH |
5802 | mc.from = from; |
5803 | mc.to = memcg; | |
5804 | mc.flags = move_flags; | |
5805 | spin_unlock(&mc.lock); | |
5806 | /* We set mc.moving_task later */ | |
5807 | ||
5808 | ret = mem_cgroup_precharge_mc(mm); | |
5809 | if (ret) | |
5810 | mem_cgroup_clear_mc(); | |
264a0ae1 TH |
5811 | } else { |
5812 | mmput(mm); | |
7dc74be0 DN |
5813 | } |
5814 | return ret; | |
5815 | } | |
5816 | ||
1f7dd3e5 | 5817 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
7dc74be0 | 5818 | { |
4e2f245d JW |
5819 | if (mc.to) |
5820 | mem_cgroup_clear_mc(); | |
7dc74be0 DN |
5821 | } |
5822 | ||
4ffef5fe DN |
5823 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5824 | unsigned long addr, unsigned long end, | |
5825 | struct mm_walk *walk) | |
7dc74be0 | 5826 | { |
4ffef5fe | 5827 | int ret = 0; |
26bcd64a | 5828 | struct vm_area_struct *vma = walk->vma; |
4ffef5fe DN |
5829 | pte_t *pte; |
5830 | spinlock_t *ptl; | |
12724850 NH |
5831 | enum mc_target_type target_type; |
5832 | union mc_target target; | |
5833 | struct page *page; | |
4ffef5fe | 5834 | |
b6ec57f4 KS |
5835 | ptl = pmd_trans_huge_lock(pmd, vma); |
5836 | if (ptl) { | |
62ade86a | 5837 | if (mc.precharge < HPAGE_PMD_NR) { |
bf929152 | 5838 | spin_unlock(ptl); |
12724850 NH |
5839 | return 0; |
5840 | } | |
5841 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); | |
5842 | if (target_type == MC_TARGET_PAGE) { | |
5843 | page = target.page; | |
5844 | if (!isolate_lru_page(page)) { | |
f627c2f5 | 5845 | if (!mem_cgroup_move_account(page, true, |
1306a85a | 5846 | mc.from, mc.to)) { |
12724850 NH |
5847 | mc.precharge -= HPAGE_PMD_NR; |
5848 | mc.moved_charge += HPAGE_PMD_NR; | |
5849 | } | |
5850 | putback_lru_page(page); | |
5851 | } | |
5852 | put_page(page); | |
c733a828 JG |
5853 | } else if (target_type == MC_TARGET_DEVICE) { |
5854 | page = target.page; | |
5855 | if (!mem_cgroup_move_account(page, true, | |
5856 | mc.from, mc.to)) { | |
5857 | mc.precharge -= HPAGE_PMD_NR; | |
5858 | mc.moved_charge += HPAGE_PMD_NR; | |
5859 | } | |
5860 | put_page(page); | |
12724850 | 5861 | } |
bf929152 | 5862 | spin_unlock(ptl); |
1a5a9906 | 5863 | return 0; |
12724850 NH |
5864 | } |
5865 | ||
45f83cef AA |
5866 | if (pmd_trans_unstable(pmd)) |
5867 | return 0; | |
4ffef5fe DN |
5868 | retry: |
5869 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5870 | for (; addr != end; addr += PAGE_SIZE) { | |
5871 | pte_t ptent = *(pte++); | |
c733a828 | 5872 | bool device = false; |
02491447 | 5873 | swp_entry_t ent; |
4ffef5fe DN |
5874 | |
5875 | if (!mc.precharge) | |
5876 | break; | |
5877 | ||
8d32ff84 | 5878 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
c733a828 JG |
5879 | case MC_TARGET_DEVICE: |
5880 | device = true; | |
e4a9bc58 | 5881 | fallthrough; |
4ffef5fe DN |
5882 | case MC_TARGET_PAGE: |
5883 | page = target.page; | |
53f9263b KS |
5884 | /* |
5885 | * We can have a part of the split pmd here. Moving it | |
5886 | * can be done but it would be too convoluted so simply | |
5887 | * ignore such a partial THP and keep it in original | |
5888 | * memcg. There should be somebody mapping the head. | |
5889 | */ | |
5890 | if (PageTransCompound(page)) | |
5891 | goto put; | |
c733a828 | 5892 | if (!device && isolate_lru_page(page)) |
4ffef5fe | 5893 | goto put; |
f627c2f5 KS |
5894 | if (!mem_cgroup_move_account(page, false, |
5895 | mc.from, mc.to)) { | |
4ffef5fe | 5896 | mc.precharge--; |
854ffa8d DN |
5897 | /* we uncharge from mc.from later. */ |
5898 | mc.moved_charge++; | |
4ffef5fe | 5899 | } |
c733a828 JG |
5900 | if (!device) |
5901 | putback_lru_page(page); | |
8d32ff84 | 5902 | put: /* get_mctgt_type() gets the page */ |
4ffef5fe DN |
5903 | put_page(page); |
5904 | break; | |
02491447 DN |
5905 | case MC_TARGET_SWAP: |
5906 | ent = target.ent; | |
e91cbb42 | 5907 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
02491447 | 5908 | mc.precharge--; |
8d22a935 HD |
5909 | mem_cgroup_id_get_many(mc.to, 1); |
5910 | /* we fixup other refcnts and charges later. */ | |
483c30b5 DN |
5911 | mc.moved_swap++; |
5912 | } | |
02491447 | 5913 | break; |
4ffef5fe DN |
5914 | default: |
5915 | break; | |
5916 | } | |
5917 | } | |
5918 | pte_unmap_unlock(pte - 1, ptl); | |
5919 | cond_resched(); | |
5920 | ||
5921 | if (addr != end) { | |
5922 | /* | |
5923 | * We have consumed all precharges we got in can_attach(). | |
5924 | * We try charge one by one, but don't do any additional | |
5925 | * charges to mc.to if we have failed in charge once in attach() | |
5926 | * phase. | |
5927 | */ | |
854ffa8d | 5928 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5929 | if (!ret) |
5930 | goto retry; | |
5931 | } | |
5932 | ||
5933 | return ret; | |
5934 | } | |
5935 | ||
7b86ac33 CH |
5936 | static const struct mm_walk_ops charge_walk_ops = { |
5937 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5938 | }; | |
5939 | ||
264a0ae1 | 5940 | static void mem_cgroup_move_charge(void) |
4ffef5fe | 5941 | { |
4ffef5fe | 5942 | lru_add_drain_all(); |
312722cb | 5943 | /* |
81f8c3a4 JW |
5944 | * Signal lock_page_memcg() to take the memcg's move_lock |
5945 | * while we're moving its pages to another memcg. Then wait | |
5946 | * for already started RCU-only updates to finish. | |
312722cb JW |
5947 | */ |
5948 | atomic_inc(&mc.from->moving_account); | |
5949 | synchronize_rcu(); | |
dfe076b0 | 5950 | retry: |
d8ed45c5 | 5951 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
dfe076b0 | 5952 | /* |
c1e8d7c6 | 5953 | * Someone who are holding the mmap_lock might be waiting in |
dfe076b0 DN |
5954 | * waitq. So we cancel all extra charges, wake up all waiters, |
5955 | * and retry. Because we cancel precharges, we might not be able | |
5956 | * to move enough charges, but moving charge is a best-effort | |
5957 | * feature anyway, so it wouldn't be a big problem. | |
5958 | */ | |
5959 | __mem_cgroup_clear_mc(); | |
5960 | cond_resched(); | |
5961 | goto retry; | |
5962 | } | |
26bcd64a NH |
5963 | /* |
5964 | * When we have consumed all precharges and failed in doing | |
5965 | * additional charge, the page walk just aborts. | |
5966 | */ | |
7b86ac33 CH |
5967 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
5968 | NULL); | |
0247f3f4 | 5969 | |
d8ed45c5 | 5970 | mmap_read_unlock(mc.mm); |
312722cb | 5971 | atomic_dec(&mc.from->moving_account); |
7dc74be0 DN |
5972 | } |
5973 | ||
264a0ae1 | 5974 | static void mem_cgroup_move_task(void) |
67e465a7 | 5975 | { |
264a0ae1 TH |
5976 | if (mc.to) { |
5977 | mem_cgroup_move_charge(); | |
a433658c | 5978 | mem_cgroup_clear_mc(); |
264a0ae1 | 5979 | } |
67e465a7 | 5980 | } |
5cfb80a7 | 5981 | #else /* !CONFIG_MMU */ |
1f7dd3e5 | 5982 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5983 | { |
5984 | return 0; | |
5985 | } | |
1f7dd3e5 | 5986 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
5cfb80a7 DN |
5987 | { |
5988 | } | |
264a0ae1 | 5989 | static void mem_cgroup_move_task(void) |
5cfb80a7 DN |
5990 | { |
5991 | } | |
5992 | #endif | |
67e465a7 | 5993 | |
677dc973 CD |
5994 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
5995 | { | |
5996 | if (value == PAGE_COUNTER_MAX) | |
5997 | seq_puts(m, "max\n"); | |
5998 | else | |
5999 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); | |
6000 | ||
6001 | return 0; | |
6002 | } | |
