| 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 | * Memory thresholds |
| 10 | * Copyright (C) 2009 Nokia Corporation |
| 11 | * Author: Kirill A. Shutemov |
| 12 | * |
| 13 | * Kernel Memory Controller |
| 14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. |
| 15 | * Authors: Glauber Costa and Suleiman Souhlal |
| 16 | * |
| 17 | * Native page reclaim |
| 18 | * Charge lifetime sanitation |
| 19 | * Lockless page tracking & accounting |
| 20 | * Unified hierarchy configuration model |
| 21 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner |
| 22 | * |
| 23 | * This program is free software; you can redistribute it and/or modify |
| 24 | * it under the terms of the GNU General Public License as published by |
| 25 | * the Free Software Foundation; either version 2 of the License, or |
| 26 | * (at your option) any later version. |
| 27 | * |
| 28 | * This program is distributed in the hope that it will be useful, |
| 29 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 30 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 31 | * GNU General Public License for more details. |
| 32 | */ |
| 33 | |
| 34 | #include <linux/page_counter.h> |
| 35 | #include <linux/memcontrol.h> |
| 36 | #include <linux/cgroup.h> |
| 37 | #include <linux/mm.h> |
| 38 | #include <linux/hugetlb.h> |
| 39 | #include <linux/pagemap.h> |
| 40 | #include <linux/smp.h> |
| 41 | #include <linux/page-flags.h> |
| 42 | #include <linux/backing-dev.h> |
| 43 | #include <linux/bit_spinlock.h> |
| 44 | #include <linux/rcupdate.h> |
| 45 | #include <linux/limits.h> |
| 46 | #include <linux/export.h> |
| 47 | #include <linux/mutex.h> |
| 48 | #include <linux/rbtree.h> |
| 49 | #include <linux/slab.h> |
| 50 | #include <linux/swap.h> |
| 51 | #include <linux/swapops.h> |
| 52 | #include <linux/spinlock.h> |
| 53 | #include <linux/eventfd.h> |
| 54 | #include <linux/poll.h> |
| 55 | #include <linux/sort.h> |
| 56 | #include <linux/fs.h> |
| 57 | #include <linux/seq_file.h> |
| 58 | #include <linux/vmpressure.h> |
| 59 | #include <linux/mm_inline.h> |
| 60 | #include <linux/swap_cgroup.h> |
| 61 | #include <linux/cpu.h> |
| 62 | #include <linux/oom.h> |
| 63 | #include <linux/lockdep.h> |
| 64 | #include <linux/file.h> |
| 65 | #include <linux/tracehook.h> |
| 66 | #include "internal.h" |
| 67 | #include <net/sock.h> |
| 68 | #include <net/ip.h> |
| 69 | #include <net/tcp_memcontrol.h> |
| 70 | #include "slab.h" |
| 71 | |
| 72 | #include <asm/uaccess.h> |
| 73 | |
| 74 | #include <trace/events/vmscan.h> |
| 75 | |
| 76 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
| 77 | EXPORT_SYMBOL(memory_cgrp_subsys); |
| 78 | |
| 79 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
| 80 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
| 81 | struct cgroup_subsys_state *mem_cgroup_root_css __read_mostly; |
| 82 | |
| 83 | /* Whether the swap controller is active */ |
| 84 | #ifdef CONFIG_MEMCG_SWAP |
| 85 | int do_swap_account __read_mostly; |
| 86 | #else |
| 87 | #define do_swap_account 0 |
| 88 | #endif |
| 89 | |
| 90 | static const char * const mem_cgroup_stat_names[] = { |
| 91 | "cache", |
| 92 | "rss", |
| 93 | "rss_huge", |
| 94 | "mapped_file", |
| 95 | "dirty", |
| 96 | "writeback", |
| 97 | "swap", |
| 98 | }; |
| 99 | |
| 100 | static const char * const mem_cgroup_events_names[] = { |
| 101 | "pgpgin", |
| 102 | "pgpgout", |
| 103 | "pgfault", |
| 104 | "pgmajfault", |
| 105 | }; |
| 106 | |
| 107 | static const char * const mem_cgroup_lru_names[] = { |
| 108 | "inactive_anon", |
| 109 | "active_anon", |
| 110 | "inactive_file", |
| 111 | "active_file", |
| 112 | "unevictable", |
| 113 | }; |
| 114 | |
| 115 | #define THRESHOLDS_EVENTS_TARGET 128 |
| 116 | #define SOFTLIMIT_EVENTS_TARGET 1024 |
| 117 | #define NUMAINFO_EVENTS_TARGET 1024 |
| 118 | |
| 119 | /* |
| 120 | * Cgroups above their limits are maintained in a RB-Tree, independent of |
| 121 | * their hierarchy representation |
| 122 | */ |
| 123 | |
| 124 | struct mem_cgroup_tree_per_zone { |
| 125 | struct rb_root rb_root; |
| 126 | spinlock_t lock; |
| 127 | }; |
| 128 | |
| 129 | struct mem_cgroup_tree_per_node { |
| 130 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; |
| 131 | }; |
| 132 | |
| 133 | struct mem_cgroup_tree { |
| 134 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; |
| 135 | }; |
| 136 | |
| 137 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; |
| 138 | |
| 139 | /* for OOM */ |
| 140 | struct mem_cgroup_eventfd_list { |
| 141 | struct list_head list; |
| 142 | struct eventfd_ctx *eventfd; |
| 143 | }; |
| 144 | |
| 145 | /* |
| 146 | * cgroup_event represents events which userspace want to receive. |
| 147 | */ |
| 148 | struct mem_cgroup_event { |
| 149 | /* |
| 150 | * memcg which the event belongs to. |
| 151 | */ |
| 152 | struct mem_cgroup *memcg; |
| 153 | /* |
| 154 | * eventfd to signal userspace about the event. |
| 155 | */ |
| 156 | struct eventfd_ctx *eventfd; |
| 157 | /* |
| 158 | * Each of these stored in a list by the cgroup. |
| 159 | */ |
| 160 | struct list_head list; |
| 161 | /* |
| 162 | * register_event() callback will be used to add new userspace |
| 163 | * waiter for changes related to this event. Use eventfd_signal() |
| 164 | * on eventfd to send notification to userspace. |
| 165 | */ |
| 166 | int (*register_event)(struct mem_cgroup *memcg, |
| 167 | struct eventfd_ctx *eventfd, const char *args); |
| 168 | /* |
| 169 | * unregister_event() callback will be called when userspace closes |
| 170 | * the eventfd or on cgroup removing. This callback must be set, |
| 171 | * if you want provide notification functionality. |
| 172 | */ |
| 173 | void (*unregister_event)(struct mem_cgroup *memcg, |
| 174 | struct eventfd_ctx *eventfd); |
| 175 | /* |
| 176 | * All fields below needed to unregister event when |
| 177 | * userspace closes eventfd. |
| 178 | */ |
| 179 | poll_table pt; |
| 180 | wait_queue_head_t *wqh; |
| 181 | wait_queue_t wait; |
| 182 | struct work_struct remove; |
| 183 | }; |
| 184 | |
| 185 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
| 186 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); |
| 187 | |
| 188 | /* Stuffs for move charges at task migration. */ |
| 189 | /* |
| 190 | * Types of charges to be moved. |
| 191 | */ |
| 192 | #define MOVE_ANON 0x1U |
| 193 | #define MOVE_FILE 0x2U |
| 194 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) |
| 195 | |
| 196 | /* "mc" and its members are protected by cgroup_mutex */ |
| 197 | static struct move_charge_struct { |
| 198 | spinlock_t lock; /* for from, to */ |
| 199 | struct mm_struct *mm; |
| 200 | struct mem_cgroup *from; |
| 201 | struct mem_cgroup *to; |
| 202 | unsigned long flags; |
| 203 | unsigned long precharge; |
| 204 | unsigned long moved_charge; |
| 205 | unsigned long moved_swap; |
| 206 | struct task_struct *moving_task; /* a task moving charges */ |
| 207 | wait_queue_head_t waitq; /* a waitq for other context */ |
| 208 | } mc = { |
| 209 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
| 210 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
| 211 | }; |
| 212 | |
| 213 | /* |
| 214 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft |
| 215 | * limit reclaim to prevent infinite loops, if they ever occur. |
| 216 | */ |
| 217 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
| 218 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
| 219 | |
| 220 | enum charge_type { |
| 221 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, |
| 222 | MEM_CGROUP_CHARGE_TYPE_ANON, |
| 223 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
| 224 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
| 225 | NR_CHARGE_TYPE, |
| 226 | }; |
| 227 | |
| 228 | /* for encoding cft->private value on file */ |
| 229 | enum res_type { |
| 230 | _MEM, |
| 231 | _MEMSWAP, |
| 232 | _OOM_TYPE, |
| 233 | _KMEM, |
| 234 | }; |
| 235 | |
| 236 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
| 237 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) |
| 238 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
| 239 | /* Used for OOM nofiier */ |
| 240 | #define OOM_CONTROL (0) |
| 241 | |
| 242 | /* |
| 243 | * The memcg_create_mutex will be held whenever a new cgroup is created. |
| 244 | * As a consequence, any change that needs to protect against new child cgroups |
| 245 | * appearing has to hold it as well. |
| 246 | */ |
| 247 | static DEFINE_MUTEX(memcg_create_mutex); |
| 248 | |
| 249 | /* Some nice accessors for the vmpressure. */ |
| 250 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) |
| 251 | { |
| 252 | if (!memcg) |
| 253 | memcg = root_mem_cgroup; |
| 254 | return &memcg->vmpressure; |
| 255 | } |
| 256 | |
| 257 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) |
| 258 | { |
| 259 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; |
| 260 | } |
| 261 | |
| 262 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
| 263 | { |
| 264 | return (memcg == root_mem_cgroup); |
| 265 | } |
| 266 | |
| 267 | /* |
| 268 | * We restrict the id in the range of [1, 65535], so it can fit into |
| 269 | * an unsigned short. |
| 270 | */ |
| 271 | #define MEM_CGROUP_ID_MAX USHRT_MAX |
| 272 | |
| 273 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
| 274 | { |
| 275 | return memcg->id.id; |
| 276 | } |
| 277 | |
| 278 | /* Writing them here to avoid exposing memcg's inner layout */ |
| 279 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
| 280 | |
| 281 | void sock_update_memcg(struct sock *sk) |
| 282 | { |
| 283 | if (mem_cgroup_sockets_enabled) { |
| 284 | struct mem_cgroup *memcg; |
| 285 | struct cg_proto *cg_proto; |
| 286 | |
| 287 | BUG_ON(!sk->sk_prot->proto_cgroup); |
| 288 | |
| 289 | /* Socket cloning can throw us here with sk_cgrp already |
| 290 | * filled. It won't however, necessarily happen from |
| 291 | * process context. So the test for root memcg given |
| 292 | * the current task's memcg won't help us in this case. |
| 293 | * |
| 294 | * Respecting the original socket's memcg is a better |
| 295 | * decision in this case. |
| 296 | */ |
| 297 | if (sk->sk_cgrp) { |
| 298 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); |
| 299 | css_get(&sk->sk_cgrp->memcg->css); |
| 300 | return; |
| 301 | } |
| 302 | |
| 303 | rcu_read_lock(); |
| 304 | memcg = mem_cgroup_from_task(current); |
| 305 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
| 306 | if (cg_proto && test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags) && |
| 307 | css_tryget_online(&memcg->css)) { |
| 308 | sk->sk_cgrp = cg_proto; |
| 309 | } |
| 310 | rcu_read_unlock(); |
| 311 | } |
| 312 | } |
| 313 | EXPORT_SYMBOL(sock_update_memcg); |
| 314 | |
| 315 | void sock_release_memcg(struct sock *sk) |
| 316 | { |
| 317 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
| 318 | struct mem_cgroup *memcg; |
| 319 | WARN_ON(!sk->sk_cgrp->memcg); |
