]>
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/limits.h> | |
31 | #include <linux/mutex.h> | |
32 | #include <linux/rbtree.h> | |
33 | #include <linux/slab.h> | |
34 | #include <linux/swap.h> | |
35 | #include <linux/spinlock.h> | |
36 | #include <linux/fs.h> | |
37 | #include <linux/seq_file.h> | |
38 | #include <linux/vmalloc.h> | |
39 | #include <linux/mm_inline.h> | |
40 | #include <linux/page_cgroup.h> | |
41 | #include <linux/cpu.h> | |
42 | #include "internal.h" | |
43 | ||
44 | #include <asm/uaccess.h> | |
45 | ||
46 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; | |
47 | #define MEM_CGROUP_RECLAIM_RETRIES 5 | |
48 | struct mem_cgroup *root_mem_cgroup __read_mostly; | |
49 | ||
50 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
51 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ | |
52 | int do_swap_account __read_mostly; | |
53 | static int really_do_swap_account __initdata = 1; /* for remember boot option*/ | |
54 | #else | |
55 | #define do_swap_account (0) | |
56 | #endif | |
57 | ||
58 | static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */ | |
59 | #define SOFTLIMIT_EVENTS_THRESH (1000) | |
60 | ||
61 | /* | |
62 | * Statistics for memory cgroup. | |
63 | */ | |
64 | enum mem_cgroup_stat_index { | |
65 | /* | |
66 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
67 | */ | |
68 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
69 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ | |
70 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ | |
71 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ | |
72 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
73 | MEM_CGROUP_STAT_EVENTS, /* sum of pagein + pageout for internal use */ | |
74 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ | |
75 | ||
76 | MEM_CGROUP_STAT_NSTATS, | |
77 | }; | |
78 | ||
79 | struct mem_cgroup_stat_cpu { | |
80 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
81 | } ____cacheline_aligned_in_smp; | |
82 | ||
83 | struct mem_cgroup_stat { | |
84 | struct mem_cgroup_stat_cpu cpustat[0]; | |
85 | }; | |
86 | ||
87 | static inline void | |
88 | __mem_cgroup_stat_reset_safe(struct mem_cgroup_stat_cpu *stat, | |
89 | enum mem_cgroup_stat_index idx) | |
90 | { | |
91 | stat->count[idx] = 0; | |
92 | } | |
93 | ||
94 | static inline s64 | |
95 | __mem_cgroup_stat_read_local(struct mem_cgroup_stat_cpu *stat, | |
96 | enum mem_cgroup_stat_index idx) | |
97 | { | |
98 | return stat->count[idx]; | |
99 | } | |
100 | ||
101 | /* | |
102 | * For accounting under irq disable, no need for increment preempt count. | |
103 | */ | |
104 | static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, | |
105 | enum mem_cgroup_stat_index idx, int val) | |
106 | { | |
107 | stat->count[idx] += val; | |
108 | } | |
109 | ||
110 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, | |
111 | enum mem_cgroup_stat_index idx) | |
112 | { | |
113 | int cpu; | |
114 | s64 ret = 0; | |
115 | for_each_possible_cpu(cpu) | |
116 | ret += stat->cpustat[cpu].count[idx]; | |
117 | return ret; | |
118 | } | |
119 | ||
120 | static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat) | |
121 | { | |
122 | s64 ret; | |
123 | ||
124 | ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE); | |
125 | ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS); | |
126 | return ret; | |
127 | } | |
128 | ||
129 | /* | |
130 | * per-zone information in memory controller. | |
131 | */ | |
132 | struct mem_cgroup_per_zone { | |
133 | /* | |
134 | * spin_lock to protect the per cgroup LRU | |
135 | */ | |
136 | struct list_head lists[NR_LRU_LISTS]; | |
137 | unsigned long count[NR_LRU_LISTS]; | |
138 | ||
139 | struct zone_reclaim_stat reclaim_stat; | |
140 | struct rb_node tree_node; /* RB tree node */ | |
141 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
142 | /* the soft limit is exceeded*/ | |
143 | bool on_tree; | |
144 | struct mem_cgroup *mem; /* Back pointer, we cannot */ | |
145 | /* use container_of */ | |
146 | }; | |
147 | /* Macro for accessing counter */ | |
148 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
149 | ||
150 | struct mem_cgroup_per_node { | |
151 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
152 | }; | |
153 | ||
154 | struct mem_cgroup_lru_info { | |
155 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
156 | }; | |
157 | ||
158 | /* | |
159 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
160 | * their hierarchy representation | |
161 | */ | |
162 | ||
163 | struct mem_cgroup_tree_per_zone { | |
164 | struct rb_root rb_root; | |
165 | spinlock_t lock; | |
166 | }; | |
167 | ||
168 | struct mem_cgroup_tree_per_node { | |
169 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
170 | }; | |
171 | ||
172 | struct mem_cgroup_tree { | |
173 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
174 | }; | |
175 | ||
176 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
177 | ||
178 | /* | |
179 | * The memory controller data structure. The memory controller controls both | |
180 | * page cache and RSS per cgroup. We would eventually like to provide | |
181 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
182 | * to help the administrator determine what knobs to tune. | |
183 | * | |
184 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
185 | * we hit the water mark. May be even add a low water mark, such that | |
186 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
187 | * a feature that will be implemented much later in the future. | |
188 | */ | |
189 | struct mem_cgroup { | |
190 | struct cgroup_subsys_state css; | |
191 | /* | |
192 | * the counter to account for memory usage | |
193 | */ | |
194 | struct res_counter res; | |
195 | /* | |
196 | * the counter to account for mem+swap usage. | |
197 | */ | |
198 | struct res_counter memsw; | |
199 | /* | |
200 | * Per cgroup active and inactive list, similar to the | |
201 | * per zone LRU lists. | |
202 | */ | |
203 | struct mem_cgroup_lru_info info; | |
204 | ||
205 | /* | |
206 | protect against reclaim related member. | |
207 | */ | |
208 | spinlock_t reclaim_param_lock; | |
209 | ||
210 | int prev_priority; /* for recording reclaim priority */ | |
211 | ||
212 | /* | |
213 | * While reclaiming in a hierarchy, we cache the last child we | |
214 | * reclaimed from. | |
215 | */ | |
216 | int last_scanned_child; | |
217 | /* | |
218 | * Should the accounting and control be hierarchical, per subtree? | |
219 | */ | |
220 | bool use_hierarchy; | |
221 | unsigned long last_oom_jiffies; | |
222 | atomic_t refcnt; | |
223 | ||
224 | unsigned int swappiness; | |
225 | ||
226 | /* set when res.limit == memsw.limit */ | |
227 | bool memsw_is_minimum; | |
228 | ||
229 | /* | |
230 | * statistics. This must be placed at the end of memcg. | |
231 | */ | |
232 | struct mem_cgroup_stat stat; | |
233 | }; | |
234 | ||
235 | /* | |
236 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
237 | * limit reclaim to prevent infinite loops, if they ever occur. | |
238 | */ | |
239 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
240 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
241 | ||
242 | enum charge_type { | |
243 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
244 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
245 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ | |
246 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ | |
247 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ | |
248 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ | |
249 | NR_CHARGE_TYPE, | |
250 | }; | |
251 | ||
252 | /* only for here (for easy reading.) */ | |
253 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
254 | #define PCGF_USED (1UL << PCG_USED) | |
255 | #define PCGF_LOCK (1UL << PCG_LOCK) | |
256 | /* Not used, but added here for completeness */ | |
257 | #define PCGF_ACCT (1UL << PCG_ACCT) | |
258 | ||
259 | /* for encoding cft->private value on file */ | |
260 | #define _MEM (0) | |
261 | #define _MEMSWAP (1) | |
262 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
263 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
264 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
265 | ||
266 | /* | |
267 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
268 | */ | |
269 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
270 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
271 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
272 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
273 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 | |
274 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) | |
275 | ||
276 | static void mem_cgroup_get(struct mem_cgroup *mem); | |
277 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
278 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); | |
279 | static void drain_all_stock_async(void); | |
280 | ||
281 | static struct mem_cgroup_per_zone * | |
282 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
283 | { | |
284 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
285 | } | |
286 | ||
287 | static struct mem_cgroup_per_zone * | |
288 | page_cgroup_zoneinfo(struct page_cgroup *pc) | |
289 | { | |
290 | struct mem_cgroup *mem = pc->mem_cgroup; | |
291 | int nid = page_cgroup_nid(pc); | |
292 | int zid = page_cgroup_zid(pc); | |
293 | ||
294 | if (!mem) | |
295 | return NULL; | |
296 | ||
297 | return mem_cgroup_zoneinfo(mem, nid, zid); | |
298 | } | |
299 | ||
300 | static struct mem_cgroup_tree_per_zone * | |
301 | soft_limit_tree_node_zone(int nid, int zid) | |
302 | { | |
303 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
304 | } | |
305 | ||
306 | static struct mem_cgroup_tree_per_zone * | |
307 | soft_limit_tree_from_page(struct page *page) | |
308 | { | |
309 | int nid = page_to_nid(page); | |
310 | int zid = page_zonenum(page); | |
311 | ||
312 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
313 | } | |
314 | ||
315 | static void | |
316 | __mem_cgroup_insert_exceeded(struct mem_cgroup *mem, | |
317 | struct mem_cgroup_per_zone *mz, | |
318 | struct mem_cgroup_tree_per_zone *mctz, | |
319 | unsigned long long new_usage_in_excess) | |
320 | { | |
321 | struct rb_node **p = &mctz->rb_root.rb_node; | |
322 | struct rb_node *parent = NULL; | |
323 | struct mem_cgroup_per_zone *mz_node; | |
324 | ||
325 | if (mz->on_tree) | |
326 | return; | |
327 | ||
328 | mz->usage_in_excess = new_usage_in_excess; | |
329 | if (!mz->usage_in_excess) | |
330 | return; | |
331 | while (*p) { | |
332 | parent = *p; | |
333 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
334 | tree_node); | |
335 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
336 | p = &(*p)->rb_left; | |
337 | /* | |
338 | * We can't avoid mem cgroups that are over their soft | |
339 | * limit by the same amount | |
340 | */ | |
341 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
342 | p = &(*p)->rb_right; | |
343 | } | |
344 | rb_link_node(&mz->tree_node, parent, p); | |
345 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
346 | mz->on_tree = true; | |
347 | } | |
348 | ||
349 | static void | |
350 | __mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
351 | struct mem_cgroup_per_zone *mz, | |
352 | struct mem_cgroup_tree_per_zone *mctz) | |
353 | { | |
354 | if (!mz->on_tree) | |
355 | return; | |
356 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
357 | mz->on_tree = false; | |
358 | } | |
359 | ||
360 | static void | |
361 | mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
362 | struct mem_cgroup_per_zone *mz, | |
363 | struct mem_cgroup_tree_per_zone *mctz) | |
364 | { | |
365 | spin_lock(&mctz->lock); | |
366 | __mem_cgroup_remove_exceeded(mem, mz, mctz); | |
367 | spin_unlock(&mctz->lock); | |
368 | } | |
369 | ||
370 | static bool mem_cgroup_soft_limit_check(struct mem_cgroup *mem) | |
371 | { | |
372 | bool ret = false; | |
373 | int cpu; | |
374 | s64 val; | |
375 | struct mem_cgroup_stat_cpu *cpustat; | |
376 | ||
377 | cpu = get_cpu(); | |
378 | cpustat = &mem->stat.cpustat[cpu]; | |
379 | val = __mem_cgroup_stat_read_local(cpustat, MEM_CGROUP_STAT_EVENTS); | |
380 | if (unlikely(val > SOFTLIMIT_EVENTS_THRESH)) { | |
381 | __mem_cgroup_stat_reset_safe(cpustat, MEM_CGROUP_STAT_EVENTS); | |
382 | ret = true; | |
383 | } | |
384 | put_cpu(); | |
385 | return ret; | |
386 | } | |
387 | ||
388 | static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page) | |
389 | { | |
390 | unsigned long long excess; | |
391 | struct mem_cgroup_per_zone *mz; | |
392 | struct mem_cgroup_tree_per_zone *mctz; | |
393 | int nid = page_to_nid(page); | |
394 | int zid = page_zonenum(page); | |
395 | mctz = soft_limit_tree_from_page(page); | |
396 | ||
397 | /* | |
398 | * Necessary to update all ancestors when hierarchy is used. | |
399 | * because their event counter is not touched. | |
400 | */ | |
401 | for (; mem; mem = parent_mem_cgroup(mem)) { | |
402 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
403 | excess = res_counter_soft_limit_excess(&mem->res); | |
404 | /* | |
405 | * We have to update the tree if mz is on RB-tree or | |
406 | * mem is over its softlimit. | |
407 | */ | |
408 | if (excess || mz->on_tree) { | |
409 | spin_lock(&mctz->lock); | |
410 | /* if on-tree, remove it */ | |
411 | if (mz->on_tree) | |
412 | __mem_cgroup_remove_exceeded(mem, mz, mctz); | |
413 | /* | |
414 | * Insert again. mz->usage_in_excess will be updated. | |
415 | * If excess is 0, no tree ops. | |
416 | */ | |
417 | __mem_cgroup_insert_exceeded(mem, mz, mctz, excess); | |
418 | spin_unlock(&mctz->lock); | |
419 | } | |
420 | } | |
421 | } | |
422 | ||
423 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem) | |
424 | { | |
425 | int node, zone; | |
426 | struct mem_cgroup_per_zone *mz; | |
427 | struct mem_cgroup_tree_per_zone *mctz; | |
428 | ||
429 | for_each_node_state(node, N_POSSIBLE) { | |
430 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
431 | mz = mem_cgroup_zoneinfo(mem, node, zone); | |
432 | mctz = soft_limit_tree_node_zone(node, zone); | |
433 | mem_cgroup_remove_exceeded(mem, mz, mctz); | |
434 | } | |
435 | } | |
436 | } | |
437 | ||
438 | static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem) | |
439 | { | |
440 | return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT; | |
441 | } | |