6003 | ||
241994ed JW |
6004 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
6005 | struct cftype *cft) | |
6006 | { | |
f5fc3c5d JW |
6007 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
6008 | ||
6009 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; | |
241994ed JW |
6010 | } |
6011 | ||
bf8d5d52 RG |
6012 | static int memory_min_show(struct seq_file *m, void *v) |
6013 | { | |
677dc973 CD |
6014 | return seq_puts_memcg_tunable(m, |
6015 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); | |
bf8d5d52 RG |
6016 | } |
6017 | ||
6018 | static ssize_t memory_min_write(struct kernfs_open_file *of, | |
6019 | char *buf, size_t nbytes, loff_t off) | |
6020 | { | |
6021 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6022 | unsigned long min; | |
6023 | int err; | |
6024 | ||
6025 | buf = strstrip(buf); | |
6026 | err = page_counter_memparse(buf, "max", &min); | |
6027 | if (err) | |
6028 | return err; | |
6029 | ||
6030 | page_counter_set_min(&memcg->memory, min); | |
6031 | ||
6032 | return nbytes; | |
6033 | } | |
6034 | ||
241994ed JW |
6035 | static int memory_low_show(struct seq_file *m, void *v) |
6036 | { | |
677dc973 CD |
6037 | return seq_puts_memcg_tunable(m, |
6038 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); | |
241994ed JW |
6039 | } |
6040 | ||
6041 | static ssize_t memory_low_write(struct kernfs_open_file *of, | |
6042 | char *buf, size_t nbytes, loff_t off) | |
6043 | { | |
6044 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6045 | unsigned long low; | |
6046 | int err; | |
6047 | ||
6048 | buf = strstrip(buf); | |
d2973697 | 6049 | err = page_counter_memparse(buf, "max", &low); |
241994ed JW |
6050 | if (err) |
6051 | return err; | |
6052 | ||
23067153 | 6053 | page_counter_set_low(&memcg->memory, low); |
241994ed JW |
6054 | |
6055 | return nbytes; | |
6056 | } | |
6057 | ||
6058 | static int memory_high_show(struct seq_file *m, void *v) | |
6059 | { | |
d1663a90 JK |
6060 | return seq_puts_memcg_tunable(m, |
6061 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); | |
241994ed JW |
6062 | } |
6063 | ||
6064 | static ssize_t memory_high_write(struct kernfs_open_file *of, | |
6065 | char *buf, size_t nbytes, loff_t off) | |
6066 | { | |
6067 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6068 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
8c8c383c | 6069 | bool drained = false; |
241994ed JW |
6070 | unsigned long high; |
6071 | int err; | |
6072 | ||
6073 | buf = strstrip(buf); | |
d2973697 | 6074 | err = page_counter_memparse(buf, "max", &high); |
241994ed JW |
6075 | if (err) |
6076 | return err; | |
6077 | ||
e82553c1 JW |
6078 | page_counter_set_high(&memcg->memory, high); |
6079 | ||
8c8c383c JW |
6080 | for (;;) { |
6081 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6082 | unsigned long reclaimed; | |
6083 | ||
6084 | if (nr_pages <= high) | |
6085 | break; | |
6086 | ||
6087 | if (signal_pending(current)) | |
6088 | break; | |
6089 | ||
6090 | if (!drained) { | |
6091 | drain_all_stock(memcg); | |
6092 | drained = true; | |
6093 | continue; | |
6094 | } | |
6095 | ||
6096 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, | |
6097 | GFP_KERNEL, true); | |
6098 | ||
6099 | if (!reclaimed && !nr_retries--) | |
6100 | break; | |
6101 | } | |
588083bb | 6102 | |
19ce33ac | 6103 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6104 | return nbytes; |
6105 | } | |
6106 | ||
6107 | static int memory_max_show(struct seq_file *m, void *v) | |
6108 | { | |
677dc973 CD |
6109 | return seq_puts_memcg_tunable(m, |
6110 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); | |
241994ed JW |
6111 | } |
6112 | ||
6113 | static ssize_t memory_max_write(struct kernfs_open_file *of, | |
6114 | char *buf, size_t nbytes, loff_t off) | |
6115 | { | |
6116 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
d977aa93 | 6117 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
b6e6edcf | 6118 | bool drained = false; |
241994ed JW |
6119 | unsigned long max; |
6120 | int err; | |
6121 | ||
6122 | buf = strstrip(buf); | |
d2973697 | 6123 | err = page_counter_memparse(buf, "max", &max); |
241994ed JW |
6124 | if (err) |
6125 | return err; | |
6126 | ||
bbec2e15 | 6127 | xchg(&memcg->memory.max, max); |
b6e6edcf JW |
6128 | |
6129 | for (;;) { | |
6130 | unsigned long nr_pages = page_counter_read(&memcg->memory); | |
6131 | ||
6132 | if (nr_pages <= max) | |
6133 | break; | |
6134 | ||
7249c9f0 | 6135 | if (signal_pending(current)) |
b6e6edcf | 6136 | break; |
b6e6edcf JW |
6137 | |
6138 | if (!drained) { | |
6139 | drain_all_stock(memcg); | |
6140 | drained = true; | |
6141 | continue; | |
6142 | } | |
6143 | ||
6144 | if (nr_reclaims) { | |
6145 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, | |
6146 | GFP_KERNEL, true)) | |
6147 | nr_reclaims--; | |
6148 | continue; | |
6149 | } | |
6150 | ||
e27be240 | 6151 | memcg_memory_event(memcg, MEMCG_OOM); |
b6e6edcf JW |
6152 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
6153 | break; | |
6154 | } | |
241994ed | 6155 | |
2529bb3a | 6156 | memcg_wb_domain_size_changed(memcg); |
241994ed JW |
6157 | return nbytes; |
6158 | } | |
6159 | ||
1e577f97 SB |
6160 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
6161 | { | |
6162 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); | |
6163 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); | |
6164 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); | |
6165 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); | |
6166 | seq_printf(m, "oom_kill %lu\n", | |
6167 | atomic_long_read(&events[MEMCG_OOM_KILL])); | |
6168 | } | |
6169 | ||
241994ed JW |
6170 | static int memory_events_show(struct seq_file *m, void *v) |
6171 | { | |
aa9694bb | 6172 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
241994ed | 6173 | |
1e577f97 SB |
6174 | __memory_events_show(m, memcg->memory_events); |
6175 | return 0; | |
6176 | } | |
6177 | ||
6178 | static int memory_events_local_show(struct seq_file *m, void *v) | |
6179 | { | |
6180 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
241994ed | 6181 | |
1e577f97 | 6182 | __memory_events_show(m, memcg->memory_events_local); |
241994ed JW |
6183 | return 0; |
6184 | } | |
6185 | ||
587d9f72 JW |
6186 | static int memory_stat_show(struct seq_file *m, void *v) |
6187 | { | |
aa9694bb | 6188 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
c8713d0b | 6189 | char *buf; |
1ff9e6e1 | 6190 | |
c8713d0b JW |
6191 | buf = memory_stat_format(memcg); |
6192 | if (!buf) | |
6193 | return -ENOMEM; | |
6194 | seq_puts(m, buf); | |
6195 | kfree(buf); | |
587d9f72 JW |
6196 | return 0; |
6197 | } | |
6198 | ||
5f9a4f4a | 6199 | #ifdef CONFIG_NUMA |
fff66b79 MS |
6200 | static inline unsigned long lruvec_page_state_output(struct lruvec *lruvec, |
6201 | int item) | |
6202 | { | |
6203 | return lruvec_page_state(lruvec, item) * memcg_page_state_unit(item); | |
6204 | } | |
6205 | ||
5f9a4f4a MS |
6206 | static int memory_numa_stat_show(struct seq_file *m, void *v) |
6207 | { | |
6208 | int i; | |
6209 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); | |
6210 | ||
6211 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { | |
6212 | int nid; | |
6213 | ||
6214 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) | |
6215 | continue; | |
