| 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/smp.h> |
| 25 | #include <linux/page-flags.h> |
| 26 | #include <linux/backing-dev.h> |
| 27 | #include <linux/bit_spinlock.h> |
| 28 | #include <linux/rcupdate.h> |
| 29 | #include <linux/slab.h> |
| 30 | #include <linux/swap.h> |
| 31 | #include <linux/spinlock.h> |
| 32 | #include <linux/fs.h> |
| 33 | #include <linux/seq_file.h> |
| 34 | #include <linux/vmalloc.h> |
| 35 | #include <linux/mm_inline.h> |
| 36 | #include <linux/page_cgroup.h> |
| 37 | |
| 38 | #include <asm/uaccess.h> |
| 39 | |
| 40 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
| 41 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
| 42 | |
| 43 | /* |
| 44 | * Statistics for memory cgroup. |
| 45 | */ |
| 46 | enum mem_cgroup_stat_index { |
| 47 | /* |
| 48 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. |
| 49 | */ |
| 50 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ |
| 51 | MEM_CGROUP_STAT_RSS, /* # of pages charged as rss */ |
| 52 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
| 53 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ |
| 54 | |
| 55 | MEM_CGROUP_STAT_NSTATS, |
| 56 | }; |
| 57 | |
| 58 | struct mem_cgroup_stat_cpu { |
| 59 | s64 count[MEM_CGROUP_STAT_NSTATS]; |
| 60 | } ____cacheline_aligned_in_smp; |
| 61 | |
| 62 | struct mem_cgroup_stat { |
| 63 | struct mem_cgroup_stat_cpu cpustat[NR_CPUS]; |
| 64 | }; |
| 65 | |
| 66 | /* |
| 67 | * For accounting under irq disable, no need for increment preempt count. |
| 68 | */ |
| 69 | static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, |
| 70 | enum mem_cgroup_stat_index idx, int val) |
| 71 | { |
| 72 | stat->count[idx] += val; |
| 73 | } |
| 74 | |
| 75 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, |
| 76 | enum mem_cgroup_stat_index idx) |
| 77 | { |
| 78 | int cpu; |
| 79 | s64 ret = 0; |
| 80 | for_each_possible_cpu(cpu) |
| 81 | ret += stat->cpustat[cpu].count[idx]; |
| 82 | return ret; |
| 83 | } |
| 84 | |
| 85 | /* |
| 86 | * per-zone information in memory controller. |
| 87 | */ |
| 88 | struct mem_cgroup_per_zone { |
| 89 | /* |
| 90 | * spin_lock to protect the per cgroup LRU |
| 91 | */ |
| 92 | spinlock_t lru_lock; |
| 93 | struct list_head lists[NR_LRU_LISTS]; |
| 94 | unsigned long count[NR_LRU_LISTS]; |
| 95 | }; |
| 96 | /* Macro for accessing counter */ |
| 97 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) |
| 98 | |
| 99 | struct mem_cgroup_per_node { |
| 100 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; |
| 101 | }; |
| 102 | |
| 103 | struct mem_cgroup_lru_info { |
| 104 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; |
| 105 | }; |
| 106 | |
| 107 | /* |
| 108 | * The memory controller data structure. The memory controller controls both |
| 109 | * page cache and RSS per cgroup. We would eventually like to provide |
| 110 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, |
| 111 | * to help the administrator determine what knobs to tune. |
| 112 | * |
| 113 | * TODO: Add a water mark for the memory controller. Reclaim will begin when |
| 114 | * we hit the water mark. May be even add a low water mark, such that |
| 115 | * no reclaim occurs from a cgroup at it's low water mark, this is |
| 116 | * a feature that will be implemented much later in the future. |
| 117 | */ |
| 118 | struct mem_cgroup { |
| 119 | struct cgroup_subsys_state css; |
| 120 | /* |
| 121 | * the counter to account for memory usage |
| 122 | */ |
| 123 | struct res_counter res; |
| 124 | /* |
| 125 | * Per cgroup active and inactive list, similar to the |
| 126 | * per zone LRU lists. |
| 127 | */ |
| 128 | struct mem_cgroup_lru_info info; |
| 129 | |
| 130 | int prev_priority; /* for recording reclaim priority */ |
| 131 | /* |
| 132 | * statistics. |
| 133 | */ |
| 134 | struct mem_cgroup_stat stat; |
| 135 | }; |
| 136 | static struct mem_cgroup init_mem_cgroup; |
| 137 | |
| 138 | enum charge_type { |
| 139 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, |
| 140 | MEM_CGROUP_CHARGE_TYPE_MAPPED, |
| 141 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
| 142 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
| 143 | NR_CHARGE_TYPE, |
| 144 | }; |
| 145 | |
| 146 | /* only for here (for easy reading.) */ |
| 147 | #define PCGF_CACHE (1UL << PCG_CACHE) |
| 148 | #define PCGF_USED (1UL << PCG_USED) |
| 149 | #define PCGF_ACTIVE (1UL << PCG_ACTIVE) |
| 150 | #define PCGF_LOCK (1UL << PCG_LOCK) |
| 151 | #define PCGF_FILE (1UL << PCG_FILE) |
| 152 | static const unsigned long |
| 153 | pcg_default_flags[NR_CHARGE_TYPE] = { |
| 154 | PCGF_CACHE | PCGF_FILE | PCGF_USED | PCGF_LOCK, /* File Cache */ |
| 155 | PCGF_ACTIVE | PCGF_USED | PCGF_LOCK, /* Anon */ |
| 156 | PCGF_ACTIVE | PCGF_CACHE | PCGF_USED | PCGF_LOCK, /* Shmem */ |