| 320 | memcg = sk->sk_cgrp->memcg; |
| 321 | css_put(&sk->sk_cgrp->memcg->css); |
| 322 | } |
| 323 | } |
| 324 | |
| 325 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) |
| 326 | { |
| 327 | if (!memcg || mem_cgroup_is_root(memcg)) |
| 328 | return NULL; |
| 329 | |
| 330 | return &memcg->tcp_mem; |
| 331 | } |
| 332 | EXPORT_SYMBOL(tcp_proto_cgroup); |
| 333 | |
| 334 | #endif |
| 335 | |
| 336 | #ifdef CONFIG_MEMCG_KMEM |
| 337 | /* |
| 338 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
| 339 | * The main reason for not using cgroup id for this: |
| 340 | * this works better in sparse environments, where we have a lot of memcgs, |
| 341 | * but only a few kmem-limited. Or also, if we have, for instance, 200 |
| 342 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a |
| 343 | * 200 entry array for that. |
| 344 | * |
| 345 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
| 346 | * will double each time we have to increase it. |
| 347 | */ |
| 348 | static DEFINE_IDA(memcg_cache_ida); |
| 349 | int memcg_nr_cache_ids; |
| 350 | |
| 351 | /* Protects memcg_nr_cache_ids */ |
| 352 | static DECLARE_RWSEM(memcg_cache_ids_sem); |
| 353 | |
| 354 | void memcg_get_cache_ids(void) |
| 355 | { |
| 356 | down_read(&memcg_cache_ids_sem); |
| 357 | } |
| 358 | |
| 359 | void memcg_put_cache_ids(void) |
| 360 | { |
| 361 | up_read(&memcg_cache_ids_sem); |
| 362 | } |
| 363 | |
| 364 | /* |
| 365 | * MIN_SIZE is different than 1, because we would like to avoid going through |
| 366 | * the alloc/free process all the time. In a small machine, 4 kmem-limited |
| 367 | * cgroups is a reasonable guess. In the future, it could be a parameter or |
| 368 | * tunable, but that is strictly not necessary. |
| 369 | * |
| 370 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
| 371 | * this constant directly from cgroup, but it is understandable that this is |
| 372 | * better kept as an internal representation in cgroup.c. In any case, the |
| 373 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
| 374 | * increase ours as well if it increases. |
| 375 | */ |
| 376 | #define MEMCG_CACHES_MIN_SIZE 4 |
| 377 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
| 378 | |
| 379 | /* |
| 380 | * A lot of the calls to the cache allocation functions are expected to be |
| 381 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are |
| 382 | * conditional to this static branch, we'll have to allow modules that does |
| 383 | * kmem_cache_alloc and the such to see this symbol as well |
| 384 | */ |
| 385 | struct static_key memcg_kmem_enabled_key; |
| 386 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
| 387 | |
| 388 | #endif /* CONFIG_MEMCG_KMEM */ |
| 389 | |
| 390 | static struct mem_cgroup_per_zone * |
| 391 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
| 392 | { |
| 393 | int nid = zone_to_nid(zone); |
| 394 | int zid = zone_idx(zone); |
| 395 | |
| 396 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
| 397 | } |
| 398 | |
| 399 | /** |
| 400 | * mem_cgroup_css_from_page - css of the memcg associated with a page |
| 401 | * @page: page of interest |
| 402 | * |
| 403 | * If memcg is bound to the default hierarchy, css of the memcg associated |
| 404 | * with @page is returned. The returned css remains associated with @page |
| 405 | * until it is released. |
| 406 | * |
| 407 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup |
| 408 | * is returned. |
| 409 | * |
| 410 | * XXX: The above description of behavior on the default hierarchy isn't |
| 411 | * strictly true yet as replace_page_cache_page() can modify the |
| 412 | * association before @page is released even on the default hierarchy; |
| 413 | * however, the current and planned usages don't mix the the two functions |
| 414 | * and replace_page_cache_page() will soon be updated to make the invariant |
| 415 | * actually true. |
| 416 | */ |
| 417 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) |
| 418 | { |
| 419 | struct mem_cgroup *memcg; |
| 420 | |
| 421 | rcu_read_lock(); |
| 422 | |
| 423 | memcg = page->mem_cgroup; |
| 424 | |
| 425 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
| 426 | memcg = root_mem_cgroup; |
| 427 | |
| 428 | rcu_read_unlock(); |
| 429 | return &memcg->css; |
| 430 | } |
| 431 | |
| 432 | /** |
| 433 | * page_cgroup_ino - return inode number of the memcg a page is charged to |
| 434 | * @page: the page |
| 435 | * |
| 436 | * Look up the closest online ancestor of the memory cgroup @page is charged to |
| 437 | * and return its inode number or 0 if @page is not charged to any cgroup. It |
| 438 | * is safe to call this function without holding a reference to @page. |
| 439 | * |
| 440 | * Note, this function is inherently racy, because there is nothing to prevent |
| 441 | * the cgroup inode from getting torn down and potentially reallocated a moment |
| 442 | * after page_cgroup_ino() returns, so it only should be used by callers that |
| 443 | * do not care (such as procfs interfaces). |
| 444 | */ |
| 445 | ino_t page_cgroup_ino(struct page *page) |
| 446 | { |
| 447 | struct mem_cgroup *memcg; |
| 448 | unsigned long ino = 0; |
| 449 | |
| 450 | rcu_read_lock(); |
| 451 | memcg = READ_ONCE(page->mem_cgroup); |
| 452 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
| 453 | memcg = parent_mem_cgroup(memcg); |
| 454 | if (memcg) |
| 455 | ino = cgroup_ino(memcg->css.cgroup); |
| 456 | rcu_read_unlock(); |
| 457 | return ino; |
| 458 | } |
| 459 | |
| 460 | static struct mem_cgroup_per_zone * |
| 461 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
| 462 | { |
| 463 | int nid = page_to_nid(page); |
| 464 | int zid = page_zonenum(page); |
| 465 | |
| 466 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
| 467 | } |
| 468 | |
| 469 | static struct mem_cgroup_tree_per_zone * |
| 470 | soft_limit_tree_node_zone(int nid, int zid) |
| 471 | { |
| 472 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 473 | } |
| 474 | |
| 475 | static struct mem_cgroup_tree_per_zone * |
| 476 | soft_limit_tree_from_page(struct page *page) |
| 477 | { |
| 478 | int nid = page_to_nid(page); |
| 479 | int zid = page_zonenum(page); |
| 480 | |
| 481 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 482 | } |
| 483 | |
| 484 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
| 485 | struct mem_cgroup_tree_per_zone *mctz, |
| 486 | unsigned long new_usage_in_excess) |
| 487 | { |
| 488 | struct rb_node **p = &mctz->rb_root.rb_node; |
| 489 | struct rb_node *parent = NULL; |
| 490 | struct mem_cgroup_per_zone *mz_node; |
| 491 | |
| 492 | if (mz->on_tree) |
| 493 | return; |
| 494 | |
| 495 | mz->usage_in_excess = new_usage_in_excess; |
| 496 | if (!mz->usage_in_excess) |
| 497 | return; |
| 498 | while (*p) { |
| 499 | parent = *p; |
| 500 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, |
| 501 | tree_node); |
| 502 | if (mz->usage_in_excess < mz_node->usage_in_excess) |
| 503 | p = &(*p)->rb_left; |
| 504 | /* |
| 505 | * We can't avoid mem cgroups that are over their soft |
| 506 | * limit by the same amount |
| 507 | */ |
| 508 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) |
| 509 | p = &(*p)->rb_right; |
| 510 | } |
| 511 | rb_link_node(&mz->tree_node, parent, p); |
| 512 | rb_insert_color(&mz->tree_node, &mctz->rb_root); |
| 513 | mz->on_tree = true; |
| 514 | } |
| 515 | |
| 516 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
| 517 | struct mem_cgroup_tree_per_zone *mctz) |
| 518 | { |
| 519 | if (!mz->on_tree) |
| 520 | return; |
| 521 | rb_erase(&mz->tree_node, &mctz->rb_root); |
| 522 | mz->on_tree = false; |
| 523 | } |
| 524 | |
| 525 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
| 526 | struct mem_cgroup_tree_per_zone *mctz) |
| 527 | { |
| 528 | unsigned long flags; |
| 529 | |
| 530 | spin_lock_irqsave(&mctz->lock, flags); |
| 531 | __mem_cgroup_remove_exceeded(mz, mctz); |
| 532 | spin_unlock_irqrestore(&mctz->lock, flags); |
| 533 | } |
| 534 | |
| 535 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
| 536 | { |
| 537 | unsigned long nr_pages = page_counter_read(&memcg->memory); |
| 538 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
| 539 | unsigned long excess = 0; |
| 540 | |
| 541 | if (nr_pages > soft_limit) |
| 542 | excess = nr_pages - soft_limit; |
| 543 | |
| 544 | return excess; |
| 545 | } |
| 546 | |
| 547 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
| 548 | { |
| 549 | unsigned long excess; |
| 550 | struct mem_cgroup_per_zone *mz; |
| 551 | struct mem_cgroup_tree_per_zone *mctz; |
| 552 | |
| 553 | mctz = soft_limit_tree_from_page(page); |
| 554 | /* |
| 555 | * Necessary to update all ancestors when hierarchy is used. |
| 556 | * because their event counter is not touched. |
| 557 | */ |
| 558 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
| 559 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
| 560 | excess = soft_limit_excess(memcg); |
| 561 | /* |
| 562 | * We have to update the tree if mz is on RB-tree or |
| 563 | * mem is over its softlimit. |
| 564 | */ |
| 565 | if (excess || mz->on_tree) { |
| 566 | unsigned long flags; |
| 567 | |
| 568 | spin_lock_irqsave(&mctz->lock, flags); |
| 569 | /* if on-tree, remove it */ |
| 570 | if (mz->on_tree) |
| 571 | __mem_cgroup_remove_exceeded(mz, mctz); |
| 572 | /* |
| 573 | * Insert again. mz->usage_in_excess will be updated. |
| 574 | * If excess is 0, no tree ops. |
| 575 | */ |
| 576 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
| 577 | spin_unlock_irqrestore(&mctz->lock, flags); |
| 578 | } |
| 579 | } |
| 580 | } |
| 581 | |
| 582 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
| 583 | { |
| 584 | struct mem_cgroup_tree_per_zone *mctz; |
| 585 | struct mem_cgroup_per_zone *mz; |
| 586 | int nid, zid; |
| 587 | |
| 588 | for_each_node(nid) { |
| 589 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 590 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
| 591 | mctz = soft_limit_tree_node_zone(nid, zid); |
| 592 | mem_cgroup_remove_exceeded(mz, mctz); |
| 593 | } |
| 594 | } |
| 595 | } |
| 596 | |
| 597 | static struct mem_cgroup_per_zone * |
| 598 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 599 | { |
| 600 | struct rb_node *rightmost = NULL; |
| 601 | struct mem_cgroup_per_zone *mz; |
| 602 | |
| 603 | retry: |
| 604 | mz = NULL; |
| 605 | rightmost = rb_last(&mctz->rb_root); |
| 606 | if (!rightmost) |
| 607 | goto done; /* Nothing to reclaim from */ |
| 608 | |
| 609 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); |
| 610 | /* |
| 611 | * Remove the node now but someone else can add it back, |
| 612 | * we will to add it back at the end of reclaim to its correct |
| 613 | * position in the tree. |
| 614 | */ |
| 615 | __mem_cgroup_remove_exceeded(mz, mctz); |
| 616 | if (!soft_limit_excess(mz->memcg) || |
| 617 | !css_tryget_online(&mz->memcg->css)) |
| 618 | goto retry; |
