]>
Commit | Line | Data |
---|---|---|
1 | /* memcontrol.c - Memory Controller | |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
6 | * Copyright 2007 OpenVZ SWsoft Inc | |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
9 | * This program is free software; you can redistribute it and/or modify | |
10 | * it under the terms of the GNU General Public License as published by | |
11 | * the Free Software Foundation; either version 2 of the License, or | |
12 | * (at your option) any later version. | |
13 | * | |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | */ | |
19 | ||
20 | #include <linux/res_counter.h> | |
21 | #include <linux/memcontrol.h> | |
22 | #include <linux/cgroup.h> | |
23 | #include <linux/mm.h> | |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/smp.h> | |
26 | #include <linux/page-flags.h> | |
27 | #include <linux/backing-dev.h> | |
28 | #include <linux/bit_spinlock.h> | |
29 | #include <linux/rcupdate.h> | |
30 | #include <linux/mutex.h> | |
31 | #include <linux/slab.h> | |
32 | #include <linux/swap.h> | |
33 | #include <linux/spinlock.h> | |
34 | #include <linux/fs.h> | |
35 | #include <linux/seq_file.h> | |
36 | #include <linux/vmalloc.h> | |
37 | #include <linux/mm_inline.h> | |
38 | #include <linux/page_cgroup.h> | |
39 | #include "internal.h" | |
40 | ||
41 | #include <asm/uaccess.h> | |
42 | ||
43 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; | |
44 | #define MEM_CGROUP_RECLAIM_RETRIES 5 | |
45 | ||
46 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
47 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 0 */ | |
48 | int do_swap_account __read_mostly; | |
49 | static int really_do_swap_account __initdata = 1; /* for remember boot option*/ | |
50 | #else | |
51 | #define do_swap_account (0) | |
52 | #endif | |
53 | ||
54 | ||
55 | /* | |
56 | * Statistics for memory cgroup. | |
57 | */ | |
58 | enum mem_cgroup_stat_index { | |
59 | /* | |
60 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
61 | */ | |
62 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
63 | MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */ | |
64 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ | |
65 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
66 | ||
67 | MEM_CGROUP_STAT_NSTATS, | |
68 | }; | |
69 | ||
70 | struct mem_cgroup_stat_cpu { | |
71 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
72 | } ____cacheline_aligned_in_smp; | |
73 | ||
74 | struct mem_cgroup_stat { | |
75 | struct mem_cgroup_stat_cpu cpustat[0]; | |
76 | }; | |
77 | ||
78 | /* | |
79 | * For accounting under irq disable, no need for increment preempt count. | |
80 | */ | |
81 | static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, | |
82 | enum mem_cgroup_stat_index idx, int val) | |
83 | { | |
84 | stat->count[idx] += val; | |
85 | } | |
86 | ||
87 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, | |
88 | enum mem_cgroup_stat_index idx) | |
89 | { | |
90 | int cpu; | |
91 | s64 ret = 0; | |
92 | for_each_possible_cpu(cpu) | |
93 | ret += stat->cpustat[cpu].count[idx]; | |
94 | return ret; | |
95 | } | |
96 | ||
97 | /* | |
98 | * per-zone information in memory controller. | |
99 | */ | |
100 | struct mem_cgroup_per_zone { | |
101 | /* | |
102 | * spin_lock to protect the per cgroup LRU | |
103 | */ | |
104 | struct list_head lists[NR_LRU_LISTS]; | |
105 | unsigned long count[NR_LRU_LISTS]; | |
106 | }; | |
107 | /* Macro for accessing counter */ | |
108 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
109 | ||
110 | struct mem_cgroup_per_node { | |
111 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
112 | }; | |
113 | ||
114 | struct mem_cgroup_lru_info { | |
115 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
116 | }; | |
117 | ||
118 | /* | |
119 | * The memory controller data structure. The memory controller controls both | |
120 | * page cache and RSS per cgroup. We would eventually like to provide | |
121 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
122 | * to help the administrator determine what knobs to tune. | |
123 | * | |
124 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
125 | * we hit the water mark. May be even add a low water mark, such that | |
126 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
127 | * a feature that will be implemented much later in the future. | |
128 | */ | |
129 | struct mem_cgroup { | |
130 | struct cgroup_subsys_state css; | |
131 | /* | |
132 | * the counter to account for memory usage | |
133 | */ | |
134 | struct res_counter res; | |
135 | /* | |
136 | * the counter to account for mem+swap usage. | |
137 | */ | |
138 | struct res_counter memsw; | |
139 | /* | |
140 | * Per cgroup active and inactive list, similar to the | |
141 | * per zone LRU lists. | |
142 | */ | |
143 | struct mem_cgroup_lru_info info; | |
144 | ||
145 | int prev_priority; /* for recording reclaim priority */ | |
146 | ||
147 | /* | |
148 | * While reclaiming in a hiearchy, we cache the last child we | |
149 | * reclaimed from. Protected by cgroup_lock() | |
150 | */ | |
151 | struct mem_cgroup *last_scanned_child; | |
152 | /* | |
153 | * Should the accounting and control be hierarchical, per subtree? | |
154 | */ | |
155 | bool use_hierarchy; | |
156 | unsigned long last_oom_jiffies; | |
157 | int obsolete; | |
158 | atomic_t refcnt; | |
159 | ||
160 | unsigned int inactive_ratio; | |
161 | ||
162 | /* | |
163 | * statistics. This must be placed at the end of memcg. | |
164 | */ | |
165 | struct mem_cgroup_stat stat; | |
166 | }; | |
167 | ||
168 | enum charge_type { | |
169 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
170 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
171 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ | |
172 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ | |
173 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ | |
174 | NR_CHARGE_TYPE, | |
175 | }; | |
176 | ||
177 | /* only for here (for easy reading.) */ | |
178 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
179 | #define PCGF_USED (1UL << PCG_USED) | |
180 | #define PCGF_LOCK (1UL << PCG_LOCK) | |
181 | static const unsigned long | |
182 | pcg_default_flags[NR_CHARGE_TYPE] = { | |
183 | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* File Cache */ | |
184 | PCGF_USED | PCGF_LOCK, /* Anon */ | |
185 | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */ | |
186 | 0, /* FORCE */ | |
187 | }; | |
188 | ||
189 | /* for encoding cft->private value on file */ | |
190 | #define _MEM (0) | |
191 | #define _MEMSWAP (1) | |
192 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
193 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
194 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
195 | ||
196 | static void mem_cgroup_get(struct mem_cgroup *mem); | |
197 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
198 | ||
199 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, | |
200 | struct page_cgroup *pc, | |
201 | bool charge) | |
202 | { | |
203 | int val = (charge)? 1 : -1; | |
204 | struct mem_cgroup_stat *stat = &mem->stat; | |
205 | struct mem_cgroup_stat_cpu *cpustat; | |
206 | int cpu = get_cpu(); | |
207 | ||
208 | cpustat = &stat->cpustat[cpu]; | |
209 | if (PageCgroupCache(pc)) | |
210 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); | |
211 | else | |
212 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); | |
213 | ||
214 | if (charge) | |
215 | __mem_cgroup_stat_add_safe(cpustat, | |
216 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); | |
217 | else | |
218 | __mem_cgroup_stat_add_safe(cpustat, | |
219 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); | |
220 | put_cpu(); | |
221 | } | |
222 | ||
223 | static struct mem_cgroup_per_zone * | |
224 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
225 | { | |
226 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
227 | } | |
228 | ||
229 | static struct mem_cgroup_per_zone * | |
230 | page_cgroup_zoneinfo(struct page_cgroup *pc) | |
231 | { | |
232 | struct mem_cgroup *mem = pc->mem_cgroup; | |
233 | int nid = page_cgroup_nid(pc); | |
234 | int zid = page_cgroup_zid(pc); | |
235 | ||
236 | if (!mem) | |
237 | return NULL; | |
238 | ||
239 | return mem_cgroup_zoneinfo(mem, nid, zid); | |
240 | } | |
241 | ||
242 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, | |
243 | enum lru_list idx) | |
244 | { | |
245 | int nid, zid; | |
246 | struct mem_cgroup_per_zone *mz; | |
247 | u64 total = 0; | |
248 | ||
249 | for_each_online_node(nid) | |
250 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