| 157 | 0, /* FORCE */ |
| 158 | }; |
| 159 | |
| 160 | /* |
| 161 | * Always modified under lru lock. Then, not necessary to preempt_disable() |
| 162 | */ |
| 163 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
| 164 | struct page_cgroup *pc, |
| 165 | bool charge) |
| 166 | { |
| 167 | int val = (charge)? 1 : -1; |
| 168 | struct mem_cgroup_stat *stat = &mem->stat; |
| 169 | struct mem_cgroup_stat_cpu *cpustat; |
| 170 | |
| 171 | VM_BUG_ON(!irqs_disabled()); |
| 172 | |
| 173 | cpustat = &stat->cpustat[smp_processor_id()]; |
| 174 | if (PageCgroupCache(pc)) |
| 175 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); |
| 176 | else |
| 177 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); |
| 178 | |
| 179 | if (charge) |
| 180 | __mem_cgroup_stat_add_safe(cpustat, |
| 181 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); |
| 182 | else |
| 183 | __mem_cgroup_stat_add_safe(cpustat, |
| 184 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); |
| 185 | } |
| 186 | |
| 187 | static struct mem_cgroup_per_zone * |
| 188 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
| 189 | { |
| 190 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
| 191 | } |
| 192 | |
| 193 | static struct mem_cgroup_per_zone * |
| 194 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
| 195 | { |
| 196 | struct mem_cgroup *mem = pc->mem_cgroup; |
| 197 | int nid = page_cgroup_nid(pc); |
| 198 | int zid = page_cgroup_zid(pc); |
| 199 | |
| 200 | return mem_cgroup_zoneinfo(mem, nid, zid); |
| 201 | } |
| 202 | |
| 203 | static unsigned long mem_cgroup_get_all_zonestat(struct mem_cgroup *mem, |
| 204 | enum lru_list idx) |
| 205 | { |
| 206 | int nid, zid; |
| 207 | struct mem_cgroup_per_zone *mz; |
| 208 | u64 total = 0; |
| 209 | |
| 210 | for_each_online_node(nid) |
| 211 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 212 | mz = mem_cgroup_zoneinfo(mem, nid, zid); |
| 213 | total += MEM_CGROUP_ZSTAT(mz, idx); |
| 214 | } |
| 215 | return total; |
| 216 | } |
| 217 | |
| 218 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
| 219 | { |
| 220 | return container_of(cgroup_subsys_state(cont, |
| 221 | mem_cgroup_subsys_id), struct mem_cgroup, |
| 222 | css); |
| 223 | } |
| 224 | |
| 225 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
| 226 | { |
| 227 | /* |
| 228 | * mm_update_next_owner() may clear mm->owner to NULL |
| 229 | * if it races with swapoff, page migration, etc. |
| 230 | * So this can be called with p == NULL. |
| 231 | */ |
| 232 | if (unlikely(!p)) |
| 233 | return NULL; |
| 234 | |
| 235 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
| 236 | struct mem_cgroup, css); |
| 237 | } |
| 238 | |
| 239 | static void __mem_cgroup_remove_list(struct mem_cgroup_per_zone *mz, |
| 240 | struct page_cgroup *pc) |
| 241 | { |
| 242 | int lru = LRU_BASE; |
| 243 | |
| 244 | if (PageCgroupUnevictable(pc)) |
| 245 | lru = LRU_UNEVICTABLE; |
| 246 | else { |
| 247 | if (PageCgroupActive(pc)) |
| 248 | lru += LRU_ACTIVE; |
| 249 | if (PageCgroupFile(pc)) |
| 250 | lru += LRU_FILE; |
| 251 | } |
| 252 | |
| 253 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
| 254 | |
| 255 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc, false); |
| 256 | list_del(&pc->lru); |
| 257 | } |
| 258 | |
| 259 | static void __mem_cgroup_add_list(struct mem_cgroup_per_zone *mz, |
| 260 | struct page_cgroup *pc) |
| 261 | { |
| 262 | int lru = LRU_BASE; |
| 263 | |
| 264 | if (PageCgroupUnevictable(pc)) |
| 265 | lru = LRU_UNEVICTABLE; |
| 266 | else { |
| 267 | if (PageCgroupActive(pc)) |
| 268 | lru += LRU_ACTIVE; |
| 269 | if (PageCgroupFile(pc)) |
| 270 | lru += LRU_FILE; |
| 271 | } |
| 272 | |
| 273 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
| 274 | list_add(&pc->lru, &mz->lists[lru]); |
| 275 | |
| 276 | mem_cgroup_charge_statistics(pc->mem_cgroup, pc, true); |
| 277 | } |
| 278 | |
| 279 | static void __mem_cgroup_move_lists(struct page_cgroup *pc, enum lru_list lru) |
| 280 | { |
| 281 | struct mem_cgroup_per_zone *mz = page_cgroup_zoneinfo(pc); |
| 282 | int active = PageCgroupActive(pc); |
| 283 | int file = PageCgroupFile(pc); |
| 284 | int unevictable = PageCgroupUnevictable(pc); |
| 285 | enum lru_list from = unevictable ? LRU_UNEVICTABLE : |
| 286 | (LRU_FILE * !!file + !!active); |
| 287 | |
| 288 | if (lru == from) |
| 289 | return; |
| 290 | |
| 291 | MEM_CGROUP_ZSTAT(mz, from) -= 1; |
| 292 | /* |
| 293 | * However this is done under mz->lru_lock, another flags, which |
| 294 | * are not related to LRU, will be modified from out-of-lock. |
| 295 | * We have to use atomic set/clear flags. |
| 296 | */ |
| 297 | if (is_unevictable_lru(lru)) { |
| 298 | ClearPageCgroupActive(pc); |
| 299 | SetPageCgroupUnevictable(pc); |