| 619 | done: |
| 620 | return mz; |
| 621 | } |
| 622 | |
| 623 | static struct mem_cgroup_per_zone * |
| 624 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 625 | { |
| 626 | struct mem_cgroup_per_zone *mz; |
| 627 | |
| 628 | spin_lock_irq(&mctz->lock); |
| 629 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
| 630 | spin_unlock_irq(&mctz->lock); |
| 631 | return mz; |
| 632 | } |
| 633 | |
| 634 | /* |
| 635 | * Return page count for single (non recursive) @memcg. |
| 636 | * |
| 637 | * Implementation Note: reading percpu statistics for memcg. |
| 638 | * |
| 639 | * Both of vmstat[] and percpu_counter has threshold and do periodic |
| 640 | * synchronization to implement "quick" read. There are trade-off between |
| 641 | * reading cost and precision of value. Then, we may have a chance to implement |
| 642 | * a periodic synchronization of counter in memcg's counter. |
| 643 | * |
| 644 | * But this _read() function is used for user interface now. The user accounts |
| 645 | * memory usage by memory cgroup and he _always_ requires exact value because |
| 646 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always |
| 647 | * have to visit all online cpus and make sum. So, for now, unnecessary |
| 648 | * synchronization is not implemented. (just implemented for cpu hotplug) |
| 649 | * |
| 650 | * If there are kernel internal actions which can make use of some not-exact |
| 651 | * value, and reading all cpu value can be performance bottleneck in some |
| 652 | * common workload, threshold and synchronization as vmstat[] should be |
| 653 | * implemented. |
| 654 | */ |
| 655 | static unsigned long |
| 656 | mem_cgroup_read_stat(struct mem_cgroup *memcg, enum mem_cgroup_stat_index idx) |
| 657 | { |
| 658 | long val = 0; |
| 659 | int cpu; |
| 660 | |
| 661 | /* Per-cpu values can be negative, use a signed accumulator */ |
| 662 | for_each_possible_cpu(cpu) |
| 663 | val += per_cpu(memcg->stat->count[idx], cpu); |
| 664 | /* |
| 665 | * Summing races with updates, so val may be negative. Avoid exposing |
| 666 | * transient negative values. |
| 667 | */ |
| 668 | if (val < 0) |
| 669 | val = 0; |
| 670 | return val; |
| 671 | } |
| 672 | |
| 673 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
| 674 | enum mem_cgroup_events_index idx) |
| 675 | { |
| 676 | unsigned long val = 0; |
| 677 | int cpu; |
| 678 | |
| 679 | for_each_possible_cpu(cpu) |
| 680 | val += per_cpu(memcg->stat->events[idx], cpu); |
| 681 | return val; |
| 682 | } |
| 683 | |
| 684 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
| 685 | struct page *page, |
| 686 | int nr_pages) |
| 687 | { |
| 688 | /* |
| 689 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is |
| 690 | * counted as CACHE even if it's on ANON LRU. |
| 691 | */ |
| 692 | if (PageAnon(page)) |
| 693 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
| 694 | nr_pages); |
| 695 | else |
| 696 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
| 697 | nr_pages); |
| 698 | |
| 699 | if (PageTransHuge(page)) |
| 700 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
| 701 | nr_pages); |
| 702 | |
| 703 | /* pagein of a big page is an event. So, ignore page size */ |
| 704 | if (nr_pages > 0) |
| 705 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
| 706 | else { |
| 707 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
| 708 | nr_pages = -nr_pages; /* for event */ |
| 709 | } |
| 710 | |
| 711 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
| 712 | } |
| 713 | |
| 714 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
| 715 | int nid, |
| 716 | unsigned int lru_mask) |
| 717 | { |
| 718 | unsigned long nr = 0; |
| 719 | int zid; |
| 720 | |
| 721 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
| 722 | |
| 723 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 724 | struct mem_cgroup_per_zone *mz; |
| 725 | enum lru_list lru; |
| 726 | |
| 727 | for_each_lru(lru) { |
| 728 | if (!(BIT(lru) & lru_mask)) |
| 729 | continue; |
| 730 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
| 731 | nr += mz->lru_size[lru]; |
| 732 | } |
| 733 | } |
| 734 | return nr; |
| 735 | } |
| 736 | |
| 737 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
| 738 | unsigned int lru_mask) |
| 739 | { |
| 740 | unsigned long nr = 0; |
| 741 | int nid; |
| 742 | |
| 743 | for_each_node_state(nid, N_MEMORY) |
| 744 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
| 745 | return nr; |
| 746 | } |
| 747 | |
| 748 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
| 749 | enum mem_cgroup_events_target target) |
| 750 | { |
| 751 | unsigned long val, next; |
| 752 | |
| 753 | val = __this_cpu_read(memcg->stat->nr_page_events); |
| 754 | next = __this_cpu_read(memcg->stat->targets[target]); |
| 755 | /* from time_after() in jiffies.h */ |
| 756 | if ((long)next - (long)val < 0) { |
| 757 | switch (target) { |
| 758 | case MEM_CGROUP_TARGET_THRESH: |
| 759 | next = val + THRESHOLDS_EVENTS_TARGET; |
| 760 | break; |
| 761 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
| 762 | next = val + SOFTLIMIT_EVENTS_TARGET; |
| 763 | break; |
| 764 | case MEM_CGROUP_TARGET_NUMAINFO: |
| 765 | next = val + NUMAINFO_EVENTS_TARGET; |
| 766 | break; |
| 767 | default: |
| 768 | break; |
| 769 | } |
| 770 | __this_cpu_write(memcg->stat->targets[target], next); |
| 771 | return true; |
| 772 | } |
| 773 | return false; |
| 774 | } |
| 775 | |
| 776 | /* |
| 777 | * Check events in order. |
| 778 | * |
| 779 | */ |
| 780 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
| 781 | { |
| 782 | /* threshold event is triggered in finer grain than soft limit */ |
| 783 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
| 784 | MEM_CGROUP_TARGET_THRESH))) { |
| 785 | bool do_softlimit; |
| 786 | bool do_numainfo __maybe_unused; |
| 787 | |
| 788 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
| 789 | MEM_CGROUP_TARGET_SOFTLIMIT); |
| 790 | #if MAX_NUMNODES > 1 |
| 791 | do_numainfo = mem_cgroup_event_ratelimit(memcg, |
| 792 | MEM_CGROUP_TARGET_NUMAINFO); |
| 793 | #endif |
| 794 | mem_cgroup_threshold(memcg); |
| 795 | if (unlikely(do_softlimit)) |
| 796 | mem_cgroup_update_tree(memcg, page); |
| 797 | #if MAX_NUMNODES > 1 |
| 798 | if (unlikely(do_numainfo)) |
| 799 | atomic_inc(&memcg->numainfo_events); |
| 800 | #endif |
| 801 | } |
| 802 | } |
| 803 | |
| 804 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
| 805 | { |
| 806 | /* |
| 807 | * mm_update_next_owner() may clear mm->owner to NULL |
| 808 | * if it races with swapoff, page migration, etc. |
| 809 | * So this can be called with p == NULL. |
| 810 | */ |
| 811 | if (unlikely(!p)) |
| 812 | return NULL; |
| 813 | |
| 814 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
| 815 | } |
| 816 | EXPORT_SYMBOL(mem_cgroup_from_task); |
| 817 | |
| 818 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
| 819 | { |
| 820 | struct mem_cgroup *memcg = NULL; |
| 821 | |
| 822 | rcu_read_lock(); |
| 823 | do { |
| 824 | /* |
| 825 | * Page cache insertions can happen withou an |
| 826 | * actual mm context, e.g. during disk probing |
| 827 | * on boot, loopback IO, acct() writes etc. |
| 828 | */ |
| 829 | if (unlikely(!mm)) |
| 830 | memcg = root_mem_cgroup; |
| 831 | else { |
| 832 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 833 | if (unlikely(!memcg)) |
| 834 | memcg = root_mem_cgroup; |
| 835 | } |
| 836 | } while (!css_tryget_online(&memcg->css)); |
| 837 | rcu_read_unlock(); |
| 838 | return memcg; |
| 839 | } |
| 840 | |
| 841 | /** |
| 842 | * mem_cgroup_iter - iterate over memory cgroup hierarchy |
| 843 | * @root: hierarchy root |
| 844 | * @prev: previously returned memcg, NULL on first invocation |
| 845 | * @reclaim: cookie for shared reclaim walks, NULL for full walks |
| 846 | * |
| 847 | * Returns references to children of the hierarchy below @root, or |
| 848 | * @root itself, or %NULL after a full round-trip. |
| 849 | * |
| 850 | * Caller must pass the return value in @prev on subsequent |
| 851 | * invocations for reference counting, or use mem_cgroup_iter_break() |
| 852 | * to cancel a hierarchy walk before the round-trip is complete. |
| 853 | * |
| 854 | * Reclaimers can specify a zone and a priority level in @reclaim to |
| 855 | * divide up the memcgs in the hierarchy among all concurrent |
| 856 | * reclaimers operating on the same zone and priority. |
| 857 | */ |
| 858 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
| 859 | struct mem_cgroup *prev, |
| 860 | struct mem_cgroup_reclaim_cookie *reclaim) |
| 861 | { |
| 862 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
| 863 | struct cgroup_subsys_state *css = NULL; |
| 864 | struct mem_cgroup *memcg = NULL; |
| 865 | struct mem_cgroup *pos = NULL; |
| 866 | |
| 867 | if (mem_cgroup_disabled()) |
| 868 | return NULL; |
| 869 | |
| 870 | if (!root) |
| 871 | root = root_mem_cgroup; |
| 872 | |
| 873 | if (prev && !reclaim) |
| 874 | pos = prev; |
| 875 | |
| 876 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
| 877 | if (prev) |
| 878 | goto out; |
| 879 | return root; |
| 880 | } |
| 881 | |
| 882 | rcu_read_lock(); |
| 883 | |
| 884 | if (reclaim) { |
| 885 | struct mem_cgroup_per_zone *mz; |
| 886 | |
| 887 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); |
| 888 | iter = &mz->iter[reclaim->priority]; |
| 889 | |
| 890 | if (prev && reclaim->generation != iter->generation) |
| 891 | goto out_unlock; |
| 892 | |
| 893 | while (1) { |
| 894 | pos = READ_ONCE(iter->position); |
| 895 | if (!pos || css_tryget(&pos->css)) |
| 896 | break; |
| 897 | /* |
| 898 | * css reference reached zero, so iter->position will |
| 899 | * be cleared by ->css_released. However, we should not |
| 900 | * rely on this happening soon, because ->css_released |
| 901 | * is called from a work queue, and by busy-waiting we |
| 902 | * might block it. So we clear iter->position right |
| 903 | * away. |
| 904 | */ |
| 905 | (void)cmpxchg(&iter->position, pos, NULL); |
| 906 | } |
| 907 | } |
| 908 | |
| 909 | if (pos) |
| 910 | css = &pos->css; |
| 911 | |
| 912 | for (;;) { |
| 913 | css = css_next_descendant_pre(css, &root->css); |
| 914 | if (!css) { |
| 915 | /* |
| 916 | * Reclaimers share the hierarchy walk, and a |
| 917 | * new one might jump in right at the end of |
| 918 | * the hierarchy - make sure they see at least |
| 919 | * one group and restart from the beginning. |
| 920 | */ |
| 921 | if (!prev) |
| 922 | continue; |
| 923 | break; |
| 924 | } |
| 925 | |
| 926 | /* |
| 927 | * Verify the css and acquire a reference. The root |
| 928 | * is provided by the caller, so we know it's alive |
| 929 | * and kicking, and don't take an extra reference. |
| 930 | */ |
| 931 | memcg = mem_cgroup_from_css(css); |
| 932 | |
| 933 | if (css == &root->css) |
| 934 | break; |
| 935 | |
| 936 | if (css_tryget(css)) { |
| 937 | /* |
| 938 | * Make sure the memcg is initialized: |
