blob: 128f45c16fa6626ac33813fdaefc239c18e9967f [file] [log] [blame]
Balbir Singh8cdea7c2008-02-07 00:13:50 -08001/* memcontrol.c - Memory Controller
2 *
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
5 *
Pavel Emelianov78fb7462008-02-07 00:13:51 -08006 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
8 *
Balbir Singh8cdea7c2008-02-07 00:13:50 -08009 * 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>
Pavel Emelianov78fb7462008-02-07 00:13:51 -080023#include <linux/mm.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080024#include <linux/page-flags.h>
Balbir Singh66e17072008-02-07 00:13:56 -080025#include <linux/backing-dev.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080026#include <linux/bit_spinlock.h>
27#include <linux/rcupdate.h>
Balbir Singh66e17072008-02-07 00:13:56 -080028#include <linux/swap.h>
29#include <linux/spinlock.h>
30#include <linux/fs.h>
Balbir Singh8cdea7c2008-02-07 00:13:50 -080031
Balbir Singh8697d332008-02-07 00:13:59 -080032#include <asm/uaccess.h>
33
Balbir Singh8cdea7c2008-02-07 00:13:50 -080034struct cgroup_subsys mem_cgroup_subsys;
Balbir Singh66e17072008-02-07 00:13:56 -080035static const int MEM_CGROUP_RECLAIM_RETRIES = 5;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080036
37/*
38 * The memory controller data structure. The memory controller controls both
39 * page cache and RSS per cgroup. We would eventually like to provide
40 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
41 * to help the administrator determine what knobs to tune.
42 *
43 * TODO: Add a water mark for the memory controller. Reclaim will begin when
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080044 * we hit the water mark. May be even add a low water mark, such that
45 * no reclaim occurs from a cgroup at it's low water mark, this is
46 * a feature that will be implemented much later in the future.
Balbir Singh8cdea7c2008-02-07 00:13:50 -080047 */
48struct mem_cgroup {
49 struct cgroup_subsys_state css;
50 /*
51 * the counter to account for memory usage
52 */
53 struct res_counter res;
Pavel Emelianov78fb7462008-02-07 00:13:51 -080054 /*
55 * Per cgroup active and inactive list, similar to the
56 * per zone LRU lists.
57 * TODO: Consider making these lists per zone
58 */
59 struct list_head active_list;
60 struct list_head inactive_list;
Balbir Singh66e17072008-02-07 00:13:56 -080061 /*
62 * spin_lock to protect the per cgroup LRU
63 */
64 spinlock_t lru_lock;
Balbir Singh8697d332008-02-07 00:13:59 -080065 unsigned long control_type; /* control RSS or RSS+Pagecache */
Balbir Singh8cdea7c2008-02-07 00:13:50 -080066};
67
68/*
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080069 * We use the lower bit of the page->page_cgroup pointer as a bit spin
70 * lock. We need to ensure that page->page_cgroup is atleast two
71 * byte aligned (based on comments from Nick Piggin)
72 */
73#define PAGE_CGROUP_LOCK_BIT 0x0
74#define PAGE_CGROUP_LOCK (1 << PAGE_CGROUP_LOCK_BIT)
75
76/*
Balbir Singh8cdea7c2008-02-07 00:13:50 -080077 * A page_cgroup page is associated with every page descriptor. The
78 * page_cgroup helps us identify information about the cgroup
79 */
80struct page_cgroup {
81 struct list_head lru; /* per cgroup LRU list */
82 struct page *page;
83 struct mem_cgroup *mem_cgroup;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080084 atomic_t ref_cnt; /* Helpful when pages move b/w */