251 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
252 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
253 | } | |
254 | return total; | |
255 | } | |
256 | ||
257 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) | |
258 | { | |
259 | return container_of(cgroup_subsys_state(cont, | |
260 | mem_cgroup_subsys_id), struct mem_cgroup, | |
261 | css); | |
262 | } | |
263 | ||
264 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) | |
265 | { | |
266 | /* | |
267 | * mm_update_next_owner() may clear mm->owner to NULL | |
268 | * if it races with swapoff, page migration, etc. | |
269 | * So this can be called with p == NULL. | |
270 | */ | |
271 | if (unlikely(!p)) | |
272 | return NULL; | |
273 | ||
274 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), | |
275 | struct mem_cgroup, css); | |
276 | } | |
277 | ||
278 | /* | |
279 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
280 | * Operations are called by routine of global LRU independently from memcg. | |
281 | * What we have to take care of here is validness of pc->mem_cgroup. | |
282 | * | |
283 | * Changes to pc->mem_cgroup happens when | |
284 | * 1. charge | |
285 | * 2. moving account | |
286 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
287 | * It is added to LRU before charge. | |
288 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
289 | * When moving account, the page is not on LRU. It's isolated. | |
290 | */ | |
291 | ||
292 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) | |
293 | { | |
294 | struct page_cgroup *pc; | |
295 | struct mem_cgroup *mem; | |
296 | struct mem_cgroup_per_zone *mz; | |
297 | ||
298 | if (mem_cgroup_disabled()) | |
299 | return; | |
300 | pc = lookup_page_cgroup(page); | |
301 | /* can happen while we handle swapcache. */ | |
302 | if (list_empty(&pc->lru)) | |
303 | return; | |
304 | mz = page_cgroup_zoneinfo(pc); | |
305 | mem = pc->mem_cgroup; | |
306 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; | |
307 | list_del_init(&pc->lru); | |
308 | return; | |
309 | } | |
310 | ||
311 | void mem_cgroup_del_lru(struct page *page) | |
312 | { | |
313 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
314 | } | |
315 | ||
316 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) | |
317 | { | |
318 | struct mem_cgroup_per_zone *mz; | |
319 | struct page_cgroup *pc; | |
320 | ||
321 | if (mem_cgroup_disabled()) | |
322 | return; | |
323 | ||
324 | pc = lookup_page_cgroup(page); | |
325 | smp_rmb(); | |
326 | /* unused page is not rotated. */ | |
327 | if (!PageCgroupUsed(pc)) | |
328 | return; | |
329 | mz = page_cgroup_zoneinfo(pc); | |
330 | list_move(&pc->lru, &mz->lists[lru]); | |
331 | } | |
332 | ||
333 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) | |
334 | { | |
335 | struct page_cgroup *pc; | |
336 | struct mem_cgroup_per_zone *mz; | |
337 | ||
338 | if (mem_cgroup_disabled()) | |
339 | return; | |
340 | pc = lookup_page_cgroup(page); | |
341 | /* barrier to sync with "charge" */ | |
342 | smp_rmb(); | |
343 | if (!PageCgroupUsed(pc)) | |
344 | return; | |
345 | ||
346 | mz = page_cgroup_zoneinfo(pc); | |
347 | MEM_CGROUP_ZSTAT(mz, lru) += 1; | |
348 | list_add(&pc->lru, &mz->lists[lru]); | |
349 | } | |
350 | /* | |
351 | * To add swapcache into LRU. Be careful to all this function. | |
352 | * zone->lru_lock shouldn't be held and irq must not be disabled. | |
353 | */ | |
354 | static void mem_cgroup_lru_fixup(struct page *page) | |
355 | { | |
356 | if (!isolate_lru_page(page)) | |
357 | putback_lru_page(page); | |
358 | } | |
359 | ||
360 | void mem_cgroup_move_lists(struct page *page, | |
361 | enum lru_list from, enum lru_list to) | |
362 | { | |
363 | if (mem_cgroup_disabled()) | |
364 | return; | |
365 | mem_cgroup_del_lru_list(page, from); | |
366 | mem_cgroup_add_lru_list(page, to); | |
367 | } | |
368 | ||
369 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) | |
370 | { | |
371 | int ret; | |
372 | ||
373 | task_lock(task); | |
374 | ret = task->mm && mm_match_cgroup(task->mm, mem); | |
375 | task_unlock(task); | |
376 | return ret; | |
377 | } | |
378 | ||
379 | /* | |
380 | * Calculate mapped_ratio under memory controller. This will be used in | |
381 | * vmscan.c for deteremining we have to reclaim mapped pages. | |
382 | */ | |
383 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) | |
384 | { | |
385 | long total, rss; | |
386 | ||
387 | /* | |
388 | * usage is recorded in bytes. But, here, we assume the number of | |
389 | * physical pages can be represented by "long" on any arch. | |
390 | */ | |
391 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; | |
392 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
393 | return (int)((rss * 100L) / total); | |
394 | } | |
395 | ||
396 | /* | |
397 | * prev_priority control...this will be used in memory reclaim path. | |
398 | */ | |
399 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
400 | { | |
401 | return mem->prev_priority; | |
402 | } | |
403 | ||
404 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
405 | { | |
406 | if (priority < mem->prev_priority) | |
407 | mem->prev_priority = priority; | |
408 | } | |
409 | ||
410 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
411 | { | |
412 | mem->prev_priority = priority; | |
413 | } | |
414 | ||
415 | /* | |
416 | * Calculate # of pages to be scanned in this priority/zone. | |
417 | * See also vmscan.c | |
418 | * | |
419 | * priority starts from "DEF_PRIORITY" and decremented in each loop. | |
420 | * (see include/linux/mmzone.h) | |
421 | */ | |
422 | ||
423 | long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone, | |
424 | int priority, enum lru_list lru) | |
425 | { | |
426 | long nr_pages; | |
427 | int nid = zone->zone_pgdat->node_id; | |
428 | int zid = zone_idx(zone); | |
429 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
430 | ||
431 | nr_pages = MEM_CGROUP_ZSTAT(mz, lru); | |
432 | ||
433 | return (nr_pages >> priority); | |
434 | } | |
435 | ||
436 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg, struct zone *zone) | |
437 | { | |
438 | unsigned long active; | |
439 | unsigned long inactive; | |
440 | ||
441 | inactive = mem_cgroup_get_all_zonestat(memcg, LRU_INACTIVE_ANON); | |
442 | active = mem_cgroup_get_all_zonestat(memcg, LRU_ACTIVE_ANON); | |
443 | ||
444 | if (inactive * memcg->inactive_ratio < active) | |
445 | return 1; | |
446 | ||
447 | return 0; | |
448 | } | |
449 | ||
450 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, | |
451 | struct zone *zone, | |
452 | enum lru_list lru) | |
453 | { | |
454 | int nid = zone->zone_pgdat->node_id; | |
455 | int zid = zone_idx(zone); | |
456 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
457 | ||
458 | return MEM_CGROUP_ZSTAT(mz, lru); | |
459 | } | |
460 | ||
461 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, | |
462 | struct list_head *dst, | |
463 | unsigned long *scanned, int order, | |
464 | int mode, struct zone *z, | |
465 | struct mem_cgroup *mem_cont, | |
466 | int active, int file) | |
467 | { | |
468 | unsigned long nr_taken = 0; | |
469 | struct page *page; | |
470 | unsigned long scan; | |
471 | LIST_HEAD(pc_list); | |
472 | struct list_head *src; | |
473 | struct page_cgroup *pc, *tmp; | |
474 | int nid = z->zone_pgdat->node_id; | |
475 | int zid = zone_idx(z); | |
476 | struct mem_cgroup_per_zone *mz; | |
477 | int lru = LRU_FILE * !!file + !!active; | |
478 | ||
479 | BUG_ON(!mem_cont); | |
480 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
481 | src = &mz->lists[lru]; | |
482 | ||
483 | scan = 0; | |
484 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
485 | if (scan >= nr_to_scan) | |
486 | break; | |
487 | ||
488 | page = pc->page; | |
489 | if (unlikely(!PageCgroupUsed(pc))) | |
490 | continue; | |
491 | if (unlikely(!PageLRU(page))) | |
492 | continue; | |
493 | ||
494 | scan++; | |
495 | if (__isolate_lru_page(page, mode, file) == 0) { | |
496 | list_move(&page->lru, dst); | |
497 | nr_taken++; | |
498 | } | |
499 | } | |
500 | ||
501 | *scanned = scan; | |
502 | return nr_taken; | |
503 | } | |
504 | ||
505 | #define mem_cgroup_from_res_counter(counter, member) \ | |
506 | container_of(counter, struct mem_cgroup, member) | |
507 | ||
508 | /* | |
509 | * This routine finds the DFS walk successor. This routine should be | |
510 | * called with cgroup_mutex held | |
511 | */ | |
512 | static struct mem_cgroup * | |
513 | mem_cgroup_get_next_node(struct mem_cgroup *curr, struct mem_cgroup *root_mem) | |
514 | { | |
515 | struct cgroup *cgroup, *curr_cgroup, *root_cgroup; | |
516 | ||
517 | curr_cgroup = curr->css.cgroup; | |