| 939 | * mem_cgroup_css_online() orders the the |
| 940 | * initialization against setting the flag. |
| 941 | */ |
| 942 | if (smp_load_acquire(&memcg->initialized)) |
| 943 | break; |
| 944 | |
| 945 | css_put(css); |
| 946 | } |
| 947 | |
| 948 | memcg = NULL; |
| 949 | } |
| 950 | |
| 951 | if (reclaim) { |
| 952 | /* |
| 953 | * The position could have already been updated by a competing |
| 954 | * thread, so check that the value hasn't changed since we read |
| 955 | * it to avoid reclaiming from the same cgroup twice. |
| 956 | */ |
| 957 | (void)cmpxchg(&iter->position, pos, memcg); |
| 958 | |
| 959 | if (pos) |
| 960 | css_put(&pos->css); |
| 961 | |
| 962 | if (!memcg) |
| 963 | iter->generation++; |
| 964 | else if (!prev) |
| 965 | reclaim->generation = iter->generation; |
| 966 | } |
| 967 | |
| 968 | out_unlock: |
| 969 | rcu_read_unlock(); |
| 970 | out: |
| 971 | if (prev && prev != root) |
| 972 | css_put(&prev->css); |
| 973 | |
| 974 | return memcg; |
| 975 | } |
| 976 | |
| 977 | /** |
| 978 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely |
| 979 | * @root: hierarchy root |
| 980 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() |
| 981 | */ |
| 982 | void mem_cgroup_iter_break(struct mem_cgroup *root, |
| 983 | struct mem_cgroup *prev) |
| 984 | { |
| 985 | if (!root) |
| 986 | root = root_mem_cgroup; |
| 987 | if (prev && prev != root) |
| 988 | css_put(&prev->css); |
| 989 | } |
| 990 | |
| 991 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
| 992 | { |
| 993 | struct mem_cgroup *memcg = dead_memcg; |
| 994 | struct mem_cgroup_reclaim_iter *iter; |
| 995 | struct mem_cgroup_per_zone *mz; |
| 996 | int nid, zid; |
| 997 | int i; |
| 998 | |
| 999 | while ((memcg = parent_mem_cgroup(memcg))) { |
| 1000 | for_each_node(nid) { |
| 1001 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 1002 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
| 1003 | for (i = 0; i <= DEF_PRIORITY; i++) { |
| 1004 | iter = &mz->iter[i]; |
| 1005 | cmpxchg(&iter->position, |
| 1006 | dead_memcg, NULL); |
| 1007 | } |
| 1008 | } |
| 1009 | } |
| 1010 | } |
| 1011 | } |
| 1012 | |
| 1013 | /* |
| 1014 | * Iteration constructs for visiting all cgroups (under a tree). If |
| 1015 | * loops are exited prematurely (break), mem_cgroup_iter_break() must |
| 1016 | * be used for reference counting. |
| 1017 | */ |
| 1018 | #define for_each_mem_cgroup_tree(iter, root) \ |
| 1019 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
| 1020 | iter != NULL; \ |
| 1021 | iter = mem_cgroup_iter(root, iter, NULL)) |
| 1022 | |
| 1023 | #define for_each_mem_cgroup(iter) \ |
| 1024 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
| 1025 | iter != NULL; \ |
| 1026 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
| 1027 | |
| 1028 | /** |
| 1029 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg |
| 1030 | * @zone: zone of the wanted lruvec |
| 1031 | * @memcg: memcg of the wanted lruvec |
| 1032 | * |
| 1033 | * Returns the lru list vector holding pages for the given @zone and |
| 1034 | * @mem. This can be the global zone lruvec, if the memory controller |
| 1035 | * is disabled. |
| 1036 | */ |
| 1037 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, |
| 1038 | struct mem_cgroup *memcg) |
| 1039 | { |
| 1040 | struct mem_cgroup_per_zone *mz; |
| 1041 | struct lruvec *lruvec; |
| 1042 | |
| 1043 | if (mem_cgroup_disabled()) { |
| 1044 | lruvec = &zone->lruvec; |
| 1045 | goto out; |
| 1046 | } |
| 1047 | |
| 1048 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
| 1049 | lruvec = &mz->lruvec; |
| 1050 | out: |
| 1051 | /* |
| 1052 | * Since a node can be onlined after the mem_cgroup was created, |
| 1053 | * we have to be prepared to initialize lruvec->zone here; |
| 1054 | * and if offlined then reonlined, we need to reinitialize it. |
| 1055 | */ |
| 1056 | if (unlikely(lruvec->zone != zone)) |
| 1057 | lruvec->zone = zone; |
| 1058 | return lruvec; |
| 1059 | } |
| 1060 | |
| 1061 | /** |
| 1062 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
| 1063 | * @page: the page |
| 1064 | * @zone: zone of the page |
| 1065 | * |
| 1066 | * This function is only safe when following the LRU page isolation |
| 1067 | * and putback protocol: the LRU lock must be held, and the page must |
| 1068 | * either be PageLRU() or the caller must have isolated/allocated it. |
| 1069 | */ |
| 1070 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
| 1071 | { |
| 1072 | struct mem_cgroup_per_zone *mz; |
| 1073 | struct mem_cgroup *memcg; |
| 1074 | struct lruvec *lruvec; |
| 1075 | |
| 1076 | if (mem_cgroup_disabled()) { |
| 1077 | lruvec = &zone->lruvec; |
| 1078 | goto out; |
| 1079 | } |
| 1080 | |
| 1081 | memcg = page->mem_cgroup; |
| 1082 | /* |
| 1083 | * Swapcache readahead pages are added to the LRU - and |
| 1084 | * possibly migrated - before they are charged. |
| 1085 | */ |
| 1086 | if (!memcg) |
| 1087 | memcg = root_mem_cgroup; |
| 1088 | |
| 1089 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
| 1090 | lruvec = &mz->lruvec; |
| 1091 | out: |
| 1092 | /* |
| 1093 | * Since a node can be onlined after the mem_cgroup was created, |
| 1094 | * we have to be prepared to initialize lruvec->zone here; |
| 1095 | * and if offlined then reonlined, we need to reinitialize it. |
| 1096 | */ |
| 1097 | if (unlikely(lruvec->zone != zone)) |
| 1098 | lruvec->zone = zone; |
| 1099 | return lruvec; |
| 1100 | } |
| 1101 | |
| 1102 | /** |
| 1103 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
| 1104 | * @lruvec: mem_cgroup per zone lru vector |
| 1105 | * @lru: index of lru list the page is sitting on |
| 1106 | * @nr_pages: positive when adding or negative when removing |
| 1107 | * |
| 1108 | * This function must be called when a page is added to or removed from an |
| 1109 | * lru list. |
| 1110 | */ |
| 1111 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
| 1112 | int nr_pages) |
| 1113 | { |
| 1114 | struct mem_cgroup_per_zone *mz; |
| 1115 | unsigned long *lru_size; |
| 1116 | |
| 1117 | if (mem_cgroup_disabled()) |
| 1118 | return; |
| 1119 | |
| 1120 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
| 1121 | lru_size = mz->lru_size + lru; |
| 1122 | *lru_size += nr_pages; |
| 1123 | VM_BUG_ON((long)(*lru_size) < 0); |
| 1124 | } |
| 1125 | |
| 1126 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
| 1127 | { |
| 1128 | struct mem_cgroup *task_memcg; |
| 1129 | struct task_struct *p; |
| 1130 | bool ret; |
| 1131 | |
| 1132 | p = find_lock_task_mm(task); |
| 1133 | if (p) { |
| 1134 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
| 1135 | task_unlock(p); |
| 1136 | } else { |
| 1137 | /* |
| 1138 | * All threads may have already detached their mm's, but the oom |
| 1139 | * killer still needs to detect if they have already been oom |
| 1140 | * killed to prevent needlessly killing additional tasks. |
| 1141 | */ |
| 1142 | rcu_read_lock(); |
| 1143 | task_memcg = mem_cgroup_from_task(task); |
| 1144 | css_get(&task_memcg->css); |
| 1145 | rcu_read_unlock(); |
| 1146 | } |
| 1147 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
| 1148 | css_put(&task_memcg->css); |
| 1149 | return ret; |
| 1150 | } |
| 1151 | |
| 1152 | #define mem_cgroup_from_counter(counter, member) \ |
| 1153 | container_of(counter, struct mem_cgroup, member) |
| 1154 | |
| 1155 | /** |
| 1156 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
| 1157 | * @memcg: the memory cgroup |
| 1158 | * |
| 1159 | * Returns the maximum amount of memory @mem can be charged with, in |
| 1160 | * pages. |
| 1161 | */ |
| 1162 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
| 1163 | { |
| 1164 | unsigned long margin = 0; |
| 1165 | unsigned long count; |
| 1166 | unsigned long limit; |
| 1167 | |
| 1168 | count = page_counter_read(&memcg->memory); |
| 1169 | limit = READ_ONCE(memcg->memory.limit); |
| 1170 | if (count < limit) |
| 1171 | margin = limit - count; |
| 1172 | |
| 1173 | if (do_swap_account) { |
| 1174 | count = page_counter_read(&memcg->memsw); |
| 1175 | limit = READ_ONCE(memcg->memsw.limit); |
| 1176 | if (count <= limit) |
| 1177 | margin = min(margin, limit - count); |
| 1178 | } |
| 1179 | |
| 1180 | return margin; |
| 1181 | } |
| 1182 | |
| 1183 | /* |
| 1184 | * A routine for checking "mem" is under move_account() or not. |
| 1185 | * |
| 1186 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
| 1187 | * moving cgroups. This is for waiting at high-memory pressure |
| 1188 | * caused by "move". |
| 1189 | */ |
| 1190 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
| 1191 | { |
| 1192 | struct mem_cgroup *from; |
| 1193 | struct mem_cgroup *to; |
| 1194 | bool ret = false; |
| 1195 | /* |
| 1196 | * Unlike task_move routines, we access mc.to, mc.from not under |
| 1197 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. |
| 1198 | */ |
| 1199 | spin_lock(&mc.lock); |
| 1200 | from = mc.from; |
| 1201 | to = mc.to; |
| 1202 | if (!from) |
| 1203 | goto unlock; |
| 1204 | |
| 1205 | ret = mem_cgroup_is_descendant(from, memcg) || |
| 1206 | mem_cgroup_is_descendant(to, memcg); |
| 1207 | unlock: |
| 1208 | spin_unlock(&mc.lock); |
| 1209 | return ret; |
| 1210 | } |
| 1211 | |
| 1212 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
| 1213 | { |
| 1214 | if (mc.moving_task && current != mc.moving_task) { |
| 1215 | if (mem_cgroup_under_move(memcg)) { |
| 1216 | DEFINE_WAIT(wait); |
| 1217 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); |
| 1218 | /* moving charge context might have finished. */ |
| 1219 | if (mc.moving_task) |
| 1220 | schedule(); |
| 1221 | finish_wait(&mc.waitq, &wait); |
| 1222 | return true; |
| 1223 | } |
| 1224 | } |
| 1225 | return false; |
| 1226 | } |
| 1227 | |
| 1228 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
| 1229 | /** |
| 1230 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
| 1231 | * @memcg: The memory cgroup that went over limit |
| 1232 | * @p: Task that is going to be killed |
| 1233 | * |
| 1234 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is |
| 1235 | * enabled |
| 1236 | */ |
| 1237 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) |
| 1238 | { |
| 1239 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
| 1240 | static DEFINE_MUTEX(oom_info_lock); |
| 1241 | struct mem_cgroup *iter; |
| 1242 | unsigned int i; |
| 1243 | |
| 1244 | mutex_lock(&oom_info_lock); |
| 1245 | rcu_read_lock(); |
| 1246 | |
| 1247 | if (p) { |
| 1248 | pr_info("Task in "); |
| 1249 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
| 1250 | pr_cont(" killed as a result of limit of "); |
| 1251 | } else { |
| 1252 | pr_info("Memory limit reached of cgroup "); |
| 1253 | } |
| 1254 | |
| 1255 | pr_cont_cgroup_path(memcg->css.cgroup); |
| 1256 | pr_cont("\n"); |
| 1257 | |
| 1258 | rcu_read_unlock(); |
| 1259 | |
| 1260 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1261 | K((u64)page_counter_read(&memcg->memory)), |