| 300 | } else { |
| 301 | if (is_active_lru(lru)) |
| 302 | SetPageCgroupActive(pc); |
| 303 | else |
| 304 | ClearPageCgroupActive(pc); |
| 305 | ClearPageCgroupUnevictable(pc); |
| 306 | } |
| 307 | |
| 308 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
| 309 | list_move(&pc->lru, &mz->lists[lru]); |
| 310 | } |
| 311 | |
| 312 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
| 313 | { |
| 314 | int ret; |
| 315 | |
| 316 | task_lock(task); |
| 317 | ret = task->mm && mm_match_cgroup(task->mm, mem); |
| 318 | task_unlock(task); |
| 319 | return ret; |
| 320 | } |
| 321 | |
| 322 | /* |
| 323 | * This routine assumes that the appropriate zone's lru lock is already held |
| 324 | */ |
| 325 | void mem_cgroup_move_lists(struct page *page, enum lru_list lru) |
| 326 | { |
| 327 | struct page_cgroup *pc; |
| 328 | struct mem_cgroup_per_zone *mz; |
| 329 | unsigned long flags; |
| 330 | |
| 331 | if (mem_cgroup_subsys.disabled) |
| 332 | return; |
| 333 | |
| 334 | /* |
| 335 | * We cannot lock_page_cgroup while holding zone's lru_lock, |
| 336 | * because other holders of lock_page_cgroup can be interrupted |
| 337 | * with an attempt to rotate_reclaimable_page. But we cannot |
| 338 | * safely get to page_cgroup without it, so just try_lock it: |
| 339 | * mem_cgroup_isolate_pages allows for page left on wrong list. |
| 340 | */ |
| 341 | pc = lookup_page_cgroup(page); |
| 342 | if (!trylock_page_cgroup(pc)) |
| 343 | return; |
| 344 | if (pc && PageCgroupUsed(pc)) { |
| 345 | mz = page_cgroup_zoneinfo(pc); |
| 346 | spin_lock_irqsave(&mz->lru_lock, flags); |
| 347 | __mem_cgroup_move_lists(pc, lru); |
| 348 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
| 349 | } |
| 350 | unlock_page_cgroup(pc); |
| 351 | } |
| 352 | |
| 353 | /* |
| 354 | * Calculate mapped_ratio under memory controller. This will be used in |
| 355 | * vmscan.c for deteremining we have to reclaim mapped pages. |
| 356 | */ |
| 357 | int mem_cgroup_calc_mapped_ratio(struct mem_cgroup *mem) |
| 358 | { |
| 359 | long total, rss; |
| 360 | |
| 361 | /* |
| 362 | * usage is recorded in bytes. But, here, we assume the number of |
| 363 | * physical pages can be represented by "long" on any arch. |
| 364 | */ |
| 365 | total = (long) (mem->res.usage >> PAGE_SHIFT) + 1L; |
| 366 | rss = (long)mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); |
| 367 | return (int)((rss * 100L) / total); |
| 368 | } |
| 369 | |
| 370 | /* |
| 371 | * prev_priority control...this will be used in memory reclaim path. |
| 372 | */ |
| 373 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) |
| 374 | { |
| 375 | return mem->prev_priority; |
| 376 | } |
| 377 | |
| 378 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) |
| 379 | { |
| 380 | if (priority < mem->prev_priority) |
| 381 | mem->prev_priority = priority; |
| 382 | } |
| 383 | |
| 384 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) |
| 385 | { |
| 386 | mem->prev_priority = priority; |
| 387 | } |
| 388 | |
| 389 | /* |
| 390 | * Calculate # of pages to be scanned in this priority/zone. |
| 391 | * See also vmscan.c |
| 392 | * |
| 393 | * priority starts from "DEF_PRIORITY" and decremented in each loop. |
| 394 | * (see include/linux/mmzone.h) |
| 395 | */ |
| 396 | |
| 397 | long mem_cgroup_calc_reclaim(struct mem_cgroup *mem, struct zone *zone, |
| 398 | int priority, enum lru_list lru) |
| 399 | { |
| 400 | long nr_pages; |
| 401 | int nid = zone->zone_pgdat->node_id; |
| 402 | int zid = zone_idx(zone); |
| 403 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(mem, nid, zid); |
| 404 | |
| 405 | nr_pages = MEM_CGROUP_ZSTAT(mz, lru); |
| 406 | |
| 407 | return (nr_pages >> priority); |
| 408 | } |
| 409 | |
| 410 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
| 411 | struct list_head *dst, |
| 412 | unsigned long *scanned, int order, |
| 413 | int mode, struct zone *z, |
| 414 | struct mem_cgroup *mem_cont, |
| 415 | int active, int file) |
| 416 | { |
| 417 | unsigned long nr_taken = 0; |
| 418 | struct page *page; |
| 419 | unsigned long scan; |
| 420 | LIST_HEAD(pc_list); |
| 421 | struct list_head *src; |
| 422 | struct page_cgroup *pc, *tmp; |
| 423 | int nid = z->zone_pgdat->node_id; |
| 424 | int zid = zone_idx(z); |
| 425 | struct mem_cgroup_per_zone *mz; |
| 426 | int lru = LRU_FILE * !!file + !!active; |
| 427 | |
| 428 | BUG_ON(!mem_cont); |
| 429 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
| 430 | src = &mz->lists[lru]; |
| 431 | |
| 432 | spin_lock(&mz->lru_lock); |
| 433 | scan = 0; |
| 434 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { |
| 435 | if (scan >= nr_to_scan) |
| 436 | break; |
| 437 | if (unlikely(!PageCgroupUsed(pc))) |
| 438 | continue; |
| 439 | page = pc->page; |
| 440 | |