6216 | ||
6217 | seq_printf(m, "%s", memory_stats[i].name); | |
6218 | for_each_node_state(nid, N_MEMORY) { | |
6219 | u64 size; | |
6220 | struct lruvec *lruvec; | |
6221 | ||
6222 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); | |
fff66b79 MS |
6223 | size = lruvec_page_state_output(lruvec, |
6224 | memory_stats[i].idx); | |
5f9a4f4a MS |
6225 | seq_printf(m, " N%d=%llu", nid, size); |
6226 | } | |
6227 | seq_putc(m, '\n'); | |
6228 | } | |
6229 | ||
6230 | return 0; | |
6231 | } | |
6232 | #endif | |
6233 | ||
3d8b38eb RG |
6234 | static int memory_oom_group_show(struct seq_file *m, void *v) |
6235 | { | |
aa9694bb | 6236 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
3d8b38eb RG |
6237 | |
6238 | seq_printf(m, "%d\n", memcg->oom_group); | |
6239 | ||
6240 | return 0; | |
6241 | } | |
6242 | ||
6243 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, | |
6244 | char *buf, size_t nbytes, loff_t off) | |
6245 | { | |
6246 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
6247 | int ret, oom_group; | |
6248 | ||
6249 | buf = strstrip(buf); | |
6250 | if (!buf) | |
6251 | return -EINVAL; | |
6252 | ||
6253 | ret = kstrtoint(buf, 0, &oom_group); | |
6254 | if (ret) | |
6255 | return ret; | |
6256 | ||
6257 | if (oom_group != 0 && oom_group != 1) | |
6258 | return -EINVAL; | |
6259 | ||
6260 | memcg->oom_group = oom_group; | |
6261 | ||
6262 | return nbytes; | |
6263 | } | |
6264 | ||
241994ed JW |
6265 | static struct cftype memory_files[] = { |
6266 | { | |
6267 | .name = "current", | |
f5fc3c5d | 6268 | .flags = CFTYPE_NOT_ON_ROOT, |
241994ed JW |
6269 | .read_u64 = memory_current_read, |
6270 | }, | |
bf8d5d52 RG |
6271 | { |
6272 | .name = "min", | |
6273 | .flags = CFTYPE_NOT_ON_ROOT, | |
6274 | .seq_show = memory_min_show, | |
6275 | .write = memory_min_write, | |
6276 | }, | |
241994ed JW |
6277 | { |
6278 | .name = "low", | |
6279 | .flags = CFTYPE_NOT_ON_ROOT, | |
6280 | .seq_show = memory_low_show, | |
6281 | .write = memory_low_write, | |
6282 | }, | |
6283 | { | |
6284 | .name = "high", | |
6285 | .flags = CFTYPE_NOT_ON_ROOT, | |
6286 | .seq_show = memory_high_show, | |
6287 | .write = memory_high_write, | |
6288 | }, | |
6289 | { | |
6290 | .name = "max", | |
6291 | .flags = CFTYPE_NOT_ON_ROOT, | |
6292 | .seq_show = memory_max_show, | |
6293 | .write = memory_max_write, | |
6294 | }, | |
6295 | { | |
6296 | .name = "events", | |
6297 | .flags = CFTYPE_NOT_ON_ROOT, | |
472912a2 | 6298 | .file_offset = offsetof(struct mem_cgroup, events_file), |
241994ed JW |
6299 | .seq_show = memory_events_show, |
6300 | }, | |
1e577f97 SB |
6301 | { |
6302 | .name = "events.local", | |
6303 | .flags = CFTYPE_NOT_ON_ROOT, | |
6304 | .file_offset = offsetof(struct mem_cgroup, events_local_file), | |
6305 | .seq_show = memory_events_local_show, | |
6306 | }, | |
587d9f72 JW |
6307 | { |
6308 | .name = "stat", | |
587d9f72 JW |
6309 | .seq_show = memory_stat_show, |
6310 | }, | |
5f9a4f4a MS |
6311 | #ifdef CONFIG_NUMA |
6312 | { | |
6313 | .name = "numa_stat", | |
6314 | .seq_show = memory_numa_stat_show, | |
6315 | }, | |
6316 | #endif | |
3d8b38eb RG |
6317 | { |
6318 | .name = "oom.group", | |
6319 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, | |
6320 | .seq_show = memory_oom_group_show, | |
6321 | .write = memory_oom_group_write, | |
6322 | }, | |
241994ed JW |
6323 | { } /* terminate */ |
6324 | }; | |
6325 | ||
073219e9 | 6326 | struct cgroup_subsys memory_cgrp_subsys = { |
92fb9748 | 6327 | .css_alloc = mem_cgroup_css_alloc, |
d142e3e6 | 6328 | .css_online = mem_cgroup_css_online, |
92fb9748 | 6329 | .css_offline = mem_cgroup_css_offline, |
6df38689 | 6330 | .css_released = mem_cgroup_css_released, |
92fb9748 | 6331 | .css_free = mem_cgroup_css_free, |
1ced953b | 6332 | .css_reset = mem_cgroup_css_reset, |
2d146aa3 | 6333 | .css_rstat_flush = mem_cgroup_css_rstat_flush, |
7dc74be0 DN |
6334 | .can_attach = mem_cgroup_can_attach, |
6335 | .cancel_attach = mem_cgroup_cancel_attach, | |
264a0ae1 | 6336 | .post_attach = mem_cgroup_move_task, |
241994ed JW |
6337 | .dfl_cftypes = memory_files, |
6338 | .legacy_cftypes = mem_cgroup_legacy_files, | |
6d12e2d8 | 6339 | .early_init = 0, |
8cdea7c0 | 6340 | }; |
c077719b | 6341 | |
bc50bcc6 JW |
6342 | /* |
6343 | * This function calculates an individual cgroup's effective | |
6344 | * protection which is derived from its own memory.min/low, its | |
6345 | * parent's and siblings' settings, as well as the actual memory | |
6346 | * distribution in the tree. | |
6347 | * | |
6348 | * The following rules apply to the effective protection values: | |
6349 | * | |
6350 | * 1. At the first level of reclaim, effective protection is equal to | |
6351 | * the declared protection in memory.min and memory.low. | |
6352 | * | |
6353 | * 2. To enable safe delegation of the protection configuration, at | |
6354 | * subsequent levels the effective protection is capped to the | |
6355 | * parent's effective protection. | |
6356 | * | |
6357 | * 3. To make complex and dynamic subtrees easier to configure, the | |
6358 | * user is allowed to overcommit the declared protection at a given | |
6359 | * level. If that is the case, the parent's effective protection is | |
6360 | * distributed to the children in proportion to how much protection | |
6361 | * they have declared and how much of it they are utilizing. | |
6362 | * | |
6363 | * This makes distribution proportional, but also work-conserving: | |
6364 | * if one cgroup claims much more protection than it uses memory, | |
6365 | * the unused remainder is available to its siblings. | |
6366 | * | |
6367 | * 4. Conversely, when the declared protection is undercommitted at a | |
6368 | * given level, the distribution of the larger parental protection | |
6369 | * budget is NOT proportional. A cgroup's protection from a sibling | |
6370 | * is capped to its own memory.min/low setting. | |
6371 | * | |
8a931f80 JW |
6372 | * 5. However, to allow protecting recursive subtrees from each other |
6373 | * without having to declare each individual cgroup's fixed share | |
6374 | * of the ancestor's claim to protection, any unutilized - | |
6375 | * "floating" - protection from up the tree is distributed in | |
6376 | * proportion to each cgroup's *usage*. This makes the protection | |
6377 | * neutral wrt sibling cgroups and lets them compete freely over | |
6378 | * the shared parental protection budget, but it protects the | |
6379 | * subtree as a whole from neighboring subtrees. | |
6380 | * | |
6381 | * Note that 4. and 5. are not in conflict: 4. is about protecting | |
6382 | * against immediate siblings whereas 5. is about protecting against | |
6383 | * neighboring subtrees. | |
bc50bcc6 JW |
6384 | */ |
6385 | static unsigned long effective_protection(unsigned long usage, | |
8a931f80 | 6386 | unsigned long parent_usage, |
bc50bcc6 JW |
6387 | unsigned long setting, |
6388 | unsigned long parent_effective, | |
6389 | unsigned long siblings_protected) | |
6390 | { | |
6391 | unsigned long protected; | |
8a931f80 | 6392 | unsigned long ep; |
bc50bcc6 JW |
6393 | |
6394 | protected = min(usage, setting); | |
6395 | /* | |
6396 | * If all cgroups at this level combined claim and use more | |
6397 | * protection then what the parent affords them, distribute | |