442 | ||
443 | static struct mem_cgroup_per_zone * | |
444 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
445 | { | |
446 | struct rb_node *rightmost = NULL; | |
447 | struct mem_cgroup_per_zone *mz; | |
448 | ||
449 | retry: | |
450 | mz = NULL; | |
451 | rightmost = rb_last(&mctz->rb_root); | |
452 | if (!rightmost) | |
453 | goto done; /* Nothing to reclaim from */ | |
454 | ||
455 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
456 | /* | |
457 | * Remove the node now but someone else can add it back, | |
458 | * we will to add it back at the end of reclaim to its correct | |
459 | * position in the tree. | |
460 | */ | |
461 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
462 | if (!res_counter_soft_limit_excess(&mz->mem->res) || | |
463 | !css_tryget(&mz->mem->css)) | |
464 | goto retry; | |
465 | done: | |
466 | return mz; | |
467 | } | |
468 | ||
469 | static struct mem_cgroup_per_zone * | |
470 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
471 | { | |
472 | struct mem_cgroup_per_zone *mz; | |
473 | ||
474 | spin_lock(&mctz->lock); | |
475 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
476 | spin_unlock(&mctz->lock); | |
477 | return mz; | |
478 | } | |
479 | ||
480 | static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, | |
481 | bool charge) | |
482 | { | |
483 | int val = (charge) ? 1 : -1; | |
484 | struct mem_cgroup_stat *stat = &mem->stat; | |
485 | struct mem_cgroup_stat_cpu *cpustat; | |
486 | int cpu = get_cpu(); | |
487 | ||
488 | cpustat = &stat->cpustat[cpu]; | |
489 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_SWAPOUT, val); | |
490 | put_cpu(); | |
491 | } | |
492 | ||
493 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, | |
494 | struct page_cgroup *pc, | |
495 | bool charge) | |
496 | { | |
497 | int val = (charge) ? 1 : -1; | |
498 | struct mem_cgroup_stat *stat = &mem->stat; | |
499 | struct mem_cgroup_stat_cpu *cpustat; | |
500 | int cpu = get_cpu(); | |
501 | ||
502 | cpustat = &stat->cpustat[cpu]; | |
503 | if (PageCgroupCache(pc)) | |
504 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); | |
505 | else | |
506 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); | |
507 | ||
508 | if (charge) | |
509 | __mem_cgroup_stat_add_safe(cpustat, | |
510 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); | |
511 | else | |
512 | __mem_cgroup_stat_add_safe(cpustat, | |
513 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); | |
514 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_EVENTS, 1); | |
515 | put_cpu(); | |
516 | } | |
517 | ||
518 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, | |
519 | enum lru_list idx) | |
520 | { | |
521 | int nid, zid; | |
522 | struct mem_cgroup_per_zone *mz; | |
523 | u64 total = 0; | |
524 | ||
525 | for_each_online_node(nid) | |
526 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
527 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
528 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
529 | } | |
530 | return total; | |
531 | } | |
532 | ||
533 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) | |
534 | { | |
535 | return container_of(cgroup_subsys_state(cont, | |
536 | mem_cgroup_subsys_id), struct mem_cgroup, | |
537 | css); | |
538 | } | |
539 | ||
540 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) | |
541 | { | |
542 | /* | |
543 | * mm_update_next_owner() may clear mm->owner to NULL | |
544 | * if it races with swapoff, page migration, etc. | |
545 | * So this can be called with p == NULL. | |
546 | */ | |
547 | if (unlikely(!p)) | |
548 | return NULL; | |
549 | ||
550 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), | |
551 | struct mem_cgroup, css); | |
552 | } | |
553 | ||
554 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) | |
555 | { | |
556 | struct mem_cgroup *mem = NULL; | |
557 | ||
558 | if (!mm) | |
559 | return NULL; | |
560 | /* | |
561 | * Because we have no locks, mm->owner's may be being moved to other | |
562 | * cgroup. We use css_tryget() here even if this looks | |
563 | * pessimistic (rather than adding locks here). | |
564 | */ | |
565 | rcu_read_lock(); | |
566 | do { | |
567 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
568 | if (unlikely(!mem)) | |
569 | break; | |
570 | } while (!css_tryget(&mem->css)); | |
571 | rcu_read_unlock(); | |
572 | return mem; | |
573 | } | |
574 | ||
575 | /* | |
576 | * Call callback function against all cgroup under hierarchy tree. | |
577 | */ | |
578 | static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data, | |
579 | int (*func)(struct mem_cgroup *, void *)) | |
580 | { | |
581 | int found, ret, nextid; | |
582 | struct cgroup_subsys_state *css; | |
583 | struct mem_cgroup *mem; | |
584 | ||
585 | if (!root->use_hierarchy) | |
586 | return (*func)(root, data); | |
587 | ||
588 | nextid = 1; | |
589 | do { | |
590 | ret = 0; | |
591 | mem = NULL; | |
592 | ||
593 | rcu_read_lock(); | |
594 | css = css_get_next(&mem_cgroup_subsys, nextid, &root->css, | |
595 | &found); | |
596 | if (css && css_tryget(css)) | |
597 | mem = container_of(css, struct mem_cgroup, css); | |
598 | rcu_read_unlock(); | |
599 | ||
600 | if (mem) { | |
601 | ret = (*func)(mem, data); | |
602 | css_put(&mem->css); | |
603 | } | |
604 | nextid = found + 1; | |
605 | } while (!ret && css); | |
606 | ||
607 | return ret; | |
608 | } | |
609 | ||
610 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) | |
611 | { | |
612 | return (mem == root_mem_cgroup); | |
613 | } | |
614 | ||
615 | /* | |
616 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
617 | * Operations are called by routine of global LRU independently from memcg. | |
618 | * What we have to take care of here is validness of pc->mem_cgroup. | |
619 | * | |
620 | * Changes to pc->mem_cgroup happens when | |
621 | * 1. charge | |
622 | * 2. moving account | |
623 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
624 | * It is added to LRU before charge. | |
625 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
626 | * When moving account, the page is not on LRU. It's isolated. | |
627 | */ | |
628 | ||
629 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) | |
630 | { | |
631 | struct page_cgroup *pc; | |
632 | struct mem_cgroup_per_zone *mz; | |
633 | ||
634 | if (mem_cgroup_disabled()) | |
635 | return; | |
636 | pc = lookup_page_cgroup(page); | |
637 | /* can happen while we handle swapcache. */ | |
638 | if (!TestClearPageCgroupAcctLRU(pc)) | |
639 | return; | |
640 | VM_BUG_ON(!pc->mem_cgroup); | |
641 | /* | |
642 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
643 | * removed from global LRU. | |
644 | */ | |
645 | mz = page_cgroup_zoneinfo(pc); | |
646 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; | |
647 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
648 | return; | |
649 | VM_BUG_ON(list_empty(&pc->lru)); | |
650 | list_del_init(&pc->lru); | |
651 | return; | |
652 | } | |
653 | ||
654 | void mem_cgroup_del_lru(struct page *page) | |
655 | { | |
656 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
657 | } | |
658 | ||
659 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) | |
660 | { | |
661 | struct mem_cgroup_per_zone *mz; | |
662 | struct page_cgroup *pc; | |
663 | ||
664 | if (mem_cgroup_disabled()) | |
665 | return; | |
666 | ||
667 | pc = lookup_page_cgroup(page); | |
668 | /* | |
669 | * Used bit is set without atomic ops but after smp_wmb(). | |
670 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
671 | */ | |
672 | smp_rmb(); | |
673 | /* unused or root page is not rotated. */ | |
674 | if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup)) | |
675 | return; | |
676 | mz = page_cgroup_zoneinfo(pc); | |
677 | list_move(&pc->lru, &mz->lists[lru]); | |
678 | } | |
679 | ||
680 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) | |
681 | { | |
682 | struct page_cgroup *pc; | |
683 | struct mem_cgroup_per_zone *mz; | |
684 | ||
685 | if (mem_cgroup_disabled()) | |
686 | return; | |
687 | pc = lookup_page_cgroup(page); | |
688 | VM_BUG_ON(PageCgroupAcctLRU(pc)); | |
689 | /* | |
690 | * Used bit is set without atomic ops but after smp_wmb(). | |
691 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
692 | */ | |
693 | smp_rmb(); | |
694 | if (!PageCgroupUsed(pc)) | |
695 | return; | |
696 | ||
697 | mz = page_cgroup_zoneinfo(pc); | |
698 | MEM_CGROUP_ZSTAT(mz, lru) += 1; | |
699 | SetPageCgroupAcctLRU(pc); | |
700 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
701 | return; | |
702 | list_add(&pc->lru, &mz->lists[lru]); | |
703 | } | |
704 | ||
705 | /* | |
706 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to | |
707 | * lru because the page may.be reused after it's fully uncharged (because of | |
708 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge | |
709 | * it again. This function is only used to charge SwapCache. It's done under | |
710 | * lock_page and expected that zone->lru_lock is never held. | |
711 | */ | |
712 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) | |
713 | { | |
714 | unsigned long flags; | |
715 | struct zone *zone = page_zone(page); | |
716 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
717 | ||
718 | spin_lock_irqsave(&zone->lru_lock, flags); | |
719 | /* | |
720 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
721 | * is guarded by lock_page() because the page is SwapCache. | |
722 | */ | |
723 | if (!PageCgroupUsed(pc)) | |
724 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
725 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
726 | } | |
727 | ||
728 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) | |
729 | { | |
730 | unsigned long flags; | |
731 | struct zone *zone = page_zone(page); | |
732 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
733 | ||
734 | spin_lock_irqsave(&zone->lru_lock, flags); | |
735 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
736 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) | |
737 | mem_cgroup_add_lru_list(page, page_lru(page)); | |
738 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
739 | } | |
740 | ||
741 | ||
742 | void mem_cgroup_move_lists(struct page *page, | |
743 | enum lru_list from, enum lru_list to) | |
744 | { | |
745 | if (mem_cgroup_disabled()) | |
746 | return; | |
747 | mem_cgroup_del_lru_list(page, from); | |
748 | mem_cgroup_add_lru_list(page, to); | |
749 | } | |
750 | ||
751 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) | |
752 | { | |
753 | int ret; | |
754 | struct mem_cgroup *curr = NULL; | |
755 | ||
756 | task_lock(task); | |
757 | rcu_read_lock(); | |
758 | curr = try_get_mem_cgroup_from_mm(task->mm); | |
759 | rcu_read_unlock(); | |
760 | task_unlock(task); | |
761 | if (!curr) | |
762 | return 0; | |
763 | if (curr->use_hierarchy) | |
764 | ret = css_is_ancestor(&curr->css, &mem->css); | |
765 | else | |
766 | ret = (curr == mem); | |
767 | css_put(&curr->css); | |
768 | return ret; | |
769 | } | |
770 | ||
771 | /* | |
772 | * prev_priority control...this will be used in memory reclaim path. | |
773 | */ | |
774 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
775 | { | |
776 | int prev_priority; | |
777 | ||
778 | spin_lock(&mem->reclaim_param_lock); | |
779 | prev_priority = mem->prev_priority; | |
780 | spin_unlock(&mem->reclaim_param_lock); | |
781 | ||
782 | return prev_priority; | |
783 | } | |
784 | ||
785 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
786 | { | |
787 | spin_lock(&mem->reclaim_param_lock); | |
788 | if (priority < mem->prev_priority) | |
789 | mem->prev_priority = priority; | |
790 | spin_unlock(&mem->reclaim_param_lock); | |
791 | } | |
792 | ||
793 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
794 | { | |
795 | spin_lock(&mem->reclaim_param_lock); | |
796 | mem->prev_priority = priority; | |
797 | spin_unlock(&mem->reclaim_param_lock); | |
798 | } | |
799 | ||
800 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) | |
801 | { | |
802 | unsigned long active; | |
803 | unsigned long inactive; | |
804 | unsigned long gb; | |
805 | unsigned long inactive_ratio; | |
806 | ||
807 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); | |
808 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); | |
809 | ||
810 | gb = (inactive + active) >> (30 - PAGE_SHIFT); | |
811 | if (gb) | |
812 | inactive_ratio = int_sqrt(10 * gb); | |
813 | else | |
814 | inactive_ratio = 1; | |
815 | ||
816 | if (present_pages) { | |
817 | present_pages[0] = inactive; | |
818 | present_pages[1] = active; | |
819 | } | |
820 | ||
821 | return inactive_ratio; | |
822 | } | |
823 | ||
824 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
825 | { | |
826 | unsigned long active; | |
827 | unsigned long inactive; | |
828 | unsigned long present_pages[2]; | |
829 | unsigned long inactive_ratio; | |
830 | ||
831 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
832 | ||
833 | inactive = present_pages[0]; | |
834 | active = present_pages[1]; | |
835 | ||
836 | if (inactive * inactive_ratio < active) | |
837 | return 1; | |
838 | ||
839 | return 0; | |
840 | } | |
841 | ||
842 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) | |
843 | { | |
844 | unsigned long active; | |
845 | unsigned long inactive; | |
846 | ||
847 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE); | |
848 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE); | |
849 | ||
850 | return (active > inactive); | |
851 | } | |
852 | ||
853 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, | |
854 | struct zone *zone, | |
855 | enum lru_list lru) | |
856 | { | |
857 | int nid = zone->zone_pgdat->node_id; | |
858 | int zid = zone_idx(zone); | |
859 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
860 | ||
861 | return MEM_CGROUP_ZSTAT(mz, lru); | |
862 | } | |
863 | ||
864 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, | |
865 | struct zone *zone) | |
866 | { | |
867 | int nid = zone->zone_pgdat->node_id; | |
868 | int zid = zone_idx(zone); | |
869 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
870 | ||
871 | return &mz->reclaim_stat; | |
872 | } | |
873 | ||
874 | struct zone_reclaim_stat * | |
875 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
876 | { | |
877 | struct page_cgroup *pc; | |
878 | struct mem_cgroup_per_zone *mz; | |
879 | ||
880 | if (mem_cgroup_disabled()) | |