| 441 | if (unlikely(!PageLRU(page))) |
| 442 | continue; |
| 443 | |
| 444 | /* |
| 445 | * TODO: play better with lumpy reclaim, grabbing anything. |
| 446 | */ |
| 447 | if (PageUnevictable(page) || |
| 448 | (PageActive(page) && !active) || |
| 449 | (!PageActive(page) && active)) { |
| 450 | __mem_cgroup_move_lists(pc, page_lru(page)); |
| 451 | continue; |
| 452 | } |
| 453 | |
| 454 | scan++; |
| 455 | list_move(&pc->lru, &pc_list); |
| 456 | |
| 457 | if (__isolate_lru_page(page, mode, file) == 0) { |
| 458 | list_move(&page->lru, dst); |
| 459 | nr_taken++; |
| 460 | } |
| 461 | } |
| 462 | |
| 463 | list_splice(&pc_list, src); |
| 464 | spin_unlock(&mz->lru_lock); |
| 465 | |
| 466 | *scanned = scan; |
| 467 | return nr_taken; |
| 468 | } |
| 469 | |
| 470 | /* |
| 471 | * Charge the memory controller for page usage. |
| 472 | * Return |
| 473 | * 0 if the charge was successful |
| 474 | * < 0 if the cgroup is over its limit |
| 475 | */ |
| 476 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, |
| 477 | gfp_t gfp_mask, enum charge_type ctype, |
| 478 | struct mem_cgroup *memcg) |
| 479 | { |
| 480 | struct mem_cgroup *mem; |
| 481 | struct page_cgroup *pc; |
| 482 | unsigned long nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
| 483 | struct mem_cgroup_per_zone *mz; |
| 484 | unsigned long flags; |
| 485 | |
| 486 | pc = lookup_page_cgroup(page); |
| 487 | /* can happen at boot */ |
| 488 | if (unlikely(!pc)) |
| 489 | return 0; |
| 490 | prefetchw(pc); |
| 491 | /* |
| 492 | * We always charge the cgroup the mm_struct belongs to. |
| 493 | * The mm_struct's mem_cgroup changes on task migration if the |
| 494 | * thread group leader migrates. It's possible that mm is not |
| 495 | * set, if so charge the init_mm (happens for pagecache usage). |
| 496 | */ |
| 497 | |
| 498 | if (likely(!memcg)) { |
| 499 | rcu_read_lock(); |
| 500 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 501 | if (unlikely(!mem)) { |
| 502 | rcu_read_unlock(); |
| 503 | return 0; |
| 504 | } |
| 505 | /* |
| 506 | * For every charge from the cgroup, increment reference count |
| 507 | */ |
| 508 | css_get(&mem->css); |
| 509 | rcu_read_unlock(); |
| 510 | } else { |
| 511 | mem = memcg; |
| 512 | css_get(&memcg->css); |
| 513 | } |
| 514 | |
| 515 | while (unlikely(res_counter_charge(&mem->res, PAGE_SIZE))) { |
| 516 | if (!(gfp_mask & __GFP_WAIT)) |
| 517 | goto out; |
| 518 | |
| 519 | if (try_to_free_mem_cgroup_pages(mem, gfp_mask)) |
| 520 | continue; |
| 521 | |
| 522 | /* |
| 523 | * try_to_free_mem_cgroup_pages() might not give us a full |
| 524 | * picture of reclaim. Some pages are reclaimed and might be |
| 525 | * moved to swap cache or just unmapped from the cgroup. |
| 526 | * Check the limit again to see if the reclaim reduced the |
| 527 | * current usage of the cgroup before giving up |
| 528 | */ |
| 529 | if (res_counter_check_under_limit(&mem->res)) |
| 530 | continue; |
| 531 | |
| 532 | if (!nr_retries--) { |
| 533 | mem_cgroup_out_of_memory(mem, gfp_mask); |
| 534 | goto out; |
| 535 | } |
| 536 | } |
| 537 | |
| 538 | |
| 539 | lock_page_cgroup(pc); |
| 540 | if (unlikely(PageCgroupUsed(pc))) { |
| 541 | unlock_page_cgroup(pc); |
| 542 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
| 543 | css_put(&mem->css); |
| 544 | |
| 545 | goto done; |
| 546 | } |
| 547 | pc->mem_cgroup = mem; |
| 548 | /* |
| 549 | * If a page is accounted as a page cache, insert to inactive list. |
| 550 | * If anon, insert to active list. |
| 551 | */ |
| 552 | pc->flags = pcg_default_flags[ctype]; |
| 553 | |
| 554 | mz = page_cgroup_zoneinfo(pc); |
| 555 | |
| 556 | spin_lock_irqsave(&mz->lru_lock, flags); |
| 557 | __mem_cgroup_add_list(mz, pc); |
| 558 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
| 559 | unlock_page_cgroup(pc); |
| 560 | |
| 561 | done: |
| 562 | return 0; |
| 563 | out: |
| 564 | css_put(&mem->css); |
| 565 | return -ENOMEM; |
| 566 | } |
| 567 | |
| 568 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
| 569 | { |
| 570 | if (mem_cgroup_subsys.disabled) |
| 571 | return 0; |
| 572 | if (PageCompound(page)) |
| 573 | return 0; |
| 574 | /* |
| 575 | * If already mapped, we don't have to account. |
| 576 | * If page cache, page->mapping has address_space. |
| 577 | * But page->mapping may have out-of-use anon_vma pointer, |
| 578 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping |
| 579 | * is NULL. |
| 580 | */ |
| 581 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) |
| 582 | return 0; |
| 583 | if (unlikely(!mm)) |
| 584 | mm = &init_mm; |
| 585 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
| 586 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
| 587 | } |
| 588 | |
| 589 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