6398 | * shares in proportion to utilization. | |
6399 | * | |
6400 | * We are using actual utilization rather than the statically | |
6401 | * claimed protection in order to be work-conserving: claimed | |
6402 | * but unused protection is available to siblings that would | |
6403 | * otherwise get a smaller chunk than what they claimed. | |
6404 | */ | |
6405 | if (siblings_protected > parent_effective) | |
6406 | return protected * parent_effective / siblings_protected; | |
6407 | ||
6408 | /* | |
6409 | * Ok, utilized protection of all children is within what the | |
6410 | * parent affords them, so we know whatever this child claims | |
6411 | * and utilizes is effectively protected. | |
6412 | * | |
6413 | * If there is unprotected usage beyond this value, reclaim | |
6414 | * will apply pressure in proportion to that amount. | |
6415 | * | |
6416 | * If there is unutilized protection, the cgroup will be fully | |
6417 | * shielded from reclaim, but we do return a smaller value for | |
6418 | * protection than what the group could enjoy in theory. This | |
6419 | * is okay. With the overcommit distribution above, effective | |
6420 | * protection is always dependent on how memory is actually | |
6421 | * consumed among the siblings anyway. | |
6422 | */ | |
8a931f80 JW |
6423 | ep = protected; |
6424 | ||
6425 | /* | |
6426 | * If the children aren't claiming (all of) the protection | |
6427 | * afforded to them by the parent, distribute the remainder in | |
6428 | * proportion to the (unprotected) memory of each cgroup. That | |
6429 | * way, cgroups that aren't explicitly prioritized wrt each | |
6430 | * other compete freely over the allowance, but they are | |
6431 | * collectively protected from neighboring trees. | |
6432 | * | |
6433 | * We're using unprotected memory for the weight so that if | |
6434 | * some cgroups DO claim explicit protection, we don't protect | |
6435 | * the same bytes twice. | |
cd324edc JW |
6436 | * |
6437 | * Check both usage and parent_usage against the respective | |
6438 | * protected values. One should imply the other, but they | |
6439 | * aren't read atomically - make sure the division is sane. | |
8a931f80 JW |
6440 | */ |
6441 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) | |
6442 | return ep; | |
cd324edc JW |
6443 | if (parent_effective > siblings_protected && |
6444 | parent_usage > siblings_protected && | |
6445 | usage > protected) { | |
8a931f80 JW |
6446 | unsigned long unclaimed; |
6447 | ||
6448 | unclaimed = parent_effective - siblings_protected; | |
6449 | unclaimed *= usage - protected; | |
6450 | unclaimed /= parent_usage - siblings_protected; | |
6451 | ||
6452 | ep += unclaimed; | |
6453 | } | |
6454 | ||
6455 | return ep; | |
bc50bcc6 JW |
6456 | } |
6457 | ||
241994ed | 6458 | /** |
bf8d5d52 | 6459 | * mem_cgroup_protected - check if memory consumption is in the normal range |
34c81057 | 6460 | * @root: the top ancestor of the sub-tree being checked |
241994ed JW |
6461 | * @memcg: the memory cgroup to check |
6462 | * | |
23067153 RG |
6463 | * WARNING: This function is not stateless! It can only be used as part |
6464 | * of a top-down tree iteration, not for isolated queries. | |
241994ed | 6465 | */ |
45c7f7e1 CD |
6466 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
6467 | struct mem_cgroup *memcg) | |
241994ed | 6468 | { |
8a931f80 | 6469 | unsigned long usage, parent_usage; |
23067153 RG |
6470 | struct mem_cgroup *parent; |
6471 | ||
241994ed | 6472 | if (mem_cgroup_disabled()) |
45c7f7e1 | 6473 | return; |
241994ed | 6474 | |
34c81057 SC |
6475 | if (!root) |
6476 | root = root_mem_cgroup; | |
22f7496f YS |
6477 | |
6478 | /* | |
6479 | * Effective values of the reclaim targets are ignored so they | |
6480 | * can be stale. Have a look at mem_cgroup_protection for more | |
6481 | * details. | |
6482 | * TODO: calculation should be more robust so that we do not need | |
6483 | * that special casing. | |
6484 | */ | |
34c81057 | 6485 | if (memcg == root) |
45c7f7e1 | 6486 | return; |
241994ed | 6487 | |
23067153 | 6488 | usage = page_counter_read(&memcg->memory); |
bf8d5d52 | 6489 | if (!usage) |
45c7f7e1 | 6490 | return; |
bf8d5d52 | 6491 | |
bf8d5d52 | 6492 | parent = parent_mem_cgroup(memcg); |
df2a4196 RG |
6493 | /* No parent means a non-hierarchical mode on v1 memcg */ |
6494 | if (!parent) | |
45c7f7e1 | 6495 | return; |
df2a4196 | 6496 | |
bc50bcc6 | 6497 | if (parent == root) { |
c3d53200 | 6498 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
03960e33 | 6499 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
45c7f7e1 | 6500 | return; |
bf8d5d52 RG |
6501 | } |
6502 | ||
8a931f80 JW |
6503 | parent_usage = page_counter_read(&parent->memory); |
6504 | ||
b3a7822e | 6505 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
c3d53200 CD |
6506 | READ_ONCE(memcg->memory.min), |
6507 | READ_ONCE(parent->memory.emin), | |
b3a7822e | 6508 | atomic_long_read(&parent->memory.children_min_usage))); |
23067153 | 6509 | |
b3a7822e | 6510 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
03960e33 CD |
6511 | READ_ONCE(memcg->memory.low), |
6512 | READ_ONCE(parent->memory.elow), | |
b3a7822e | 6513 | atomic_long_read(&parent->memory.children_low_usage))); |
241994ed JW |
6514 | } |
6515 | ||
0add0c77 SB |
6516 | static int __mem_cgroup_charge(struct page *page, struct mem_cgroup *memcg, |
6517 | gfp_t gfp) | |
6518 | { | |
6519 | unsigned int nr_pages = thp_nr_pages(page); | |
6520 | int ret; | |
6521 | ||
6522 | ret = try_charge(memcg, gfp, nr_pages); | |
6523 | if (ret) | |
6524 | goto out; | |
6525 | ||
6526 | css_get(&memcg->css); | |
6527 | commit_charge(page, memcg); | |
6528 | ||
6529 | local_irq_disable(); | |
6530 | mem_cgroup_charge_statistics(memcg, page, nr_pages); | |
6531 | memcg_check_events(memcg, page); | |
6532 | local_irq_enable(); | |
6533 | out: | |
6534 | return ret; | |
6535 | } | |
6536 | ||
00501b53 | 6537 | /** |
f0e45fb4 | 6538 | * mem_cgroup_charge - charge a newly allocated page to a cgroup |
00501b53 JW |
6539 | * @page: page to charge |
6540 | * @mm: mm context of the victim | |
6541 | * @gfp_mask: reclaim mode | |
00501b53 JW |
6542 | * |
6543 | * Try to charge @page to the memcg that @mm belongs to, reclaiming | |
6544 | * pages according to @gfp_mask if necessary. | |
6545 | * | |
0add0c77 SB |
6546 | * Do not use this for pages allocated for swapin. |
6547 | * | |
f0e45fb4 | 6548 | * Returns 0 on success. Otherwise, an error code is returned. |
00501b53 | 6549 | */ |
d9eb1ea2 | 6550 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
00501b53 | 6551 | { |
0add0c77 SB |
6552 | struct mem_cgroup *memcg; |
6553 | int ret; | |
00501b53 JW |
6554 | |
6555 | if (mem_cgroup_disabled()) | |
0add0c77 | 6556 | return 0; |
00501b53 | 6557 | |
0add0c77 SB |
6558 | memcg = get_mem_cgroup_from_mm(mm); |
6559 | ret = __mem_cgroup_charge(page, memcg, gfp_mask); | |
6560 | css_put(&memcg->css); | |
2d1c4980 | 6561 | |
0add0c77 SB |
6562 | return ret; |
6563 | } | |
e993d905 | 6564 | |
0add0c77 SB |
6565 | /** |
6566 | * mem_cgroup_swapin_charge_page - charge a newly allocated page for swapin | |
6567 | * @page: page to charge | |
6568 | * @mm: mm context of the victim | |
6569 | * @gfp: reclaim mode | |
6570 | * @entry: swap entry for which the page is allocated | |