| 1262 | K((u64)memcg->memory.limit), memcg->memory.failcnt); |
| 1263 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1264 | K((u64)page_counter_read(&memcg->memsw)), |
| 1265 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); |
| 1266 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1267 | K((u64)page_counter_read(&memcg->kmem)), |
| 1268 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); |
| 1269 | |
| 1270 | for_each_mem_cgroup_tree(iter, memcg) { |
| 1271 | pr_info("Memory cgroup stats for "); |
| 1272 | pr_cont_cgroup_path(iter->css.cgroup); |
| 1273 | pr_cont(":"); |
| 1274 | |
| 1275 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
| 1276 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
| 1277 | continue; |
| 1278 | pr_cont(" %s:%luKB", mem_cgroup_stat_names[i], |
| 1279 | K(mem_cgroup_read_stat(iter, i))); |
| 1280 | } |
| 1281 | |
| 1282 | for (i = 0; i < NR_LRU_LISTS; i++) |
| 1283 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], |
| 1284 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); |
| 1285 | |
| 1286 | pr_cont("\n"); |
| 1287 | } |
| 1288 | mutex_unlock(&oom_info_lock); |
| 1289 | } |
| 1290 | |
| 1291 | /* |
| 1292 | * This function returns the number of memcg under hierarchy tree. Returns |
| 1293 | * 1(self count) if no children. |
| 1294 | */ |
| 1295 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
| 1296 | { |
| 1297 | int num = 0; |
| 1298 | struct mem_cgroup *iter; |
| 1299 | |
| 1300 | for_each_mem_cgroup_tree(iter, memcg) |
| 1301 | num++; |
| 1302 | return num; |
| 1303 | } |
| 1304 | |
| 1305 | /* |
| 1306 | * Return the memory (and swap, if configured) limit for a memcg. |
| 1307 | */ |
| 1308 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
| 1309 | { |
| 1310 | unsigned long limit; |
| 1311 | |
| 1312 | limit = memcg->memory.limit; |
| 1313 | if (mem_cgroup_swappiness(memcg)) { |
| 1314 | unsigned long memsw_limit; |
| 1315 | |
| 1316 | memsw_limit = memcg->memsw.limit; |
| 1317 | limit = min(limit + total_swap_pages, memsw_limit); |
| 1318 | } |
| 1319 | return limit; |
| 1320 | } |
| 1321 | |
| 1322 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
| 1323 | int order) |
| 1324 | { |
| 1325 | struct oom_control oc = { |
| 1326 | .zonelist = NULL, |
| 1327 | .nodemask = NULL, |
| 1328 | .gfp_mask = gfp_mask, |
| 1329 | .order = order, |
| 1330 | }; |
| 1331 | struct mem_cgroup *iter; |
| 1332 | unsigned long chosen_points = 0; |
| 1333 | unsigned long totalpages; |
| 1334 | unsigned int points = 0; |
| 1335 | struct task_struct *chosen = NULL; |
| 1336 | |
| 1337 | mutex_lock(&oom_lock); |
| 1338 | |
| 1339 | /* |
| 1340 | * If current has a pending SIGKILL or is exiting, then automatically |
| 1341 | * select it. The goal is to allow it to allocate so that it may |
| 1342 | * quickly exit and free its memory. |
| 1343 | */ |
| 1344 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
| 1345 | mark_oom_victim(current); |
| 1346 | goto unlock; |
| 1347 | } |
| 1348 | |
| 1349 | check_panic_on_oom(&oc, CONSTRAINT_MEMCG, memcg); |
| 1350 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
| 1351 | for_each_mem_cgroup_tree(iter, memcg) { |
| 1352 | struct css_task_iter it; |
| 1353 | struct task_struct *task; |
| 1354 | |
| 1355 | css_task_iter_start(&iter->css, &it); |
| 1356 | while ((task = css_task_iter_next(&it))) { |
| 1357 | switch (oom_scan_process_thread(&oc, task, totalpages)) { |
| 1358 | case OOM_SCAN_SELECT: |
| 1359 | if (chosen) |
| 1360 | put_task_struct(chosen); |
| 1361 | chosen = task; |
| 1362 | chosen_points = ULONG_MAX; |
| 1363 | get_task_struct(chosen); |
| 1364 | /* fall through */ |
| 1365 | case OOM_SCAN_CONTINUE: |
| 1366 | continue; |
| 1367 | case OOM_SCAN_ABORT: |
| 1368 | css_task_iter_end(&it); |
| 1369 | mem_cgroup_iter_break(memcg, iter); |
| 1370 | if (chosen) |
| 1371 | put_task_struct(chosen); |
| 1372 | goto unlock; |
| 1373 | case OOM_SCAN_OK: |
| 1374 | break; |
| 1375 | }; |
| 1376 | points = oom_badness(task, memcg, NULL, totalpages); |
| 1377 | if (!points || points < chosen_points) |
| 1378 | continue; |
| 1379 | /* Prefer thread group leaders for display purposes */ |
| 1380 | if (points == chosen_points && |
| 1381 | thread_group_leader(chosen)) |
| 1382 | continue; |
| 1383 | |
| 1384 | if (chosen) |
| 1385 | put_task_struct(chosen); |
| 1386 | chosen = task; |
| 1387 | chosen_points = points; |
| 1388 | get_task_struct(chosen); |
| 1389 | } |
| 1390 | css_task_iter_end(&it); |
| 1391 | } |
| 1392 | |
| 1393 | if (chosen) { |
| 1394 | points = chosen_points * 1000 / totalpages; |
| 1395 | oom_kill_process(&oc, chosen, points, totalpages, memcg, |
| 1396 | "Memory cgroup out of memory"); |
| 1397 | } |
| 1398 | unlock: |
| 1399 | mutex_unlock(&oom_lock); |
| 1400 | return chosen; |
| 1401 | } |
| 1402 | |
| 1403 | #if MAX_NUMNODES > 1 |
| 1404 | |
| 1405 | /** |
| 1406 | * test_mem_cgroup_node_reclaimable |
| 1407 | * @memcg: the target memcg |
| 1408 | * @nid: the node ID to be checked. |
| 1409 | * @noswap : specify true here if the user wants flle only information. |
| 1410 | * |
| 1411 | * This function returns whether the specified memcg contains any |
| 1412 | * reclaimable pages on a node. Returns true if there are any reclaimable |
| 1413 | * pages in the node. |
| 1414 | */ |
| 1415 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
| 1416 | int nid, bool noswap) |
| 1417 | { |
| 1418 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
| 1419 | return true; |
| 1420 | if (noswap || !total_swap_pages) |
| 1421 | return false; |
| 1422 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
| 1423 | return true; |
| 1424 | return false; |
| 1425 | |
| 1426 | } |
| 1427 | |
| 1428 | /* |
| 1429 | * Always updating the nodemask is not very good - even if we have an empty |
| 1430 | * list or the wrong list here, we can start from some node and traverse all |
| 1431 | * nodes based on the zonelist. So update the list loosely once per 10 secs. |
| 1432 | * |
| 1433 | */ |
| 1434 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
| 1435 | { |
| 1436 | int nid; |
| 1437 | /* |
| 1438 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET |
| 1439 | * pagein/pageout changes since the last update. |
| 1440 | */ |
| 1441 | if (!atomic_read(&memcg->numainfo_events)) |
| 1442 | return; |
| 1443 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
| 1444 | return; |
| 1445 | |
| 1446 | /* make a nodemask where this memcg uses memory from */ |
| 1447 | memcg->scan_nodes = node_states[N_MEMORY]; |
| 1448 | |
| 1449 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
| 1450 | |
| 1451 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
| 1452 | node_clear(nid, memcg->scan_nodes); |
| 1453 | } |
| 1454 | |
| 1455 | atomic_set(&memcg->numainfo_events, 0); |
| 1456 | atomic_set(&memcg->numainfo_updating, 0); |
| 1457 | } |
| 1458 | |
| 1459 | /* |
| 1460 | * Selecting a node where we start reclaim from. Because what we need is just |
| 1461 | * reducing usage counter, start from anywhere is O,K. Considering |
| 1462 | * memory reclaim from current node, there are pros. and cons. |
| 1463 | * |
| 1464 | * Freeing memory from current node means freeing memory from a node which |
| 1465 | * we'll use or we've used. So, it may make LRU bad. And if several threads |
| 1466 | * hit limits, it will see a contention on a node. But freeing from remote |
| 1467 | * node means more costs for memory reclaim because of memory latency. |
| 1468 | * |
| 1469 | * Now, we use round-robin. Better algorithm is welcomed. |
| 1470 | */ |
| 1471 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
| 1472 | { |
| 1473 | int node; |
| 1474 | |
| 1475 | mem_cgroup_may_update_nodemask(memcg); |
| 1476 | node = memcg->last_scanned_node; |
| 1477 | |
| 1478 | node = next_node(node, memcg->scan_nodes); |
| 1479 | if (node == MAX_NUMNODES) |
| 1480 | node = first_node(memcg->scan_nodes); |
| 1481 | /* |
| 1482 | * We call this when we hit limit, not when pages are added to LRU. |
| 1483 | * No LRU may hold pages because all pages are UNEVICTABLE or |
| 1484 | * memcg is too small and all pages are not on LRU. In that case, |
| 1485 | * we use curret node. |
| 1486 | */ |
| 1487 | if (unlikely(node == MAX_NUMNODES)) |
| 1488 | node = numa_node_id(); |
| 1489 | |
| 1490 | memcg->last_scanned_node = node; |
| 1491 | return node; |
| 1492 | } |
| 1493 | #else |
| 1494 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
| 1495 | { |
| 1496 | return 0; |
| 1497 | } |
| 1498 | #endif |
| 1499 | |
| 1500 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
| 1501 | struct zone *zone, |
| 1502 | gfp_t gfp_mask, |
| 1503 | unsigned long *total_scanned) |
| 1504 | { |
| 1505 | struct mem_cgroup *victim = NULL; |
| 1506 | int total = 0; |
| 1507 | int loop = 0; |
| 1508 | unsigned long excess; |
| 1509 | unsigned long nr_scanned; |
| 1510 | struct mem_cgroup_reclaim_cookie reclaim = { |
| 1511 | .zone = zone, |
| 1512 | .priority = 0, |
| 1513 | }; |
| 1514 | |
| 1515 | excess = soft_limit_excess(root_memcg); |
| 1516 | |
| 1517 | while (1) { |
| 1518 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
| 1519 | if (!victim) { |
| 1520 | loop++; |
| 1521 | if (loop >= 2) { |
| 1522 | /* |
| 1523 | * If we have not been able to reclaim |
| 1524 | * anything, it might because there are |
| 1525 | * no reclaimable pages under this hierarchy |
| 1526 | */ |
| 1527 | if (!total) |
| 1528 | break; |
| 1529 | /* |
| 1530 | * We want to do more targeted reclaim. |
| 1531 | * excess >> 2 is not to excessive so as to |
| 1532 | * reclaim too much, nor too less that we keep |
| 1533 | * coming back to reclaim from this cgroup |
| 1534 | */ |
| 1535 | if (total >= (excess >> 2) || |
| 1536 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
| 1537 | break; |
| 1538 | } |
| 1539 | continue; |
| 1540 | } |
| 1541 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
| 1542 | zone, &nr_scanned); |
| 1543 | *total_scanned += nr_scanned; |
| 1544 | if (!soft_limit_excess(root_memcg)) |
| 1545 | break; |
| 1546 | } |
| 1547 | mem_cgroup_iter_break(root_memcg, victim); |
| 1548 | return total; |
| 1549 | } |
| 1550 | |
| 1551 | #ifdef CONFIG_LOCKDEP |
| 1552 | static struct lockdep_map memcg_oom_lock_dep_map = { |
| 1553 | .name = "memcg_oom_lock", |
| 1554 | }; |
| 1555 | #endif |
| 1556 | |
| 1557 | static DEFINE_SPINLOCK(memcg_oom_lock); |
| 1558 | |
| 1559 | /* |
| 1560 | * Check OOM-Killer is already running under our hierarchy. |
| 1561 | * If someone is running, return false. |
| 1562 | */ |
| 1563 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
| 1564 | { |
| 1565 | struct mem_cgroup *iter, *failed = NULL; |
| 1566 | |
| 1567 | spin_lock(&memcg_oom_lock); |
| 1568 | |
| 1569 | for_each_mem_cgroup_tree(iter, memcg) { |
| 1570 | if (iter->oom_lock) { |
| 1571 | /* |