| 590 | gfp_t gfp_mask) |
| 591 | { |
| 592 | if (mem_cgroup_subsys.disabled) |
| 593 | return 0; |
| 594 | if (PageCompound(page)) |
| 595 | return 0; |
| 596 | /* |
| 597 | * Corner case handling. This is called from add_to_page_cache() |
| 598 | * in usual. But some FS (shmem) precharges this page before calling it |
| 599 | * and call add_to_page_cache() with GFP_NOWAIT. |
| 600 | * |
| 601 | * For GFP_NOWAIT case, the page may be pre-charged before calling |
| 602 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call |
| 603 | * charge twice. (It works but has to pay a bit larger cost.) |
| 604 | */ |
| 605 | if (!(gfp_mask & __GFP_WAIT)) { |
| 606 | struct page_cgroup *pc; |
| 607 | |
| 608 | |
| 609 | pc = lookup_page_cgroup(page); |
| 610 | if (!pc) |
| 611 | return 0; |
| 612 | lock_page_cgroup(pc); |
| 613 | if (PageCgroupUsed(pc)) { |
| 614 | unlock_page_cgroup(pc); |
| 615 | return 0; |
| 616 | } |
| 617 | unlock_page_cgroup(pc); |
| 618 | } |
| 619 | |
| 620 | if (unlikely(!mm)) |
| 621 | mm = &init_mm; |
| 622 | |
| 623 | if (page_is_file_cache(page)) |
| 624 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
| 625 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
| 626 | else |
| 627 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
| 628 | MEM_CGROUP_CHARGE_TYPE_SHMEM, NULL); |
| 629 | } |
| 630 | |
| 631 | /* |
| 632 | * uncharge if !page_mapped(page) |
| 633 | */ |
| 634 | static void |
| 635 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
| 636 | { |
| 637 | struct page_cgroup *pc; |
| 638 | struct mem_cgroup *mem; |
| 639 | struct mem_cgroup_per_zone *mz; |
| 640 | unsigned long flags; |
| 641 | |
| 642 | if (mem_cgroup_subsys.disabled) |
| 643 | return; |
| 644 | |
| 645 | /* |
| 646 | * Check if our page_cgroup is valid |
| 647 | */ |
| 648 | pc = lookup_page_cgroup(page); |
| 649 | if (unlikely(!pc || !PageCgroupUsed(pc))) |
| 650 | return; |
| 651 | |
| 652 | lock_page_cgroup(pc); |
| 653 | if ((ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED && page_mapped(page)) |
| 654 | || !PageCgroupUsed(pc)) { |
| 655 | /* This happens at race in zap_pte_range() and do_swap_page()*/ |
| 656 | unlock_page_cgroup(pc); |
| 657 | return; |
| 658 | } |
| 659 | ClearPageCgroupUsed(pc); |
| 660 | mem = pc->mem_cgroup; |
| 661 | |
| 662 | mz = page_cgroup_zoneinfo(pc); |
| 663 | spin_lock_irqsave(&mz->lru_lock, flags); |
| 664 | __mem_cgroup_remove_list(mz, pc); |
| 665 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
| 666 | unlock_page_cgroup(pc); |
| 667 | |
| 668 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
| 669 | css_put(&mem->css); |
| 670 | |
| 671 | return; |
| 672 | } |
| 673 | |
| 674 | void mem_cgroup_uncharge_page(struct page *page) |
| 675 | { |
| 676 | /* early check. */ |
| 677 | if (page_mapped(page)) |
| 678 | return; |
| 679 | if (page->mapping && !PageAnon(page)) |
| 680 | return; |
| 681 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
| 682 | } |
| 683 | |
| 684 | void mem_cgroup_uncharge_cache_page(struct page *page) |
| 685 | { |
| 686 | VM_BUG_ON(page_mapped(page)); |
| 687 | VM_BUG_ON(page->mapping); |
| 688 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
| 689 | } |
| 690 | |
| 691 | /* |
| 692 | * Before starting migration, account against new page. |
| 693 | */ |
| 694 | int mem_cgroup_prepare_migration(struct page *page, struct page *newpage) |
| 695 | { |
| 696 | struct page_cgroup *pc; |
| 697 | struct mem_cgroup *mem = NULL; |
| 698 | enum charge_type ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; |
| 699 | int ret = 0; |
| 700 | |
| 701 | if (mem_cgroup_subsys.disabled) |
| 702 | return 0; |
| 703 | |
| 704 | pc = lookup_page_cgroup(page); |
| 705 | lock_page_cgroup(pc); |
| 706 | if (PageCgroupUsed(pc)) { |
| 707 | mem = pc->mem_cgroup; |
| 708 | css_get(&mem->css); |
| 709 | if (PageCgroupCache(pc)) { |
| 710 | if (page_is_file_cache(page)) |
| 711 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
| 712 | else |
| 713 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; |
| 714 | } |
| 715 | } |
| 716 | unlock_page_cgroup(pc); |
| 717 | if (mem) { |
| 718 | ret = mem_cgroup_charge_common(newpage, NULL, GFP_KERNEL, |
| 719 | ctype, mem); |
| 720 | css_put(&mem->css); |
| 721 | } |
| 722 | return ret; |
| 723 | } |
| 724 | |
| 725 | /* remove redundant charge if migration failed*/ |
| 726 | void mem_cgroup_end_migration(struct page *newpage) |
| 727 | { |
| 728 | /* |
| 729 | * At success, page->mapping is not NULL. |
| 730 | * special rollback care is necessary when |
| 731 | * 1. at migration failure. (newpage->mapping is cleared in this case) |
| 732 | * 2. the newpage was moved but not remapped again because the task |
| 733 | * exits and the newpage is obsolete. In this case, the new page |