85 /* mapped and cached states */
Balbir Singh8cdea7c2008-02-07 00:13:50 -080086};
87
Balbir Singh8697d332008-02-07 00:13:59 -080088enum {
89 MEM_CGROUP_TYPE_UNSPEC = 0,
90 MEM_CGROUP_TYPE_MAPPED,
91 MEM_CGROUP_TYPE_CACHED,
92 MEM_CGROUP_TYPE_ALL,
93 MEM_CGROUP_TYPE_MAX,
94};
95
96static struct mem_cgroup init_mem_cgroup;
Balbir Singh8cdea7c2008-02-07 00:13:50 -080097
98static inline
99struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
100{
101 return container_of(cgroup_subsys_state(cont,
102 mem_cgroup_subsys_id), struct mem_cgroup,
103 css);
104}
105
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800106static inline
107struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
108{
109 return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
110 struct mem_cgroup, css);
111}
112
113void mm_init_cgroup(struct mm_struct *mm, struct task_struct *p)
114{
115 struct mem_cgroup *mem;
116
117 mem = mem_cgroup_from_task(p);
118 css_get(&mem->css);
119 mm->mem_cgroup = mem;
120}
121
122void mm_free_cgroup(struct mm_struct *mm)
123{
124 css_put(&mm->mem_cgroup->css);
125}
126
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800127static inline int page_cgroup_locked(struct page *page)
128{
129 return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT,
130 &page->page_cgroup);
131}
132
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800133void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc)
134{
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800135 int locked;
136
137 /*
138 * While resetting the page_cgroup we might not hold the
139 * page_cgroup lock. free_hot_cold_page() is an example
140 * of such a scenario
141 */
142 if (pc)
143 VM_BUG_ON(!page_cgroup_locked(page));
144 locked = (page->page_cgroup & PAGE_CGROUP_LOCK);
145 page->page_cgroup = ((unsigned long)pc | locked);
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800146}
147
148struct page_cgroup *page_get_page_cgroup(struct page *page)
149{
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800150 return (struct page_cgroup *)
151 (page->page_cgroup & ~PAGE_CGROUP_LOCK);
152}
153
Balbir Singh8697d332008-02-07 00:13:59 -0800154static void __always_inline lock_page_cgroup(struct page *page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800155{
156 bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
157 VM_BUG_ON(!page_cgroup_locked(page));
158}
159
Balbir Singh8697d332008-02-07 00:13:59 -0800160static void __always_inline unlock_page_cgroup(struct page *page)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800161{
162 bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
163}
164
KAMEZAWA Hiroyuki9175e032008-02-07 00:14:08 -0800165/*
166 * Tie new page_cgroup to struct page under lock_page_cgroup()
167 * This can fail if the page has been tied to a page_cgroup.
168 * If success, returns 0.
169 */
170static inline int
171page_cgroup_assign_new_page_cgroup(struct page *page, struct page_cgroup *pc)
172{
173 int ret = 0;
174
175 lock_page_cgroup(page);
176 if (!page_get_page_cgroup(page))
177 page_assign_page_cgroup(page, pc);
178 else /* A page is tied to other pc. */
179 ret = 1;
180 unlock_page_cgroup(page);
181 return ret;
182}
183
184/*
185 * Clear page->page_cgroup member under lock_page_cgroup().
186 * If given "pc" value is different from one page->page_cgroup,
187 * page->cgroup is not cleared.
188 * Returns a value of page->page_cgroup at lock taken.