518 | root_cgroup = root_mem->css.cgroup; | |
519 | ||
520 | if (!list_empty(&curr_cgroup->children)) { | |
521 | /* | |
522 | * Walk down to children | |
523 | */ | |
524 | mem_cgroup_put(curr); | |
525 | cgroup = list_entry(curr_cgroup->children.next, | |
526 | struct cgroup, sibling); | |
527 | curr = mem_cgroup_from_cont(cgroup); | |
528 | mem_cgroup_get(curr); | |
529 | goto done; | |
530 | } | |
531 | ||
532 | visit_parent: | |
533 | if (curr_cgroup == root_cgroup) { | |
534 | mem_cgroup_put(curr); | |
535 | curr = root_mem; | |
536 | mem_cgroup_get(curr); | |
537 | goto done; | |
538 | } | |
539 | ||
540 | /* | |
541 | * Goto next sibling | |
542 | */ | |
543 | if (curr_cgroup->sibling.next != &curr_cgroup->parent->children) { | |
544 | mem_cgroup_put(curr); | |
545 | cgroup = list_entry(curr_cgroup->sibling.next, struct cgroup, | |
546 | sibling); | |
547 | curr = mem_cgroup_from_cont(cgroup); | |
548 | mem_cgroup_get(curr); | |
549 | goto done; | |
550 | } | |
551 | ||
552 | /* | |
553 | * Go up to next parent and next parent's sibling if need be | |
554 | */ | |
555 | curr_cgroup = curr_cgroup->parent; | |
556 | goto visit_parent; | |
557 | ||
558 | done: | |
559 | root_mem->last_scanned_child = curr; | |
560 | return curr; | |
561 | } | |
562 | ||
563 | /* | |
564 | * Visit the first child (need not be the first child as per the ordering | |
565 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
566 | * that to reclaim free pages from. | |
567 | */ | |
568 | static struct mem_cgroup * | |
569 | mem_cgroup_get_first_node(struct mem_cgroup *root_mem) | |
570 | { | |
571 | struct cgroup *cgroup; | |
572 | struct mem_cgroup *ret; | |
573 | bool obsolete = (root_mem->last_scanned_child && | |
574 | root_mem->last_scanned_child->obsolete); | |
575 | ||
576 | /* | |
577 | * Scan all children under the mem_cgroup mem | |
578 | */ | |
579 | cgroup_lock(); | |
580 | if (list_empty(&root_mem->css.cgroup->children)) { | |
581 | ret = root_mem; | |
582 | goto done; | |
583 | } | |
584 | ||
585 | if (!root_mem->last_scanned_child || obsolete) { | |
586 | ||
587 | if (obsolete) | |
588 | mem_cgroup_put(root_mem->last_scanned_child); | |
589 | ||
590 | cgroup = list_first_entry(&root_mem->css.cgroup->children, | |
591 | struct cgroup, sibling); | |
592 | ret = mem_cgroup_from_cont(cgroup); | |
593 | mem_cgroup_get(ret); | |
594 | } else | |
595 | ret = mem_cgroup_get_next_node(root_mem->last_scanned_child, | |
596 | root_mem); | |
597 | ||
598 | done: | |
599 | root_mem->last_scanned_child = ret; | |
600 | cgroup_unlock(); | |
601 | return ret; | |
602 | } | |
603 | ||
604 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) | |
605 | { | |
606 | if (do_swap_account) { | |
607 | if (res_counter_check_under_limit(&mem->res) && | |
608 | res_counter_check_under_limit(&mem->memsw)) | |
609 | return true; | |
610 | } else | |
611 | if (res_counter_check_under_limit(&mem->res)) | |
612 | return true; | |
613 | return false; | |
614 | } | |
615 | ||
616 | /* | |
617 | * Dance down the hierarchy if needed to reclaim memory. We remember the | |
618 | * last child we reclaimed from, so that we don't end up penalizing | |
619 | * one child extensively based on its position in the children list. | |
620 | * | |
621 | * root_mem is the original ancestor that we've been reclaim from. | |
622 | */ | |
623 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
624 | gfp_t gfp_mask, bool noswap) | |
625 | { | |
626 | struct mem_cgroup *next_mem; | |
627 | int ret = 0; | |
628 | ||
629 | /* | |
630 | * Reclaim unconditionally and don't check for return value. | |
631 | * We need to reclaim in the current group and down the tree. | |
632 | * One might think about checking for children before reclaiming, | |
633 | * but there might be left over accounting, even after children | |
634 | * have left. | |
635 | */ | |
636 | ret = try_to_free_mem_cgroup_pages(root_mem, gfp_mask, noswap); | |
637 | if (mem_cgroup_check_under_limit(root_mem)) | |
638 | return 0; | |
639 | if (!root_mem->use_hierarchy) | |
640 | return ret; | |
641 | ||
642 | next_mem = mem_cgroup_get_first_node(root_mem); | |
643 | ||
644 | while (next_mem != root_mem) { | |
645 | if (next_mem->obsolete) { | |
646 | mem_cgroup_put(next_mem); | |
647 | cgroup_lock(); | |
648 | next_mem = mem_cgroup_get_first_node(root_mem); | |
649 | cgroup_unlock(); | |
650 | continue; | |
651 | } | |
652 | ret = try_to_free_mem_cgroup_pages(next_mem, gfp_mask, noswap); | |
653 | if (mem_cgroup_check_under_limit(root_mem)) | |
654 | return 0; | |
655 | cgroup_lock(); | |
656 | next_mem = mem_cgroup_get_next_node(next_mem, root_mem); | |
657 | cgroup_unlock(); | |
658 | } | |
659 | return ret; | |
660 | } | |
661 | ||
662 | bool mem_cgroup_oom_called(struct task_struct *task) | |
663 | { | |
664 | bool ret = false; | |
665 | struct mem_cgroup *mem; | |
666 | struct mm_struct *mm; | |
667 | ||
668 | rcu_read_lock(); | |
669 | mm = task->mm; | |
670 | if (!mm) | |
671 | mm = &init_mm; | |
672 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
673 | if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10)) | |
674 | ret = true; | |
675 | rcu_read_unlock(); | |
676 | return ret; | |
677 | } | |
678 | /* | |
679 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
680 | * oom-killer can be invoked. | |
681 | */ | |
682 | static int __mem_cgroup_try_charge(struct mm_struct *mm, | |
683 | gfp_t gfp_mask, struct mem_cgroup **memcg, | |
684 | bool oom) | |
685 | { | |
686 | struct mem_cgroup *mem, *mem_over_limit; | |
687 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
688 | struct res_counter *fail_res; | |
689 | ||
690 | if (unlikely(test_thread_flag(TIF_MEMDIE))) { | |
691 | /* Don't account this! */ | |
692 | *memcg = NULL; | |
693 | return 0; | |
694 | } | |
695 | ||
696 | /* | |
697 | * We always charge the cgroup the mm_struct belongs to. | |
698 | * The mm_struct's mem_cgroup changes on task migration if the | |
699 | * thread group leader migrates. It's possible that mm is not | |
700 | * set, if so charge the init_mm (happens for pagecache usage). | |
701 | */ | |
702 | if (likely(!*memcg)) { | |
703 | rcu_read_lock(); | |
704 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
705 | if (unlikely(!mem)) { | |
706 | rcu_read_unlock(); | |
707 | return 0; | |
708 | } | |
709 | /* | |
710 | * For every charge from the cgroup, increment reference count | |
711 | */ | |
712 | css_get(&mem->css); | |
713 | *memcg = mem; | |
714 | rcu_read_unlock(); | |
715 | } else { | |
716 | mem = *memcg; | |
717 | css_get(&mem->css); | |
718 | } | |
719 | ||
720 | while (1) { | |
721 | int ret; | |
722 | bool noswap = false; | |
723 | ||
724 | ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res); | |
725 | if (likely(!ret)) { | |
726 | if (!do_swap_account) | |
727 | break; | |
728 | ret = res_counter_charge(&mem->memsw, PAGE_SIZE, | |
729 | &fail_res); | |
730 | if (likely(!ret)) | |
731 | break; | |
732 | /* mem+swap counter fails */ | |
733 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
734 | noswap = true; | |
735 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, | |
736 | memsw); | |
737 | } else | |
738 | /* mem counter fails */ | |
739 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, | |
740 | res); | |
741 | ||
742 | if (!(gfp_mask & __GFP_WAIT)) | |
743 | goto nomem; | |
744 | ||
745 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, gfp_mask, | |
746 | noswap); | |
747 | ||
748 | /* | |
749 | * try_to_free_mem_cgroup_pages() might not give us a full | |
750 | * picture of reclaim. Some pages are reclaimed and might be | |
751 | * moved to swap cache or just unmapped from the cgroup. | |
752 | * Check the limit again to see if the reclaim reduced the | |
753 | * current usage of the cgroup before giving up | |
754 | * | |
755 | */ | |
756 | if (mem_cgroup_check_under_limit(mem_over_limit)) | |
757 | continue; | |
758 | ||
759 | if (!nr_retries--) { | |
760 | if (oom) { | |
761 | mem_cgroup_out_of_memory(mem_over_limit, gfp_mask); | |
762 | mem_over_limit->last_oom_jiffies = jiffies; | |
763 | } | |
764 | goto nomem; | |
765 | } | |
766 | } | |
767 | return 0; | |
768 | nomem: | |
769 | css_put(&mem->css); | |
770 | return -ENOMEM; | |
771 | } | |
772 | ||
773 | /** | |
774 | * mem_cgroup_try_charge - get charge of PAGE_SIZE. | |
775 | * @mm: an mm_struct which is charged against. (when *memcg is NULL) | |
776 | * @gfp_mask: gfp_mask for reclaim. | |
777 | * @memcg: a pointer to memory cgroup which is charged against. | |
778 | * | |
779 | * charge against memory cgroup pointed by *memcg. if *memcg == NULL, estimated | |