| 734 | * may be a swapcache. So, we just call mem_cgroup_uncharge_page() |
| 735 | * always for avoiding mess. The page_cgroup will be removed if |
| 736 | * unnecessary. File cache pages is still on radix-tree. Don't |
| 737 | * care it. |
| 738 | */ |
| 739 | if (!newpage->mapping) |
| 740 | __mem_cgroup_uncharge_common(newpage, |
| 741 | MEM_CGROUP_CHARGE_TYPE_FORCE); |
| 742 | else if (PageAnon(newpage)) |
| 743 | mem_cgroup_uncharge_page(newpage); |
| 744 | } |
| 745 | |
| 746 | /* |
| 747 | * A call to try to shrink memory usage under specified resource controller. |
| 748 | * This is typically used for page reclaiming for shmem for reducing side |
| 749 | * effect of page allocation from shmem, which is used by some mem_cgroup. |
| 750 | */ |
| 751 | int mem_cgroup_shrink_usage(struct mm_struct *mm, gfp_t gfp_mask) |
| 752 | { |
| 753 | struct mem_cgroup *mem; |
| 754 | int progress = 0; |
| 755 | int retry = MEM_CGROUP_RECLAIM_RETRIES; |
| 756 | |
| 757 | if (mem_cgroup_subsys.disabled) |
| 758 | return 0; |
| 759 | if (!mm) |
| 760 | return 0; |
| 761 | |
| 762 | rcu_read_lock(); |
| 763 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 764 | if (unlikely(!mem)) { |
| 765 | rcu_read_unlock(); |
| 766 | return 0; |
| 767 | } |
| 768 | css_get(&mem->css); |
| 769 | rcu_read_unlock(); |
| 770 | |
| 771 | do { |
| 772 | progress = try_to_free_mem_cgroup_pages(mem, gfp_mask); |
| 773 | progress += res_counter_check_under_limit(&mem->res); |
| 774 | } while (!progress && --retry); |
| 775 | |
| 776 | css_put(&mem->css); |
| 777 | if (!retry) |
| 778 | return -ENOMEM; |
| 779 | return 0; |
| 780 | } |
| 781 | |
| 782 | int mem_cgroup_resize_limit(struct mem_cgroup *memcg, unsigned long long val) |
| 783 | { |
| 784 | |
| 785 | int retry_count = MEM_CGROUP_RECLAIM_RETRIES; |
| 786 | int progress; |
| 787 | int ret = 0; |
| 788 | |
| 789 | while (res_counter_set_limit(&memcg->res, val)) { |
| 790 | if (signal_pending(current)) { |
| 791 | ret = -EINTR; |
| 792 | break; |
| 793 | } |
| 794 | if (!retry_count) { |
| 795 | ret = -EBUSY; |
| 796 | break; |
| 797 | } |
| 798 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL); |
| 799 | if (!progress) |
| 800 | retry_count--; |
| 801 | } |
| 802 | return ret; |
| 803 | } |
| 804 | |
| 805 | |
| 806 | /* |
| 807 | * This routine traverse page_cgroup in given list and drop them all. |
| 808 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
| 809 | */ |
| 810 | #define FORCE_UNCHARGE_BATCH (128) |
| 811 | static void mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
| 812 | struct mem_cgroup_per_zone *mz, |
| 813 | enum lru_list lru) |
| 814 | { |
| 815 | struct page_cgroup *pc; |
| 816 | struct page *page; |
| 817 | int count = FORCE_UNCHARGE_BATCH; |
| 818 | unsigned long flags; |
| 819 | struct list_head *list; |
| 820 | |
| 821 | list = &mz->lists[lru]; |
| 822 | |
| 823 | spin_lock_irqsave(&mz->lru_lock, flags); |
| 824 | while (!list_empty(list)) { |
| 825 | pc = list_entry(list->prev, struct page_cgroup, lru); |
| 826 | page = pc->page; |
| 827 | if (!PageCgroupUsed(pc)) |
| 828 | break; |
| 829 | get_page(page); |
| 830 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
| 831 | /* |
| 832 | * Check if this page is on LRU. !LRU page can be found |
| 833 | * if it's under page migration. |
| 834 | */ |
| 835 | if (PageLRU(page)) { |
| 836 | __mem_cgroup_uncharge_common(page, |
| 837 | MEM_CGROUP_CHARGE_TYPE_FORCE); |
| 838 | put_page(page); |
| 839 | if (--count <= 0) { |
| 840 | count = FORCE_UNCHARGE_BATCH; |
| 841 | cond_resched(); |
| 842 | } |
| 843 | } else { |
| 844 | spin_lock_irqsave(&mz->lru_lock, flags); |
| 845 | break; |
| 846 | } |
| 847 | spin_lock_irqsave(&mz->lru_lock, flags); |
| 848 | } |
| 849 | spin_unlock_irqrestore(&mz->lru_lock, flags); |
| 850 | } |
| 851 | |
| 852 | /* |
| 853 | * make mem_cgroup's charge to be 0 if there is no task. |
| 854 | * This enables deleting this mem_cgroup. |
| 855 | */ |
| 856 | static int mem_cgroup_force_empty(struct mem_cgroup *mem) |
| 857 | { |
| 858 | int ret = -EBUSY; |
| 859 | int node, zid; |
| 860 | |
| 861 | css_get(&mem->css); |
| 862 | /* |
| 863 | * page reclaim code (kswapd etc..) will move pages between |
| 864 | * active_list <-> inactive_list while we don't take a lock. |
| 865 | * So, we have to do loop here until all lists are empty. |
| 866 | */ |
| 867 | while (mem->res.usage > 0) { |
| 868 | if (atomic_read(&mem->css.cgroup->count) > 0) |
| 869 | goto out; |
| 870 | /* This is for making all *used* pages to be on LRU. */ |
| 871 | lru_add_drain_all(); |
| 872 | for_each_node_state(node, N_POSSIBLE) |
| 873 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 874 | struct mem_cgroup_per_zone *mz; |