881 | return NULL; | |
882 | ||
883 | pc = lookup_page_cgroup(page); | |
884 | /* | |
885 | * Used bit is set without atomic ops but after smp_wmb(). | |
886 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
887 | */ | |
888 | smp_rmb(); | |
889 | if (!PageCgroupUsed(pc)) | |
890 | return NULL; | |
891 | ||
892 | mz = page_cgroup_zoneinfo(pc); | |
893 | if (!mz) | |
894 | return NULL; | |
895 | ||
896 | return &mz->reclaim_stat; | |
897 | } | |
898 | ||
899 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, | |
900 | struct list_head *dst, | |
901 | unsigned long *scanned, int order, | |
902 | int mode, struct zone *z, | |
903 | struct mem_cgroup *mem_cont, | |
904 | int active, int file) | |
905 | { | |
906 | unsigned long nr_taken = 0; | |
907 | struct page *page; | |
908 | unsigned long scan; | |
909 | LIST_HEAD(pc_list); | |
910 | struct list_head *src; | |
911 | struct page_cgroup *pc, *tmp; | |
912 | int nid = z->zone_pgdat->node_id; | |
913 | int zid = zone_idx(z); | |
914 | struct mem_cgroup_per_zone *mz; | |
915 | int lru = LRU_FILE * file + active; | |
916 | int ret; | |
917 | ||
918 | BUG_ON(!mem_cont); | |
919 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
920 | src = &mz->lists[lru]; | |
921 | ||
922 | scan = 0; | |
923 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
924 | if (scan >= nr_to_scan) | |
925 | break; | |
926 | ||
927 | page = pc->page; | |
928 | if (unlikely(!PageCgroupUsed(pc))) | |
929 | continue; | |
930 | if (unlikely(!PageLRU(page))) | |
931 | continue; | |
932 | ||
933 | scan++; | |
934 | ret = __isolate_lru_page(page, mode, file); | |
935 | switch (ret) { | |
936 | case 0: | |
937 | list_move(&page->lru, dst); | |
938 | mem_cgroup_del_lru(page); | |
939 | nr_taken++; | |
940 | break; | |
941 | case -EBUSY: | |
942 | /* we don't affect global LRU but rotate in our LRU */ | |
943 | mem_cgroup_rotate_lru_list(page, page_lru(page)); | |
944 | break; | |
945 | default: | |
946 | break; | |
947 | } | |
948 | } | |
949 | ||
950 | *scanned = scan; | |
951 | return nr_taken; | |
952 | } | |
953 | ||
954 | #define mem_cgroup_from_res_counter(counter, member) \ | |
955 | container_of(counter, struct mem_cgroup, member) | |
956 | ||
957 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) | |
958 | { | |
959 | if (do_swap_account) { | |
960 | if (res_counter_check_under_limit(&mem->res) && | |
961 | res_counter_check_under_limit(&mem->memsw)) | |
962 | return true; | |
963 | } else | |
964 | if (res_counter_check_under_limit(&mem->res)) | |
965 | return true; | |
966 | return false; | |
967 | } | |
968 | ||
969 | static unsigned int get_swappiness(struct mem_cgroup *memcg) | |
970 | { | |
971 | struct cgroup *cgrp = memcg->css.cgroup; | |
972 | unsigned int swappiness; | |
973 | ||
974 | /* root ? */ | |
975 | if (cgrp->parent == NULL) | |
976 | return vm_swappiness; | |
977 | ||
978 | spin_lock(&memcg->reclaim_param_lock); | |
979 | swappiness = memcg->swappiness; | |
980 | spin_unlock(&memcg->reclaim_param_lock); | |
981 | ||
982 | return swappiness; | |
983 | } | |
984 | ||
985 | static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data) | |
986 | { | |
987 | int *val = data; | |
988 | (*val)++; | |
989 | return 0; | |
990 | } | |
991 | ||
992 | /** | |
993 | * mem_cgroup_print_mem_info: Called from OOM with tasklist_lock held in read mode. | |
994 | * @memcg: The memory cgroup that went over limit | |
995 | * @p: Task that is going to be killed | |
996 | * | |
997 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
998 | * enabled | |
999 | */ | |
1000 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1001 | { | |
1002 | struct cgroup *task_cgrp; | |
1003 | struct cgroup *mem_cgrp; | |
1004 | /* | |
1005 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1006 | * on the assumption that OOM is serialized for memory controller. | |
1007 | * If this assumption is broken, revisit this code. | |
1008 | */ | |
1009 | static char memcg_name[PATH_MAX]; | |
1010 | int ret; | |
1011 | ||
1012 | if (!memcg) | |
1013 | return; | |
1014 | ||
1015 | ||
1016 | rcu_read_lock(); | |
1017 | ||
1018 | mem_cgrp = memcg->css.cgroup; | |
1019 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1020 | ||
1021 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1022 | if (ret < 0) { | |
1023 | /* | |
1024 | * Unfortunately, we are unable to convert to a useful name | |
1025 | * But we'll still print out the usage information | |
1026 | */ | |
1027 | rcu_read_unlock(); | |
1028 | goto done; | |
1029 | } | |
1030 | rcu_read_unlock(); | |
1031 | ||
1032 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1033 | ||
1034 | rcu_read_lock(); | |
1035 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1036 | if (ret < 0) { | |
1037 | rcu_read_unlock(); | |
1038 | goto done; | |
1039 | } | |
1040 | rcu_read_unlock(); | |
1041 | ||
1042 | /* | |
1043 | * Continues from above, so we don't need an KERN_ level | |
1044 | */ | |
1045 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1046 | done: | |
1047 | ||
1048 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1049 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1050 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1051 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1052 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1053 | "failcnt %llu\n", | |
1054 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1055 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1056 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1057 | } | |
1058 | ||
1059 | /* | |
1060 | * This function returns the number of memcg under hierarchy tree. Returns | |
1061 | * 1(self count) if no children. | |
1062 | */ | |
1063 | static int mem_cgroup_count_children(struct mem_cgroup *mem) | |
1064 | { | |
1065 | int num = 0; | |
1066 | mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb); | |
1067 | return num; | |
1068 | } | |
1069 | ||
1070 | /* | |
1071 | * Visit the first child (need not be the first child as per the ordering | |
1072 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
1073 | * that to reclaim free pages from. | |
1074 | */ | |
1075 | static struct mem_cgroup * | |
1076 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) | |
1077 | { | |
1078 | struct mem_cgroup *ret = NULL; | |
1079 | struct cgroup_subsys_state *css; | |
1080 | int nextid, found; | |
1081 | ||
1082 | if (!root_mem->use_hierarchy) { | |
1083 | css_get(&root_mem->css); | |
1084 | ret = root_mem; | |
1085 | } | |
1086 | ||
1087 | while (!ret) { | |
1088 | rcu_read_lock(); | |
1089 | nextid = root_mem->last_scanned_child + 1; | |
1090 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, | |
1091 | &found); | |
1092 | if (css && css_tryget(css)) | |
1093 | ret = container_of(css, struct mem_cgroup, css); | |
1094 | ||
1095 | rcu_read_unlock(); | |
1096 | /* Updates scanning parameter */ | |
1097 | spin_lock(&root_mem->reclaim_param_lock); | |
1098 | if (!css) { | |
1099 | /* this means start scan from ID:1 */ | |
1100 | root_mem->last_scanned_child = 0; | |
1101 | } else | |
1102 | root_mem->last_scanned_child = found; | |
1103 | spin_unlock(&root_mem->reclaim_param_lock); | |
1104 | } | |
1105 | ||
1106 | return ret; | |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
1111 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
1112 | * based on its position in the children list. | |
1113 | * | |
1114 | * root_mem is the original ancestor that we've been reclaim from. | |
1115 | * | |
1116 | * We give up and return to the caller when we visit root_mem twice. | |
1117 | * (other groups can be removed while we're walking....) | |
1118 | * | |
1119 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
1120 | */ | |
1121 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
1122 | struct zone *zone, | |
1123 | gfp_t gfp_mask, | |
1124 | unsigned long reclaim_options) | |
1125 | { | |
1126 | struct mem_cgroup *victim; | |
1127 | int ret, total = 0; | |
1128 | int loop = 0; | |
1129 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; | |
1130 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; | |
1131 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; | |
1132 | unsigned long excess = mem_cgroup_get_excess(root_mem); | |
1133 | ||
1134 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ | |
1135 | if (root_mem->memsw_is_minimum) | |
1136 | noswap = true; | |
1137 | ||
1138 | while (1) { | |
1139 | victim = mem_cgroup_select_victim(root_mem); | |
1140 | if (victim == root_mem) { | |
1141 | loop++; | |
1142 | if (loop >= 1) | |
1143 | drain_all_stock_async(); | |
1144 | if (loop >= 2) { | |
1145 | /* | |
1146 | * If we have not been able to reclaim | |
1147 | * anything, it might because there are | |
1148 | * no reclaimable pages under this hierarchy | |
1149 | */ | |
1150 | if (!check_soft || !total) { | |
1151 | css_put(&victim->css); | |
1152 | break; | |
1153 | } | |
1154 | /* | |
1155 | * We want to do more targetted reclaim. | |
1156 | * excess >> 2 is not to excessive so as to | |
1157 | * reclaim too much, nor too less that we keep | |
1158 | * coming back to reclaim from this cgroup | |
1159 | */ | |
1160 | if (total >= (excess >> 2) || | |
1161 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { | |
1162 | css_put(&victim->css); | |
1163 | break; | |
1164 | } | |
1165 | } | |
1166 | } | |
1167 | if (!mem_cgroup_local_usage(&victim->stat)) { | |
1168 | /* this cgroup's local usage == 0 */ | |
1169 | css_put(&victim->css); | |
1170 | continue; | |
1171 | } | |
1172 | /* we use swappiness of local cgroup */ | |
1173 | if (check_soft) | |
1174 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, | |
1175 | noswap, get_swappiness(victim), zone, | |
1176 | zone->zone_pgdat->node_id); | |
1177 | else | |
1178 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, | |
1179 | noswap, get_swappiness(victim)); | |
1180 | css_put(&victim->css); | |
1181 | /* | |
1182 | * At shrinking usage, we can't check we should stop here or | |
1183 | * reclaim more. It's depends on callers. last_scanned_child | |
1184 | * will work enough for keeping fairness under tree. | |
1185 | */ | |
1186 | if (shrink) | |
1187 | return ret; | |
1188 | total += ret; | |
1189 | if (check_soft) { | |
1190 | if (res_counter_check_under_soft_limit(&root_mem->res)) | |
1191 | return total; | |
1192 | } else if (mem_cgroup_check_under_limit(root_mem)) | |
1193 | return 1 + total; | |
1194 | } | |
1195 | return total; | |
1196 | } | |
1197 | ||
1198 | bool mem_cgroup_oom_called(struct task_struct *task) | |
1199 | { | |
1200 | bool ret = false; | |
1201 | struct mem_cgroup *mem; | |
1202 | struct mm_struct *mm; | |
1203 | ||
1204 | rcu_read_lock(); | |
1205 | mm = task->mm; | |
1206 | if (!mm) | |
1207 | mm = &init_mm; | |
1208 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1209 | if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10)) | |
1210 | ret = true; | |
1211 | rcu_read_unlock(); | |
1212 | return ret; | |
1213 | } | |
1214 | ||
1215 | static int record_last_oom_cb(struct mem_cgroup *mem, void *data) | |
1216 | { | |
1217 | mem->last_oom_jiffies = jiffies; | |
1218 | return 0; | |
1219 | } | |
1220 | ||
1221 | static void record_last_oom(struct mem_cgroup *mem) | |
1222 | { | |
1223 | mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb); | |
1224 | } | |
1225 | ||
1226 | /* | |
1227 | * Currently used to update mapped file statistics, but the routine can be | |
1228 | * generalized to update other statistics as well. | |
1229 | */ | |
1230 | void mem_cgroup_update_file_mapped(struct page *page, int val) | |
1231 | { | |
1232 | struct mem_cgroup *mem; | |
1233 | struct mem_cgroup_stat *stat; | |
1234 | struct mem_cgroup_stat_cpu *cpustat; | |
1235 | int cpu; | |
1236 | struct page_cgroup *pc; | |
1237 | ||
1238 | pc = lookup_page_cgroup(page); | |
1239 | if (unlikely(!pc)) | |
1240 | return; | |
1241 | ||
1242 | lock_page_cgroup(pc); | |
1243 | mem = pc->mem_cgroup; | |
1244 | if (!mem) | |
1245 | goto done; | |
1246 | ||
1247 | if (!PageCgroupUsed(pc)) | |
1248 | goto done; | |
1249 | ||
1250 | /* | |
1251 | * Preemption is already disabled, we don't need get_cpu() | |
1252 | */ | |
1253 | cpu = smp_processor_id(); | |
1254 | stat = &mem->stat; | |
1255 | cpustat = &stat->cpustat[cpu]; | |
1256 | ||
1257 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED, val); | |
1258 | done: | |
1259 | unlock_page_cgroup(pc); | |
1260 | } | |
1261 | ||
1262 | /* | |
1263 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1264 | * TODO: maybe necessary to use big numbers in big irons. | |
1265 | */ | |
1266 | #define CHARGE_SIZE (32 * PAGE_SIZE) | |
1267 | struct memcg_stock_pcp { | |
1268 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
1269 | int charge; | |
1270 | struct work_struct work; | |
1271 | }; | |
1272 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
1273 | static atomic_t memcg_drain_count; | |
1274 | ||
1275 | /* | |
1276 | * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed | |
1277 | * from local stock and true is returned. If the stock is 0 or charges from a | |
1278 | * cgroup which is not current target, returns false. This stock will be | |
1279 | * refilled. | |
1280 | */ | |
1281 | static bool consume_stock(struct mem_cgroup *mem) | |
1282 | { | |
1283 | struct memcg_stock_pcp *stock; | |
1284 | bool ret = true; | |
1285 | ||
1286 | stock = &get_cpu_var(memcg_stock); | |
1287 | if (mem == stock->cached && stock->charge) | |
1288 | stock->charge -= PAGE_SIZE; | |
1289 | else /* need to call res_counter_charge */ | |
1290 | ret = false; | |
1291 | put_cpu_var(memcg_stock); | |
1292 | return ret; | |
1293 | } | |
1294 | ||
1295 | /* | |
1296 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
1297 | */ | |
1298 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1299 | { | |
1300 | struct mem_cgroup *old = stock->cached; | |
1301 | ||
1302 | if (stock->charge) { | |
1303 | res_counter_uncharge(&old->res, stock->charge); | |
1304 | if (do_swap_account) | |
1305 | res_counter_uncharge(&old->memsw, stock->charge); | |
1306 | } | |
1307 | stock->cached = NULL; | |
1308 | stock->charge = 0; | |
1309 | } | |
1310 | ||
1311 | /* | |
1312 | * This must be called under preempt disabled or must be called by | |
1313 | * a thread which is pinned to local cpu. | |