780 | * memory cgroup from @mm is got and stored in *memcg. | |
781 | * | |
782 | * Returns 0 if success. -ENOMEM at failure. | |
783 | * This call can invoke OOM-Killer. | |
784 | */ | |
785 | ||
786 | int mem_cgroup_try_charge(struct mm_struct *mm, | |
787 | gfp_t mask, struct mem_cgroup **memcg) | |
788 | { | |
789 | return __mem_cgroup_try_charge(mm, mask, memcg, true); | |
790 | } | |
791 | ||
792 | /* | |
793 | * commit a charge got by mem_cgroup_try_charge() and makes page_cgroup to be | |
794 | * USED state. If already USED, uncharge and return. | |
795 | */ | |
796 | ||
797 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
798 | struct page_cgroup *pc, | |
799 | enum charge_type ctype) | |
800 | { | |
801 | /* try_charge() can return NULL to *memcg, taking care of it. */ | |
802 | if (!mem) | |
803 | return; | |
804 | ||
805 | lock_page_cgroup(pc); | |
806 | if (unlikely(PageCgroupUsed(pc))) { | |
807 | unlock_page_cgroup(pc); | |
808 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
809 | if (do_swap_account) | |
810 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
811 | css_put(&mem->css); | |
812 | return; | |
813 | } | |
814 | pc->mem_cgroup = mem; | |
815 | smp_wmb(); | |
816 | pc->flags = pcg_default_flags[ctype]; | |
817 | ||
818 | mem_cgroup_charge_statistics(mem, pc, true); | |
819 | ||
820 | unlock_page_cgroup(pc); | |
821 | } | |
822 | ||
823 | /** | |
824 | * mem_cgroup_move_account - move account of the page | |
825 | * @pc: page_cgroup of the page. | |
826 | * @from: mem_cgroup which the page is moved from. | |
827 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
828 | * | |
829 | * The caller must confirm following. | |
830 | * - page is not on LRU (isolate_page() is useful.) | |
831 | * | |
832 | * returns 0 at success, | |
833 | * returns -EBUSY when lock is busy or "pc" is unstable. | |
834 | * | |
835 | * This function does "uncharge" from old cgroup but doesn't do "charge" to | |
836 | * new cgroup. It should be done by a caller. | |
837 | */ | |
838 | ||
839 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
840 | struct mem_cgroup *from, struct mem_cgroup *to) | |
841 | { | |
842 | struct mem_cgroup_per_zone *from_mz, *to_mz; | |
843 | int nid, zid; | |
844 | int ret = -EBUSY; | |
845 | ||
846 | VM_BUG_ON(from == to); | |
847 | VM_BUG_ON(PageLRU(pc->page)); | |
848 | ||
849 | nid = page_cgroup_nid(pc); | |
850 | zid = page_cgroup_zid(pc); | |
851 | from_mz = mem_cgroup_zoneinfo(from, nid, zid); | |
852 | to_mz = mem_cgroup_zoneinfo(to, nid, zid); | |
853 | ||
854 | if (!trylock_page_cgroup(pc)) | |
855 | return ret; | |
856 | ||
857 | if (!PageCgroupUsed(pc)) | |
858 | goto out; | |
859 | ||
860 | if (pc->mem_cgroup != from) | |
861 | goto out; | |
862 | ||
863 | css_put(&from->css); | |
864 | res_counter_uncharge(&from->res, PAGE_SIZE); | |
865 | mem_cgroup_charge_statistics(from, pc, false); | |
866 | if (do_swap_account) | |
867 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
868 | pc->mem_cgroup = to; | |
869 | mem_cgroup_charge_statistics(to, pc, true); | |
870 | css_get(&to->css); | |
871 | ret = 0; | |
872 | out: | |
873 | unlock_page_cgroup(pc); | |
874 | return ret; | |
875 | } | |
876 | ||
877 | /* | |
878 | * move charges to its parent. | |
879 | */ | |
880 | ||
881 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
882 | struct mem_cgroup *child, | |
883 | gfp_t gfp_mask) | |
884 | { | |
885 | struct page *page = pc->page; | |
886 | struct cgroup *cg = child->css.cgroup; | |
887 | struct cgroup *pcg = cg->parent; | |
888 | struct mem_cgroup *parent; | |
889 | int ret; | |
890 | ||
891 | /* Is ROOT ? */ | |
892 | if (!pcg) | |
893 | return -EINVAL; | |
894 | ||
895 | ||
896 | parent = mem_cgroup_from_cont(pcg); | |
897 | ||
898 | ||
899 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false); | |
900 | if (ret || !parent) | |
901 | return ret; | |
902 | ||
903 | if (!get_page_unless_zero(page)) | |
904 | return -EBUSY; | |
905 | ||
906 | ret = isolate_lru_page(page); | |
907 | ||
908 | if (ret) | |
909 | goto cancel; | |
910 | ||
911 | ret = mem_cgroup_move_account(pc, child, parent); | |
912 | ||
913 | /* drop extra refcnt by try_charge() (move_account increment one) */ | |
914 | css_put(&parent->css); | |
915 | putback_lru_page(page); | |
916 | if (!ret) { | |
917 | put_page(page); | |
918 | return 0; | |
919 | } | |
920 | /* uncharge if move fails */ | |
921 | cancel: | |
922 | res_counter_uncharge(&parent->res, PAGE_SIZE); | |
923 | if (do_swap_account) | |
924 | res_counter_uncharge(&parent->memsw, PAGE_SIZE); | |
925 | put_page(page); | |
926 | return ret; | |
927 | } | |
928 | ||
929 | /* | |
930 | * Charge the memory controller for page usage. | |
931 | * Return | |
932 | * 0 if the charge was successful | |
933 | * < 0 if the cgroup is over its limit | |
934 | */ | |
935 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
936 | gfp_t gfp_mask, enum charge_type ctype, | |
937 | struct mem_cgroup *memcg) | |
938 | { | |
939 | struct mem_cgroup *mem; | |
940 | struct page_cgroup *pc; | |
941 | int ret; | |
942 | ||
943 | pc = lookup_page_cgroup(page); | |
944 | /* can happen at boot */ | |
945 | if (unlikely(!pc)) | |
946 | return 0; | |
947 | prefetchw(pc); | |
948 | ||
949 | mem = memcg; | |
950 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true); | |
951 | if (ret || !mem) | |
952 | return ret; | |
953 | ||
954 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
955 | return 0; | |
956 | } | |
957 | ||
958 | int mem_cgroup_newpage_charge(struct page *page, | |
959 | struct mm_struct *mm, gfp_t gfp_mask) | |
960 | { | |
961 | if (mem_cgroup_disabled()) | |
962 | return 0; | |
963 | if (PageCompound(page)) | |
964 | return 0; | |
965 | /* | |
966 | * If already mapped, we don't have to account. | |
967 | * If page cache, page->mapping has address_space. | |
968 | * But page->mapping may have out-of-use anon_vma pointer, | |
969 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
970 | * is NULL. | |
971 | */ | |
972 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
973 | return 0; | |
974 | if (unlikely(!mm)) | |
975 | mm = &init_mm; | |
976 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
977 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); | |
978 | } | |
979 | ||
980 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, | |
981 | gfp_t gfp_mask) | |
982 | { | |
983 | if (mem_cgroup_disabled()) | |
984 | return 0; | |
985 | if (PageCompound(page)) | |
986 | return 0; | |
987 | /* | |
988 | * Corner case handling. This is called from add_to_page_cache() | |
989 | * in usual. But some FS (shmem) precharges this page before calling it | |
990 | * and call add_to_page_cache() with GFP_NOWAIT. | |
991 | * | |
992 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
993 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
994 | * charge twice. (It works but has to pay a bit larger cost.) | |
995 | */ | |
996 | if (!(gfp_mask & __GFP_WAIT)) { | |
997 | struct page_cgroup *pc; | |
998 | ||
999 | ||
1000 | pc = lookup_page_cgroup(page); | |
1001 | if (!pc) | |
1002 | return 0; | |
1003 | lock_page_cgroup(pc); | |
1004 | if (PageCgroupUsed(pc)) { | |
1005 | unlock_page_cgroup(pc); | |
1006 | return 0; | |
1007 | } | |
1008 | unlock_page_cgroup(pc); | |
1009 | } | |
1010 | ||
1011 | if (unlikely(!mm)) | |
1012 | mm = &init_mm; | |
1013 | ||
1014 | if (page_is_file_cache(page)) | |
1015 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
1016 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); | |
1017 | else | |
1018 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
1019 | MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL); | |
1020 | } | |
1021 | ||
1022 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, | |
1023 | struct page *page, | |
1024 | gfp_t mask, struct mem_cgroup **ptr) | |
1025 | { | |
1026 | struct mem_cgroup *mem; | |
1027 | swp_entry_t ent; | |
1028 | ||
1029 | if (mem_cgroup_disabled()) | |
1030 | return 0; | |
1031 | ||
1032 | if (!do_swap_account) | |
1033 | goto charge_cur_mm; | |
1034 | ||
1035 | /* | |
1036 | * A racing thread's fault, or swapoff, may have already updated | |
1037 | * the pte, and even removed page from swap cache: return success | |
1038 | * to go on to do_swap_page()'s pte_same() test, which should fail. | |
1039 | */ | |
1040 | if (!PageSwapCache(page)) | |
1041 | return 0; | |
1042 | ||
1043 | ent.val = page_private(page); | |
1044 | ||
1045 | mem = lookup_swap_cgroup(ent); | |
1046 | if (!mem || mem->obsolete) | |
1047 | goto charge_cur_mm; | |
1048 | *ptr = mem; | |