189 * A can can detect failure of clearing by following
190 * clear_page_cgroup(page, pc) == pc
191 */
192
193static inline struct page_cgroup *
194clear_page_cgroup(struct page *page, struct page_cgroup *pc)
195{
196 struct page_cgroup *ret;
197 /* lock and clear */
198 lock_page_cgroup(page);
199 ret = page_get_page_cgroup(page);
200 if (likely(ret == pc))
201 page_assign_page_cgroup(page, NULL);
202 unlock_page_cgroup(page);
203 return ret;
204}
205
206
Balbir Singh8697d332008-02-07 00:13:59 -0800207static void __mem_cgroup_move_lists(struct page_cgroup *pc, bool active)
Balbir Singh66e17072008-02-07 00:13:56 -0800208{
209 if (active)
210 list_move(&pc->lru, &pc->mem_cgroup->active_list);
211 else
212 list_move(&pc->lru, &pc->mem_cgroup->inactive_list);
213}
214
David Rientjes4c4a2212008-02-07 00:14:06 -0800215int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem)
216{
217 int ret;
218
219 task_lock(task);
220 ret = task->mm && mm_cgroup(task->mm) == mem;
221 task_unlock(task);
222 return ret;
223}
224
Balbir Singh66e17072008-02-07 00:13:56 -0800225/*
226 * This routine assumes that the appropriate zone's lru lock is already held
227 */
228void mem_cgroup_move_lists(struct page_cgroup *pc, bool active)
229{
230 struct mem_cgroup *mem;
231 if (!pc)
232 return;
233
234 mem = pc->mem_cgroup;
235
236 spin_lock(&mem->lru_lock);
237 __mem_cgroup_move_lists(pc, active);
238 spin_unlock(&mem->lru_lock);
239}
240
241unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
242 struct list_head *dst,
243 unsigned long *scanned, int order,
244 int mode, struct zone *z,
245 struct mem_cgroup *mem_cont,
246 int active)
247{
248 unsigned long nr_taken = 0;
249 struct page *page;
250 unsigned long scan;
251 LIST_HEAD(pc_list);
252 struct list_head *src;
253 struct page_cgroup *pc;
254
255 if (active)
256 src = &mem_cont->active_list;
257 else
258 src = &mem_cont->inactive_list;
259
260 spin_lock(&mem_cont->lru_lock);
261 for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) {
262 pc = list_entry(src->prev, struct page_cgroup, lru);
263 page = pc->page;
264 VM_BUG_ON(!pc);
265
266 if (PageActive(page) && !active) {
267 __mem_cgroup_move_lists(pc, true);
268 scan--;
269 continue;
270 }
271 if (!PageActive(page) && active) {
272 __mem_cgroup_move_lists(pc, false);
273 scan--;
274 continue;
275 }
276
277 /*
278 * Reclaim, per zone
279 * TODO: make the active/inactive lists per zone
280 */
281 if (page_zone(page) != z)
282 continue;
283
284 /*
285 * Check if the meta page went away from under us
286 */
287 if (!list_empty(&pc->lru))
288 list_move(&pc->lru, &pc_list);
289 else
290 continue;
291
292 if (__isolate_lru_page(page, mode) == 0) {
293 list_move(&page->lru, dst);
294 nr_taken++;
295 }
296 }
297
298 list_splice(&pc_list, src);
299 spin_unlock(&mem_cont->lru_lock);
300
301 *scanned = scan;
302 return nr_taken;
303}
304
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800305/*
306 * Charge the memory controller for page usage.
307 * Return
308 * 0 if the charge was successful
309 * < 0 if the cgroup is over its limit
310 */
Balbir Singhe1a1cd52008-02-07 00:14:02 -0800311int mem_cgroup_charge(struct page *page, struct mm_struct *mm,
312 gfp_t gfp_mask)
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800313{
314 struct mem_cgroup *mem;
KAMEZAWA Hiroyuki9175e032008-02-07 00:14:08 -0800315 struct page_cgroup *pc;
Balbir Singh66e17072008-02-07 00:13:56 -0800316 unsigned long flags;
317 unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800318
319 /*
320 * Should page_cgroup's go to their own slab?