6571 | * | |
6572 | * This function charges a page allocated for swapin. Please call this before | |
6573 | * adding the page to the swapcache. | |
6574 | * | |
6575 | * Returns 0 on success. Otherwise, an error code is returned. | |
6576 | */ | |
6577 | int mem_cgroup_swapin_charge_page(struct page *page, struct mm_struct *mm, | |
6578 | gfp_t gfp, swp_entry_t entry) | |
6579 | { | |
6580 | struct mem_cgroup *memcg; | |
6581 | unsigned short id; | |
6582 | int ret; | |
00501b53 | 6583 | |
0add0c77 SB |
6584 | if (mem_cgroup_disabled()) |
6585 | return 0; | |
00501b53 | 6586 | |
0add0c77 SB |
6587 | id = lookup_swap_cgroup_id(entry); |
6588 | rcu_read_lock(); | |
6589 | memcg = mem_cgroup_from_id(id); | |
6590 | if (!memcg || !css_tryget_online(&memcg->css)) | |
6591 | memcg = get_mem_cgroup_from_mm(mm); | |
6592 | rcu_read_unlock(); | |
00501b53 | 6593 | |
0add0c77 | 6594 | ret = __mem_cgroup_charge(page, memcg, gfp); |
6abb5a86 | 6595 | |
0add0c77 SB |
6596 | css_put(&memcg->css); |
6597 | return ret; | |
6598 | } | |
00501b53 | 6599 | |
0add0c77 SB |
6600 | /* |
6601 | * mem_cgroup_swapin_uncharge_swap - uncharge swap slot | |
6602 | * @entry: swap entry for which the page is charged | |
6603 | * | |
6604 | * Call this function after successfully adding the charged page to swapcache. | |
6605 | * | |
6606 | * Note: This function assumes the page for which swap slot is being uncharged | |
6607 | * is order 0 page. | |
6608 | */ | |
6609 | void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry) | |
6610 | { | |
cae3af62 MS |
6611 | /* |
6612 | * Cgroup1's unified memory+swap counter has been charged with the | |
6613 | * new swapcache page, finish the transfer by uncharging the swap | |
6614 | * slot. The swap slot would also get uncharged when it dies, but | |
6615 | * it can stick around indefinitely and we'd count the page twice | |
6616 | * the entire time. | |
6617 | * | |
6618 | * Cgroup2 has separate resource counters for memory and swap, | |
6619 | * so this is a non-issue here. Memory and swap charge lifetimes | |
6620 | * correspond 1:1 to page and swap slot lifetimes: we charge the | |
6621 | * page to memory here, and uncharge swap when the slot is freed. | |
6622 | */ | |
0add0c77 | 6623 | if (!mem_cgroup_disabled() && do_memsw_account()) { |
00501b53 JW |
6624 | /* |
6625 | * The swap entry might not get freed for a long time, | |
6626 | * let's not wait for it. The page already received a | |
6627 | * memory+swap charge, drop the swap entry duplicate. | |
6628 | */ | |
0add0c77 | 6629 | mem_cgroup_uncharge_swap(entry, 1); |
00501b53 | 6630 | } |
3fea5a49 JW |
6631 | } |
6632 | ||
a9d5adee JG |
6633 | struct uncharge_gather { |
6634 | struct mem_cgroup *memcg; | |
b4e0b68f | 6635 | unsigned long nr_memory; |
a9d5adee | 6636 | unsigned long pgpgout; |
a9d5adee | 6637 | unsigned long nr_kmem; |
a9d5adee JG |
6638 | struct page *dummy_page; |
6639 | }; | |
6640 | ||
6641 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) | |
747db954 | 6642 | { |
a9d5adee JG |
6643 | memset(ug, 0, sizeof(*ug)); |
6644 | } | |
6645 | ||
6646 | static void uncharge_batch(const struct uncharge_gather *ug) | |
6647 | { | |
747db954 JW |
6648 | unsigned long flags; |
6649 | ||
b4e0b68f MS |
6650 | if (ug->nr_memory) { |
6651 | page_counter_uncharge(&ug->memcg->memory, ug->nr_memory); | |
7941d214 | 6652 | if (do_memsw_account()) |
b4e0b68f | 6653 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_memory); |
a9d5adee JG |
6654 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) |
6655 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); | |
6656 | memcg_oom_recover(ug->memcg); | |
ce00a967 | 6657 | } |
747db954 JW |
6658 | |
6659 | local_irq_save(flags); | |
c9019e9b | 6660 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
b4e0b68f | 6661 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_memory); |
a9d5adee | 6662 | memcg_check_events(ug->memcg, ug->dummy_page); |
747db954 | 6663 | local_irq_restore(flags); |
f1796544 MH |
6664 | |
6665 | /* drop reference from uncharge_page */ | |
6666 | css_put(&ug->memcg->css); | |
a9d5adee JG |
6667 | } |
6668 | ||
6669 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) | |
6670 | { | |
9f762dbe | 6671 | unsigned long nr_pages; |
b4e0b68f MS |
6672 | struct mem_cgroup *memcg; |
6673 | struct obj_cgroup *objcg; | |
9f762dbe | 6674 | |
a9d5adee | 6675 | VM_BUG_ON_PAGE(PageLRU(page), page); |
a9d5adee | 6676 | |
a9d5adee JG |
6677 | /* |
6678 | * Nobody should be changing or seriously looking at | |
b4e0b68f | 6679 | * page memcg or objcg at this point, we have fully |
a9d5adee JG |
6680 | * exclusive access to the page. |
6681 | */ | |
b4e0b68f MS |
6682 | if (PageMemcgKmem(page)) { |
6683 | objcg = __page_objcg(page); | |
6684 | /* | |
6685 | * This get matches the put at the end of the function and | |
6686 | * kmem pages do not hold memcg references anymore. | |
6687 | */ | |
6688 | memcg = get_mem_cgroup_from_objcg(objcg); | |
6689 | } else { | |
6690 | memcg = __page_memcg(page); | |
6691 | } | |
a9d5adee | 6692 | |
b4e0b68f MS |
6693 | if (!memcg) |
6694 | return; | |
6695 | ||
6696 | if (ug->memcg != memcg) { | |
a9d5adee JG |
6697 | if (ug->memcg) { |
6698 | uncharge_batch(ug); | |
6699 | uncharge_gather_clear(ug); | |
6700 | } | |
b4e0b68f | 6701 | ug->memcg = memcg; |
7ab345a8 | 6702 | ug->dummy_page = page; |
f1796544 MH |
6703 | |
6704 | /* pairs with css_put in uncharge_batch */ | |
b4e0b68f | 6705 | css_get(&memcg->css); |
a9d5adee JG |
6706 | } |
6707 | ||
9f762dbe | 6708 | nr_pages = compound_nr(page); |
a9d5adee | 6709 | |
b4e0b68f MS |
6710 | if (PageMemcgKmem(page)) { |
6711 | ug->nr_memory += nr_pages; | |
9f762dbe | 6712 | ug->nr_kmem += nr_pages; |
b4e0b68f MS |
6713 | |
6714 | page->memcg_data = 0; | |
6715 | obj_cgroup_put(objcg); | |
6716 | } else { | |
6717 | /* LRU pages aren't accounted at the root level */ | |
6718 | if (!mem_cgroup_is_root(memcg)) | |
6719 | ug->nr_memory += nr_pages; | |
18b2db3b | 6720 | ug->pgpgout++; |
a9d5adee | 6721 | |
b4e0b68f MS |
6722 | page->memcg_data = 0; |
6723 | } | |
6724 | ||
6725 | css_put(&memcg->css); | |
747db954 JW |
6726 | } |
6727 | ||
0a31bc97 JW |
6728 | /** |
6729 | * mem_cgroup_uncharge - uncharge a page | |
6730 | * @page: page to uncharge | |
6731 | * | |
f0e45fb4 | 6732 | * Uncharge a page previously charged with mem_cgroup_charge(). |
0a31bc97 JW |
6733 | */ |
6734 | void mem_cgroup_uncharge(struct page *page) | |
6735 | { | |
a9d5adee JG |
6736 | struct uncharge_gather ug; |
6737 | ||
0a31bc97 JW |
6738 | if (mem_cgroup_disabled()) |
6739 | return; | |
6740 | ||
747db954 | 6741 | /* Don't touch page->lru of any random page, pre-check: */ |
bcfe06bf | 6742 | if (!page_memcg(page)) |
0a31bc97 JW |
6743 | return; |
6744 | ||
a9d5adee JG |
6745 | uncharge_gather_clear(&ug); |
6746 | uncharge_page(page, &ug); | |
6747 | uncharge_batch(&ug); | |
747db954 | 6748 | } |
0a31bc97 | 6749 | |
747db954 JW |
6750 | /** |
6751 | * mem_cgroup_uncharge_list - uncharge a list of page | |
6752 | * @page_list: list of pages to uncharge | |
6753 | * | |
6754 | * Uncharge a list of pages previously charged with | |
f0e45fb4 | 6755 | * mem_cgroup_charge(). |
747db954 JW |
6756 | */ |
6757 | void mem_cgroup_uncharge_list(struct list_head *page_list) | |