1049 | return __mem_cgroup_try_charge(NULL, mask, ptr, true); | |
1050 | charge_cur_mm: | |
1051 | if (unlikely(!mm)) | |
1052 | mm = &init_mm; | |
1053 | return __mem_cgroup_try_charge(mm, mask, ptr, true); | |
1054 | } | |
1055 | ||
1056 | #ifdef CONFIG_SWAP | |
1057 | ||
1058 | int mem_cgroup_cache_charge_swapin(struct page *page, | |
1059 | struct mm_struct *mm, gfp_t mask, bool locked) | |
1060 | { | |
1061 | int ret = 0; | |
1062 | ||
1063 | if (mem_cgroup_disabled()) | |
1064 | return 0; | |
1065 | if (unlikely(!mm)) | |
1066 | mm = &init_mm; | |
1067 | if (!locked) | |
1068 | lock_page(page); | |
1069 | /* | |
1070 | * If not locked, the page can be dropped from SwapCache until | |
1071 | * we reach here. | |
1072 | */ | |
1073 | if (PageSwapCache(page)) { | |
1074 | struct mem_cgroup *mem = NULL; | |
1075 | swp_entry_t ent; | |
1076 | ||
1077 | ent.val = page_private(page); | |
1078 | if (do_swap_account) { | |
1079 | mem = lookup_swap_cgroup(ent); | |
1080 | if (mem && mem->obsolete) | |
1081 | mem = NULL; | |
1082 | if (mem) | |
1083 | mm = NULL; | |
1084 | } | |
1085 | ret = mem_cgroup_charge_common(page, mm, mask, | |
1086 | MEM_CGROUP_CHARGE_TYPE_SHMEM, mem); | |
1087 | ||
1088 | if (!ret && do_swap_account) { | |
1089 | /* avoid double counting */ | |
1090 | mem = swap_cgroup_record(ent, NULL); | |
1091 | if (mem) { | |
1092 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
1093 | mem_cgroup_put(mem); | |
1094 | } | |
1095 | } | |
1096 | } | |
1097 | if (!locked) | |
1098 | unlock_page(page); | |
1099 | /* add this page(page_cgroup) to the LRU we want. */ | |
1100 | mem_cgroup_lru_fixup(page); | |
1101 | ||
1102 | return ret; | |
1103 | } | |
1104 | #endif | |
1105 | ||
1106 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) | |
1107 | { | |
1108 | struct page_cgroup *pc; | |
1109 | ||
1110 | if (mem_cgroup_disabled()) | |
1111 | return; | |
1112 | if (!ptr) | |
1113 | return; | |
1114 | pc = lookup_page_cgroup(page); | |
1115 | __mem_cgroup_commit_charge(ptr, pc, MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
1116 | /* | |
1117 | * Now swap is on-memory. This means this page may be | |
1118 | * counted both as mem and swap....double count. | |
1119 | * Fix it by uncharging from memsw. This SwapCache is stable | |
1120 | * because we're still under lock_page(). | |
1121 | */ | |
1122 | if (do_swap_account) { | |
1123 | swp_entry_t ent = {.val = page_private(page)}; | |
1124 | struct mem_cgroup *memcg; | |
1125 | memcg = swap_cgroup_record(ent, NULL); | |
1126 | if (memcg) { | |
1127 | /* If memcg is obsolete, memcg can be != ptr */ | |
1128 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); | |
1129 | mem_cgroup_put(memcg); | |
1130 | } | |
1131 | ||
1132 | } | |
1133 | /* add this page(page_cgroup) to the LRU we want. */ | |
1134 | mem_cgroup_lru_fixup(page); | |
1135 | } | |
1136 | ||
1137 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) | |
1138 | { | |
1139 | if (mem_cgroup_disabled()) | |
1140 | return; | |
1141 | if (!mem) | |
1142 | return; | |
1143 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
1144 | if (do_swap_account) | |
1145 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
1146 | css_put(&mem->css); | |
1147 | } | |
1148 | ||
1149 | ||
1150 | /* | |
1151 | * uncharge if !page_mapped(page) | |
1152 | */ | |
1153 | static struct mem_cgroup * | |
1154 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) | |
1155 | { | |
1156 | struct page_cgroup *pc; | |
1157 | struct mem_cgroup *mem = NULL; | |
1158 | struct mem_cgroup_per_zone *mz; | |
1159 | ||
1160 | if (mem_cgroup_disabled()) | |
1161 | return NULL; | |
1162 | ||
1163 | if (PageSwapCache(page)) | |
1164 | return NULL; | |
1165 | ||
1166 | /* | |
1167 | * Check if our page_cgroup is valid | |
1168 | */ | |
1169 | pc = lookup_page_cgroup(page); | |
1170 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
1171 | return NULL; | |
1172 | ||
1173 | lock_page_cgroup(pc); | |
1174 | ||
1175 | mem = pc->mem_cgroup; | |
1176 | ||
1177 | if (!PageCgroupUsed(pc)) | |
1178 | goto unlock_out; | |
1179 | ||
1180 | switch (ctype) { | |
1181 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
1182 | if (page_mapped(page)) | |
1183 | goto unlock_out; | |
1184 | break; | |
1185 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
1186 | if (!PageAnon(page)) { /* Shared memory */ | |
1187 | if (page->mapping && !page_is_file_cache(page)) | |
1188 | goto unlock_out; | |
1189 | } else if (page_mapped(page)) /* Anon */ | |
1190 | goto unlock_out; | |
1191 | break; | |
1192 | default: | |
1193 | break; | |
1194 | } | |
1195 | ||
1196 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
1197 | if (do_swap_account && (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)) | |
1198 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
1199 | ||
1200 | mem_cgroup_charge_statistics(mem, pc, false); | |
1201 | ClearPageCgroupUsed(pc); | |
1202 | ||
1203 | mz = page_cgroup_zoneinfo(pc); | |
1204 | unlock_page_cgroup(pc); | |
1205 | ||
1206 | /* at swapout, this memcg will be accessed to record to swap */ | |
1207 | if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
1208 | css_put(&mem->css); | |
1209 | ||
1210 | return mem; | |
1211 | ||
1212 | unlock_out: | |
1213 | unlock_page_cgroup(pc); | |
1214 | return NULL; | |
1215 | } | |
1216 | ||
1217 | void mem_cgroup_uncharge_page(struct page *page) | |
1218 | { | |
1219 | /* early check. */ | |
1220 | if (page_mapped(page)) | |
1221 | return; | |
1222 | if (page->mapping && !PageAnon(page)) | |
1223 | return; | |
1224 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
1225 | } | |
1226 | ||
1227 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
1228 | { | |
1229 | VM_BUG_ON(page_mapped(page)); | |
1230 | VM_BUG_ON(page->mapping); | |
1231 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); | |
1232 | } | |
1233 | ||
1234 | /* | |
1235 | * called from __delete_from_swap_cache() and drop "page" account. | |
1236 | * memcg information is recorded to swap_cgroup of "ent" | |
1237 | */ | |
1238 | void mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent) | |
1239 | { | |
1240 | struct mem_cgroup *memcg; | |
1241 | ||
1242 | memcg = __mem_cgroup_uncharge_common(page, | |
1243 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT); | |
1244 | /* record memcg information */ | |
1245 | if (do_swap_account && memcg) { | |
1246 | swap_cgroup_record(ent, memcg); | |
1247 | mem_cgroup_get(memcg); | |
1248 | } | |
1249 | if (memcg) | |
1250 | css_put(&memcg->css); | |
1251 | } | |
1252 | ||
1253 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
1254 | /* | |
1255 | * called from swap_entry_free(). remove record in swap_cgroup and | |
1256 | * uncharge "memsw" account. | |
1257 | */ | |
1258 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
1259 | { | |
1260 | struct mem_cgroup *memcg; | |
1261 | ||
1262 | if (!do_swap_account) | |
1263 | return; | |
1264 | ||
1265 | memcg = swap_cgroup_record(ent, NULL); | |
1266 | if (memcg) { | |
1267 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); | |
1268 | mem_cgroup_put(memcg); | |
1269 | } | |
1270 | } | |
1271 | #endif | |
1272 | ||
1273 | /* | |
1274 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old | |
1275 | * page belongs to. | |
1276 | */ | |
1277 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) | |
1278 | { | |
1279 | struct page_cgroup *pc; | |
1280 | struct mem_cgroup *mem = NULL; | |
1281 | int ret = 0; | |
1282 | ||
1283 | if (mem_cgroup_disabled()) | |
1284 | return 0; | |
1285 | ||
1286 | pc = lookup_page_cgroup(page); | |
1287 | lock_page_cgroup(pc); | |
1288 | if (PageCgroupUsed(pc)) { | |
1289 | mem = pc->mem_cgroup; | |
1290 | css_get(&mem->css); | |
1291 | } | |
1292 | unlock_page_cgroup(pc); | |
1293 | ||
1294 | if (mem) { | |
1295 | ret = mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem); | |
1296 | css_put(&mem->css); | |
1297 | } | |
1298 | *ptr = mem; | |
1299 | return ret; | |
1300 | } | |
1301 | ||
1302 | /* remove redundant charge if migration failed*/ | |
1303 | void mem_cgroup_end_migration(struct mem_cgroup *mem, | |
1304 | struct page *oldpage, struct page *newpage) | |
1305 | { | |
1306 | struct page *target, *unused; | |
1307 | struct page_cgroup *pc; | |
1308 | enum charge_type ctype; | |
1309 | ||
1310 | if (!mem) | |
1311 | return; | |
1312 | ||
1313 | /* at migration success, oldpage->mapping is NULL. */ | |
1314 | if (oldpage->mapping) { | |
1315 | target = oldpage; | |
1316 | unused = NULL; | |
1317 | } else { | |
1318 | target = newpage; | |
1319 | unused = oldpage; | |