| 875 | enum lru_list l; |
| 876 | mz = mem_cgroup_zoneinfo(mem, node, zid); |
| 877 | for_each_lru(l) |
| 878 | mem_cgroup_force_empty_list(mem, mz, l); |
| 879 | } |
| 880 | cond_resched(); |
| 881 | } |
| 882 | ret = 0; |
| 883 | out: |
| 884 | css_put(&mem->css); |
| 885 | return ret; |
| 886 | } |
| 887 | |
| 888 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
| 889 | { |
| 890 | return res_counter_read_u64(&mem_cgroup_from_cont(cont)->res, |
| 891 | cft->private); |
| 892 | } |
| 893 | /* |
| 894 | * The user of this function is... |
| 895 | * RES_LIMIT. |
| 896 | */ |
| 897 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
| 898 | const char *buffer) |
| 899 | { |
| 900 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
| 901 | unsigned long long val; |
| 902 | int ret; |
| 903 | |
| 904 | switch (cft->private) { |
| 905 | case RES_LIMIT: |
| 906 | /* This function does all necessary parse...reuse it */ |
| 907 | ret = res_counter_memparse_write_strategy(buffer, &val); |
| 908 | if (!ret) |
| 909 | ret = mem_cgroup_resize_limit(memcg, val); |
| 910 | break; |
| 911 | default: |
| 912 | ret = -EINVAL; /* should be BUG() ? */ |
| 913 | break; |
| 914 | } |
| 915 | return ret; |
| 916 | } |
| 917 | |
| 918 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
| 919 | { |
| 920 | struct mem_cgroup *mem; |
| 921 | |
| 922 | mem = mem_cgroup_from_cont(cont); |
| 923 | switch (event) { |
| 924 | case RES_MAX_USAGE: |
| 925 | res_counter_reset_max(&mem->res); |
| 926 | break; |
| 927 | case RES_FAILCNT: |
| 928 | res_counter_reset_failcnt(&mem->res); |
| 929 | break; |
| 930 | } |
| 931 | return 0; |
| 932 | } |
| 933 | |
| 934 | static int mem_force_empty_write(struct cgroup *cont, unsigned int event) |
| 935 | { |
| 936 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont)); |
| 937 | } |
| 938 | |
| 939 | static const struct mem_cgroup_stat_desc { |
| 940 | const char *msg; |
| 941 | u64 unit; |
| 942 | } mem_cgroup_stat_desc[] = { |
| 943 | [MEM_CGROUP_STAT_CACHE] = { "cache", PAGE_SIZE, }, |
| 944 | [MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, }, |
| 945 | [MEM_CGROUP_STAT_PGPGIN_COUNT] = {"pgpgin", 1, }, |
| 946 | [MEM_CGROUP_STAT_PGPGOUT_COUNT] = {"pgpgout", 1, }, |
| 947 | }; |
| 948 | |
| 949 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
| 950 | struct cgroup_map_cb *cb) |
| 951 | { |
| 952 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
| 953 | struct mem_cgroup_stat *stat = &mem_cont->stat; |
| 954 | int i; |
| 955 | |
| 956 | for (i = 0; i < ARRAY_SIZE(stat->cpustat[0].count); i++) { |
| 957 | s64 val; |
| 958 | |
| 959 | val = mem_cgroup_read_stat(stat, i); |
| 960 | val *= mem_cgroup_stat_desc[i].unit; |
| 961 | cb->fill(cb, mem_cgroup_stat_desc[i].msg, val); |
| 962 | } |
| 963 | /* showing # of active pages */ |
| 964 | { |
| 965 | unsigned long active_anon, inactive_anon; |
| 966 | unsigned long active_file, inactive_file; |
| 967 | unsigned long unevictable; |
| 968 | |
| 969 | inactive_anon = mem_cgroup_get_all_zonestat(mem_cont, |
| 970 | LRU_INACTIVE_ANON); |
| 971 | active_anon = mem_cgroup_get_all_zonestat(mem_cont, |
| 972 | LRU_ACTIVE_ANON); |
| 973 | inactive_file = mem_cgroup_get_all_zonestat(mem_cont, |
| 974 | LRU_INACTIVE_FILE); |
| 975 | active_file = mem_cgroup_get_all_zonestat(mem_cont, |
| 976 | LRU_ACTIVE_FILE); |
| 977 | unevictable = mem_cgroup_get_all_zonestat(mem_cont, |
| 978 | LRU_UNEVICTABLE); |
| 979 | |
| 980 | cb->fill(cb, "active_anon", (active_anon) * PAGE_SIZE); |
| 981 | cb->fill(cb, "inactive_anon", (inactive_anon) * PAGE_SIZE); |
| 982 | cb->fill(cb, "active_file", (active_file) * PAGE_SIZE); |
| 983 | cb->fill(cb, "inactive_file", (inactive_file) * PAGE_SIZE); |
| 984 | cb->fill(cb, "unevictable", unevictable * PAGE_SIZE); |
| 985 | |
| 986 | } |
| 987 | return 0; |
| 988 | } |
| 989 | |
| 990 | static struct cftype mem_cgroup_files[] = { |
| 991 | { |
| 992 | .name = "usage_in_bytes", |
| 993 | .private = RES_USAGE, |
| 994 | .read_u64 = mem_cgroup_read, |
| 995 | }, |
| 996 | { |
| 997 | .name = "max_usage_in_bytes", |
| 998 | .private = RES_MAX_USAGE, |
| 999 | .trigger = mem_cgroup_reset, |
| 1000 | .read_u64 = mem_cgroup_read, |
| 1001 | }, |
| 1002 | { |
| 1003 | .name = "limit_in_bytes", |
| 1004 | .private = RES_LIMIT, |
| 1005 | .write_string = mem_cgroup_write, |
| 1006 | .read_u64 = mem_cgroup_read, |
| 1007 | }, |
| 1008 | { |
| 1009 | .name = "failcnt", |
| 1010 | .private = RES_FAILCNT, |
| 1011 | .trigger = mem_cgroup_reset, |
| 1012 | .read_u64 = mem_cgroup_read, |
| 1013 | }, |
| 1014 | { |
| 1015 | .name = "force_empty", |
| 1016 | .trigger = mem_force_empty_write, |
| 1017 | }, |
| 1018 | { |
| 1019 | .name = "stat", |
| 1020 | .read_map = mem_control_stat_show, |
| 1021 | }, |