1314 | */ | |
1315 | static void drain_local_stock(struct work_struct *dummy) | |
1316 | { | |
1317 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
1318 | drain_stock(stock); | |
1319 | } | |
1320 | ||
1321 | /* | |
1322 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
1323 | * This will be consumed by consumt_stock() function, later. | |
1324 | */ | |
1325 | static void refill_stock(struct mem_cgroup *mem, int val) | |
1326 | { | |
1327 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1328 | ||
1329 | if (stock->cached != mem) { /* reset if necessary */ | |
1330 | drain_stock(stock); | |
1331 | stock->cached = mem; | |
1332 | } | |
1333 | stock->charge += val; | |
1334 | put_cpu_var(memcg_stock); | |
1335 | } | |
1336 | ||
1337 | /* | |
1338 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
1339 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
1340 | * expects some charges will be back to res_counter later but cannot wait for | |
1341 | * it. | |
1342 | */ | |
1343 | static void drain_all_stock_async(void) | |
1344 | { | |
1345 | int cpu; | |
1346 | /* This function is for scheduling "drain" in asynchronous way. | |
1347 | * The result of "drain" is not directly handled by callers. Then, | |
1348 | * if someone is calling drain, we don't have to call drain more. | |
1349 | * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if | |
1350 | * there is a race. We just do loose check here. | |
1351 | */ | |
1352 | if (atomic_read(&memcg_drain_count)) | |
1353 | return; | |
1354 | /* Notify other cpus that system-wide "drain" is running */ | |
1355 | atomic_inc(&memcg_drain_count); | |
1356 | get_online_cpus(); | |
1357 | for_each_online_cpu(cpu) { | |
1358 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
1359 | schedule_work_on(cpu, &stock->work); | |
1360 | } | |
1361 | put_online_cpus(); | |
1362 | atomic_dec(&memcg_drain_count); | |
1363 | /* We don't wait for flush_work */ | |
1364 | } | |
1365 | ||
1366 | /* This is a synchronous drain interface. */ | |
1367 | static void drain_all_stock_sync(void) | |
1368 | { | |
1369 | /* called when force_empty is called */ | |
1370 | atomic_inc(&memcg_drain_count); | |
1371 | schedule_on_each_cpu(drain_local_stock); | |
1372 | atomic_dec(&memcg_drain_count); | |
1373 | } | |
1374 | ||
1375 | static int __cpuinit memcg_stock_cpu_callback(struct notifier_block *nb, | |
1376 | unsigned long action, | |
1377 | void *hcpu) | |
1378 | { | |
1379 | int cpu = (unsigned long)hcpu; | |
1380 | struct memcg_stock_pcp *stock; | |
1381 | ||
1382 | if (action != CPU_DEAD) | |
1383 | return NOTIFY_OK; | |
1384 | stock = &per_cpu(memcg_stock, cpu); | |
1385 | drain_stock(stock); | |
1386 | return NOTIFY_OK; | |
1387 | } | |
1388 | ||
1389 | /* | |
1390 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
1391 | * oom-killer can be invoked. | |
1392 | */ | |
1393 | static int __mem_cgroup_try_charge(struct mm_struct *mm, | |
1394 | gfp_t gfp_mask, struct mem_cgroup **memcg, | |
1395 | bool oom, struct page *page) | |
1396 | { | |
1397 | struct mem_cgroup *mem, *mem_over_limit; | |
1398 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
1399 | struct res_counter *fail_res; | |
1400 | int csize = CHARGE_SIZE; | |
1401 | ||
1402 | if (unlikely(test_thread_flag(TIF_MEMDIE))) { | |
1403 | /* Don't account this! */ | |
1404 | *memcg = NULL; | |
1405 | return 0; | |
1406 | } | |
1407 | ||
1408 | /* | |
1409 | * We always charge the cgroup the mm_struct belongs to. | |
1410 | * The mm_struct's mem_cgroup changes on task migration if the | |
1411 | * thread group leader migrates. It's possible that mm is not | |
1412 | * set, if so charge the init_mm (happens for pagecache usage). | |
1413 | */ | |
1414 | mem = *memcg; | |
1415 | if (likely(!mem)) { | |
1416 | mem = try_get_mem_cgroup_from_mm(mm); | |
1417 | *memcg = mem; | |
1418 | } else { | |
1419 | css_get(&mem->css); | |
1420 | } | |
1421 | if (unlikely(!mem)) | |
1422 | return 0; | |
1423 | ||
1424 | VM_BUG_ON(css_is_removed(&mem->css)); | |
1425 | if (mem_cgroup_is_root(mem)) | |
1426 | goto done; | |
1427 | ||
1428 | while (1) { | |
1429 | int ret = 0; | |
1430 | unsigned long flags = 0; | |
1431 | ||
1432 | if (consume_stock(mem)) | |
1433 | goto charged; | |
1434 | ||
1435 | ret = res_counter_charge(&mem->res, csize, &fail_res); | |
1436 | if (likely(!ret)) { | |
1437 | if (!do_swap_account) | |
1438 | break; | |
1439 | ret = res_counter_charge(&mem->memsw, csize, &fail_res); | |
1440 | if (likely(!ret)) | |
1441 | break; | |
1442 | /* mem+swap counter fails */ | |
1443 | res_counter_uncharge(&mem->res, csize); | |
1444 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
1445 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, | |
1446 | memsw); | |
1447 | } else | |
1448 | /* mem counter fails */ | |
1449 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, | |
1450 | res); | |
1451 | ||
1452 | /* reduce request size and retry */ | |
1453 | if (csize > PAGE_SIZE) { | |
1454 | csize = PAGE_SIZE; | |
1455 | continue; | |
1456 | } | |
1457 | if (!(gfp_mask & __GFP_WAIT)) | |
1458 | goto nomem; | |
1459 | ||
1460 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, | |
1461 | gfp_mask, flags); | |
1462 | if (ret) | |
1463 | continue; | |
1464 | ||
1465 | /* | |
1466 | * try_to_free_mem_cgroup_pages() might not give us a full | |
1467 | * picture of reclaim. Some pages are reclaimed and might be | |
1468 | * moved to swap cache or just unmapped from the cgroup. | |
1469 | * Check the limit again to see if the reclaim reduced the | |
1470 | * current usage of the cgroup before giving up | |
1471 | * | |
1472 | */ | |
1473 | if (mem_cgroup_check_under_limit(mem_over_limit)) | |
1474 | continue; | |
1475 | ||
1476 | if (!nr_retries--) { | |
1477 | if (oom) { | |
1478 | mutex_lock(&memcg_tasklist); | |
1479 | mem_cgroup_out_of_memory(mem_over_limit, gfp_mask); | |
1480 | mutex_unlock(&memcg_tasklist); | |
1481 | record_last_oom(mem_over_limit); | |
1482 | } | |
1483 | goto nomem; | |
1484 | } | |
1485 | } | |
1486 | if (csize > PAGE_SIZE) | |
1487 | refill_stock(mem, csize - PAGE_SIZE); | |
1488 | charged: | |
1489 | /* | |
1490 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
1491 | * if they exceeds softlimit. | |
1492 | */ | |
1493 | if (mem_cgroup_soft_limit_check(mem)) | |
1494 | mem_cgroup_update_tree(mem, page); | |
1495 | done: | |
1496 | return 0; | |
1497 | nomem: | |
1498 | css_put(&mem->css); | |
1499 | return -ENOMEM; | |
1500 | } | |
1501 | ||
1502 | /* | |
1503 | * Somemtimes we have to undo a charge we got by try_charge(). | |
1504 | * This function is for that and do uncharge, put css's refcnt. | |
1505 | * gotten by try_charge(). | |
1506 | */ | |
1507 | static void mem_cgroup_cancel_charge(struct mem_cgroup *mem) | |
1508 | { | |
1509 | if (!mem_cgroup_is_root(mem)) { | |
1510 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
1511 | if (do_swap_account) | |
1512 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
1513 | } | |
1514 | css_put(&mem->css); | |
1515 | } | |
1516 | ||
1517 | /* | |
1518 | * A helper function to get mem_cgroup from ID. must be called under | |
1519 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
1520 | * it's concern. (dropping refcnt from swap can be called against removed | |
1521 | * memcg.) | |
1522 | */ | |
1523 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
1524 | { | |
1525 | struct cgroup_subsys_state *css; | |
1526 | ||
1527 | /* ID 0 is unused ID */ | |
1528 | if (!id) | |
1529 | return NULL; | |
1530 | css = css_lookup(&mem_cgroup_subsys, id); | |
1531 | if (!css) | |
1532 | return NULL; | |
1533 | return container_of(css, struct mem_cgroup, css); | |
1534 | } | |
1535 | ||
1536 | static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page) | |
1537 | { | |
1538 | struct mem_cgroup *mem; | |
1539 | struct page_cgroup *pc; | |
1540 | unsigned short id; | |
1541 | swp_entry_t ent; | |
1542 | ||
1543 | VM_BUG_ON(!PageLocked(page)); | |
1544 | ||
1545 | if (!PageSwapCache(page)) | |
1546 | return NULL; | |
1547 | ||
1548 | pc = lookup_page_cgroup(page); | |
1549 | lock_page_cgroup(pc); | |
1550 | if (PageCgroupUsed(pc)) { | |
1551 | mem = pc->mem_cgroup; | |
1552 | if (mem && !css_tryget(&mem->css)) | |
1553 | mem = NULL; | |
1554 | } else { | |
1555 | ent.val = page_private(page); | |
1556 | id = lookup_swap_cgroup(ent); | |
1557 | rcu_read_lock(); | |
1558 | mem = mem_cgroup_lookup(id); | |
1559 | if (mem && !css_tryget(&mem->css)) | |
1560 | mem = NULL; | |
1561 | rcu_read_unlock(); | |
1562 | } | |
1563 | unlock_page_cgroup(pc); | |
1564 | return mem; | |
1565 | } | |
1566 | ||
1567 | /* | |
1568 | * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be | |
1569 | * USED state. If already USED, uncharge and return. | |
1570 | */ | |
1571 | ||
1572 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
1573 | struct page_cgroup *pc, | |
1574 | enum charge_type ctype) | |
1575 | { | |
1576 | /* try_charge() can return NULL to *memcg, taking care of it. */ | |
1577 | if (!mem) | |
1578 | return; | |
1579 | ||
1580 | lock_page_cgroup(pc); | |
1581 | if (unlikely(PageCgroupUsed(pc))) { | |
1582 | unlock_page_cgroup(pc); | |
1583 | mem_cgroup_cancel_charge(mem); | |
1584 | return; | |
1585 | } | |
1586 | ||
1587 | pc->mem_cgroup = mem; | |
1588 | /* | |
1589 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
1590 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
1591 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
1592 | * before USED bit, we need memory barrier here. | |
1593 | * See mem_cgroup_add_lru_list(), etc. | |
1594 | */ | |
1595 | smp_wmb(); | |
1596 | switch (ctype) { | |
1597 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
1598 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
1599 | SetPageCgroupCache(pc); | |
1600 | SetPageCgroupUsed(pc); | |
1601 | break; | |
1602 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
1603 | ClearPageCgroupCache(pc); | |
1604 | SetPageCgroupUsed(pc); | |
1605 | break; | |
1606 | default: | |
1607 | break; | |
1608 | } | |
1609 | ||
1610 | mem_cgroup_charge_statistics(mem, pc, true); | |
1611 | ||
1612 | unlock_page_cgroup(pc); | |
1613 | } | |
1614 | ||
1615 | /** | |
1616 | * mem_cgroup_move_account - move account of the page | |
1617 | * @pc: page_cgroup of the page. | |
1618 | * @from: mem_cgroup which the page is moved from. | |
1619 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
1620 | * | |
1621 | * The caller must confirm following. | |
1622 | * - page is not on LRU (isolate_page() is useful.) | |
1623 | * | |
1624 | * returns 0 at success, | |
1625 | * returns -EBUSY when lock is busy or "pc" is unstable. | |
1626 | * | |
1627 | * This function does "uncharge" from old cgroup but doesn't do "charge" to | |
1628 | * new cgroup. It should be done by a caller. | |
1629 | */ | |
1630 | ||
1631 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
1632 | struct mem_cgroup *from, struct mem_cgroup *to) | |
1633 | { | |
1634 | struct mem_cgroup_per_zone *from_mz, *to_mz; | |
1635 | int nid, zid; | |
1636 | int ret = -EBUSY; | |
1637 | struct page *page; | |
1638 | int cpu; | |
1639 | struct mem_cgroup_stat *stat; | |
1640 | struct mem_cgroup_stat_cpu *cpustat; | |
1641 | ||
1642 | VM_BUG_ON(from == to); | |
1643 | VM_BUG_ON(PageLRU(pc->page)); | |
1644 | ||
1645 | nid = page_cgroup_nid(pc); | |
1646 | zid = page_cgroup_zid(pc); | |
1647 | from_mz = mem_cgroup_zoneinfo(from, nid, zid); | |
1648 | to_mz = mem_cgroup_zoneinfo(to, nid, zid); | |
1649 | ||
1650 | if (!trylock_page_cgroup(pc)) | |
1651 | return ret; | |
1652 | ||
1653 | if (!PageCgroupUsed(pc)) | |
1654 | goto out; | |
1655 | ||
1656 | if (pc->mem_cgroup != from) | |
1657 | goto out; | |
1658 | ||
1659 | if (!mem_cgroup_is_root(from)) | |
1660 | res_counter_uncharge(&from->res, PAGE_SIZE); | |
1661 | mem_cgroup_charge_statistics(from, pc, false); | |
1662 | ||
1663 | page = pc->page; | |
1664 | if (page_mapped(page) && !PageAnon(page)) { | |
1665 | cpu = smp_processor_id(); | |
1666 | /* Update mapped_file data for mem_cgroup "from" */ | |
1667 | stat = &from->stat; | |
1668 | cpustat = &stat->cpustat[cpu]; | |
1669 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED, | |
1670 | -1); | |
1671 | ||
1672 | /* Update mapped_file data for mem_cgroup "to" */ | |
1673 | stat = &to->stat; | |
1674 | cpustat = &stat->cpustat[cpu]; | |
1675 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED, | |
1676 | 1); | |
1677 | } | |
1678 | ||
1679 | if (do_swap_account && !mem_cgroup_is_root(from)) | |
1680 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
1681 | css_put(&from->css); | |
1682 | ||
1683 | css_get(&to->css); | |
1684 | pc->mem_cgroup = to; | |
1685 | mem_cgroup_charge_statistics(to, pc, true); | |
1686 | ret = 0; | |
1687 | out: | |
1688 | unlock_page_cgroup(pc); | |
1689 | /* | |
1690 | * We charges against "to" which may not have any tasks. Then, "to" | |
1691 | * can be under rmdir(). But in current implementation, caller of | |
1692 | * this function is just force_empty() and it's garanteed that | |
1693 | * "to" is never removed. So, we don't check rmdir status here. | |
1694 | */ | |
1695 | return ret; | |
1696 | } | |
1697 | ||
1698 | /* | |
1699 | * move charges to its parent. | |
1700 | */ | |
1701 | ||
1702 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
1703 | struct mem_cgroup *child, | |
1704 | gfp_t gfp_mask) | |
1705 | { | |
1706 | struct page *page = pc->page; | |
1707 | struct cgroup *cg = child->css.cgroup; | |
1708 | struct cgroup *pcg = cg->parent; | |
1709 | struct mem_cgroup *parent; | |
1710 | int ret; | |
1711 | ||
1712 | /* Is ROOT ? */ | |
1713 | if (!pcg) | |
1714 | return -EINVAL; | |
1715 | ||
1716 | ||
1717 | parent = mem_cgroup_from_cont(pcg); | |
1718 | ||
1719 | ||
1720 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false, page); | |
1721 | if (ret || !parent) | |
1722 | return ret; | |
1723 | ||
1724 | if (!get_page_unless_zero(page)) { | |
1725 | ret = -EBUSY; | |
1726 | goto uncharge; | |
1727 | } | |
1728 | ||
1729 | ret = isolate_lru_page(page); | |
1730 | ||
1731 | if (ret) | |
1732 | goto cancel; | |
1733 | ||
1734 | ret = mem_cgroup_move_account(pc, child, parent); | |
1735 | ||
1736 | putback_lru_page(page); | |
1737 | if (!ret) { | |