| 1572 | * this subtree of our hierarchy is already locked |
| 1573 | * so we cannot give a lock. |
| 1574 | */ |
| 1575 | failed = iter; |
| 1576 | mem_cgroup_iter_break(memcg, iter); |
| 1577 | break; |
| 1578 | } else |
| 1579 | iter->oom_lock = true; |
| 1580 | } |
| 1581 | |
| 1582 | if (failed) { |
| 1583 | /* |
| 1584 | * OK, we failed to lock the whole subtree so we have |
| 1585 | * to clean up what we set up to the failing subtree |
| 1586 | */ |
| 1587 | for_each_mem_cgroup_tree(iter, memcg) { |
| 1588 | if (iter == failed) { |
| 1589 | mem_cgroup_iter_break(memcg, iter); |
| 1590 | break; |
| 1591 | } |
| 1592 | iter->oom_lock = false; |
| 1593 | } |
| 1594 | } else |
| 1595 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); |
| 1596 | |
| 1597 | spin_unlock(&memcg_oom_lock); |
| 1598 | |
| 1599 | return !failed; |
| 1600 | } |
| 1601 | |
| 1602 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
| 1603 | { |
| 1604 | struct mem_cgroup *iter; |
| 1605 | |
| 1606 | spin_lock(&memcg_oom_lock); |
| 1607 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
| 1608 | for_each_mem_cgroup_tree(iter, memcg) |
| 1609 | iter->oom_lock = false; |
| 1610 | spin_unlock(&memcg_oom_lock); |
| 1611 | } |
| 1612 | |
| 1613 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
| 1614 | { |
| 1615 | struct mem_cgroup *iter; |
| 1616 | |
| 1617 | spin_lock(&memcg_oom_lock); |
| 1618 | for_each_mem_cgroup_tree(iter, memcg) |
| 1619 | iter->under_oom++; |
| 1620 | spin_unlock(&memcg_oom_lock); |
| 1621 | } |
| 1622 | |
| 1623 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
| 1624 | { |
| 1625 | struct mem_cgroup *iter; |
| 1626 | |
| 1627 | /* |
| 1628 | * When a new child is created while the hierarchy is under oom, |
| 1629 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
| 1630 | */ |
| 1631 | spin_lock(&memcg_oom_lock); |
| 1632 | for_each_mem_cgroup_tree(iter, memcg) |
| 1633 | if (iter->under_oom > 0) |
| 1634 | iter->under_oom--; |
| 1635 | spin_unlock(&memcg_oom_lock); |
| 1636 | } |
| 1637 | |
| 1638 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
| 1639 | |
| 1640 | struct oom_wait_info { |
| 1641 | struct mem_cgroup *memcg; |
| 1642 | wait_queue_t wait; |
| 1643 | }; |
| 1644 | |
| 1645 | static int memcg_oom_wake_function(wait_queue_t *wait, |
| 1646 | unsigned mode, int sync, void *arg) |
| 1647 | { |
| 1648 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
| 1649 | struct mem_cgroup *oom_wait_memcg; |
| 1650 | struct oom_wait_info *oom_wait_info; |
| 1651 | |
| 1652 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); |
| 1653 | oom_wait_memcg = oom_wait_info->memcg; |
| 1654 | |
| 1655 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
| 1656 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) |
| 1657 | return 0; |
| 1658 | return autoremove_wake_function(wait, mode, sync, arg); |
| 1659 | } |
| 1660 | |
| 1661 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
| 1662 | { |
| 1663 | /* |
| 1664 | * For the following lockless ->under_oom test, the only required |
| 1665 | * guarantee is that it must see the state asserted by an OOM when |
| 1666 | * this function is called as a result of userland actions |
| 1667 | * triggered by the notification of the OOM. This is trivially |
| 1668 | * achieved by invoking mem_cgroup_mark_under_oom() before |
| 1669 | * triggering notification. |
| 1670 | */ |
| 1671 | if (memcg && memcg->under_oom) |
| 1672 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
| 1673 | } |
| 1674 | |
| 1675 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
| 1676 | { |
| 1677 | if (!current->memcg_may_oom) |
| 1678 | return; |
| 1679 | /* |
| 1680 | * We are in the middle of the charge context here, so we |
| 1681 | * don't want to block when potentially sitting on a callstack |
| 1682 | * that holds all kinds of filesystem and mm locks. |
| 1683 | * |
| 1684 | * Also, the caller may handle a failed allocation gracefully |
| 1685 | * (like optional page cache readahead) and so an OOM killer |
| 1686 | * invocation might not even be necessary. |
| 1687 | * |
| 1688 | * That's why we don't do anything here except remember the |
| 1689 | * OOM context and then deal with it at the end of the page |
| 1690 | * fault when the stack is unwound, the locks are released, |
| 1691 | * and when we know whether the fault was overall successful. |
| 1692 | */ |
| 1693 | css_get(&memcg->css); |
| 1694 | current->memcg_in_oom = memcg; |
| 1695 | current->memcg_oom_gfp_mask = mask; |
| 1696 | current->memcg_oom_order = order; |
| 1697 | } |
| 1698 | |
| 1699 | /** |
| 1700 | * mem_cgroup_oom_synchronize - complete memcg OOM handling |
| 1701 | * @handle: actually kill/wait or just clean up the OOM state |
| 1702 | * |
| 1703 | * This has to be called at the end of a page fault if the memcg OOM |
| 1704 | * handler was enabled. |
| 1705 | * |
| 1706 | * Memcg supports userspace OOM handling where failed allocations must |
| 1707 | * sleep on a waitqueue until the userspace task resolves the |
| 1708 | * situation. Sleeping directly in the charge context with all kinds |
| 1709 | * of locks held is not a good idea, instead we remember an OOM state |
| 1710 | * in the task and mem_cgroup_oom_synchronize() has to be called at |
| 1711 | * the end of the page fault to complete the OOM handling. |
| 1712 | * |
| 1713 | * Returns %true if an ongoing memcg OOM situation was detected and |
| 1714 | * completed, %false otherwise. |
| 1715 | */ |
| 1716 | bool mem_cgroup_oom_synchronize(bool handle) |
| 1717 | { |
| 1718 | struct mem_cgroup *memcg = current->memcg_in_oom; |
| 1719 | struct oom_wait_info owait; |
| 1720 | bool locked; |
| 1721 | |
| 1722 | /* OOM is global, do not handle */ |
| 1723 | if (!memcg) |
| 1724 | return false; |
| 1725 | |
| 1726 | if (!handle || oom_killer_disabled) |
| 1727 | goto cleanup; |
| 1728 | |
| 1729 | owait.memcg = memcg; |
| 1730 | owait.wait.flags = 0; |
| 1731 | owait.wait.func = memcg_oom_wake_function; |
| 1732 | owait.wait.private = current; |
| 1733 | INIT_LIST_HEAD(&owait.wait.task_list); |
| 1734 | |
| 1735 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
| 1736 | mem_cgroup_mark_under_oom(memcg); |
| 1737 | |
| 1738 | locked = mem_cgroup_oom_trylock(memcg); |
| 1739 | |
| 1740 | if (locked) |
| 1741 | mem_cgroup_oom_notify(memcg); |
| 1742 | |
| 1743 | if (locked && !memcg->oom_kill_disable) { |
| 1744 | mem_cgroup_unmark_under_oom(memcg); |
| 1745 | finish_wait(&memcg_oom_waitq, &owait.wait); |
| 1746 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
| 1747 | current->memcg_oom_order); |
| 1748 | } else { |
| 1749 | schedule(); |
| 1750 | mem_cgroup_unmark_under_oom(memcg); |
| 1751 | finish_wait(&memcg_oom_waitq, &owait.wait); |
| 1752 | } |
| 1753 | |
| 1754 | if (locked) { |
| 1755 | mem_cgroup_oom_unlock(memcg); |
| 1756 | /* |
| 1757 | * There is no guarantee that an OOM-lock contender |
| 1758 | * sees the wakeups triggered by the OOM kill |
| 1759 | * uncharges. Wake any sleepers explicitely. |
| 1760 | */ |
| 1761 | memcg_oom_recover(memcg); |
| 1762 | } |
| 1763 | cleanup: |
| 1764 | current->memcg_in_oom = NULL; |
| 1765 | css_put(&memcg->css); |
| 1766 | return true; |
| 1767 | } |
| 1768 | |
| 1769 | /** |
| 1770 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction |
| 1771 | * @page: page that is going to change accounted state |
| 1772 | * |
| 1773 | * This function must mark the beginning of an accounted page state |
| 1774 | * change to prevent double accounting when the page is concurrently |
| 1775 | * being moved to another memcg: |
| 1776 | * |
| 1777 | * memcg = mem_cgroup_begin_page_stat(page); |
| 1778 | * if (TestClearPageState(page)) |
| 1779 | * mem_cgroup_update_page_stat(memcg, state, -1); |
| 1780 | * mem_cgroup_end_page_stat(memcg); |
| 1781 | */ |
| 1782 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page) |
| 1783 | { |
| 1784 | struct mem_cgroup *memcg; |
| 1785 | unsigned long flags; |
| 1786 | |
| 1787 | /* |
| 1788 | * The RCU lock is held throughout the transaction. The fast |
| 1789 | * path can get away without acquiring the memcg->move_lock |
| 1790 | * because page moving starts with an RCU grace period. |
| 1791 | * |
| 1792 | * The RCU lock also protects the memcg from being freed when |
| 1793 | * the page state that is going to change is the only thing |
| 1794 | * preventing the page from being uncharged. |
| 1795 | * E.g. end-writeback clearing PageWriteback(), which allows |
| 1796 | * migration to go ahead and uncharge the page before the |
| 1797 | * account transaction might be complete. |
| 1798 | */ |
| 1799 | rcu_read_lock(); |
| 1800 | |
| 1801 | if (mem_cgroup_disabled()) |
| 1802 | return NULL; |
| 1803 | again: |
| 1804 | memcg = page->mem_cgroup; |
| 1805 | if (unlikely(!memcg)) |
| 1806 | return NULL; |
| 1807 | |
| 1808 | if (atomic_read(&memcg->moving_account) <= 0) |
| 1809 | return memcg; |
| 1810 | |
| 1811 | spin_lock_irqsave(&memcg->move_lock, flags); |
| 1812 | if (memcg != page->mem_cgroup) { |
| 1813 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
| 1814 | goto again; |
| 1815 | } |
| 1816 | |
| 1817 | /* |
| 1818 | * When charge migration first begins, we can have locked and |
| 1819 | * unlocked page stat updates happening concurrently. Track |
| 1820 | * the task who has the lock for mem_cgroup_end_page_stat(). |
| 1821 | */ |
| 1822 | memcg->move_lock_task = current; |
| 1823 | memcg->move_lock_flags = flags; |
| 1824 | |
| 1825 | return memcg; |
| 1826 | } |
| 1827 | EXPORT_SYMBOL(mem_cgroup_begin_page_stat); |
| 1828 | |
| 1829 | /** |
| 1830 | * mem_cgroup_end_page_stat - finish a page state statistics transaction |
| 1831 | * @memcg: the memcg that was accounted against |
| 1832 | */ |
| 1833 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg) |
| 1834 | { |
| 1835 | if (memcg && memcg->move_lock_task == current) { |
| 1836 | unsigned long flags = memcg->move_lock_flags; |
| 1837 | |
| 1838 | memcg->move_lock_task = NULL; |
| 1839 | memcg->move_lock_flags = 0; |
| 1840 | |
| 1841 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
| 1842 | } |
| 1843 | |
| 1844 | rcu_read_unlock(); |
| 1845 | } |
| 1846 | EXPORT_SYMBOL(mem_cgroup_end_page_stat); |
| 1847 | |
| 1848 | /* |
| 1849 | * size of first charge trial. "32" comes from vmscan.c's magic value. |
| 1850 | * TODO: maybe necessary to use big numbers in big irons. |
| 1851 | */ |
| 1852 | #define CHARGE_BATCH 32U |
| 1853 | struct memcg_stock_pcp { |
| 1854 | struct mem_cgroup *cached; /* this never be root cgroup */ |
| 1855 | unsigned int nr_pages; |
| 1856 | struct work_struct work; |
| 1857 | unsigned long flags; |
| 1858 | #define FLUSHING_CACHED_CHARGE 0 |
| 1859 | }; |
| 1860 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); |
| 1861 | static DEFINE_MUTEX(percpu_charge_mutex); |
| 1862 | |
| 1863 | /** |