6758 | { | |
c41a40b6 MS |
6759 | struct uncharge_gather ug; |
6760 | struct page *page; | |
6761 | ||
747db954 JW |
6762 | if (mem_cgroup_disabled()) |
6763 | return; | |
0a31bc97 | 6764 | |
c41a40b6 MS |
6765 | uncharge_gather_clear(&ug); |
6766 | list_for_each_entry(page, page_list, lru) | |
6767 | uncharge_page(page, &ug); | |
6768 | if (ug.memcg) | |
6769 | uncharge_batch(&ug); | |
0a31bc97 JW |
6770 | } |
6771 | ||
6772 | /** | |
6a93ca8f JW |
6773 | * mem_cgroup_migrate - charge a page's replacement |
6774 | * @oldpage: currently circulating page | |
6775 | * @newpage: replacement page | |
0a31bc97 | 6776 | * |
6a93ca8f JW |
6777 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
6778 | * be uncharged upon free. | |
0a31bc97 JW |
6779 | * |
6780 | * Both pages must be locked, @newpage->mapping must be set up. | |
6781 | */ | |
6a93ca8f | 6782 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
0a31bc97 | 6783 | { |
29833315 | 6784 | struct mem_cgroup *memcg; |
44b7a8d3 | 6785 | unsigned int nr_pages; |
d93c4130 | 6786 | unsigned long flags; |
0a31bc97 JW |
6787 | |
6788 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); | |
6789 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
0a31bc97 | 6790 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
6abb5a86 JW |
6791 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
6792 | newpage); | |
0a31bc97 JW |
6793 | |
6794 | if (mem_cgroup_disabled()) | |
6795 | return; | |
6796 | ||
6797 | /* Page cache replacement: new page already charged? */ | |
bcfe06bf | 6798 | if (page_memcg(newpage)) |
0a31bc97 JW |
6799 | return; |
6800 | ||
bcfe06bf | 6801 | memcg = page_memcg(oldpage); |
a4055888 | 6802 | VM_WARN_ON_ONCE_PAGE(!memcg, oldpage); |
29833315 | 6803 | if (!memcg) |
0a31bc97 JW |
6804 | return; |
6805 | ||
44b7a8d3 | 6806 | /* Force-charge the new page. The old one will be freed soon */ |
6c357848 | 6807 | nr_pages = thp_nr_pages(newpage); |
44b7a8d3 JW |
6808 | |
6809 | page_counter_charge(&memcg->memory, nr_pages); | |
6810 | if (do_memsw_account()) | |
6811 | page_counter_charge(&memcg->memsw, nr_pages); | |
0a31bc97 | 6812 | |
1a3e1f40 | 6813 | css_get(&memcg->css); |
d9eb1ea2 | 6814 | commit_charge(newpage, memcg); |
44b7a8d3 | 6815 | |
d93c4130 | 6816 | local_irq_save(flags); |
3fba69a5 | 6817 | mem_cgroup_charge_statistics(memcg, newpage, nr_pages); |
44b7a8d3 | 6818 | memcg_check_events(memcg, newpage); |
d93c4130 | 6819 | local_irq_restore(flags); |
0a31bc97 JW |
6820 | } |
6821 | ||
ef12947c | 6822 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
11092087 JW |
6823 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
6824 | ||
2d758073 | 6825 | void mem_cgroup_sk_alloc(struct sock *sk) |
11092087 JW |
6826 | { |
6827 | struct mem_cgroup *memcg; | |
6828 | ||
2d758073 JW |
6829 | if (!mem_cgroup_sockets_enabled) |
6830 | return; | |
6831 | ||
e876ecc6 SB |
6832 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
6833 | if (in_interrupt()) | |
6834 | return; | |
6835 | ||
11092087 JW |
6836 | rcu_read_lock(); |
6837 | memcg = mem_cgroup_from_task(current); | |
f7e1cb6e JW |
6838 | if (memcg == root_mem_cgroup) |
6839 | goto out; | |
0db15298 | 6840 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
f7e1cb6e | 6841 | goto out; |
8965aa28 | 6842 | if (css_tryget(&memcg->css)) |
11092087 | 6843 | sk->sk_memcg = memcg; |
f7e1cb6e | 6844 | out: |
11092087 JW |
6845 | rcu_read_unlock(); |
6846 | } | |
11092087 | 6847 | |
2d758073 | 6848 | void mem_cgroup_sk_free(struct sock *sk) |
11092087 | 6849 | { |
2d758073 JW |
6850 | if (sk->sk_memcg) |
6851 | css_put(&sk->sk_memcg->css); | |
11092087 JW |
6852 | } |
6853 | ||
6854 | /** | |
6855 | * mem_cgroup_charge_skmem - charge socket memory | |
6856 | * @memcg: memcg to charge | |
6857 | * @nr_pages: number of pages to charge | |
6858 | * | |
6859 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within | |
6860 | * @memcg's configured limit, %false if the charge had to be forced. | |
6861 | */ | |
6862 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6863 | { | |
f7e1cb6e | 6864 | gfp_t gfp_mask = GFP_KERNEL; |
11092087 | 6865 | |
f7e1cb6e | 6866 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6867 | struct page_counter *fail; |
f7e1cb6e | 6868 | |
0db15298 JW |
6869 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
6870 | memcg->tcpmem_pressure = 0; | |
f7e1cb6e JW |
6871 | return true; |
6872 | } | |
0db15298 JW |
6873 | page_counter_charge(&memcg->tcpmem, nr_pages); |
6874 | memcg->tcpmem_pressure = 1; | |
f7e1cb6e | 6875 | return false; |
11092087 | 6876 | } |
d886f4e4 | 6877 | |
f7e1cb6e JW |
6878 | /* Don't block in the packet receive path */ |
6879 | if (in_softirq()) | |
6880 | gfp_mask = GFP_NOWAIT; | |
6881 | ||
c9019e9b | 6882 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
b2807f07 | 6883 | |
f7e1cb6e JW |
6884 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
6885 | return true; | |
6886 | ||
6887 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); | |
11092087 JW |
6888 | return false; |
6889 | } | |
6890 | ||
6891 | /** | |
6892 | * mem_cgroup_uncharge_skmem - uncharge socket memory | |
b7701a5f MR |
6893 | * @memcg: memcg to uncharge |
6894 | * @nr_pages: number of pages to uncharge | |
11092087 JW |
6895 | */ |
6896 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) | |
6897 | { | |
f7e1cb6e | 6898 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
0db15298 | 6899 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
f7e1cb6e JW |
6900 | return; |
6901 | } | |
d886f4e4 | 6902 | |
c9019e9b | 6903 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
b2807f07 | 6904 | |
475d0487 | 6905 | refill_stock(memcg, nr_pages); |
11092087 JW |
6906 | } |
6907 | ||
f7e1cb6e JW |
6908 | static int __init cgroup_memory(char *s) |
6909 | { | |
6910 | char *token; | |
6911 | ||
6912 | while ((token = strsep(&s, ",")) != NULL) { | |
6913 | if (!*token) | |
6914 | continue; | |
6915 | if (!strcmp(token, "nosocket")) | |
6916 | cgroup_memory_nosocket = true; | |
04823c83 VD |
6917 | if (!strcmp(token, "nokmem")) |
6918 | cgroup_memory_nokmem = true; | |
f7e1cb6e JW |
6919 | } |
6920 | return 0; | |
6921 | } | |
6922 | __setup("cgroup.memory=", cgroup_memory); | |
11092087 | 6923 | |
2d11085e | 6924 | /* |
1081312f MH |
6925 | * subsys_initcall() for memory controller. |
6926 | * | |
308167fc SAS |
6927 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
6928 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but | |
6929 | * basically everything that doesn't depend on a specific mem_cgroup structure | |
6930 | * should be initialized from here. | |
2d11085e MH |
6931 | */ |
6932 | static int __init mem_cgroup_init(void) | |
6933 | { | |
95a045f6 JW |
6934 | int cpu, node; |
6935 | ||
f3344adf MS |
6936 | /* |
6937 | * Currently s32 type (can refer to struct batched_lruvec_stat) is | |
6938 | * used for per-memcg-per-cpu caching of per-node statistics. In order | |
6939 | * to work fine, we should make sure that the overfill threshold can't | |
6940 | * exceed S32_MAX / PAGE_SIZE. | |
6941 | */ | |
6942 | BUILD_BUG_ON(MEMCG_CHARGE_BATCH > S32_MAX / PAGE_SIZE); | |
6943 | ||
308167fc SAS |
6944 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
6945 | memcg_hotplug_cpu_dead); | |