321 * One could optimize the performance of the charging routine
322 * by saving a bit in the page_flags and using it as a lock
323 * to see if the cgroup page already has a page_cgroup associated
324 * with it
325 */
Balbir Singh66e17072008-02-07 00:13:56 -0800326retry:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800327 lock_page_cgroup(page);
328 pc = page_get_page_cgroup(page);
329 /*
330 * The page_cgroup exists and the page has already been accounted
331 */
332 if (pc) {
Balbir Singh66e17072008-02-07 00:13:56 -0800333 if (unlikely(!atomic_inc_not_zero(&pc->ref_cnt))) {
334 /* this page is under being uncharged ? */
335 unlock_page_cgroup(page);
336 cpu_relax();
337 goto retry;
KAMEZAWA Hiroyuki9175e032008-02-07 00:14:08 -0800338 } else {
339 unlock_page_cgroup(page);
Balbir Singh66e17072008-02-07 00:13:56 -0800340 goto done;
KAMEZAWA Hiroyuki9175e032008-02-07 00:14:08 -0800341 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800342 }
343
344 unlock_page_cgroup(page);
345
Balbir Singhe1a1cd52008-02-07 00:14:02 -0800346 pc = kzalloc(sizeof(struct page_cgroup), gfp_mask);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800347 if (pc == NULL)
348 goto err;
349
350 rcu_read_lock();
351 /*
352 * We always charge the cgroup the mm_struct belongs to
353 * the mm_struct's mem_cgroup changes on task migration if the
354 * thread group leader migrates. It's possible that mm is not
355 * set, if so charge the init_mm (happens for pagecache usage).
356 */
357 if (!mm)
358 mm = &init_mm;
359
360 mem = rcu_dereference(mm->mem_cgroup);
361 /*
362 * For every charge from the cgroup, increment reference
363 * count
364 */
365 css_get(&mem->css);
366 rcu_read_unlock();
367
368 /*
369 * If we created the page_cgroup, we should free it on exceeding
370 * the cgroup limit.
371 */
Balbir Singh0eea1032008-02-07 00:13:57 -0800372 while (res_counter_charge(&mem->res, PAGE_SIZE)) {
Balbir Singhe1a1cd52008-02-07 00:14:02 -0800373 bool is_atomic = gfp_mask & GFP_ATOMIC;
374 /*
375 * We cannot reclaim under GFP_ATOMIC, fail the charge
376 */
377 if (is_atomic)
378 goto noreclaim;
379
380 if (try_to_free_mem_cgroup_pages(mem, gfp_mask))
Balbir Singh66e17072008-02-07 00:13:56 -0800381 continue;
382
383 /*
384 * try_to_free_mem_cgroup_pages() might not give us a full
385 * picture of reclaim. Some pages are reclaimed and might be
386 * moved to swap cache or just unmapped from the cgroup.
387 * Check the limit again to see if the reclaim reduced the
388 * current usage of the cgroup before giving up
389 */
390 if (res_counter_check_under_limit(&mem->res))
391 continue;
392 /*
393 * Since we control both RSS and cache, we end up with a
394 * very interesting scenario where we end up reclaiming
395 * memory (essentially RSS), since the memory is pushed
396 * to swap cache, we eventually end up adding those
397 * pages back to our list. Hence we give ourselves a
398 * few chances before we fail
399 */
400 else if (nr_retries--) {
401 congestion_wait(WRITE, HZ/10);
402 continue;
403 }
Balbir Singhe1a1cd52008-02-07 00:14:02 -0800404noreclaim:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800405 css_put(&mem->css);
Balbir Singhe1a1cd52008-02-07 00:14:02 -0800406 if (!is_atomic)
407 mem_cgroup_out_of_memory(mem, GFP_KERNEL);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800408 goto free_pc;
409 }
410
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800411 atomic_set(&pc->ref_cnt, 1);
412 pc->mem_cgroup = mem;
413 pc->page = page;
KAMEZAWA Hiroyuki9175e032008-02-07 00:14:08 -0800414 if (page_cgroup_assign_new_page_cgroup(page, pc)) {
415 /*
416 * an another charge is added to this page already.
417 * we do take lock_page_cgroup(page) again and read
418 * page->cgroup, increment refcnt.... just retry is OK.