1320 | } | |
1321 | ||
1322 | if (PageAnon(target)) | |
1323 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
1324 | else if (page_is_file_cache(target)) | |
1325 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
1326 | else | |
1327 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
1328 | ||
1329 | /* unused page is not on radix-tree now. */ | |
1330 | if (unused) | |
1331 | __mem_cgroup_uncharge_common(unused, ctype); | |
1332 | ||
1333 | pc = lookup_page_cgroup(target); | |
1334 | /* | |
1335 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. | |
1336 | * So, double-counting is effectively avoided. | |
1337 | */ | |
1338 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
1339 | ||
1340 | /* | |
1341 | * Both of oldpage and newpage are still under lock_page(). | |
1342 | * Then, we don't have to care about race in radix-tree. | |
1343 | * But we have to be careful that this page is unmapped or not. | |
1344 | * | |
1345 | * There is a case for !page_mapped(). At the start of | |
1346 | * migration, oldpage was mapped. But now, it's zapped. | |
1347 | * But we know *target* page is not freed/reused under us. | |
1348 | * mem_cgroup_uncharge_page() does all necessary checks. | |
1349 | */ | |
1350 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) | |
1351 | mem_cgroup_uncharge_page(target); | |
1352 | } | |
1353 | ||
1354 | /* | |
1355 | * A call to try to shrink memory usage under specified resource controller. | |
1356 | * This is typically used for page reclaiming for shmem for reducing side | |
1357 | * effect of page allocation from shmem, which is used by some mem_cgroup. | |
1358 | */ | |
1359 | int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask) | |
1360 | { | |
1361 | struct mem_cgroup *mem; | |
1362 | int progress = 0; | |
1363 | int retry = MEM_CGROUP_RECLAIM_RETRIES; | |
1364 | ||
1365 | if (mem_cgroup_disabled()) | |
1366 | return 0; | |
1367 | if (!mm) | |
1368 | return 0; | |
1369 | ||
1370 | rcu_read_lock(); | |
1371 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1372 | if (unlikely(!mem)) { | |
1373 | rcu_read_unlock(); | |
1374 | return 0; | |
1375 | } | |
1376 | css_get(&mem->css); | |
1377 | rcu_read_unlock(); | |
1378 | ||
1379 | do { | |
1380 | progress = try_to_free_mem_cgroup_pages(mem, gfp_mask, true); | |
1381 | progress += mem_cgroup_check_under_limit(mem); | |
1382 | } while (!progress && --retry); | |
1383 | ||
1384 | css_put(&mem->css); | |
1385 | if (!retry) | |
1386 | return -ENOMEM; | |
1387 | return 0; | |
1388 | } | |
1389 | ||
1390 | /* | |
1391 | * The inactive anon list should be small enough that the VM never has to | |
1392 | * do too much work, but large enough that each inactive page has a chance | |
1393 | * to be referenced again before it is swapped out. | |
1394 | * | |
1395 | * this calculation is straightforward porting from | |
1396 | * page_alloc.c::setup_per_zone_inactive_ratio(). | |
1397 | * it describe more detail. | |
1398 | */ | |
1399 | static void mem_cgroup_set_inactive_ratio(struct mem_cgroup *memcg) | |
1400 | { | |
1401 | unsigned int gb, ratio; | |
1402 | ||
1403 | gb = res_counter_read_u64(&memcg->res, RES_LIMIT) >> 30; | |
1404 | if (gb) | |
1405 | ratio = int_sqrt(10 * gb); | |
1406 | else | |
1407 | ratio = 1; | |
1408 | ||
1409 | memcg->inactive_ratio = ratio; | |
1410 | ||
1411 | } | |
1412 | ||
1413 | static DEFINE_MUTEX(set_limit_mutex); | |
1414 | ||
1415 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, | |
1416 | unsigned long long val) | |
1417 | { | |
1418 | ||
1419 | int retry_count = MEM_CGROUP_RECLAIM_RETRIES; | |
1420 | int progress; | |
1421 | u64 memswlimit; | |
1422 | int ret = 0; | |
1423 | ||
1424 | while (retry_count) { | |
1425 | if (signal_pending(current)) { | |
1426 | ret = -EINTR; | |
1427 | break; | |
1428 | } | |
1429 | /* | |
1430 | * Rather than hide all in some function, I do this in | |
1431 | * open coded manner. You see what this really does. | |
1432 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
1433 | */ | |
1434 | mutex_lock(&set_limit_mutex); | |
1435 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
1436 | if (memswlimit < val) { | |
1437 | ret = -EINVAL; | |
1438 | mutex_unlock(&set_limit_mutex); | |
1439 | break; | |
1440 | } | |
1441 | ret = res_counter_set_limit(&memcg->res, val); | |
1442 | mutex_unlock(&set_limit_mutex); | |
1443 | ||
1444 | if (!ret) | |
1445 | break; | |
1446 | ||
1447 | progress = try_to_free_mem_cgroup_pages(memcg, | |
1448 | GFP_KERNEL, false); | |
1449 | if (!progress) retry_count--; | |
1450 | } | |
1451 | ||
1452 | if (!ret) | |
1453 | mem_cgroup_set_inactive_ratio(memcg); | |
1454 | ||
1455 | return ret; | |
1456 | } | |
1457 | ||
1458 | int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, | |
1459 | unsigned long long val) | |
1460 | { | |
1461 | int retry_count = MEM_CGROUP_RECLAIM_RETRIES; | |
1462 | u64 memlimit, oldusage, curusage; | |
1463 | int ret; | |
1464 | ||
1465 | if (!do_swap_account) | |
1466 | return -EINVAL; | |
1467 | ||
1468 | while (retry_count) { | |
1469 | if (signal_pending(current)) { | |
1470 | ret = -EINTR; | |
1471 | break; | |
1472 | } | |
1473 | /* | |
1474 | * Rather than hide all in some function, I do this in | |
1475 | * open coded manner. You see what this really does. | |
1476 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
1477 | */ | |
1478 | mutex_lock(&set_limit_mutex); | |
1479 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
1480 | if (memlimit > val) { | |
1481 | ret = -EINVAL; | |
1482 | mutex_unlock(&set_limit_mutex); | |
1483 | break; | |
1484 | } | |
1485 | ret = res_counter_set_limit(&memcg->memsw, val); | |
1486 | mutex_unlock(&set_limit_mutex); | |
1487 | ||
1488 | if (!ret) | |
1489 | break; | |
1490 | ||
1491 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
1492 | try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, true); | |
1493 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
1494 | if (curusage >= oldusage) | |
1495 | retry_count--; | |
1496 | } | |
1497 | return ret; | |
1498 | } | |
1499 | ||
1500 | /* | |
1501 | * This routine traverse page_cgroup in given list and drop them all. | |
1502 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. | |
1503 | */ | |
1504 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, | |
1505 | int node, int zid, enum lru_list lru) | |
1506 | { | |
1507 | struct zone *zone; | |
1508 | struct mem_cgroup_per_zone *mz; | |
1509 | struct page_cgroup *pc, *busy; | |
1510 | unsigned long flags, loop; | |
1511 | struct list_head *list; | |
1512 | int ret = 0; | |
1513 | ||
1514 | zone = &NODE_DATA(node)->node_zones[zid]; | |
1515 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
1516 | list = &mz->lists[lru]; | |
1517 | ||
1518 | loop = MEM_CGROUP_ZSTAT(mz, lru); | |
1519 | /* give some margin against EBUSY etc...*/ | |
1520 | loop += 256; | |
1521 | busy = NULL; | |
1522 | while (loop--) { | |
1523 | ret = 0; | |
1524 | spin_lock_irqsave(&zone->lru_lock, flags); | |
1525 | if (list_empty(list)) { | |
1526 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
1527 | break; | |
1528 | } | |
1529 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
1530 | if (busy == pc) { | |
1531 | list_move(&pc->lru, list); | |
1532 | busy = 0; | |
1533 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
1534 | continue; | |
1535 | } | |
1536 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
1537 | ||
1538 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); | |
1539 | if (ret == -ENOMEM) | |
1540 | break; | |
1541 | ||
1542 | if (ret == -EBUSY || ret == -EINVAL) { | |
1543 | /* found lock contention or "pc" is obsolete. */ | |
1544 | busy = pc; | |
1545 | cond_resched(); | |
1546 | } else | |
1547 | busy = NULL; | |
1548 | } | |
1549 | ||
1550 | if (!ret && !list_empty(list)) | |
1551 | return -EBUSY; | |
1552 | return ret; | |
1553 | } | |
1554 | ||
1555 | /* | |
1556 | * make mem_cgroup's charge to be 0 if there is no task. | |
1557 | * This enables deleting this mem_cgroup. | |
1558 | */ | |
1559 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) | |
1560 | { | |
1561 | int ret; | |
1562 | int node, zid, shrink; | |
1563 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
1564 | struct cgroup *cgrp = mem->css.cgroup; | |
1565 | ||
1566 | css_get(&mem->css); | |
1567 | ||
1568 | shrink = 0; | |
1569 | /* should free all ? */ | |
1570 | if (free_all) | |
1571 | goto try_to_free; | |
1572 | move_account: | |
1573 | while (mem->res.usage > 0) { | |
1574 | ret = -EBUSY; | |
1575 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) | |