95a045f6 JW |
6946 | |
6947 | for_each_possible_cpu(cpu) | |
6948 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, | |
6949 | drain_local_stock); | |
6950 | ||
6951 | for_each_node(node) { | |
6952 | struct mem_cgroup_tree_per_node *rtpn; | |
95a045f6 JW |
6953 | |
6954 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, | |
6955 | node_online(node) ? node : NUMA_NO_NODE); | |
6956 | ||
ef8f2327 | 6957 | rtpn->rb_root = RB_ROOT; |
fa90b2fd | 6958 | rtpn->rb_rightmost = NULL; |
ef8f2327 | 6959 | spin_lock_init(&rtpn->lock); |
95a045f6 JW |
6960 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
6961 | } | |
6962 | ||
2d11085e MH |
6963 | return 0; |
6964 | } | |
6965 | subsys_initcall(mem_cgroup_init); | |
21afa38e JW |
6966 | |
6967 | #ifdef CONFIG_MEMCG_SWAP | |
358c07fc AB |
6968 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
6969 | { | |
1c2d479a | 6970 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
358c07fc AB |
6971 | /* |
6972 | * The root cgroup cannot be destroyed, so it's refcount must | |
6973 | * always be >= 1. | |
6974 | */ | |
6975 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { | |
6976 | VM_BUG_ON(1); | |
6977 | break; | |
6978 | } | |
6979 | memcg = parent_mem_cgroup(memcg); | |
6980 | if (!memcg) | |
6981 | memcg = root_mem_cgroup; | |
6982 | } | |
6983 | return memcg; | |
6984 | } | |
6985 | ||
21afa38e JW |
6986 | /** |
6987 | * mem_cgroup_swapout - transfer a memsw charge to swap | |
6988 | * @page: page whose memsw charge to transfer | |
6989 | * @entry: swap entry to move the charge to | |
6990 | * | |
6991 | * Transfer the memsw charge of @page to @entry. | |
6992 | */ | |
6993 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) | |
6994 | { | |
1f47b61f | 6995 | struct mem_cgroup *memcg, *swap_memcg; |
d6810d73 | 6996 | unsigned int nr_entries; |
21afa38e JW |
6997 | unsigned short oldid; |
6998 | ||
6999 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
7000 | VM_BUG_ON_PAGE(page_count(page), page); | |
7001 | ||
76358ab5 AS |
7002 | if (mem_cgroup_disabled()) |
7003 | return; | |
7004 | ||
2d1c4980 | 7005 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
21afa38e JW |
7006 | return; |
7007 | ||
bcfe06bf | 7008 | memcg = page_memcg(page); |
21afa38e | 7009 | |
a4055888 | 7010 | VM_WARN_ON_ONCE_PAGE(!memcg, page); |
21afa38e JW |
7011 | if (!memcg) |
7012 | return; | |
7013 | ||
1f47b61f VD |
7014 | /* |
7015 | * In case the memcg owning these pages has been offlined and doesn't | |
7016 | * have an ID allocated to it anymore, charge the closest online | |
7017 | * ancestor for the swap instead and transfer the memory+swap charge. | |
7018 | */ | |
7019 | swap_memcg = mem_cgroup_id_get_online(memcg); | |
6c357848 | 7020 | nr_entries = thp_nr_pages(page); |
d6810d73 HY |
7021 | /* Get references for the tail pages, too */ |
7022 | if (nr_entries > 1) | |
7023 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); | |
7024 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), | |
7025 | nr_entries); | |
21afa38e | 7026 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7027 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
21afa38e | 7028 | |
bcfe06bf | 7029 | page->memcg_data = 0; |
21afa38e JW |
7030 | |
7031 | if (!mem_cgroup_is_root(memcg)) | |
d6810d73 | 7032 | page_counter_uncharge(&memcg->memory, nr_entries); |
21afa38e | 7033 | |
2d1c4980 | 7034 | if (!cgroup_memory_noswap && memcg != swap_memcg) { |
1f47b61f | 7035 | if (!mem_cgroup_is_root(swap_memcg)) |
d6810d73 HY |
7036 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
7037 | page_counter_uncharge(&memcg->memsw, nr_entries); | |
1f47b61f VD |
7038 | } |
7039 | ||
ce9ce665 SAS |
7040 | /* |
7041 | * Interrupts should be disabled here because the caller holds the | |
b93b0163 | 7042 | * i_pages lock which is taken with interrupts-off. It is |
ce9ce665 | 7043 | * important here to have the interrupts disabled because it is the |
b93b0163 | 7044 | * only synchronisation we have for updating the per-CPU variables. |
ce9ce665 SAS |
7045 | */ |
7046 | VM_BUG_ON(!irqs_disabled()); | |
3fba69a5 | 7047 | mem_cgroup_charge_statistics(memcg, page, -nr_entries); |
21afa38e | 7048 | memcg_check_events(memcg, page); |
73f576c0 | 7049 | |
1a3e1f40 | 7050 | css_put(&memcg->css); |
21afa38e JW |
7051 | } |
7052 | ||
38d8b4e6 HY |
7053 | /** |
7054 | * mem_cgroup_try_charge_swap - try charging swap space for a page | |
37e84351 VD |
7055 | * @page: page being added to swap |
7056 | * @entry: swap entry to charge | |
7057 | * | |
38d8b4e6 | 7058 | * Try to charge @page's memcg for the swap space at @entry. |
37e84351 VD |
7059 | * |
7060 | * Returns 0 on success, -ENOMEM on failure. | |
7061 | */ | |
7062 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) | |
7063 | { | |
6c357848 | 7064 | unsigned int nr_pages = thp_nr_pages(page); |
37e84351 | 7065 | struct page_counter *counter; |
38d8b4e6 | 7066 | struct mem_cgroup *memcg; |
37e84351 VD |
7067 | unsigned short oldid; |
7068 | ||
76358ab5 AS |
7069 | if (mem_cgroup_disabled()) |
7070 | return 0; | |
7071 | ||
2d1c4980 | 7072 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
37e84351 VD |
7073 | return 0; |
7074 | ||
bcfe06bf | 7075 | memcg = page_memcg(page); |
37e84351 | 7076 | |
a4055888 | 7077 | VM_WARN_ON_ONCE_PAGE(!memcg, page); |
37e84351 VD |
7078 | if (!memcg) |
7079 | return 0; | |
7080 | ||
f3a53a3a TH |
7081 | if (!entry.val) { |
7082 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
bb98f2c5 | 7083 | return 0; |
f3a53a3a | 7084 | } |
bb98f2c5 | 7085 | |
1f47b61f VD |
7086 | memcg = mem_cgroup_id_get_online(memcg); |
7087 | ||
2d1c4980 | 7088 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) && |
38d8b4e6 | 7089 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
f3a53a3a TH |
7090 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
7091 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); | |
1f47b61f | 7092 | mem_cgroup_id_put(memcg); |
37e84351 | 7093 | return -ENOMEM; |
1f47b61f | 7094 | } |
37e84351 | 7095 | |
38d8b4e6 HY |
7096 | /* Get references for the tail pages, too */ |
7097 | if (nr_pages > 1) | |
7098 | mem_cgroup_id_get_many(memcg, nr_pages - 1); | |
7099 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); | |
37e84351 | 7100 | VM_BUG_ON_PAGE(oldid, page); |
c9019e9b | 7101 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
37e84351 | 7102 | |
37e84351 VD |
7103 | return 0; |
7104 | } | |
7105 | ||
21afa38e | 7106 | /** |
38d8b4e6 | 7107 | * mem_cgroup_uncharge_swap - uncharge swap space |
21afa38e | 7108 | * @entry: swap entry to uncharge |
38d8b4e6 | 7109 | * @nr_pages: the amount of swap space to uncharge |
21afa38e | 7110 | */ |
38d8b4e6 | 7111 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
21afa38e JW |
7112 | { |
7113 | struct mem_cgroup *memcg; | |
7114 | unsigned short id; | |
7115 | ||
38d8b4e6 | 7116 | id = swap_cgroup_record(entry, 0, nr_pages); |
21afa38e | 7117 | rcu_read_lock(); |
adbe427b | 7118 | memcg = mem_cgroup_from_id(id); |
21afa38e | 7119 | if (memcg) { |
2d1c4980 | 7120 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) { |
37e84351 | 7121 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
38d8b4e6 | 7122 | page_counter_uncharge(&memcg->swap, nr_pages); |
37e84351 | 7123 | else |