1738 | put_page(page); | |
1739 | /* drop extra refcnt by try_charge() */ | |
1740 | css_put(&parent->css); | |
1741 | return 0; | |
1742 | } | |
1743 | ||
1744 | cancel: | |
1745 | put_page(page); | |
1746 | uncharge: | |
1747 | mem_cgroup_cancel_charge(parent); | |
1748 | return ret; | |
1749 | } | |
1750 | ||
1751 | /* | |
1752 | * Charge the memory controller for page usage. | |
1753 | * Return | |
1754 | * 0 if the charge was successful | |
1755 | * < 0 if the cgroup is over its limit | |
1756 | */ | |
1757 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
1758 | gfp_t gfp_mask, enum charge_type ctype, | |
1759 | struct mem_cgroup *memcg) | |
1760 | { | |
1761 | struct mem_cgroup *mem; | |
1762 | struct page_cgroup *pc; | |
1763 | int ret; | |
1764 | ||
1765 | pc = lookup_page_cgroup(page); | |
1766 | /* can happen at boot */ | |
1767 | if (unlikely(!pc)) | |
1768 | return 0; | |
1769 | prefetchw(pc); | |
1770 | ||
1771 | mem = memcg; | |
1772 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true, page); | |
1773 | if (ret || !mem) | |
1774 | return ret; | |
1775 | ||
1776 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
1777 | return 0; | |
1778 | } | |
1779 | ||
1780 | int mem_cgroup_newpage_charge(struct page *page, | |
1781 | struct mm_struct *mm, gfp_t gfp_mask) | |
1782 | { | |
1783 | if (mem_cgroup_disabled()) | |
1784 | return 0; | |
1785 | if (PageCompound(page)) | |
1786 | return 0; | |
1787 | /* | |
1788 | * If already mapped, we don't have to account. | |
1789 | * If page cache, page->mapping has address_space. | |
1790 | * But page->mapping may have out-of-use anon_vma pointer, | |
1791 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
1792 | * is NULL. | |
1793 | */ | |
1794 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
1795 | return 0; | |
1796 | if (unlikely(!mm)) | |
1797 | mm = &init_mm; | |
1798 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
1799 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); | |
1800 | } | |
1801 | ||
1802 | static void | |
1803 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
1804 | enum charge_type ctype); | |
1805 | ||
1806 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, | |
1807 | gfp_t gfp_mask) | |
1808 | { | |
1809 | struct mem_cgroup *mem = NULL; | |
1810 | int ret; | |
1811 | ||
1812 | if (mem_cgroup_disabled()) | |
1813 | return 0; | |
1814 | if (PageCompound(page)) | |
1815 | return 0; | |
1816 | /* | |
1817 | * Corner case handling. This is called from add_to_page_cache() | |
1818 | * in usual. But some FS (shmem) precharges this page before calling it | |
1819 | * and call add_to_page_cache() with GFP_NOWAIT. | |
1820 | * | |
1821 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
1822 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
1823 | * charge twice. (It works but has to pay a bit larger cost.) | |
1824 | * And when the page is SwapCache, it should take swap information | |
1825 | * into account. This is under lock_page() now. | |
1826 | */ | |
1827 | if (!(gfp_mask & __GFP_WAIT)) { | |
1828 | struct page_cgroup *pc; | |
1829 | ||
1830 | ||
1831 | pc = lookup_page_cgroup(page); | |
1832 | if (!pc) | |
1833 | return 0; | |
1834 | lock_page_cgroup(pc); | |
1835 | if (PageCgroupUsed(pc)) { | |
1836 | unlock_page_cgroup(pc); | |
1837 | return 0; | |
1838 | } | |
1839 | unlock_page_cgroup(pc); | |
1840 | } | |
1841 | ||
1842 | if (unlikely(!mm && !mem)) | |
1843 | mm = &init_mm; | |
1844 | ||
1845 | if (page_is_file_cache(page)) | |
1846 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
1847 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); | |
1848 | ||
1849 | /* shmem */ | |
1850 | if (PageSwapCache(page)) { | |
1851 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); | |
1852 | if (!ret) | |
1853 | __mem_cgroup_commit_charge_swapin(page, mem, | |
1854 | MEM_CGROUP_CHARGE_TYPE_SHMEM); | |
1855 | } else | |
1856 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
1857 | MEM_CGROUP_CHARGE_TYPE_SHMEM, mem); | |
1858 | ||
1859 | return ret; | |
1860 | } | |
1861 | ||
1862 | /* | |
1863 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
1864 | * And when try_charge() successfully returns, one refcnt to memcg without | |
1865 | * struct page_cgroup is acquired. This refcnt will be consumed by | |
1866 | * "commit()" or removed by "cancel()" | |
1867 | */ | |
1868 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, | |
1869 | struct page *page, | |
1870 | gfp_t mask, struct mem_cgroup **ptr) | |
1871 | { | |
1872 | struct mem_cgroup *mem; | |
1873 | int ret; | |
1874 | ||
1875 | if (mem_cgroup_disabled()) | |
1876 | return 0; | |
1877 | ||
1878 | if (!do_swap_account) | |
1879 | goto charge_cur_mm; | |
1880 | /* | |
1881 | * A racing thread's fault, or swapoff, may have already updated | |
1882 | * the pte, and even removed page from swap cache: in those cases | |
1883 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
1884 | * KSM case which does need to charge the page. | |
1885 | */ | |
1886 | if (!PageSwapCache(page)) | |
1887 | goto charge_cur_mm; | |
1888 | mem = try_get_mem_cgroup_from_swapcache(page); | |
1889 | if (!mem) | |
1890 | goto charge_cur_mm; | |
1891 | *ptr = mem; | |
1892 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, page); | |
1893 | /* drop extra refcnt from tryget */ | |
1894 | css_put(&mem->css); | |
1895 | return ret; | |
1896 | charge_cur_mm: | |
1897 | if (unlikely(!mm)) | |
1898 | mm = &init_mm; | |
1899 | return __mem_cgroup_try_charge(mm, mask, ptr, true, page); | |
1900 | } | |
1901 | ||
1902 | static void | |
1903 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
1904 | enum charge_type ctype) | |
1905 | { | |
1906 | struct page_cgroup *pc; | |
1907 | ||
1908 | if (mem_cgroup_disabled()) | |
1909 | return; | |
1910 | if (!ptr) | |
1911 | return; | |
1912 | cgroup_exclude_rmdir(&ptr->css); | |
1913 | pc = lookup_page_cgroup(page); | |
1914 | mem_cgroup_lru_del_before_commit_swapcache(page); | |
1915 | __mem_cgroup_commit_charge(ptr, pc, ctype); | |
1916 | mem_cgroup_lru_add_after_commit_swapcache(page); | |
1917 | /* | |
1918 | * Now swap is on-memory. This means this page may be | |
1919 | * counted both as mem and swap....double count. | |
1920 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable | |
1921 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
1922 | * may call delete_from_swap_cache() before reach here. | |
1923 | */ | |
1924 | if (do_swap_account && PageSwapCache(page)) { | |
1925 | swp_entry_t ent = {.val = page_private(page)}; | |
1926 | unsigned short id; | |
1927 | struct mem_cgroup *memcg; | |
1928 | ||
1929 | id = swap_cgroup_record(ent, 0); | |
1930 | rcu_read_lock(); | |
1931 | memcg = mem_cgroup_lookup(id); | |
1932 | if (memcg) { | |
1933 | /* | |
1934 | * This recorded memcg can be obsolete one. So, avoid | |
1935 | * calling css_tryget | |
1936 | */ | |
1937 | if (!mem_cgroup_is_root(memcg)) | |
1938 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); | |
1939 | mem_cgroup_swap_statistics(memcg, false); | |
1940 | mem_cgroup_put(memcg); | |
1941 | } | |
1942 | rcu_read_unlock(); | |
1943 | } | |
1944 | /* | |
1945 | * At swapin, we may charge account against cgroup which has no tasks. | |
1946 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
1947 | * In that case, we need to call pre_destroy() again. check it here. | |
1948 | */ | |
1949 | cgroup_release_and_wakeup_rmdir(&ptr->css); | |
1950 | } | |
1951 | ||
1952 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) | |
1953 | { | |
1954 | __mem_cgroup_commit_charge_swapin(page, ptr, | |
1955 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
1956 | } | |
1957 | ||
1958 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) | |
1959 | { | |
1960 | if (mem_cgroup_disabled()) | |
1961 | return; | |
1962 | if (!mem) | |
1963 | return; | |
1964 | mem_cgroup_cancel_charge(mem); | |
1965 | } | |
1966 | ||
1967 | static void | |
1968 | __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype) | |
1969 | { | |
1970 | struct memcg_batch_info *batch = NULL; | |
1971 | bool uncharge_memsw = true; | |
1972 | /* If swapout, usage of swap doesn't decrease */ | |
1973 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
1974 | uncharge_memsw = false; | |
1975 | /* | |
1976 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
1977 | * In those cases, all pages freed continously can be expected to be in | |
1978 | * the same cgroup and we have chance to coalesce uncharges. | |
1979 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
1980 | * because we want to do uncharge as soon as possible. | |
1981 | */ | |
1982 | if (!current->memcg_batch.do_batch || test_thread_flag(TIF_MEMDIE)) | |
1983 | goto direct_uncharge; | |
1984 | ||
1985 | batch = ¤t->memcg_batch; | |
1986 | /* | |
1987 | * In usual, we do css_get() when we remember memcg pointer. | |
1988 | * But in this case, we keep res->usage until end of a series of | |
1989 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
1990 | */ | |
1991 | if (!batch->memcg) | |
1992 | batch->memcg = mem; | |
1993 | /* | |
1994 | * In typical case, batch->memcg == mem. This means we can | |
1995 | * merge a series of uncharges to an uncharge of res_counter. | |
1996 | * If not, we uncharge res_counter ony by one. | |
1997 | */ | |
1998 | if (batch->memcg != mem) | |
1999 | goto direct_uncharge; | |
2000 | /* remember freed charge and uncharge it later */ | |
2001 | batch->bytes += PAGE_SIZE; | |
2002 | if (uncharge_memsw) | |
2003 | batch->memsw_bytes += PAGE_SIZE; | |
2004 | return; | |
2005 | direct_uncharge: | |
2006 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
2007 | if (uncharge_memsw) | |
2008 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
2009 | return; | |
2010 | } | |
2011 | ||
2012 | /* | |
2013 | * uncharge if !page_mapped(page) | |
2014 | */ | |
2015 | static struct mem_cgroup * | |
2016 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) | |
2017 | { | |
2018 | struct page_cgroup *pc; | |
2019 | struct mem_cgroup *mem = NULL; | |
2020 | struct mem_cgroup_per_zone *mz; | |
2021 | ||
2022 | if (mem_cgroup_disabled()) | |
2023 | return NULL; | |
2024 | ||
2025 | if (PageSwapCache(page)) | |
2026 | return NULL; | |
2027 | ||
2028 | /* | |
2029 | * Check if our page_cgroup is valid | |
2030 | */ | |
2031 | pc = lookup_page_cgroup(page); | |
2032 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
2033 | return NULL; | |
2034 | ||
2035 | lock_page_cgroup(pc); | |
2036 | ||
2037 | mem = pc->mem_cgroup; | |
2038 | ||
2039 | if (!PageCgroupUsed(pc)) | |
2040 | goto unlock_out; | |
2041 | ||
2042 | switch (ctype) { | |
2043 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
2044 | case MEM_CGROUP_CHARGE_TYPE_DROP: | |
2045 | if (page_mapped(page)) | |
2046 | goto unlock_out; | |
2047 | break; | |
2048 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
2049 | if (!PageAnon(page)) { /* Shared memory */ | |
2050 | if (page->mapping && !page_is_file_cache(page)) | |
2051 | goto unlock_out; | |
2052 | } else if (page_mapped(page)) /* Anon */ | |
2053 | goto unlock_out; | |
2054 | break; | |
2055 | default: | |
2056 | break; | |
2057 | } | |
2058 | ||
2059 | if (!mem_cgroup_is_root(mem)) | |
2060 | __do_uncharge(mem, ctype); | |
2061 | if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2062 | mem_cgroup_swap_statistics(mem, true); | |
2063 | mem_cgroup_charge_statistics(mem, pc, false); | |
2064 | ||
2065 | ClearPageCgroupUsed(pc); | |
2066 | /* | |
2067 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
2068 | * freed from LRU. This is safe because uncharged page is expected not | |
2069 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
2070 | * special functions. | |
2071 | */ | |
2072 | ||
2073 | mz = page_cgroup_zoneinfo(pc); | |
2074 | unlock_page_cgroup(pc); | |
2075 | ||
2076 | if (mem_cgroup_soft_limit_check(mem)) | |
2077 | mem_cgroup_update_tree(mem, page); | |
2078 | /* at swapout, this memcg will be accessed to record to swap */ | |
2079 | if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2080 | css_put(&mem->css); | |
2081 | ||
2082 | return mem; | |
2083 | ||
2084 | unlock_out: | |
2085 | unlock_page_cgroup(pc); | |
2086 | return NULL; | |
2087 | } | |
2088 | ||
2089 | void mem_cgroup_uncharge_page(struct page *page) | |
2090 | { | |
2091 | /* early check. */ | |
2092 | if (page_mapped(page)) | |
2093 | return; | |
2094 | if (page->mapping && !PageAnon(page)) | |
2095 | return; | |
2096 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
2097 | } | |
2098 | ||
2099 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
2100 | { | |
2101 | VM_BUG_ON(page_mapped(page)); | |
2102 | VM_BUG_ON(page->mapping); | |
2103 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); | |
2104 | } | |
2105 | ||
2106 | /* | |
2107 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
2108 | * In that cases, pages are freed continuously and we can expect pages | |
2109 | * are in the same memcg. All these calls itself limits the number of | |
2110 | * pages freed at once, then uncharge_start/end() is called properly. | |
2111 | * This may be called prural(2) times in a context, | |
2112 | */ | |
2113 | ||
2114 | void mem_cgroup_uncharge_start(void) | |
2115 | { | |
2116 | current->memcg_batch.do_batch++; | |
2117 | /* We can do nest. */ | |
2118 | if (current->memcg_batch.do_batch == 1) { | |
2119 | current->memcg_batch.memcg = NULL; | |
2120 | current->memcg_batch.bytes = 0; | |
2121 | current->memcg_batch.memsw_bytes = 0; | |
2122 | } | |
2123 | } | |
2124 | ||
2125 | void mem_cgroup_uncharge_end(void) | |
2126 | { | |
2127 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
2128 | ||
2129 | if (!batch->do_batch) | |
2130 | return; | |
2131 | ||
2132 | batch->do_batch--; | |
2133 | if (batch->do_batch) /* If stacked, do nothing. */ | |
2134 | return; | |
2135 | ||
2136 | if (!batch->memcg) | |
2137 | return; | |
2138 | /* | |
2139 | * This "batch->memcg" is valid without any css_get/put etc... | |
2140 | * bacause we hide charges behind us. | |
2141 | */ | |
2142 | if (batch->bytes) | |
2143 | res_counter_uncharge(&batch->memcg->res, batch->bytes); | |
2144 | if (batch->memsw_bytes) | |
2145 | res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes); | |
2146 | /* forget this pointer (for sanity check) */ | |