| 1864 | * consume_stock: Try to consume stocked charge on this cpu. |
| 1865 | * @memcg: memcg to consume from. |
| 1866 | * @nr_pages: how many pages to charge. |
| 1867 | * |
| 1868 | * The charges will only happen if @memcg matches the current cpu's memcg |
| 1869 | * stock, and at least @nr_pages are available in that stock. Failure to |
| 1870 | * service an allocation will refill the stock. |
| 1871 | * |
| 1872 | * returns true if successful, false otherwise. |
| 1873 | */ |
| 1874 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
| 1875 | { |
| 1876 | struct memcg_stock_pcp *stock; |
| 1877 | bool ret = false; |
| 1878 | |
| 1879 | if (nr_pages > CHARGE_BATCH) |
| 1880 | return ret; |
| 1881 | |
| 1882 | stock = &get_cpu_var(memcg_stock); |
| 1883 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
| 1884 | stock->nr_pages -= nr_pages; |
| 1885 | ret = true; |
| 1886 | } |
| 1887 | put_cpu_var(memcg_stock); |
| 1888 | return ret; |
| 1889 | } |
| 1890 | |
| 1891 | /* |
| 1892 | * Returns stocks cached in percpu and reset cached information. |
| 1893 | */ |
| 1894 | static void drain_stock(struct memcg_stock_pcp *stock) |
| 1895 | { |
| 1896 | struct mem_cgroup *old = stock->cached; |
| 1897 | |
| 1898 | if (stock->nr_pages) { |
| 1899 | page_counter_uncharge(&old->memory, stock->nr_pages); |
| 1900 | if (do_swap_account) |
| 1901 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
| 1902 | css_put_many(&old->css, stock->nr_pages); |
| 1903 | stock->nr_pages = 0; |
| 1904 | } |
| 1905 | stock->cached = NULL; |
| 1906 | } |
| 1907 | |
| 1908 | /* |
| 1909 | * This must be called under preempt disabled or must be called by |
| 1910 | * a thread which is pinned to local cpu. |
| 1911 | */ |
| 1912 | static void drain_local_stock(struct work_struct *dummy) |
| 1913 | { |
| 1914 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
| 1915 | drain_stock(stock); |
| 1916 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
| 1917 | } |
| 1918 | |
| 1919 | /* |
| 1920 | * Cache charges(val) to local per_cpu area. |
| 1921 | * This will be consumed by consume_stock() function, later. |
| 1922 | */ |
| 1923 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
| 1924 | { |
| 1925 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); |
| 1926 | |
| 1927 | if (stock->cached != memcg) { /* reset if necessary */ |
| 1928 | drain_stock(stock); |
| 1929 | stock->cached = memcg; |
| 1930 | } |
| 1931 | stock->nr_pages += nr_pages; |
| 1932 | put_cpu_var(memcg_stock); |
| 1933 | } |
| 1934 | |
| 1935 | /* |
| 1936 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
| 1937 | * of the hierarchy under it. |
| 1938 | */ |
| 1939 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
| 1940 | { |
| 1941 | int cpu, curcpu; |
| 1942 | |
| 1943 | /* If someone's already draining, avoid adding running more workers. */ |
| 1944 | if (!mutex_trylock(&percpu_charge_mutex)) |
| 1945 | return; |
| 1946 | /* Notify other cpus that system-wide "drain" is running */ |
| 1947 | get_online_cpus(); |
| 1948 | curcpu = get_cpu(); |
| 1949 | for_each_online_cpu(cpu) { |
| 1950 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); |
| 1951 | struct mem_cgroup *memcg; |
| 1952 | |
| 1953 | memcg = stock->cached; |
| 1954 | if (!memcg || !stock->nr_pages) |
| 1955 | continue; |
| 1956 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
| 1957 | continue; |
| 1958 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
| 1959 | if (cpu == curcpu) |
| 1960 | drain_local_stock(&stock->work); |
| 1961 | else |
| 1962 | schedule_work_on(cpu, &stock->work); |
| 1963 | } |
| 1964 | } |
| 1965 | put_cpu(); |
| 1966 | put_online_cpus(); |
| 1967 | mutex_unlock(&percpu_charge_mutex); |
| 1968 | } |
| 1969 | |
| 1970 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
| 1971 | unsigned long action, |
| 1972 | void *hcpu) |
| 1973 | { |
| 1974 | int cpu = (unsigned long)hcpu; |
| 1975 | struct memcg_stock_pcp *stock; |
| 1976 | |
| 1977 | if (action == CPU_ONLINE) |
| 1978 | return NOTIFY_OK; |
| 1979 | |
| 1980 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
| 1981 | return NOTIFY_OK; |
| 1982 | |
| 1983 | stock = &per_cpu(memcg_stock, cpu); |
| 1984 | drain_stock(stock); |
| 1985 | return NOTIFY_OK; |
| 1986 | } |
| 1987 | |
| 1988 | /* |
| 1989 | * Scheduled by try_charge() to be executed from the userland return path |
| 1990 | * and reclaims memory over the high limit. |
| 1991 | */ |
| 1992 | void mem_cgroup_handle_over_high(void) |
| 1993 | { |
| 1994 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
| 1995 | struct mem_cgroup *memcg, *pos; |
| 1996 | |
| 1997 | if (likely(!nr_pages)) |
| 1998 | return; |
| 1999 | |
| 2000 | pos = memcg = get_mem_cgroup_from_mm(current->mm); |
| 2001 | |
| 2002 | do { |
| 2003 | if (page_counter_read(&pos->memory) <= pos->high) |
| 2004 | continue; |
| 2005 | mem_cgroup_events(pos, MEMCG_HIGH, 1); |
| 2006 | try_to_free_mem_cgroup_pages(pos, nr_pages, GFP_KERNEL, true); |
| 2007 | } while ((pos = parent_mem_cgroup(pos))); |
| 2008 | |
| 2009 | css_put(&memcg->css); |
| 2010 | current->memcg_nr_pages_over_high = 0; |
| 2011 | } |
| 2012 | |
| 2013 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
| 2014 | unsigned int nr_pages) |
| 2015 | { |
| 2016 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
| 2017 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
| 2018 | struct mem_cgroup *mem_over_limit; |
| 2019 | struct page_counter *counter; |
| 2020 | unsigned long nr_reclaimed; |
| 2021 | bool may_swap = true; |
| 2022 | bool drained = false; |
| 2023 | |
| 2024 | if (mem_cgroup_is_root(memcg)) |
| 2025 | return 0; |
| 2026 | retry: |
| 2027 | if (consume_stock(memcg, nr_pages)) |
| 2028 | return 0; |
| 2029 | |
| 2030 | if (!do_swap_account || |
| 2031 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
| 2032 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) |
| 2033 | goto done_restock; |
| 2034 | if (do_swap_account) |
| 2035 | page_counter_uncharge(&memcg->memsw, batch); |
| 2036 | mem_over_limit = mem_cgroup_from_counter(counter, memory); |
| 2037 | } else { |
| 2038 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
| 2039 | may_swap = false; |
| 2040 | } |
| 2041 | |
| 2042 | if (batch > nr_pages) { |
| 2043 | batch = nr_pages; |
| 2044 | goto retry; |
| 2045 | } |
| 2046 | |
| 2047 | /* |
| 2048 | * Unlike in global OOM situations, memcg is not in a physical |
| 2049 | * memory shortage. Allow dying and OOM-killed tasks to |
| 2050 | * bypass the last charges so that they can exit quickly and |
| 2051 | * free their memory. |
| 2052 | */ |
| 2053 | if (unlikely(test_thread_flag(TIF_MEMDIE) || |
| 2054 | fatal_signal_pending(current) || |
| 2055 | current->flags & PF_EXITING)) |
| 2056 | goto force; |
| 2057 | |
| 2058 | /* |
| 2059 | * Prevent unbounded recursion when reclaim operations need to |
| 2060 | * allocate memory. This might exceed the limits temporarily, |
| 2061 | * but we prefer facilitating memory reclaim and getting back |
| 2062 | * under the limit over triggering OOM kills in these cases. |
| 2063 | */ |
| 2064 | if (unlikely(current->flags & PF_MEMALLOC)) |
| 2065 | goto force; |
| 2066 | |
| 2067 | if (unlikely(task_in_memcg_oom(current))) |
| 2068 | goto nomem; |
| 2069 | |
| 2070 | if (!gfpflags_allow_blocking(gfp_mask)) |
| 2071 | goto nomem; |
| 2072 | |
| 2073 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
| 2074 | |
| 2075 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
| 2076 | gfp_mask, may_swap); |
| 2077 | |
| 2078 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
| 2079 | goto retry; |
| 2080 | |
| 2081 | if (!drained) { |
| 2082 | drain_all_stock(mem_over_limit); |
| 2083 | drained = true; |
| 2084 | goto retry; |
| 2085 | } |
| 2086 | |
| 2087 | if (gfp_mask & __GFP_NORETRY) |
| 2088 | goto nomem; |
| 2089 | /* |
| 2090 | * Even though the limit is exceeded at this point, reclaim |
| 2091 | * may have been able to free some pages. Retry the charge |
| 2092 | * before killing the task. |
| 2093 | * |
| 2094 | * Only for regular pages, though: huge pages are rather |
| 2095 | * unlikely to succeed so close to the limit, and we fall back |
| 2096 | * to regular pages anyway in case of failure. |
| 2097 | */ |
| 2098 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
| 2099 | goto retry; |
| 2100 | /* |
| 2101 | * At task move, charge accounts can be doubly counted. So, it's |
| 2102 | * better to wait until the end of task_move if something is going on. |
| 2103 | */ |
| 2104 | if (mem_cgroup_wait_acct_move(mem_over_limit)) |
| 2105 | goto retry; |
| 2106 | |
| 2107 | if (nr_retries--) |
| 2108 | goto retry; |
| 2109 | |
| 2110 | if (gfp_mask & __GFP_NOFAIL) |
| 2111 | goto force; |
| 2112 | |
| 2113 | if (fatal_signal_pending(current)) |
| 2114 | goto force; |
| 2115 | |
| 2116 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
| 2117 | |
| 2118 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
| 2119 | get_order(nr_pages * PAGE_SIZE)); |
| 2120 | nomem: |
| 2121 | if (!(gfp_mask & __GFP_NOFAIL)) |
| 2122 | return -ENOMEM; |
| 2123 | force: |
| 2124 | /* |
| 2125 | * The allocation either can't fail or will lead to more memory |
| 2126 | * being freed very soon. Allow memory usage go over the limit |
| 2127 | * temporarily by force charging it. |
| 2128 | */ |
| 2129 | page_counter_charge(&memcg->memory, nr_pages); |
| 2130 | if (do_swap_account) |
| 2131 | page_counter_charge(&memcg->memsw, nr_pages); |
| 2132 | css_get_many(&memcg->css, nr_pages); |
| 2133 | |
| 2134 | return 0; |
| 2135 | |
| 2136 | done_restock: |
| 2137 | css_get_many(&memcg->css, batch); |
| 2138 | if (batch > nr_pages) |
| 2139 | refill_stock(memcg, batch - nr_pages); |
| 2140 | |
| 2141 | /* |
| 2142 | * If the hierarchy is above the normal consumption range, schedule |
| 2143 | * reclaim on returning to userland. We can perform reclaim here |
| 2144 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
| 2145 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
| 2146 | * not recorded as it most likely matches current's and won't |
| 2147 | * change in the meantime. As high limit is checked again before |
| 2148 | * reclaim, the cost of mismatch is negligible. |
| 2149 | */ |
| 2150 | do { |
| 2151 | if (page_counter_read(&memcg->memory) > memcg->high) { |
| 2152 | current->memcg_nr_pages_over_high += batch; |
| 2153 | set_notify_resume(current); |
| 2154 | break; |
| 2155 | } |
| 2156 | } while ((memcg = parent_mem_cgroup(memcg))); |
| 2157 | |
| 2158 | return 0; |
| 2159 | } |
| 2160 | |
| 2161 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
| 2162 | { |
| 2163 | if (mem_cgroup_is_root(memcg)) |
| 2164 | return; |
| 2165 | |
| 2166 | page_counter_uncharge(&memcg->memory, nr_pages); |
| 2167 | if (do_swap_account) |
| 2168 | page_counter_uncharge(&memcg->memsw, nr_pages); |
| 2169 | |
| 2170 | css_put_many(&memcg->css, nr_pages); |