419 */
420 res_counter_uncharge(&mem->res, PAGE_SIZE);
421 css_put(&mem->css);
422 kfree(pc);
423 goto retry;
424 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800425
Balbir Singh66e17072008-02-07 00:13:56 -0800426 spin_lock_irqsave(&mem->lru_lock, flags);
427 list_add(&pc->lru, &mem->active_list);
428 spin_unlock_irqrestore(&mem->lru_lock, flags);
429
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800430done:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800431 return 0;
432free_pc:
433 kfree(pc);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800434err:
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800435 return -ENOMEM;
436}
437
438/*
Balbir Singh8697d332008-02-07 00:13:59 -0800439 * See if the cached pages should be charged at all?
440 */
Balbir Singhe1a1cd52008-02-07 00:14:02 -0800441int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
442 gfp_t gfp_mask)
Balbir Singh8697d332008-02-07 00:13:59 -0800443{
444 struct mem_cgroup *mem;
445 if (!mm)
446 mm = &init_mm;
447
448 mem = rcu_dereference(mm->mem_cgroup);
449 if (mem->control_type == MEM_CGROUP_TYPE_ALL)
Balbir Singhe1a1cd52008-02-07 00:14:02 -0800450 return mem_cgroup_charge(page, mm, gfp_mask);
Balbir Singh8697d332008-02-07 00:13:59 -0800451 else
452 return 0;
453}
454
455/*
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800456 * Uncharging is always a welcome operation, we never complain, simply
457 * uncharge.
458 */
459void mem_cgroup_uncharge(struct page_cgroup *pc)
460{
461 struct mem_cgroup *mem;
462 struct page *page;
Balbir Singh66e17072008-02-07 00:13:56 -0800463 unsigned long flags;
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800464
Balbir Singh8697d332008-02-07 00:13:59 -0800465 /*
466 * This can handle cases when a page is not charged at all and we
467 * are switching between handling the control_type.
468 */
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800469 if (!pc)
470 return;
471
472 if (atomic_dec_and_test(&pc->ref_cnt)) {
473 page = pc->page;
KAMEZAWA Hiroyuki9175e032008-02-07 00:14:08 -0800474 /*
475 * get page->cgroup and clear it under lock.
476 */
477 if (clear_page_cgroup(page, pc) == pc) {
478 mem = pc->mem_cgroup;
479 css_put(&mem->css);
480 res_counter_uncharge(&mem->res, PAGE_SIZE);
481 spin_lock_irqsave(&mem->lru_lock, flags);
482 list_del_init(&pc->lru);
483 spin_unlock_irqrestore(&mem->lru_lock, flags);
484 kfree(pc);
485 } else {
486 /*
487 * Note:This will be removed when force-empty patch is
488 * applied. just show warning here.
489 */
490 printk(KERN_ERR "Race in mem_cgroup_uncharge() ?");
491 dump_stack();
492 }
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800493 }
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800494}
KAMEZAWA Hiroyukiae41be32008-02-07 00:14:10 -0800495/*
496 * Returns non-zero if a page (under migration) has valid page_cgroup member.
497 * Refcnt of page_cgroup is incremented.
498 */
499
500int mem_cgroup_prepare_migration(struct page *page)
501{
502 struct page_cgroup *pc;
503 int ret = 0;
504 lock_page_cgroup(page);
505 pc = page_get_page_cgroup(page);
506 if (pc && atomic_inc_not_zero(&pc->ref_cnt))
507 ret = 1;
508 unlock_page_cgroup(page);
509 return ret;
510}
511
512void mem_cgroup_end_migration(struct page *page)
513{
514 struct page_cgroup *pc = page_get_page_cgroup(page);
515 mem_cgroup_uncharge(pc);
516}
517/*
518 * We know both *page* and *newpage* are now not-on-LRU and Pg_locked.