2147 | batch->memcg = NULL; | |
2148 | } | |
2149 | ||
2150 | #ifdef CONFIG_SWAP | |
2151 | /* | |
2152 | * called after __delete_from_swap_cache() and drop "page" account. | |
2153 | * memcg information is recorded to swap_cgroup of "ent" | |
2154 | */ | |
2155 | void | |
2156 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
2157 | { | |
2158 | struct mem_cgroup *memcg; | |
2159 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; | |
2160 | ||
2161 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
2162 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
2163 | ||
2164 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
2165 | ||
2166 | /* record memcg information */ | |
2167 | if (do_swap_account && swapout && memcg) { | |
2168 | swap_cgroup_record(ent, css_id(&memcg->css)); | |
2169 | mem_cgroup_get(memcg); | |
2170 | } | |
2171 | if (swapout && memcg) | |
2172 | css_put(&memcg->css); | |
2173 | } | |
2174 | #endif | |
2175 | ||
2176 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
2177 | /* | |
2178 | * called from swap_entry_free(). remove record in swap_cgroup and | |
2179 | * uncharge "memsw" account. | |
2180 | */ | |
2181 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
2182 | { | |
2183 | struct mem_cgroup *memcg; | |
2184 | unsigned short id; | |
2185 | ||
2186 | if (!do_swap_account) | |
2187 | return; | |
2188 | ||
2189 | id = swap_cgroup_record(ent, 0); | |
2190 | rcu_read_lock(); | |
2191 | memcg = mem_cgroup_lookup(id); | |
2192 | if (memcg) { | |
2193 | /* | |
2194 | * We uncharge this because swap is freed. | |
2195 | * This memcg can be obsolete one. We avoid calling css_tryget | |
2196 | */ | |
2197 | if (!mem_cgroup_is_root(memcg)) | |
2198 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); | |
2199 | mem_cgroup_swap_statistics(memcg, false); | |
2200 | mem_cgroup_put(memcg); | |
2201 | } | |
2202 | rcu_read_unlock(); | |
2203 | } | |
2204 | #endif | |
2205 | ||
2206 | /* | |
2207 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old | |
2208 | * page belongs to. | |
2209 | */ | |
2210 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) | |
2211 | { | |
2212 | struct page_cgroup *pc; | |
2213 | struct mem_cgroup *mem = NULL; | |
2214 | int ret = 0; | |
2215 | ||
2216 | if (mem_cgroup_disabled()) | |
2217 | return 0; | |
2218 | ||
2219 | pc = lookup_page_cgroup(page); | |
2220 | lock_page_cgroup(pc); | |
2221 | if (PageCgroupUsed(pc)) { | |
2222 | mem = pc->mem_cgroup; | |
2223 | css_get(&mem->css); | |
2224 | } | |
2225 | unlock_page_cgroup(pc); | |
2226 | ||
2227 | if (mem) { | |
2228 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false, | |
2229 | page); | |
2230 | css_put(&mem->css); | |
2231 | } | |
2232 | *ptr = mem; | |
2233 | return ret; | |
2234 | } | |
2235 | ||
2236 | /* remove redundant charge if migration failed*/ | |
2237 | void mem_cgroup_end_migration(struct mem_cgroup *mem, | |
2238 | struct page *oldpage, struct page *newpage) | |
2239 | { | |
2240 | struct page *target, *unused; | |
2241 | struct page_cgroup *pc; | |
2242 | enum charge_type ctype; | |
2243 | ||
2244 | if (!mem) | |
2245 | return; | |
2246 | cgroup_exclude_rmdir(&mem->css); | |
2247 | /* at migration success, oldpage->mapping is NULL. */ | |
2248 | if (oldpage->mapping) { | |
2249 | target = oldpage; | |
2250 | unused = NULL; | |
2251 | } else { | |
2252 | target = newpage; | |
2253 | unused = oldpage; | |
2254 | } | |
2255 | ||
2256 | if (PageAnon(target)) | |
2257 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
2258 | else if (page_is_file_cache(target)) | |
2259 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
2260 | else | |
2261 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
2262 | ||
2263 | /* unused page is not on radix-tree now. */ | |
2264 | if (unused) | |
2265 | __mem_cgroup_uncharge_common(unused, ctype); | |
2266 | ||
2267 | pc = lookup_page_cgroup(target); | |
2268 | /* | |
2269 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. | |
2270 | * So, double-counting is effectively avoided. | |
2271 | */ | |
2272 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
2273 | ||
2274 | /* | |
2275 | * Both of oldpage and newpage are still under lock_page(). | |
2276 | * Then, we don't have to care about race in radix-tree. | |
2277 | * But we have to be careful that this page is unmapped or not. | |
2278 | * | |
2279 | * There is a case for !page_mapped(). At the start of | |
2280 | * migration, oldpage was mapped. But now, it's zapped. | |
2281 | * But we know *target* page is not freed/reused under us. | |
2282 | * mem_cgroup_uncharge_page() does all necessary checks. | |
2283 | */ | |
2284 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) | |
2285 | mem_cgroup_uncharge_page(target); | |
2286 | /* | |
2287 | * At migration, we may charge account against cgroup which has no tasks | |
2288 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2289 | * In that case, we need to call pre_destroy() again. check it here. | |
2290 | */ | |
2291 | cgroup_release_and_wakeup_rmdir(&mem->css); | |
2292 | } | |
2293 | ||
2294 | /* | |
2295 | * A call to try to shrink memory usage on charge failure at shmem's swapin. | |
2296 | * Calling hierarchical_reclaim is not enough because we should update | |
2297 | * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM. | |
2298 | * Moreover considering hierarchy, we should reclaim from the mem_over_limit, | |
2299 | * not from the memcg which this page would be charged to. | |
2300 | * try_charge_swapin does all of these works properly. | |
2301 | */ | |
2302 | int mem_cgroup_shmem_charge_fallback(struct page *page, | |
2303 | struct mm_struct *mm, | |
2304 | gfp_t gfp_mask) | |
2305 | { | |
2306 | struct mem_cgroup *mem = NULL; | |
2307 | int ret; | |
2308 | ||
2309 | if (mem_cgroup_disabled()) | |
2310 | return 0; | |
2311 | ||
2312 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); | |
2313 | if (!ret) | |
2314 | mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */ | |
2315 | ||
2316 | return ret; | |
2317 | } | |
2318 | ||
2319 | static DEFINE_MUTEX(set_limit_mutex); | |
2320 | ||
2321 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, | |
2322 | unsigned long long val) | |
2323 | { | |
2324 | int retry_count; | |
2325 | int progress; | |
2326 | u64 memswlimit; | |
2327 | int ret = 0; | |
2328 | int children = mem_cgroup_count_children(memcg); | |
2329 | u64 curusage, oldusage; | |
2330 | ||
2331 | /* | |
2332 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2333 | * is depends on callers. We set our retry-count to be function | |
2334 | * of # of children which we should visit in this loop. | |
2335 | */ | |
2336 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
2337 | ||
2338 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
2339 | ||
2340 | while (retry_count) { | |
2341 | if (signal_pending(current)) { | |
2342 | ret = -EINTR; | |
2343 | break; | |
2344 | } | |
2345 | /* | |
2346 | * Rather than hide all in some function, I do this in | |
2347 | * open coded manner. You see what this really does. | |
2348 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
2349 | */ | |
2350 | mutex_lock(&set_limit_mutex); | |
2351 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2352 | if (memswlimit < val) { | |
2353 | ret = -EINVAL; | |
2354 | mutex_unlock(&set_limit_mutex); | |
2355 | break; | |
2356 | } | |
2357 | ret = res_counter_set_limit(&memcg->res, val); | |
2358 | if (!ret) { | |
2359 | if (memswlimit == val) | |
2360 | memcg->memsw_is_minimum = true; | |
2361 | else | |
2362 | memcg->memsw_is_minimum = false; | |
2363 | } | |
2364 | mutex_unlock(&set_limit_mutex); | |
2365 | ||
2366 | if (!ret) | |
2367 | break; | |
2368 | ||
2369 | progress = mem_cgroup_hierarchical_reclaim(memcg, NULL, | |
2370 | GFP_KERNEL, | |
2371 | MEM_CGROUP_RECLAIM_SHRINK); | |
2372 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
2373 | /* Usage is reduced ? */ | |
2374 | if (curusage >= oldusage) | |
2375 | retry_count--; | |
2376 | else | |
2377 | oldusage = curusage; | |
2378 | } | |
2379 | ||
2380 | return ret; | |
2381 | } | |
2382 | ||
2383 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, | |
2384 | unsigned long long val) | |
2385 | { | |
2386 | int retry_count; | |
2387 | u64 memlimit, oldusage, curusage; | |
2388 | int children = mem_cgroup_count_children(memcg); | |
2389 | int ret = -EBUSY; | |
2390 | ||
2391 | /* see mem_cgroup_resize_res_limit */ | |
2392 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
2393 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
2394 | while (retry_count) { | |
2395 | if (signal_pending(current)) { | |
2396 | ret = -EINTR; | |
2397 | break; | |
2398 | } | |
2399 | /* | |
2400 | * Rather than hide all in some function, I do this in | |
2401 | * open coded manner. You see what this really does. | |
2402 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
2403 | */ | |
2404 | mutex_lock(&set_limit_mutex); | |
2405 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2406 | if (memlimit > val) { | |
2407 | ret = -EINVAL; | |
2408 | mutex_unlock(&set_limit_mutex); | |
2409 | break; | |
2410 | } | |
2411 | ret = res_counter_set_limit(&memcg->memsw, val); | |
2412 | if (!ret) { | |
2413 | if (memlimit == val) | |
2414 | memcg->memsw_is_minimum = true; | |
2415 | else | |
2416 | memcg->memsw_is_minimum = false; | |
2417 | } | |
2418 | mutex_unlock(&set_limit_mutex); | |
2419 | ||
2420 | if (!ret) | |
2421 | break; | |
2422 | ||
2423 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, | |
2424 | MEM_CGROUP_RECLAIM_NOSWAP | | |
2425 | MEM_CGROUP_RECLAIM_SHRINK); | |
2426 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
2427 | /* Usage is reduced ? */ | |
2428 | if (curusage >= oldusage) | |
2429 | retry_count--; | |
2430 | else | |
2431 | oldusage = curusage; | |
2432 | } | |
2433 | return ret; | |
2434 | } | |
2435 | ||
2436 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, | |
2437 | gfp_t gfp_mask, int nid, | |
2438 | int zid) | |
2439 | { | |
2440 | unsigned long nr_reclaimed = 0; | |
2441 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2442 | unsigned long reclaimed; | |
2443 | int loop = 0; | |
2444 | struct mem_cgroup_tree_per_zone *mctz; | |
2445 | unsigned long long excess; | |
2446 | ||
2447 | if (order > 0) | |
2448 | return 0; | |
2449 | ||
2450 | mctz = soft_limit_tree_node_zone(nid, zid); | |
2451 | /* | |
2452 | * This loop can run a while, specially if mem_cgroup's continuously | |
2453 | * keep exceeding their soft limit and putting the system under | |
2454 | * pressure | |
2455 | */ | |
2456 | do { | |
2457 | if (next_mz) | |
2458 | mz = next_mz; | |
2459 | else | |
2460 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2461 | if (!mz) | |
2462 | break; | |
2463 | ||
2464 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, | |
2465 | gfp_mask, | |
2466 | MEM_CGROUP_RECLAIM_SOFT); | |
2467 | nr_reclaimed += reclaimed; | |
2468 | spin_lock(&mctz->lock); | |
2469 | ||
2470 | /* | |
2471 | * If we failed to reclaim anything from this memory cgroup | |
2472 | * it is time to move on to the next cgroup | |
2473 | */ | |
2474 | next_mz = NULL; | |
2475 | if (!reclaimed) { | |
2476 | do { | |
2477 | /* | |
2478 | * Loop until we find yet another one. | |
2479 | * | |
2480 | * By the time we get the soft_limit lock | |
2481 | * again, someone might have aded the | |
2482 | * group back on the RB tree. Iterate to | |
2483 | * make sure we get a different mem. | |
2484 | * mem_cgroup_largest_soft_limit_node returns | |
2485 | * NULL if no other cgroup is present on | |
2486 | * the tree | |
2487 | */ | |
2488 | next_mz = | |
2489 | __mem_cgroup_largest_soft_limit_node(mctz); | |
2490 | if (next_mz == mz) { | |
2491 | css_put(&next_mz->mem->css); | |
2492 | next_mz = NULL; | |
2493 | } else /* next_mz == NULL or other memcg */ | |
2494 | break; | |
2495 | } while (1); | |
2496 | } | |
2497 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
2498 | excess = res_counter_soft_limit_excess(&mz->mem->res); | |
2499 | /* | |
2500 | * One school of thought says that we should not add | |
2501 | * back the node to the tree if reclaim returns 0. | |
2502 | * But our reclaim could return 0, simply because due | |
2503 | * to priority we are exposing a smaller subset of | |
2504 | * memory to reclaim from. Consider this as a longer | |
2505 | * term TODO. | |
2506 | */ | |
2507 | /* If excess == 0, no tree ops */ | |
2508 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); | |
2509 | spin_unlock(&mctz->lock); | |
2510 | css_put(&mz->mem->css); | |
2511 | loop++; | |
2512 | /* | |
2513 | * Could not reclaim anything and there are no more | |
2514 | * mem cgroups to try or we seem to be looping without | |
2515 | * reclaiming anything. | |
2516 | */ | |
2517 | if (!nr_reclaimed && | |
2518 | (next_mz == NULL || | |
2519 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2520 | break; | |
2521 | } while (!nr_reclaimed); | |
2522 | if (next_mz) | |
2523 | css_put(&next_mz->mem->css); | |
2524 | return nr_reclaimed; | |
2525 | } | |
2526 | ||
2527 | /* | |
2528 | * This routine traverse page_cgroup in given list and drop them all. | |
2529 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. | |
2530 | */ | |
2531 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, | |
2532 | int node, int zid, enum lru_list lru) | |
2533 | { | |
2534 | struct zone *zone; | |
2535 | struct mem_cgroup_per_zone *mz; | |
2536 | struct page_cgroup *pc, *busy; | |
2537 | unsigned long flags, loop; | |
2538 | struct list_head *list; | |
2539 | int ret = 0; | |
2540 | ||
2541 | zone = &NODE_DATA(node)->node_zones[zid]; | |
2542 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
2543 | list = &mz->lists[lru]; | |
2544 | ||
2545 | loop = MEM_CGROUP_ZSTAT(mz, lru); | |
2546 | /* give some margin against EBUSY etc...*/ | |
2547 | loop += 256; | |
2548 | busy = NULL; | |
2549 | while (loop--) { | |
2550 | ret = 0; | |
2551 | spin_lock_irqsave(&zone->lru_lock, flags); | |
2552 | if (list_empty(list)) { | |
2553 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
2554 | break; | |
2555 | } | |
2556 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
2557 | if (busy == pc) { | |
2558 | list_move(&pc->lru, list); | |
2559 | busy = 0; | |