1576 | goto out; | |
1577 | ret = -EINTR; | |
1578 | if (signal_pending(current)) | |
1579 | goto out; | |
1580 | /* This is for making all *used* pages to be on LRU. */ | |
1581 | lru_add_drain_all(); | |
1582 | ret = 0; | |
1583 | for_each_node_state(node, N_POSSIBLE) { | |
1584 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { | |
1585 | enum lru_list l; | |
1586 | for_each_lru(l) { | |
1587 | ret = mem_cgroup_force_empty_list(mem, | |
1588 | node, zid, l); | |
1589 | if (ret) | |
1590 | break; | |
1591 | } | |
1592 | } | |
1593 | if (ret) | |
1594 | break; | |
1595 | } | |
1596 | /* it seems parent cgroup doesn't have enough mem */ | |
1597 | if (ret == -ENOMEM) | |
1598 | goto try_to_free; | |
1599 | cond_resched(); | |
1600 | } | |
1601 | ret = 0; | |
1602 | out: | |
1603 | css_put(&mem->css); | |
1604 | return ret; | |
1605 | ||
1606 | try_to_free: | |
1607 | /* returns EBUSY if there is a task or if we come here twice. */ | |
1608 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
1609 | ret = -EBUSY; | |
1610 | goto out; | |
1611 | } | |
1612 | /* we call try-to-free pages for make this cgroup empty */ | |
1613 | lru_add_drain_all(); | |
1614 | /* try to free all pages in this cgroup */ | |
1615 | shrink = 1; | |
1616 | while (nr_retries && mem->res.usage > 0) { | |
1617 | int progress; | |
1618 | ||
1619 | if (signal_pending(current)) { | |
1620 | ret = -EINTR; | |
1621 | goto out; | |
1622 | } | |
1623 | progress = try_to_free_mem_cgroup_pages(mem, | |
1624 | GFP_KERNEL, false); | |
1625 | if (!progress) { | |
1626 | nr_retries--; | |
1627 | /* maybe some writeback is necessary */ | |
1628 | congestion_wait(WRITE, HZ/10); | |
1629 | } | |
1630 | ||
1631 | } | |
1632 | lru_add_drain(); | |
1633 | /* try move_account...there may be some *locked* pages. */ | |
1634 | if (mem->res.usage) | |
1635 | goto move_account; | |
1636 | ret = 0; | |
1637 | goto out; | |
1638 | } | |
1639 | ||
1640 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) | |
1641 | { | |
1642 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
1643 | } | |
1644 | ||
1645 | ||
1646 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) | |
1647 | { | |
1648 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
1649 | } | |
1650 | ||
1651 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
1652 | u64 val) | |
1653 | { | |
1654 | int retval = 0; | |
1655 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1656 | struct cgroup *parent = cont->parent; | |
1657 | struct mem_cgroup *parent_mem = NULL; | |
1658 | ||
1659 | if (parent) | |
1660 | parent_mem = mem_cgroup_from_cont(parent); | |
1661 | ||
1662 | cgroup_lock(); | |
1663 | /* | |
1664 | * If parent's use_hiearchy is set, we can't make any modifications | |
1665 | * in the child subtrees. If it is unset, then the change can | |
1666 | * occur, provided the current cgroup has no children. | |
1667 | * | |
1668 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
1669 | * set if there are no children. | |
1670 | */ | |
1671 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
1672 | (val == 1 || val == 0)) { | |
1673 | if (list_empty(&cont->children)) | |
1674 | mem->use_hierarchy = val; | |
1675 | else | |
1676 | retval = -EBUSY; | |
1677 | } else | |
1678 | retval = -EINVAL; | |
1679 | cgroup_unlock(); | |
1680 | ||
1681 | return retval; | |
1682 | } | |
1683 | ||
1684 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) | |
1685 | { | |
1686 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
1687 | u64 val = 0; | |
1688 | int type, name; | |
1689 | ||
1690 | type = MEMFILE_TYPE(cft->private); | |
1691 | name = MEMFILE_ATTR(cft->private); | |
1692 | switch (type) { | |
1693 | case _MEM: | |
1694 | val = res_counter_read_u64(&mem->res, name); | |
1695 | break; | |
1696 | case _MEMSWAP: | |
1697 | if (do_swap_account) | |
1698 | val = res_counter_read_u64(&mem->memsw, name); | |
1699 | break; | |
1700 | default: | |
1701 | BUG(); | |
1702 | break; | |
1703 | } | |
1704 | return val; | |
1705 | } | |
1706 | /* | |
1707 | * The user of this function is... | |
1708 | * RES_LIMIT. | |
1709 | */ | |
1710 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, | |
1711 | const char *buffer) | |
1712 | { | |
1713 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); | |
1714 | int type, name; | |
1715 | unsigned long long val; | |
1716 | int ret; | |
1717 | ||
1718 | type = MEMFILE_TYPE(cft->private); | |
1719 | name = MEMFILE_ATTR(cft->private); | |
1720 | switch (name) { | |
1721 | case RES_LIMIT: | |
1722 | /* This function does all necessary parse...reuse it */ | |
1723 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
1724 | if (ret) | |
1725 | break; | |
1726 | if (type == _MEM) | |
1727 | ret = mem_cgroup_resize_limit(memcg, val); | |
1728 | else | |
1729 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
1730 | break; | |
1731 | default: | |
1732 | ret = -EINVAL; /* should be BUG() ? */ | |
1733 | break; | |
1734 | } | |
1735 | return ret; | |
1736 | } | |
1737 | ||
1738 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) | |
1739 | { | |
1740 | struct mem_cgroup *mem; | |
1741 | int type, name; | |
1742 | ||
1743 | mem = mem_cgroup_from_cont(cont); | |
1744 | type = MEMFILE_TYPE(event); | |
1745 | name = MEMFILE_ATTR(event); | |
1746 | switch (name) { | |
1747 | case RES_MAX_USAGE: | |
1748 | if (type == _MEM) | |
1749 | res_counter_reset_max(&mem->res); | |
1750 | else | |
1751 | res_counter_reset_max(&mem->memsw); | |
1752 | break; | |
1753 | case RES_FAILCNT: | |
1754 | if (type == _MEM) | |
1755 | res_counter_reset_failcnt(&mem->res); | |
1756 | else | |
1757 | res_counter_reset_failcnt(&mem->memsw); | |
1758 | break; | |
1759 | } | |
1760 | return 0; | |
1761 | } | |
1762 | ||
1763 | static const struct mem_cgroup_stat_desc { | |
1764 | const char *msg; | |
1765 | u64 unit; | |
1766 | } mem_cgroup_stat_desc[] = { | |
1767 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, | |
1768 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, | |
1769 | [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, }, | |
1770 | [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, }, | |
1771 | }; | |
1772 | ||
1773 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, | |
1774 | struct cgroup_map_cb *cb) | |
1775 | { | |
1776 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); | |
1777 | struct mem_cgroup_stat *stat = &mem_cont->stat; | |
1778 | int i; | |
1779 | ||
1780 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { | |
1781 | s64 val; | |
1782 | ||
1783 | val = mem_cgroup_read_stat(stat, i); | |
1784 | val *= mem_cgroup_stat_desc[i].unit; | |
1785 | cb->fill(cb, mem_cgroup_stat_desc[i].msg, val); | |
1786 | } | |
1787 | /* showing # of active pages */ | |
1788 | { | |
1789 | unsigned long active_anon, inactive_anon; | |
1790 | unsigned long active_file, inactive_file; | |
1791 | unsigned long unevictable; | |
1792 | ||
1793 | inactive_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1794 | LRU_INACTIVE_ANON); | |
1795 | active_anon = mem_cgroup_get_all_zonestat(mem_cont, | |
1796 | LRU_ACTIVE_ANON); | |
1797 | inactive_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1798 | LRU_INACTIVE_FILE); | |
1799 | active_file = mem_cgroup_get_all_zonestat(mem_cont, | |
1800 | LRU_ACTIVE_FILE); | |
1801 | unevictable = mem_cgroup_get_all_zonestat(mem_cont, | |
1802 | LRU_UNEVICTABLE); | |
1803 | ||
1804 | cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE); | |
1805 | cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE); | |
1806 | cb->fill(cb, "active_file", (active_file) * PAGE_SIZE); | |
1807 | cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE); | |
1808 | cb->fill(cb, "unevictable", unevictable * PAGE_SIZE); | |
1809 | ||
1810 | } | |
1811 | return 0; | |
1812 | } | |
1813 | ||
1814 | ||
1815 | static struct cftype mem_cgroup_files[] = { | |
1816 | { | |
1817 | .name = "usage_in_bytes", | |
1818 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), | |
1819 | .read_u64 = mem_cgroup_read, | |
1820 | }, | |
1821 | { | |
1822 | .name = "max_usage_in_bytes", | |
1823 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), | |
1824 | .trigger = mem_cgroup_reset, | |
1825 | .read_u64 = mem_cgroup_read, | |
1826 | }, | |
1827 | { | |
1828 | .name = "limit_in_bytes", | |
1829 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), | |
1830 | .write_string = mem_cgroup_write, | |
1831 | .read_u64 = mem_cgroup_read, | |
1832 | }, | |
1833 | { | |
1834 | .name = "failcnt", | |
1835 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), | |