38d8b4e6 | 7124 | page_counter_uncharge(&memcg->memsw, nr_pages); |
37e84351 | 7125 | } |
c9019e9b | 7126 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
38d8b4e6 | 7127 | mem_cgroup_id_put_many(memcg, nr_pages); |
21afa38e JW |
7128 | } |
7129 | rcu_read_unlock(); | |
7130 | } | |
7131 | ||
d8b38438 VD |
7132 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
7133 | { | |
7134 | long nr_swap_pages = get_nr_swap_pages(); | |
7135 | ||
eccb52e7 | 7136 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
d8b38438 VD |
7137 | return nr_swap_pages; |
7138 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) | |
7139 | nr_swap_pages = min_t(long, nr_swap_pages, | |
bbec2e15 | 7140 | READ_ONCE(memcg->swap.max) - |
d8b38438 VD |
7141 | page_counter_read(&memcg->swap)); |
7142 | return nr_swap_pages; | |
7143 | } | |
7144 | ||
5ccc5aba VD |
7145 | bool mem_cgroup_swap_full(struct page *page) |
7146 | { | |
7147 | struct mem_cgroup *memcg; | |
7148 | ||
7149 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
7150 | ||
7151 | if (vm_swap_full()) | |
7152 | return true; | |
eccb52e7 | 7153 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
5ccc5aba VD |
7154 | return false; |
7155 | ||
bcfe06bf | 7156 | memcg = page_memcg(page); |
5ccc5aba VD |
7157 | if (!memcg) |
7158 | return false; | |
7159 | ||
4b82ab4f JK |
7160 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
7161 | unsigned long usage = page_counter_read(&memcg->swap); | |
7162 | ||
7163 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || | |
7164 | usage * 2 >= READ_ONCE(memcg->swap.max)) | |
5ccc5aba | 7165 | return true; |
4b82ab4f | 7166 | } |
5ccc5aba VD |
7167 | |
7168 | return false; | |
7169 | } | |
7170 | ||
eccb52e7 | 7171 | static int __init setup_swap_account(char *s) |
21afa38e JW |
7172 | { |
7173 | if (!strcmp(s, "1")) | |
5ab92901 | 7174 | cgroup_memory_noswap = false; |
21afa38e | 7175 | else if (!strcmp(s, "0")) |
5ab92901 | 7176 | cgroup_memory_noswap = true; |
21afa38e JW |
7177 | return 1; |
7178 | } | |
eccb52e7 | 7179 | __setup("swapaccount=", setup_swap_account); |
21afa38e | 7180 | |
37e84351 VD |
7181 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
7182 | struct cftype *cft) | |
7183 | { | |
7184 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); | |
7185 | ||
7186 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; | |
7187 | } | |
7188 | ||
4b82ab4f JK |
7189 | static int swap_high_show(struct seq_file *m, void *v) |
7190 | { | |
7191 | return seq_puts_memcg_tunable(m, | |
7192 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); | |
7193 | } | |
7194 | ||
7195 | static ssize_t swap_high_write(struct kernfs_open_file *of, | |
7196 | char *buf, size_t nbytes, loff_t off) | |
7197 | { | |
7198 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7199 | unsigned long high; | |
7200 | int err; | |
7201 | ||
7202 | buf = strstrip(buf); | |
7203 | err = page_counter_memparse(buf, "max", &high); | |
7204 | if (err) | |
7205 | return err; | |
7206 | ||
7207 | page_counter_set_high(&memcg->swap, high); | |
7208 | ||
7209 | return nbytes; | |
7210 | } | |
7211 | ||
37e84351 VD |
7212 | static int swap_max_show(struct seq_file *m, void *v) |
7213 | { | |
677dc973 CD |
7214 | return seq_puts_memcg_tunable(m, |
7215 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); | |
37e84351 VD |
7216 | } |
7217 | ||
7218 | static ssize_t swap_max_write(struct kernfs_open_file *of, | |
7219 | char *buf, size_t nbytes, loff_t off) | |
7220 | { | |
7221 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); | |
7222 | unsigned long max; | |
7223 | int err; | |
7224 | ||
7225 | buf = strstrip(buf); | |
7226 | err = page_counter_memparse(buf, "max", &max); | |
7227 | if (err) | |
7228 | return err; | |
7229 | ||
be09102b | 7230 | xchg(&memcg->swap.max, max); |
37e84351 VD |
7231 | |
7232 | return nbytes; | |
7233 | } | |
7234 | ||
f3a53a3a TH |
7235 | static int swap_events_show(struct seq_file *m, void *v) |
7236 | { | |
aa9694bb | 7237 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
f3a53a3a | 7238 | |
4b82ab4f JK |
7239 | seq_printf(m, "high %lu\n", |
7240 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); | |
f3a53a3a TH |
7241 | seq_printf(m, "max %lu\n", |
7242 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); | |
7243 | seq_printf(m, "fail %lu\n", | |
7244 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); | |
7245 | ||
7246 | return 0; | |
7247 | } | |
7248 | ||
37e84351 VD |
7249 | static struct cftype swap_files[] = { |
7250 | { | |
7251 | .name = "swap.current", | |
7252 | .flags = CFTYPE_NOT_ON_ROOT, | |
7253 | .read_u64 = swap_current_read, | |
7254 | }, | |
4b82ab4f JK |
7255 | { |
7256 | .name = "swap.high", | |
7257 | .flags = CFTYPE_NOT_ON_ROOT, | |
7258 | .seq_show = swap_high_show, | |
7259 | .write = swap_high_write, | |
7260 | }, | |
37e84351 VD |
7261 | { |
7262 | .name = "swap.max", | |
7263 | .flags = CFTYPE_NOT_ON_ROOT, | |
7264 | .seq_show = swap_max_show, | |
7265 | .write = swap_max_write, | |
7266 | }, | |
f3a53a3a TH |
7267 | { |
7268 | .name = "swap.events", | |
7269 | .flags = CFTYPE_NOT_ON_ROOT, | |
7270 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), | |
7271 | .seq_show = swap_events_show, | |
7272 | }, | |
37e84351 VD |
7273 | { } /* terminate */ |
7274 | }; | |
7275 | ||
eccb52e7 | 7276 | static struct cftype memsw_files[] = { |
21afa38e JW |
7277 | { |
7278 | .name = "memsw.usage_in_bytes", | |
7279 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
7280 | .read_u64 = mem_cgroup_read_u64, | |
7281 | }, | |
7282 | { | |
7283 | .name = "memsw.max_usage_in_bytes", | |
7284 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
7285 | .write = mem_cgroup_reset, | |
7286 | .read_u64 = mem_cgroup_read_u64, | |
7287 | }, | |
7288 | { | |
7289 | .name = "memsw.limit_in_bytes", | |
7290 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
7291 | .write = mem_cgroup_write, | |
7292 | .read_u64 = mem_cgroup_read_u64, | |
7293 | }, | |
7294 | { | |
7295 | .name = "memsw.failcnt", | |
7296 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
7297 | .write = mem_cgroup_reset, | |
7298 | .read_u64 = mem_cgroup_read_u64, | |
7299 | }, | |
7300 | { }, /* terminate */ | |
7301 | }; | |
7302 | ||
82ff165c BS |
7303 | /* |
7304 | * If mem_cgroup_swap_init() is implemented as a subsys_initcall() | |
7305 | * instead of a core_initcall(), this could mean cgroup_memory_noswap still | |
7306 | * remains set to false even when memcg is disabled via "cgroup_disable=memory" | |
7307 | * boot parameter. This may result in premature OOPS inside | |
7308 | * mem_cgroup_get_nr_swap_pages() function in corner cases. | |
7309 | */ | |
21afa38e JW |
7310 | static int __init mem_cgroup_swap_init(void) |
7311 | { | |
2d1c4980 JW |
7312 | /* No memory control -> no swap control */ |
7313 | if (mem_cgroup_disabled()) | |
7314 | cgroup_memory_noswap = true; | |
7315 | ||
7316 | if (cgroup_memory_noswap) | |
eccb52e7 JW |
7317 | return 0; |
7318 | ||
7319 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); | |
7320 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); | |
7321 | ||
21afa38e JW |
7322 | return 0; |
7323 | } | |
82ff165c | 7324 | core_initcall(mem_cgroup_swap_init); |
21afa38e JW |
7325 | |
7326 | #endif /* CONFIG_MEMCG_SWAP */ |