2560 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
2561 | continue; | |
2562 | } | |
2563 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
2564 | ||
2565 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); | |
2566 | if (ret == -ENOMEM) | |
2567 | break; | |
2568 | ||
2569 | if (ret == -EBUSY || ret == -EINVAL) { | |
2570 | /* found lock contention or "pc" is obsolete. */ | |
2571 | busy = pc; | |
2572 | cond_resched(); | |
2573 | } else | |
2574 | busy = NULL; | |
2575 | } | |
2576 | ||
2577 | if (!ret && !list_empty(list)) | |
2578 | return -EBUSY; | |
2579 | return ret; | |
2580 | } | |
2581 | ||
2582 | /* | |
2583 | * make mem_cgroup's charge to be 0 if there is no task. | |
2584 | * This enables deleting this mem_cgroup. | |
2585 | */ | |
2586 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) | |
2587 | { | |
2588 | int ret; | |
2589 | int node, zid, shrink; | |
2590 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
2591 | struct cgroup *cgrp = mem->css.cgroup; | |
2592 | ||
2593 | css_get(&mem->css); | |
2594 | ||
2595 | shrink = 0; | |
2596 | /* should free all ? */ | |
2597 | if (free_all) | |
2598 | goto try_to_free; | |
2599 | move_account: | |
2600 | while (mem->res.usage > 0) { | |
2601 | ret = -EBUSY; | |
2602 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) | |
2603 | goto out; | |
2604 | ret = -EINTR; | |
2605 | if (signal_pending(current)) | |
2606 | goto out; | |
2607 | /* This is for making all *used* pages to be on LRU. */ | |
2608 | lru_add_drain_all(); | |
2609 | drain_all_stock_sync(); | |
2610 | ret = 0; | |
2611 | for_each_node_state(node, N_HIGH_MEMORY) { | |
2612 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { | |
2613 | enum lru_list l; | |
2614 | for_each_lru(l) { | |
2615 | ret = mem_cgroup_force_empty_list(mem, | |
2616 | node, zid, l); | |
2617 | if (ret) | |
2618 | break; | |
2619 | } | |
2620 | } | |
2621 | if (ret) | |
2622 | break; | |
2623 | } | |
2624 | /* it seems parent cgroup doesn't have enough mem */ | |
2625 | if (ret == -ENOMEM) | |
2626 | goto try_to_free; | |
2627 | cond_resched(); | |
2628 | } | |
2629 | ret = 0; | |
2630 | out: | |
2631 | css_put(&mem->css); | |
2632 | return ret; | |
2633 | ||
2634 | try_to_free: | |
2635 | /* returns EBUSY if there is a task or if we come here twice. */ | |
2636 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
2637 | ret = -EBUSY; | |
2638 | goto out; | |
2639 | } | |
2640 | /* we call try-to-free pages for make this cgroup empty */ | |
2641 | lru_add_drain_all(); | |
2642 | /* try to free all pages in this cgroup */ | |
2643 | shrink = 1; | |
2644 | while (nr_retries && mem->res.usage > 0) { | |
2645 | int progress; | |
2646 | ||
2647 | if (signal_pending(current)) { | |
2648 | ret = -EINTR; | |
2649 | goto out; | |
2650 | } | |
2651 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, | |
2652 | false, get_swappiness(mem)); | |
2653 | if (!progress) { | |
2654 | nr_retries--; | |
2655 | /* maybe some writeback is necessary */ | |
2656 | congestion_wait(BLK_RW_ASYNC, HZ/10); | |
2657 | } | |
2658 | ||
2659 | } | |
2660 | lru_add_drain(); | |
2661 | /* try move_account...there may be some *locked* pages. */ | |
2662 | if (mem->res.usage) | |
2663 | goto move_account; | |
2664 | ret = 0; | |
2665 | goto out; | |
2666 | } | |
2667 | ||
2668 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) | |
2669 | { | |
2670 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
2671 | } | |
2672 | ||
2673 | ||
2674 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) | |
2675 | { | |
2676 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
2677 | } | |
2678 | ||
2679 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
2680 | u64 val) | |
2681 | { | |
2682 | int retval = 0; | |
2683 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
2684 | struct cgroup *parent = cont->parent; | |
2685 | struct mem_cgroup *parent_mem = NULL; | |
2686 | ||
2687 | if (parent) | |
2688 | parent_mem = mem_cgroup_from_cont(parent); | |
2689 | ||
2690 | cgroup_lock(); | |
2691 | /* | |
2692 | * If parent's use_hierarchy is set, we can't make any modifications | |
2693 | * in the child subtrees. If it is unset, then the change can | |
2694 | * occur, provided the current cgroup has no children. | |
2695 | * | |
2696 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2697 | * set if there are no children. | |
2698 | */ | |
2699 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
2700 | (val == 1 || val == 0)) { | |
2701 | if (list_empty(&cont->children)) | |
2702 | mem->use_hierarchy = val; | |
2703 | else | |
2704 | retval = -EBUSY; | |
2705 | } else | |
2706 | retval = -EINVAL; | |
2707 | cgroup_unlock(); | |
2708 | ||
2709 | return retval; | |
2710 | } | |
2711 | ||
2712 | struct mem_cgroup_idx_data { | |
2713 | s64 val; | |
2714 | enum mem_cgroup_stat_index idx; | |
2715 | }; | |
2716 | ||
2717 | static int | |
2718 | mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data) | |
2719 | { | |
2720 | struct mem_cgroup_idx_data *d = data; | |
2721 | d->val += mem_cgroup_read_stat(&mem->stat, d->idx); | |
2722 | return 0; | |
2723 | } | |
2724 | ||
2725 | static void | |
2726 | mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem, | |
2727 | enum mem_cgroup_stat_index idx, s64 *val) | |
2728 | { | |
2729 | struct mem_cgroup_idx_data d; | |
2730 | d.idx = idx; | |
2731 | d.val = 0; | |
2732 | mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat); | |
2733 | *val = d.val; | |
2734 | } | |
2735 | ||
2736 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) | |
2737 | { | |
2738 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
2739 | u64 idx_val, val; | |
2740 | int type, name; | |
2741 | ||
2742 | type = MEMFILE_TYPE(cft->private); | |
2743 | name = MEMFILE_ATTR(cft->private); | |
2744 | switch (type) { | |
2745 | case _MEM: | |
2746 | if (name == RES_USAGE && mem_cgroup_is_root(mem)) { | |
2747 | mem_cgroup_get_recursive_idx_stat(mem, | |
2748 | MEM_CGROUP_STAT_CACHE, &idx_val); | |
2749 | val = idx_val; | |
2750 | mem_cgroup_get_recursive_idx_stat(mem, | |
2751 | MEM_CGROUP_STAT_RSS, &idx_val); | |
2752 | val += idx_val; | |
2753 | val <<= PAGE_SHIFT; | |
2754 | } else | |
2755 | val = res_counter_read_u64(&mem->res, name); | |
2756 | break; | |
2757 | case _MEMSWAP: | |
2758 | if (name == RES_USAGE && mem_cgroup_is_root(mem)) { | |
2759 | mem_cgroup_get_recursive_idx_stat(mem, | |
2760 | MEM_CGROUP_STAT_CACHE, &idx_val); | |
2761 | val = idx_val; | |
2762 | mem_cgroup_get_recursive_idx_stat(mem, | |
2763 | MEM_CGROUP_STAT_RSS, &idx_val); | |
2764 | val += idx_val; | |
2765 | mem_cgroup_get_recursive_idx_stat(mem, | |
2766 | MEM_CGROUP_STAT_SWAPOUT, &idx_val); | |
2767 | val += idx_val; | |
2768 | val <<= PAGE_SHIFT; | |
2769 | } else | |
2770 | val = res_counter_read_u64(&mem->memsw, name); | |
2771 | break; | |
2772 | default: | |
2773 | BUG(); | |
2774 | break; | |
2775 | } | |
2776 | return val; | |
2777 | } | |
2778 | /* | |
2779 | * The user of this function is... | |
2780 | * RES_LIMIT. | |
2781 | */ | |
2782 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, | |
2783 | const char *buffer) | |
2784 | { | |
2785 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); | |
2786 | int type, name; | |
2787 | unsigned long long val; | |
2788 | int ret; | |
2789 | ||
2790 | type = MEMFILE_TYPE(cft->private); | |
2791 | name = MEMFILE_ATTR(cft->private); | |
2792 | switch (name) { | |
2793 | case RES_LIMIT: | |
2794 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ | |
2795 | ret = -EINVAL; | |
2796 | break; | |
2797 | } | |
2798 | /* This function does all necessary parse...reuse it */ | |
2799 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
2800 | if (ret) | |
2801 | break; | |
2802 | if (type == _MEM) | |
2803 | ret = mem_cgroup_resize_limit(memcg, val); | |
2804 | else | |
2805 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
2806 | break; | |
2807 | case RES_SOFT_LIMIT: | |
2808 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
2809 | if (ret) | |
2810 | break; | |
2811 | /* | |
2812 | * For memsw, soft limits are hard to implement in terms | |
2813 | * of semantics, for now, we support soft limits for | |
2814 | * control without swap | |
2815 | */ | |
2816 | if (type == _MEM) | |
2817 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
2818 | else | |
2819 | ret = -EINVAL; | |
2820 | break; | |
2821 | default: | |
2822 | ret = -EINVAL; /* should be BUG() ? */ | |
2823 | break; | |
2824 | } | |
2825 | return ret; | |
2826 | } | |
2827 | ||
2828 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, | |
2829 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
2830 | { | |
2831 | struct cgroup *cgroup; | |
2832 | unsigned long long min_limit, min_memsw_limit, tmp; | |
2833 | ||
2834 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2835 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2836 | cgroup = memcg->css.cgroup; | |
2837 | if (!memcg->use_hierarchy) | |
2838 | goto out; | |
2839 | ||
2840 | while (cgroup->parent) { | |
2841 | cgroup = cgroup->parent; | |
2842 | memcg = mem_cgroup_from_cont(cgroup); | |
2843 | if (!memcg->use_hierarchy) | |
2844 | break; | |
2845 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2846 | min_limit = min(min_limit, tmp); | |
2847 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2848 | min_memsw_limit = min(min_memsw_limit, tmp); | |
2849 | } | |
2850 | out: | |
2851 | *mem_limit = min_limit; | |
2852 | *memsw_limit = min_memsw_limit; | |
2853 | return; | |
2854 | } | |
2855 | ||
2856 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) | |
2857 | { | |
2858 | struct mem_cgroup *mem; | |
2859 | int type, name; | |
2860 | ||
2861 | mem = mem_cgroup_from_cont(cont); | |
2862 | type = MEMFILE_TYPE(event); | |
2863 | name = MEMFILE_ATTR(event); | |
2864 | switch (name) { | |
2865 | case RES_MAX_USAGE: | |
2866 | if (type == _MEM) | |
2867 | res_counter_reset_max(&mem->res); | |
2868 | else | |
2869 | res_counter_reset_max(&mem->memsw); | |
2870 | break; | |
2871 | case RES_FAILCNT: | |
2872 | if (type == _MEM) | |
2873 | res_counter_reset_failcnt(&mem->res); | |
2874 | else | |
2875 | res_counter_reset_failcnt(&mem->memsw); | |
2876 | break; | |
2877 | } | |
2878 | ||
2879 | return 0; | |
2880 | } | |
2881 | ||
2882 | ||
2883 | /* For read statistics */ | |
2884 | enum { | |
2885 | MCS_CACHE, | |
2886 | MCS_RSS, | |
2887 | MCS_FILE_MAPPED, | |
2888 | MCS_PGPGIN, | |
2889 | MCS_PGPGOUT, | |
2890 | MCS_SWAP, | |
2891 | MCS_INACTIVE_ANON, | |
2892 | MCS_ACTIVE_ANON, | |
2893 | MCS_INACTIVE_FILE, | |
2894 | MCS_ACTIVE_FILE, | |
2895 | MCS_UNEVICTABLE, | |
2896 | NR_MCS_STAT, | |
2897 | }; | |
2898 | ||
2899 | struct mcs_total_stat { | |
2900 | s64 stat[NR_MCS_STAT]; | |
2901 | }; | |
2902 | ||
2903 | struct { | |
2904 | char *local_name; | |
2905 | char *total_name; | |
2906 | } memcg_stat_strings[NR_MCS_STAT] = { | |
2907 | {"cache", "total_cache"}, | |
2908 | {"rss", "total_rss"}, | |
2909 | {"mapped_file", "total_mapped_file"}, | |
2910 | {"pgpgin", "total_pgpgin"}, | |
2911 | {"pgpgout", "total_pgpgout"}, | |
2912 | {"swap", "total_swap"}, | |
2913 | {"inactive_anon", "total_inactive_anon"}, | |
2914 | {"active_anon", "total_active_anon"}, | |
2915 | {"inactive_file", "total_inactive_file"}, | |
2916 | {"active_file", "total_active_file"}, | |
2917 | {"unevictable", "total_unevictable"} | |
2918 | }; | |
2919 | ||
2920 | ||
2921 | static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data) | |
2922 | { | |
2923 | struct mcs_total_stat *s = data; | |
2924 | s64 val; | |
2925 | ||
2926 | /* per cpu stat */ | |
2927 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE); | |
2928 | s->stat[MCS_CACHE] += val * PAGE_SIZE; | |
2929 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
2930 | s->stat[MCS_RSS] += val * PAGE_SIZE; | |
2931 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_FILE_MAPPED); | |
2932 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; | |
2933 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT); | |
2934 | s->stat[MCS_PGPGIN] += val; | |
2935 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT); | |
2936 | s->stat[MCS_PGPGOUT] += val; | |
2937 | if (do_swap_account) { | |
2938 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_SWAPOUT); | |
2939 | s->stat[MCS_SWAP] += val * PAGE_SIZE; | |
2940 | } | |
2941 | ||
2942 | /* per zone stat */ | |
2943 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); | |
2944 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; | |
2945 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); | |
2946 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; | |
2947 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); | |
2948 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; | |
2949 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); | |
2950 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; | |
2951 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); | |
2952 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; | |
2953 | return 0; | |
2954 | } | |
2955 | ||
2956 | static void | |
2957 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
2958 | { | |
2959 | mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat); | |
2960 | } | |
2961 | ||
2962 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, | |
2963 | struct cgroup_map_cb *cb) | |
2964 | { | |
2965 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); | |
2966 | struct mcs_total_stat mystat; | |
2967 | int i; | |
2968 | ||
2969 | memset(&mystat, 0, sizeof(mystat)); | |
2970 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
2971 | ||
2972 | for (i = 0; i < NR_MCS_STAT; i++) { | |
2973 | if (i == MCS_SWAP && !do_swap_account) | |
2974 | continue; | |
2975 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); | |
2976 | } | |
2977 | ||
2978 | /* Hierarchical information */ | |
2979 | { | |
2980 | unsigned long long limit, memsw_limit; | |
2981 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
2982 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
2983 | if (do_swap_account) | |
2984 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
2985 | } | |
2986 | ||
2987 | memset(&mystat, 0, sizeof(mystat)); | |
2988 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
2989 | for (i = 0; i < NR_MCS_STAT; i++) { | |