519 * And no race with uncharge() routines because page_cgroup for *page*
520 * has extra one reference by mem_cgroup_prepare_migration.
521 */
522
523void mem_cgroup_page_migration(struct page *page, struct page *newpage)
524{
525 struct page_cgroup *pc;
526retry:
527 pc = page_get_page_cgroup(page);
528 if (!pc)
529 return;
530 if (clear_page_cgroup(page, pc) != pc)
531 goto retry;
532 pc->page = newpage;
533 lock_page_cgroup(newpage);
534 page_assign_page_cgroup(newpage, pc);
535 unlock_page_cgroup(newpage);
536 return;
537}
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800538
Balbir Singh0eea1032008-02-07 00:13:57 -0800539int mem_cgroup_write_strategy(char *buf, unsigned long long *tmp)
540{
541 *tmp = memparse(buf, &buf);
542 if (*buf != '\0')
543 return -EINVAL;
544
545 /*
546 * Round up the value to the closest page size
547 */
548 *tmp = ((*tmp + PAGE_SIZE - 1) >> PAGE_SHIFT) << PAGE_SHIFT;
549 return 0;
550}
551
552static ssize_t mem_cgroup_read(struct cgroup *cont,
553 struct cftype *cft, struct file *file,
554 char __user *userbuf, size_t nbytes, loff_t *ppos)
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800555{
556 return res_counter_read(&mem_cgroup_from_cont(cont)->res,
Balbir Singh0eea1032008-02-07 00:13:57 -0800557 cft->private, userbuf, nbytes, ppos,
558 NULL);
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800559}
560
561static ssize_t mem_cgroup_write(struct cgroup *cont, struct cftype *cft,
562 struct file *file, const char __user *userbuf,
563 size_t nbytes, loff_t *ppos)
564{
565 return res_counter_write(&mem_cgroup_from_cont(cont)->res,
Balbir Singh0eea1032008-02-07 00:13:57 -0800566 cft->private, userbuf, nbytes, ppos,
567 mem_cgroup_write_strategy);
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800568}
569
Balbir Singh8697d332008-02-07 00:13:59 -0800570static ssize_t mem_control_type_write(struct cgroup *cont,
571 struct cftype *cft, struct file *file,
572 const char __user *userbuf,
573 size_t nbytes, loff_t *pos)
574{
575 int ret;
576 char *buf, *end;
577 unsigned long tmp;
578 struct mem_cgroup *mem;
579
580 mem = mem_cgroup_from_cont(cont);
581 buf = kmalloc(nbytes + 1, GFP_KERNEL);
582 ret = -ENOMEM;
583 if (buf == NULL)
584 goto out;
585
586 buf[nbytes] = 0;
587 ret = -EFAULT;
588 if (copy_from_user(buf, userbuf, nbytes))
589 goto out_free;
590
591 ret = -EINVAL;
592 tmp = simple_strtoul(buf, &end, 10);
593 if (*end != '\0')
594 goto out_free;
595
596 if (tmp <= MEM_CGROUP_TYPE_UNSPEC || tmp >= MEM_CGROUP_TYPE_MAX)
597 goto out_free;
598
599 mem->control_type = tmp;
600 ret = nbytes;
601out_free:
602 kfree(buf);
603out:
604 return ret;
605}
606
607static ssize_t mem_control_type_read(struct cgroup *cont,
608 struct cftype *cft,
609 struct file *file, char __user *userbuf,
610 size_t nbytes, loff_t *ppos)
611{
612 unsigned long val;
613 char buf[64], *s;
614 struct mem_cgroup *mem;
615
616 mem = mem_cgroup_from_cont(cont);
617 s = buf;
618 val = mem->control_type;
619 s += sprintf(s, "%lu\n", val);
620 return simple_read_from_buffer((void __user *)userbuf, nbytes,
621 ppos, buf, s - buf);
622}
623
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800624static struct cftype mem_cgroup_files[] = {
625 {
Balbir Singh0eea1032008-02-07 00:13:57 -0800626 .name = "usage_in_bytes",