| 1022 | }; |
| 1023 | |
| 1024 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
| 1025 | { |
| 1026 | struct mem_cgroup_per_node *pn; |
| 1027 | struct mem_cgroup_per_zone *mz; |
| 1028 | enum lru_list l; |
| 1029 | int zone, tmp = node; |
| 1030 | /* |
| 1031 | * This routine is called against possible nodes. |
| 1032 | * But it's BUG to call kmalloc() against offline node. |
| 1033 | * |
| 1034 | * TODO: this routine can waste much memory for nodes which will |
| 1035 | * never be onlined. It's better to use memory hotplug callback |
| 1036 | * function. |
| 1037 | */ |
| 1038 | if (!node_state(node, N_NORMAL_MEMORY)) |
| 1039 | tmp = -1; |
| 1040 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
| 1041 | if (!pn) |
| 1042 | return 1; |
| 1043 | |
| 1044 | mem->info.nodeinfo[node] = pn; |
| 1045 | memset(pn, 0, sizeof(*pn)); |
| 1046 | |
| 1047 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 1048 | mz = &pn->zoneinfo[zone]; |
| 1049 | spin_lock_init(&mz->lru_lock); |
| 1050 | for_each_lru(l) |
| 1051 | INIT_LIST_HEAD(&mz->lists[l]); |
| 1052 | } |
| 1053 | return 0; |
| 1054 | } |
| 1055 | |
| 1056 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
| 1057 | { |
| 1058 | kfree(mem->info.nodeinfo[node]); |
| 1059 | } |
| 1060 | |
| 1061 | static struct mem_cgroup *mem_cgroup_alloc(void) |
| 1062 | { |
| 1063 | struct mem_cgroup *mem; |
| 1064 | |
| 1065 | if (sizeof(*mem) < PAGE_SIZE) |
| 1066 | mem = kmalloc(sizeof(*mem), GFP_KERNEL); |
| 1067 | else |
| 1068 | mem = vmalloc(sizeof(*mem)); |
| 1069 | |
| 1070 | if (mem) |
| 1071 | memset(mem, 0, sizeof(*mem)); |
| 1072 | return mem; |
| 1073 | } |
| 1074 | |
| 1075 | static void mem_cgroup_free(struct mem_cgroup *mem) |
| 1076 | { |
| 1077 | if (sizeof(*mem) < PAGE_SIZE) |
| 1078 | kfree(mem); |
| 1079 | else |
| 1080 | vfree(mem); |
| 1081 | } |
| 1082 | |
| 1083 | |
| 1084 | static struct cgroup_subsys_state * |
| 1085 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
| 1086 | { |
| 1087 | struct mem_cgroup *mem; |
| 1088 | int node; |
| 1089 | |
| 1090 | if (unlikely((cont->parent) == NULL)) { |
| 1091 | mem = &init_mem_cgroup; |
| 1092 | } else { |
| 1093 | mem = mem_cgroup_alloc(); |
| 1094 | if (!mem) |
| 1095 | return ERR_PTR(-ENOMEM); |
| 1096 | } |
| 1097 | |
| 1098 | res_counter_init(&mem->res); |
| 1099 | |
| 1100 | for_each_node_state(node, N_POSSIBLE) |
| 1101 | if (alloc_mem_cgroup_per_zone_info(mem, node)) |
| 1102 | goto free_out; |
| 1103 | |
| 1104 | return &mem->css; |
| 1105 | free_out: |
| 1106 | for_each_node_state(node, N_POSSIBLE) |
| 1107 | free_mem_cgroup_per_zone_info(mem, node); |
| 1108 | if (cont->parent != NULL) |
| 1109 | mem_cgroup_free(mem); |
| 1110 | return ERR_PTR(-ENOMEM); |
| 1111 | } |
| 1112 | |
| 1113 | static void mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
| 1114 | struct cgroup *cont) |
| 1115 | { |
| 1116 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
| 1117 | mem_cgroup_force_empty(mem); |
| 1118 | } |
| 1119 | |
| 1120 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
| 1121 | struct cgroup *cont) |
| 1122 | { |
| 1123 | int node; |
| 1124 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
| 1125 | |
| 1126 | for_each_node_state(node, N_POSSIBLE) |
| 1127 | free_mem_cgroup_per_zone_info(mem, node); |
| 1128 | |
| 1129 | mem_cgroup_free(mem_cgroup_from_cont(cont)); |
| 1130 | } |
| 1131 | |
| 1132 | static int mem_cgroup_populate(struct cgroup_subsys *ss, |
| 1133 | struct cgroup *cont) |
| 1134 | { |
| 1135 | return cgroup_add_files(cont, ss, mem_cgroup_files, |
| 1136 | ARRAY_SIZE(mem_cgroup_files)); |
| 1137 | } |
| 1138 | |
| 1139 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
| 1140 | struct cgroup *cont, |
| 1141 | struct cgroup *old_cont, |
| 1142 | struct task_struct *p) |
| 1143 | { |
| 1144 | struct mm_struct *mm; |
| 1145 | struct mem_cgroup *mem, *old_mem; |
| 1146 | |
| 1147 | mm = get_task_mm(p); |
| 1148 | if (mm == NULL) |
| 1149 | return; |
| 1150 | |
| 1151 | mem = mem_cgroup_from_cont(cont); |
| 1152 | old_mem = mem_cgroup_from_cont(old_cont); |
| 1153 | |
| 1154 | /* |
| 1155 | * Only thread group leaders are allowed to migrate, the mm_struct is |
| 1156 | * in effect owned by the leader |
| 1157 | */ |
| 1158 | if (!thread_group_leader(p)) |
| 1159 | goto out; |
| 1160 | |
| 1161 | out: |
| 1162 | mmput(mm); |
| 1163 | } |
| 1164 | |
| 1165 | struct cgroup_subsys mem_cgroup_subsys = { |
| 1166 | .name = "memory", |
| 1167 | .subsys_id = mem_cgroup_subsys_id, |
| 1168 | .create = mem_cgroup_create, |
| 1169 | .pre_destroy = mem_cgroup_pre_destroy, |
| 1170 | .destroy = mem_cgroup_destroy, |
| 1171 | .populate = mem_cgroup_populate, |
| 1172 | .attach = mem_cgroup_move_task, |
| 1173 | .early_init = 0, |
| 1174 | }; |