2990 | if (i == MCS_SWAP && !do_swap_account) | |
2991 | continue; | |
2992 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); | |
2993 | } | |
2994 | ||
2995 | #ifdef CONFIG_DEBUG_VM | |
2996 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); | |
2997 | ||
2998 | { | |
2999 | int nid, zid; | |
3000 | struct mem_cgroup_per_zone *mz; | |
3001 | unsigned long recent_rotated[2] = {0, 0}; | |
3002 | unsigned long recent_scanned[2] = {0, 0}; | |
3003 | ||
3004 | for_each_online_node(nid) | |
3005 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
3006 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
3007 | ||
3008 | recent_rotated[0] += | |
3009 | mz->reclaim_stat.recent_rotated[0]; | |
3010 | recent_rotated[1] += | |
3011 | mz->reclaim_stat.recent_rotated[1]; | |
3012 | recent_scanned[0] += | |
3013 | mz->reclaim_stat.recent_scanned[0]; | |
3014 | recent_scanned[1] += | |
3015 | mz->reclaim_stat.recent_scanned[1]; | |
3016 | } | |
3017 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
3018 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
3019 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
3020 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
3021 | } | |
3022 | #endif | |
3023 | ||
3024 | return 0; | |
3025 | } | |
3026 | ||
3027 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) | |
3028 | { | |
3029 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3030 | ||
3031 | return get_swappiness(memcg); | |
3032 | } | |
3033 | ||
3034 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
3035 | u64 val) | |
3036 | { | |
3037 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3038 | struct mem_cgroup *parent; | |
3039 | ||
3040 | if (val > 100) | |
3041 | return -EINVAL; | |
3042 | ||
3043 | if (cgrp->parent == NULL) | |
3044 | return -EINVAL; | |
3045 | ||
3046 | parent = mem_cgroup_from_cont(cgrp->parent); | |
3047 | ||
3048 | cgroup_lock(); | |
3049 | ||
3050 | /* If under hierarchy, only empty-root can set this value */ | |
3051 | if ((parent->use_hierarchy) || | |
3052 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { | |
3053 | cgroup_unlock(); | |
3054 | return -EINVAL; | |
3055 | } | |
3056 | ||
3057 | spin_lock(&memcg->reclaim_param_lock); | |
3058 | memcg->swappiness = val; | |
3059 | spin_unlock(&memcg->reclaim_param_lock); | |
3060 | ||
3061 | cgroup_unlock(); | |
3062 | ||
3063 | return 0; | |
3064 | } | |
3065 | ||
3066 | ||
3067 | static struct cftype mem_cgroup_files[] = { | |
3068 | { | |
3069 | .name = "usage_in_bytes", | |
3070 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), | |
3071 | .read_u64 = mem_cgroup_read, | |
3072 | }, | |
3073 | { | |
3074 | .name = "max_usage_in_bytes", | |
3075 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), | |
3076 | .trigger = mem_cgroup_reset, | |
3077 | .read_u64 = mem_cgroup_read, | |
3078 | }, | |
3079 | { | |
3080 | .name = "limit_in_bytes", | |
3081 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), | |
3082 | .write_string = mem_cgroup_write, | |
3083 | .read_u64 = mem_cgroup_read, | |
3084 | }, | |
3085 | { | |
3086 | .name = "soft_limit_in_bytes", | |
3087 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
3088 | .write_string = mem_cgroup_write, | |
3089 | .read_u64 = mem_cgroup_read, | |
3090 | }, | |
3091 | { | |
3092 | .name = "failcnt", | |
3093 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), | |
3094 | .trigger = mem_cgroup_reset, | |
3095 | .read_u64 = mem_cgroup_read, | |
3096 | }, | |
3097 | { | |
3098 | .name = "stat", | |
3099 | .read_map = mem_control_stat_show, | |
3100 | }, | |
3101 | { | |
3102 | .name = "force_empty", | |
3103 | .trigger = mem_cgroup_force_empty_write, | |
3104 | }, | |
3105 | { | |
3106 | .name = "use_hierarchy", | |
3107 | .write_u64 = mem_cgroup_hierarchy_write, | |
3108 | .read_u64 = mem_cgroup_hierarchy_read, | |
3109 | }, | |
3110 | { | |
3111 | .name = "swappiness", | |
3112 | .read_u64 = mem_cgroup_swappiness_read, | |
3113 | .write_u64 = mem_cgroup_swappiness_write, | |
3114 | }, | |
3115 | }; | |
3116 | ||
3117 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
3118 | static struct cftype memsw_cgroup_files[] = { | |
3119 | { | |
3120 | .name = "memsw.usage_in_bytes", | |
3121 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
3122 | .read_u64 = mem_cgroup_read, | |
3123 | }, | |
3124 | { | |
3125 | .name = "memsw.max_usage_in_bytes", | |
3126 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
3127 | .trigger = mem_cgroup_reset, | |
3128 | .read_u64 = mem_cgroup_read, | |
3129 | }, | |
3130 | { | |
3131 | .name = "memsw.limit_in_bytes", | |
3132 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
3133 | .write_string = mem_cgroup_write, | |
3134 | .read_u64 = mem_cgroup_read, | |
3135 | }, | |
3136 | { | |
3137 | .name = "memsw.failcnt", | |
3138 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
3139 | .trigger = mem_cgroup_reset, | |
3140 | .read_u64 = mem_cgroup_read, | |
3141 | }, | |
3142 | }; | |
3143 | ||
3144 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
3145 | { | |
3146 | if (!do_swap_account) | |
3147 | return 0; | |
3148 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
3149 | ARRAY_SIZE(memsw_cgroup_files)); | |
3150 | }; | |
3151 | #else | |
3152 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
3153 | { | |
3154 | return 0; | |
3155 | } | |
3156 | #endif | |
3157 | ||
3158 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) | |
3159 | { | |
3160 | struct mem_cgroup_per_node *pn; | |
3161 | struct mem_cgroup_per_zone *mz; | |
3162 | enum lru_list l; | |
3163 | int zone, tmp = node; | |
3164 | /* | |
3165 | * This routine is called against possible nodes. | |
3166 | * But it's BUG to call kmalloc() against offline node. | |
3167 | * | |
3168 | * TODO: this routine can waste much memory for nodes which will | |
3169 | * never be onlined. It's better to use memory hotplug callback | |
3170 | * function. | |
3171 | */ | |
3172 | if (!node_state(node, N_NORMAL_MEMORY)) | |
3173 | tmp = -1; | |
3174 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
3175 | if (!pn) | |
3176 | return 1; | |
3177 | ||
3178 | mem->info.nodeinfo[node] = pn; | |
3179 | memset(pn, 0, sizeof(*pn)); | |
3180 | ||
3181 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
3182 | mz = &pn->zoneinfo[zone]; | |
3183 | for_each_lru(l) | |
3184 | INIT_LIST_HEAD(&mz->lists[l]); | |
3185 | mz->usage_in_excess = 0; | |
3186 | mz->on_tree = false; | |
3187 | mz->mem = mem; | |
3188 | } | |
3189 | return 0; | |
3190 | } | |
3191 | ||
3192 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) | |
3193 | { | |
3194 | kfree(mem->info.nodeinfo[node]); | |
3195 | } | |
3196 | ||
3197 | static int mem_cgroup_size(void) | |
3198 | { | |
3199 | int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu); | |
3200 | return sizeof(struct mem_cgroup) + cpustat_size; | |
3201 | } | |
3202 | ||
3203 | static struct mem_cgroup *mem_cgroup_alloc(void) | |
3204 | { | |
3205 | struct mem_cgroup *mem; | |
3206 | int size = mem_cgroup_size(); | |
3207 | ||
3208 | if (size < PAGE_SIZE) | |
3209 | mem = kmalloc(size, GFP_KERNEL); | |
3210 | else | |
3211 | mem = vmalloc(size); | |
3212 | ||
3213 | if (mem) | |
3214 | memset(mem, 0, size); | |
3215 | return mem; | |
3216 | } | |
3217 | ||
3218 | /* | |
3219 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
3220 | * (scanning all at force_empty is too costly...) | |
3221 | * | |
3222 | * Instead of clearing all references at force_empty, we remember | |
3223 | * the number of reference from swap_cgroup and free mem_cgroup when | |
3224 | * it goes down to 0. | |
3225 | * | |
3226 | * Removal of cgroup itself succeeds regardless of refs from swap. | |
3227 | */ | |
3228 | ||
3229 | static void __mem_cgroup_free(struct mem_cgroup *mem) | |
3230 | { | |
3231 | int node; | |
3232 | ||
3233 | mem_cgroup_remove_from_trees(mem); | |
3234 | free_css_id(&mem_cgroup_subsys, &mem->css); | |
3235 | ||
3236 | for_each_node_state(node, N_POSSIBLE) | |
3237 | free_mem_cgroup_per_zone_info(mem, node); | |
3238 | ||
3239 | if (mem_cgroup_size() < PAGE_SIZE) | |
3240 | kfree(mem); | |
3241 | else | |
3242 | vfree(mem); | |
3243 | } | |
3244 | ||
3245 | static void mem_cgroup_get(struct mem_cgroup *mem) | |
3246 | { | |
3247 | atomic_inc(&mem->refcnt); | |
3248 | } | |
3249 | ||
3250 | static void mem_cgroup_put(struct mem_cgroup *mem) | |
3251 | { | |
3252 | if (atomic_dec_and_test(&mem->refcnt)) { | |
3253 | struct mem_cgroup *parent = parent_mem_cgroup(mem); | |
3254 | __mem_cgroup_free(mem); | |
3255 | if (parent) | |
3256 | mem_cgroup_put(parent); | |
3257 | } | |
3258 | } | |
3259 | ||
3260 | /* | |
3261 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
3262 | */ | |
3263 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) | |
3264 | { | |
3265 | if (!mem->res.parent) | |
3266 | return NULL; | |
3267 | return mem_cgroup_from_res_counter(mem->res.parent, res); | |
3268 | } | |
3269 | ||
3270 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
3271 | static void __init enable_swap_cgroup(void) | |
3272 | { | |
3273 | if (!mem_cgroup_disabled() && really_do_swap_account) | |
3274 | do_swap_account = 1; | |
3275 | } | |
3276 | #else | |
3277 | static void __init enable_swap_cgroup(void) | |
3278 | { | |
3279 | } | |
3280 | #endif | |
3281 | ||
3282 | static int mem_cgroup_soft_limit_tree_init(void) | |
3283 | { | |
3284 | struct mem_cgroup_tree_per_node *rtpn; | |
3285 | struct mem_cgroup_tree_per_zone *rtpz; | |
3286 | int tmp, node, zone; | |
3287 | ||
3288 | for_each_node_state(node, N_POSSIBLE) { | |
3289 | tmp = node; | |
3290 | if (!node_state(node, N_NORMAL_MEMORY)) | |
3291 | tmp = -1; | |
3292 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
3293 | if (!rtpn) | |
3294 | return 1; | |
3295 | ||
3296 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
3297 | ||
3298 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
3299 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
3300 | rtpz->rb_root = RB_ROOT; | |
3301 | spin_lock_init(&rtpz->lock); | |
3302 | } | |
3303 | } | |
3304 | return 0; | |
3305 | } | |
3306 | ||
3307 | static struct cgroup_subsys_state * __ref | |
3308 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) | |
3309 | { | |
3310 | struct mem_cgroup *mem, *parent; | |
3311 | long error = -ENOMEM; | |
3312 | int node; | |
3313 | ||
3314 | mem = mem_cgroup_alloc(); | |
3315 | if (!mem) | |
3316 | return ERR_PTR(error); | |
3317 | ||
3318 | for_each_node_state(node, N_POSSIBLE) | |
3319 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
3320 | goto free_out; | |
3321 | ||
3322 | /* root ? */ | |
3323 | if (cont->parent == NULL) { | |
3324 | int cpu; | |
3325 | enable_swap_cgroup(); | |
3326 | parent = NULL; | |
3327 | root_mem_cgroup = mem; | |
3328 | if (mem_cgroup_soft_limit_tree_init()) | |
3329 | goto free_out; | |
3330 | for_each_possible_cpu(cpu) { | |
3331 | struct memcg_stock_pcp *stock = | |
3332 | &per_cpu(memcg_stock, cpu); | |
3333 | INIT_WORK(&stock->work, drain_local_stock); | |
3334 | } | |
3335 | hotcpu_notifier(memcg_stock_cpu_callback, 0); | |
3336 | ||
3337 | } else { | |
3338 | parent = mem_cgroup_from_cont(cont->parent); | |
3339 | mem->use_hierarchy = parent->use_hierarchy; | |
3340 | } | |
3341 | ||
3342 | if (parent && parent->use_hierarchy) { | |
3343 | res_counter_init(&mem->res, &parent->res); | |
3344 | res_counter_init(&mem->memsw, &parent->memsw); | |
3345 | /* | |
3346 | * We increment refcnt of the parent to ensure that we can | |
3347 | * safely access it on res_counter_charge/uncharge. | |
3348 | * This refcnt will be decremented when freeing this | |
3349 | * mem_cgroup(see mem_cgroup_put). | |
3350 | */ | |
3351 | mem_cgroup_get(parent); | |
3352 | } else { | |
3353 | res_counter_init(&mem->res, NULL); | |
3354 | res_counter_init(&mem->memsw, NULL); | |
3355 | } | |
3356 | mem->last_scanned_child = 0; | |
3357 | spin_lock_init(&mem->reclaim_param_lock); | |
3358 | ||
3359 | if (parent) | |
3360 | mem->swappiness = get_swappiness(parent); | |
3361 | atomic_set(&mem->refcnt, 1); | |
3362 | return &mem->css; | |
3363 | free_out: | |
3364 | __mem_cgroup_free(mem); | |
3365 | root_mem_cgroup = NULL; | |
3366 | return ERR_PTR(error); | |
3367 | } | |
3368 | ||
3369 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, | |
3370 | struct cgroup *cont) | |
3371 | { | |
3372 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
3373 | ||
3374 | return mem_cgroup_force_empty(mem, false); | |
3375 | } | |
3376 | ||
3377 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, | |
3378 | struct cgroup *cont) | |
3379 | { | |
3380 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
3381 | ||
3382 | mem_cgroup_put(mem); | |
3383 | } | |
3384 | ||
3385 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
3386 | struct cgroup *cont) | |
3387 | { | |
3388 | int ret; | |
3389 | ||
3390 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
3391 | ARRAY_SIZE(mem_cgroup_files)); | |
3392 | ||
3393 | if (!ret) | |
3394 | ret = register_memsw_files(cont, ss); | |
3395 | return ret; | |
3396 | } | |
3397 | ||
3398 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, | |
3399 | struct cgroup *cont, | |
3400 | struct cgroup *old_cont, | |
3401 | struct task_struct *p, | |
3402 | bool threadgroup) | |
3403 | { | |
3404 | mutex_lock(&memcg_tasklist); | |
3405 | /* | |
3406 | * FIXME: It's better to move charges of this process from old | |
3407 | * memcg to new memcg. But it's just on TODO-List now. | |
3408 | */ | |
3409 | mutex_unlock(&memcg_tasklist); | |
3410 | } | |
3411 | ||
3412 | struct cgroup_subsys mem_cgroup_subsys = { | |
3413 | .name = "memory", | |
3414 | .subsys_id = mem_cgroup_subsys_id, | |
3415 | .create = mem_cgroup_create, | |
3416 | .pre_destroy = mem_cgroup_pre_destroy, | |
3417 | .destroy = mem_cgroup_destroy, | |
3418 | .populate = mem_cgroup_populate, | |
3419 | .attach = mem_cgroup_move_task, | |
3420 | .early_init = 0, | |
3421 | .use_id = 1, | |
3422 | }; | |
3423 | ||
3424 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
3425 | ||
3426 | static int __init disable_swap_account(char *s) | |
3427 | { | |
3428 | really_do_swap_account = 0; | |
3429 | return 1; | |
3430 | } | |
3431 | __setup("noswapaccount", disable_swap_account); | |
3432 | #endif |