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800627 .private = RES_USAGE,
628 .read = mem_cgroup_read,
629 },
630 {
Balbir Singh0eea1032008-02-07 00:13:57 -0800631 .name = "limit_in_bytes",
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800632 .private = RES_LIMIT,
633 .write = mem_cgroup_write,
634 .read = mem_cgroup_read,
635 },
636 {
637 .name = "failcnt",
638 .private = RES_FAILCNT,
639 .read = mem_cgroup_read,
640 },
Balbir Singh8697d332008-02-07 00:13:59 -0800641 {
642 .name = "control_type",
643 .write = mem_control_type_write,
644 .read = mem_control_type_read,
645 },
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800646};
647
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800648static struct mem_cgroup init_mem_cgroup;
649
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800650static struct cgroup_subsys_state *
651mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
652{
653 struct mem_cgroup *mem;
654
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800655 if (unlikely((cont->parent) == NULL)) {
656 mem = &init_mem_cgroup;
657 init_mm.mem_cgroup = mem;
658 } else
659 mem = kzalloc(sizeof(struct mem_cgroup), GFP_KERNEL);
660
661 if (mem == NULL)
662 return NULL;
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800663
664 res_counter_init(&mem->res);
Balbir Singh8a9f3cc2008-02-07 00:13:53 -0800665 INIT_LIST_HEAD(&mem->active_list);
666 INIT_LIST_HEAD(&mem->inactive_list);
Balbir Singh66e17072008-02-07 00:13:56 -0800667 spin_lock_init(&mem->lru_lock);
Balbir Singh8697d332008-02-07 00:13:59 -0800668 mem->control_type = MEM_CGROUP_TYPE_ALL;
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800669 return &mem->css;
670}
671
672static void mem_cgroup_destroy(struct cgroup_subsys *ss,
673 struct cgroup *cont)
674{
675 kfree(mem_cgroup_from_cont(cont));
676}
677
678static int mem_cgroup_populate(struct cgroup_subsys *ss,
679 struct cgroup *cont)
680{
681 return cgroup_add_files(cont, ss, mem_cgroup_files,
682 ARRAY_SIZE(mem_cgroup_files));
683}
684
Balbir Singh67e465a2008-02-07 00:13:54 -0800685static void mem_cgroup_move_task(struct cgroup_subsys *ss,
686 struct cgroup *cont,
687 struct cgroup *old_cont,
688 struct task_struct *p)
689{
690 struct mm_struct *mm;
691 struct mem_cgroup *mem, *old_mem;
692
693 mm = get_task_mm(p);
694 if (mm == NULL)
695 return;
696
697 mem = mem_cgroup_from_cont(cont);
698 old_mem = mem_cgroup_from_cont(old_cont);
699
700 if (mem == old_mem)
701 goto out;
702
703 /*
704 * Only thread group leaders are allowed to migrate, the mm_struct is
705 * in effect owned by the leader
706 */
707 if (p->tgid != p->pid)
708 goto out;
709
710 css_get(&mem->css);
711 rcu_assign_pointer(mm->mem_cgroup, mem);
712 css_put(&old_mem->css);
713
714out:
715 mmput(mm);
716 return;
717}
718
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800719struct cgroup_subsys mem_cgroup_subsys = {
720 .name = "memory",
721 .subsys_id = mem_cgroup_subsys_id,
722 .create = mem_cgroup_create,
723 .destroy = mem_cgroup_destroy,
724 .populate = mem_cgroup_populate,
Balbir Singh67e465a2008-02-07 00:13:54 -0800725 .attach = mem_cgroup_move_task,
Pavel Emelianov78fb7462008-02-07 00:13:51 -0800726 .early_init = 1,
Balbir Singh8cdea7c2008-02-07 00:13:50 -0800727};