1836 | .trigger = mem_cgroup_reset, | |
1837 | .read_u64 = mem_cgroup_read, | |
1838 | }, | |
1839 | { | |
1840 | .name = "stat", | |
1841 | .read_map = mem_control_stat_show, | |
1842 | }, | |
1843 | { | |
1844 | .name = "force_empty", | |
1845 | .trigger = mem_cgroup_force_empty_write, | |
1846 | }, | |
1847 | { | |
1848 | .name = "use_hierarchy", | |
1849 | .write_u64 = mem_cgroup_hierarchy_write, | |
1850 | .read_u64 = mem_cgroup_hierarchy_read, | |
1851 | }, | |
1852 | }; | |
1853 | ||
1854 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
1855 | static struct cftype memsw_cgroup_files[] = { | |
1856 | { | |
1857 | .name = "memsw.usage_in_bytes", | |
1858 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
1859 | .read_u64 = mem_cgroup_read, | |
1860 | }, | |
1861 | { | |
1862 | .name = "memsw.max_usage_in_bytes", | |
1863 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
1864 | .trigger = mem_cgroup_reset, | |
1865 | .read_u64 = mem_cgroup_read, | |
1866 | }, | |
1867 | { | |
1868 | .name = "memsw.limit_in_bytes", | |
1869 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
1870 | .write_string = mem_cgroup_write, | |
1871 | .read_u64 = mem_cgroup_read, | |
1872 | }, | |
1873 | { | |
1874 | .name = "memsw.failcnt", | |
1875 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
1876 | .trigger = mem_cgroup_reset, | |
1877 | .read_u64 = mem_cgroup_read, | |
1878 | }, | |
1879 | }; | |
1880 | ||
1881 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
1882 | { | |
1883 | if (!do_swap_account) | |
1884 | return 0; | |
1885 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
1886 | ARRAY_SIZE(memsw_cgroup_files)); | |
1887 | }; | |
1888 | #else | |
1889 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
1890 | { | |
1891 | return 0; | |
1892 | } | |
1893 | #endif | |
1894 | ||
1895 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) | |
1896 | { | |
1897 | struct mem_cgroup_per_node *pn; | |
1898 | struct mem_cgroup_per_zone *mz; | |
1899 | enum lru_list l; | |
1900 | int zone, tmp = node; | |
1901 | /* | |
1902 | * This routine is called against possible nodes. | |
1903 | * But it's BUG to call kmalloc() against offline node. | |
1904 | * | |
1905 | * TODO: this routine can waste much memory for nodes which will | |
1906 | * never be onlined. It's better to use memory hotplug callback | |
1907 | * function. | |
1908 | */ | |
1909 | if (!node_state(node, N_NORMAL_MEMORY)) | |
1910 | tmp = -1; | |
1911 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
1912 | if (!pn) | |
1913 | return 1; | |
1914 | ||
1915 | mem->info.nodeinfo[node] = pn; | |
1916 | memset(pn, 0, sizeof(*pn)); | |
1917 | ||
1918 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
1919 | mz = &pn->zoneinfo[zone]; | |
1920 | for_each_lru(l) | |
1921 | INIT_LIST_HEAD(&mz->lists[l]); | |
1922 | } | |
1923 | return 0; | |
1924 | } | |
1925 | ||
1926 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) | |
1927 | { | |
1928 | kfree(mem->info.nodeinfo[node]); | |
1929 | } | |
1930 | ||
1931 | static int mem_cgroup_size(void) | |
1932 | { | |
1933 | int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu); | |
1934 | return sizeof(struct mem_cgroup) + cpustat_size; | |
1935 | } | |
1936 | ||
1937 | static struct mem_cgroup *mem_cgroup_alloc(void) | |
1938 | { | |
1939 | struct mem_cgroup *mem; | |
1940 | int size = mem_cgroup_size(); | |
1941 | ||
1942 | if (size < PAGE_SIZE) | |
1943 | mem = kmalloc(size, GFP_KERNEL); | |
1944 | else | |
1945 | mem = vmalloc(size); | |
1946 | ||
1947 | if (mem) | |
1948 | memset(mem, 0, size); | |
1949 | return mem; | |
1950 | } | |
1951 | ||
1952 | /* | |
1953 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
1954 | * (scanning all at force_empty is too costly...) | |
1955 | * | |
1956 | * Instead of clearing all references at force_empty, we remember | |
1957 | * the number of reference from swap_cgroup and free mem_cgroup when | |
1958 | * it goes down to 0. | |
1959 | * | |
1960 | * When mem_cgroup is destroyed, mem->obsolete will be set to 0 and | |
1961 | * entry which points to this memcg will be ignore at swapin. | |
1962 | * | |
1963 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
1964 | */ | |
1965 | ||
1966 | static void mem_cgroup_free(struct mem_cgroup *mem) | |
1967 | { | |
1968 | int node; | |
1969 | ||
1970 | if (atomic_read(&mem->refcnt) > 0) | |
1971 | return; | |
1972 | ||
1973 | ||
1974 | for_each_node_state(node, N_POSSIBLE) | |
1975 | free_mem_cgroup_per_zone_info(mem, node); | |
1976 | ||
1977 | if (mem_cgroup_size() < PAGE_SIZE) | |
1978 | kfree(mem); | |
1979 | else | |
1980 | vfree(mem); | |
1981 | } | |
1982 | ||
1983 | static void mem_cgroup_get(struct mem_cgroup *mem) | |
1984 | { | |
1985 | atomic_inc(&mem->refcnt); | |
1986 | } | |
1987 | ||
1988 | static void mem_cgroup_put(struct mem_cgroup *mem) | |
1989 | { | |
1990 | if (atomic_dec_and_test(&mem->refcnt)) { | |
1991 | if (!mem->obsolete) | |
1992 | return; | |
1993 | mem_cgroup_free(mem); | |
1994 | } | |
1995 | } | |
1996 | ||
1997 | ||
1998 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
1999 | static void __init enable_swap_cgroup(void) | |
2000 | { | |
2001 | if (!mem_cgroup_disabled() && really_do_swap_account) | |
2002 | do_swap_account = 1; | |
2003 | } | |
2004 | #else | |
2005 | static void __init enable_swap_cgroup(void) | |
2006 | { | |
2007 | } | |
2008 | #endif | |
2009 | ||
2010 | static struct cgroup_subsys_state * | |
2011 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
2012 | { | |
2013 | struct mem_cgroup *mem, *parent; | |
2014 | int node; | |
2015 | ||
2016 | mem = mem_cgroup_alloc(); | |
2017 | if (!mem) | |
2018 | return ERR_PTR(-ENOMEM); | |
2019 | ||
2020 | for_each_node_state(node, N_POSSIBLE) | |
2021 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
2022 | goto free_out; | |
2023 | /* root ? */ | |
2024 | if (cont->parent == NULL) { | |
2025 | enable_swap_cgroup(); | |
2026 | parent = NULL; | |
2027 | } else { | |
2028 | parent = mem_cgroup_from_cont(cont->parent); | |
2029 | mem->use_hierarchy = parent->use_hierarchy; | |
2030 | } | |
2031 | ||
2032 | if (parent && parent->use_hierarchy) { | |
2033 | res_counter_init(&mem->res, &parent->res); | |
2034 | res_counter_init(&mem->memsw, &parent->memsw); | |
2035 | } else { | |
2036 | res_counter_init(&mem->res, NULL); | |
2037 | res_counter_init(&mem->memsw, NULL); | |
2038 | } | |
2039 | mem_cgroup_set_inactive_ratio(mem); | |
2040 | mem->last_scanned_child = NULL; | |
2041 | ||
2042 | return &mem->css; | |
2043 | free_out: | |
2044 | for_each_node_state(node, N_POSSIBLE) | |
2045 | free_mem_cgroup_per_zone_info(mem, node); | |
2046 | mem_cgroup_free(mem); | |
2047 | return ERR_PTR(-ENOMEM); | |
2048 | } | |
2049 | ||
2050 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, | |
2051 | struct cgroup *cont) | |
2052 | { | |
2053 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
2054 | mem->obsolete = 1; | |
2055 | mem_cgroup_force_empty(mem, false); | |
2056 | } | |
2057 | ||
2058 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, | |
2059 | struct cgroup *cont) | |
2060 | { | |
2061 | mem_cgroup_free(mem_cgroup_from_cont(cont)); | |
2062 | } | |
2063 | ||
2064 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
2065 | struct cgroup *cont) | |
2066 | { | |
2067 | int ret; | |
2068 | ||
2069 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
2070 | ARRAY_SIZE(mem_cgroup_files)); | |
2071 | ||
2072 | if (!ret) | |
2073 | ret = register_memsw_files(cont, ss); | |
2074 | return ret; | |
2075 | } | |
2076 | ||
2077 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, | |
2078 | struct cgroup *cont, | |
2079 | struct cgroup *old_cont, | |
2080 | struct task_struct *p) | |
2081 | { | |
2082 | /* | |
2083 | * FIXME: It's better to move charges of this process from old | |
2084 | * memcg to new memcg. But it's just on TODO-List now. | |
2085 | */ | |
2086 | } | |
2087 | ||
2088 | struct cgroup_subsys mem_cgroup_subsys = { | |
2089 | .name = "memory", | |
2090 | .subsys_id = mem_cgroup_subsys_id, | |
2091 | .create = mem_cgroup_create, | |
2092 | .pre_destroy = mem_cgroup_pre_destroy, | |
2093 | .destroy = mem_cgroup_destroy, | |
2094 | .populate = mem_cgroup_populate, | |
2095 | .attach = mem_cgroup_move_task, | |
2096 | .early_init = 0, | |
2097 | }; | |
2098 | ||
2099 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
2100 | ||
2101 | static int __init disable_swap_account(char *s) | |
2102 | { | |
2103 | really_do_swap_account = 0; | |
2104 | return 1; | |
2105 | } | |
2106 | __setup("noswapaccount", disable_swap_account); | |
2107 | #endif |