| 2171 | } |
| 2172 | |
| 2173 | static void lock_page_lru(struct page *page, int *isolated) |
| 2174 | { |
| 2175 | struct zone *zone = page_zone(page); |
| 2176 | |
| 2177 | spin_lock_irq(&zone->lru_lock); |
| 2178 | if (PageLRU(page)) { |
| 2179 | struct lruvec *lruvec; |
| 2180 | |
| 2181 | lruvec = mem_cgroup_page_lruvec(page, zone); |
| 2182 | ClearPageLRU(page); |
| 2183 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
| 2184 | *isolated = 1; |
| 2185 | } else |
| 2186 | *isolated = 0; |
| 2187 | } |
| 2188 | |
| 2189 | static void unlock_page_lru(struct page *page, int isolated) |
| 2190 | { |
| 2191 | struct zone *zone = page_zone(page); |
| 2192 | |
| 2193 | if (isolated) { |
| 2194 | struct lruvec *lruvec; |
| 2195 | |
| 2196 | lruvec = mem_cgroup_page_lruvec(page, zone); |
| 2197 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 2198 | SetPageLRU(page); |
| 2199 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
| 2200 | } |
| 2201 | spin_unlock_irq(&zone->lru_lock); |
| 2202 | } |
| 2203 | |
| 2204 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
| 2205 | bool lrucare) |
| 2206 | { |
| 2207 | int isolated; |
| 2208 | |
| 2209 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
| 2210 | |
| 2211 | /* |
| 2212 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page |
| 2213 | * may already be on some other mem_cgroup's LRU. Take care of it. |
| 2214 | */ |
| 2215 | if (lrucare) |
| 2216 | lock_page_lru(page, &isolated); |
| 2217 | |
| 2218 | /* |
| 2219 | * Nobody should be changing or seriously looking at |
| 2220 | * page->mem_cgroup at this point: |
| 2221 | * |
| 2222 | * - the page is uncharged |
| 2223 | * |
| 2224 | * - the page is off-LRU |
| 2225 | * |
| 2226 | * - an anonymous fault has exclusive page access, except for |
| 2227 | * a locked page table |
| 2228 | * |
| 2229 | * - a page cache insertion, a swapin fault, or a migration |
| 2230 | * have the page locked |
| 2231 | */ |
| 2232 | page->mem_cgroup = memcg; |
| 2233 | |
| 2234 | if (lrucare) |
| 2235 | unlock_page_lru(page, isolated); |
| 2236 | } |
| 2237 | |
| 2238 | #ifdef CONFIG_MEMCG_KMEM |
| 2239 | static int memcg_alloc_cache_id(void) |
| 2240 | { |
| 2241 | int id, size; |
| 2242 | int err; |
| 2243 | |
| 2244 | id = ida_simple_get(&memcg_cache_ida, |
| 2245 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
| 2246 | if (id < 0) |
| 2247 | return id; |
| 2248 | |
| 2249 | if (id < memcg_nr_cache_ids) |
| 2250 | return id; |
| 2251 | |
| 2252 | /* |
| 2253 | * There's no space for the new id in memcg_caches arrays, |
| 2254 | * so we have to grow them. |
| 2255 | */ |
| 2256 | down_write(&memcg_cache_ids_sem); |
| 2257 | |
| 2258 | size = 2 * (id + 1); |
| 2259 | if (size < MEMCG_CACHES_MIN_SIZE) |
| 2260 | size = MEMCG_CACHES_MIN_SIZE; |
| 2261 | else if (size > MEMCG_CACHES_MAX_SIZE) |
| 2262 | size = MEMCG_CACHES_MAX_SIZE; |
| 2263 | |
| 2264 | err = memcg_update_all_caches(size); |
| 2265 | if (!err) |
| 2266 | err = memcg_update_all_list_lrus(size); |
| 2267 | if (!err) |
| 2268 | memcg_nr_cache_ids = size; |
| 2269 | |
| 2270 | up_write(&memcg_cache_ids_sem); |
| 2271 | |
| 2272 | if (err) { |
| 2273 | ida_simple_remove(&memcg_cache_ida, id); |
| 2274 | return err; |
| 2275 | } |
| 2276 | return id; |
| 2277 | } |
| 2278 | |
| 2279 | static void memcg_free_cache_id(int id) |
| 2280 | { |
| 2281 | ida_simple_remove(&memcg_cache_ida, id); |
| 2282 | } |
| 2283 | |
| 2284 | struct memcg_kmem_cache_create_work { |
| 2285 | struct mem_cgroup *memcg; |
| 2286 | struct kmem_cache *cachep; |
| 2287 | struct work_struct work; |
| 2288 | }; |
| 2289 | |
| 2290 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
| 2291 | { |
| 2292 | struct memcg_kmem_cache_create_work *cw = |
| 2293 | container_of(w, struct memcg_kmem_cache_create_work, work); |
| 2294 | struct mem_cgroup *memcg = cw->memcg; |
| 2295 | struct kmem_cache *cachep = cw->cachep; |
| 2296 | |
| 2297 | memcg_create_kmem_cache(memcg, cachep); |
| 2298 | |
| 2299 | css_put(&memcg->css); |
| 2300 | kfree(cw); |
| 2301 | } |
| 2302 | |
| 2303 | /* |
| 2304 | * Enqueue the creation of a per-memcg kmem_cache. |
| 2305 | */ |
| 2306 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
| 2307 | struct kmem_cache *cachep) |
| 2308 | { |
| 2309 | struct memcg_kmem_cache_create_work *cw; |
| 2310 | |
| 2311 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
| 2312 | if (!cw) |
| 2313 | return; |
| 2314 | |
| 2315 | css_get(&memcg->css); |
| 2316 | |
| 2317 | cw->memcg = memcg; |
| 2318 | cw->cachep = cachep; |
| 2319 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
| 2320 | |
| 2321 | schedule_work(&cw->work); |
| 2322 | } |
| 2323 | |
| 2324 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
| 2325 | struct kmem_cache *cachep) |
| 2326 | { |
| 2327 | /* |
| 2328 | * We need to stop accounting when we kmalloc, because if the |
| 2329 | * corresponding kmalloc cache is not yet created, the first allocation |
| 2330 | * in __memcg_schedule_kmem_cache_create will recurse. |
| 2331 | * |
| 2332 | * However, it is better to enclose the whole function. Depending on |
| 2333 | * the debugging options enabled, INIT_WORK(), for instance, can |
| 2334 | * trigger an allocation. This too, will make us recurse. Because at |
| 2335 | * this point we can't allow ourselves back into memcg_kmem_get_cache, |
| 2336 | * the safest choice is to do it like this, wrapping the whole function. |
| 2337 | */ |
| 2338 | current->memcg_kmem_skip_account = 1; |
| 2339 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
| 2340 | current->memcg_kmem_skip_account = 0; |
| 2341 | } |
| 2342 | |
| 2343 | /* |
| 2344 | * Return the kmem_cache we're supposed to use for a slab allocation. |
| 2345 | * We try to use the current memcg's version of the cache. |
| 2346 | * |
| 2347 | * If the cache does not exist yet, if we are the first user of it, |
| 2348 | * we either create it immediately, if possible, or create it asynchronously |
| 2349 | * in a workqueue. |
| 2350 | * In the latter case, we will let the current allocation go through with |
| 2351 | * the original cache. |
| 2352 | * |
| 2353 | * Can't be called in interrupt context or from kernel threads. |
| 2354 | * This function needs to be called with rcu_read_lock() held. |
| 2355 | */ |
| 2356 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep) |
| 2357 | { |
| 2358 | struct mem_cgroup *memcg; |
| 2359 | struct kmem_cache *memcg_cachep; |
| 2360 | int kmemcg_id; |
| 2361 | |
| 2362 | VM_BUG_ON(!is_root_cache(cachep)); |
| 2363 | |
| 2364 | if (current->memcg_kmem_skip_account) |
| 2365 | return cachep; |
| 2366 | |
| 2367 | memcg = get_mem_cgroup_from_mm(current->mm); |
| 2368 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
| 2369 | if (kmemcg_id < 0) |
| 2370 | goto out; |
| 2371 | |
| 2372 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
| 2373 | if (likely(memcg_cachep)) |
| 2374 | return memcg_cachep; |
| 2375 | |
| 2376 | /* |
| 2377 | * If we are in a safe context (can wait, and not in interrupt |
| 2378 | * context), we could be be predictable and return right away. |
| 2379 | * This would guarantee that the allocation being performed |
| 2380 | * already belongs in the new cache. |
| 2381 | * |
| 2382 | * However, there are some clashes that can arrive from locking. |
| 2383 | * For instance, because we acquire the slab_mutex while doing |
| 2384 | * memcg_create_kmem_cache, this means no further allocation |
| 2385 | * could happen with the slab_mutex held. So it's better to |
| 2386 | * defer everything. |
| 2387 | */ |
| 2388 | memcg_schedule_kmem_cache_create(memcg, cachep); |
| 2389 | out: |
| 2390 | css_put(&memcg->css); |
| 2391 | return cachep; |
| 2392 | } |
| 2393 | |
| 2394 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
| 2395 | { |
| 2396 | if (!is_root_cache(cachep)) |
| 2397 | css_put(&cachep->memcg_params.memcg->css); |
| 2398 | } |
| 2399 | |
| 2400 | int __memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
| 2401 | struct mem_cgroup *memcg) |
| 2402 | { |
| 2403 | unsigned int nr_pages = 1 << order; |
| 2404 | struct page_counter *counter; |
| 2405 | int ret; |
| 2406 | |
| 2407 | if (!memcg_kmem_is_active(memcg)) |
| 2408 | return 0; |
| 2409 | |
| 2410 | if (!page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) |
| 2411 | return -ENOMEM; |
| 2412 | |
| 2413 | ret = try_charge(memcg, gfp, nr_pages); |
| 2414 | if (ret) { |
| 2415 | page_counter_uncharge(&memcg->kmem, nr_pages); |
| 2416 | return ret; |
| 2417 | } |
| 2418 | |
| 2419 | page->mem_cgroup = memcg; |
| 2420 | |
| 2421 | return 0; |
| 2422 | } |
| 2423 | |
| 2424 | int __memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
| 2425 | { |
| 2426 | struct mem_cgroup *memcg; |
| 2427 | int ret; |
| 2428 | |
| 2429 | memcg = get_mem_cgroup_from_mm(current->mm); |
| 2430 | ret = __memcg_kmem_charge_memcg(page, gfp, order, memcg); |
| 2431 | css_put(&memcg->css); |
| 2432 | return ret; |
| 2433 | } |
| 2434 | |
| 2435 | void __memcg_kmem_uncharge(struct page *page, int order) |
| 2436 | { |
| 2437 | struct mem_cgroup *memcg = page->mem_cgroup; |
| 2438 | unsigned int nr_pages = 1 << order; |
| 2439 | |
| 2440 | if (!memcg) |
| 2441 | return; |
| 2442 | |
| 2443 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
| 2444 | |
| 2445 | page_counter_uncharge(&memcg->kmem, nr_pages); |
| 2446 | page_counter_uncharge(&memcg->memory, nr_pages); |
| 2447 | if (do_swap_account) |
| 2448 | page_counter_uncharge(&memcg->memsw, nr_pages); |
| 2449 | |
| 2450 | page->mem_cgroup = NULL; |
| 2451 | css_put_many(&memcg->css, nr_pages); |
| 2452 | } |
| 2453 | #endif /* CONFIG_MEMCG_KMEM */ |
| 2454 | |
| 2455 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 2456 | |
| 2457 | /* |
| 2458 | * Because tail pages are not marked as "used", set it. We're under |
| 2459 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
| 2460 | * charge/uncharge will be never happen and move_account() is done under |
| 2461 | * compound_lock(), so we don't have to take care of races. |
| 2462 | */ |
| 2463 | void mem_cgroup_split_huge_fixup(struct page *head) |
| 2464 | { |
| 2465 | int i; |
| 2466 | |
| 2467 | if (mem_cgroup_disabled()) |
| 2468 | return; |
| 2469 | |
| 2470 | for (i = 1; i < HPAGE_PMD_NR; i++) |
| 2471 | head[i].mem_cgroup = head->mem_cgroup; |
| 2472 | |
| 2473 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
| 2474 | HPAGE_PMD_NR); |
| 2475 | } |
| 2476 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
| 2477 | |
| 2478 | #ifdef CONFIG_MEMCG_SWAP |
| 2479 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
| 2480 | bool charge) |
| 2481 | { |
| 2482 | int val = (charge) ? 1 : -1; |
| 2483 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
| 2484 | } |
| 2485 | |
| 2486 | /** |
| 2487 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. |
| 2488 | * @entry: swap entry to be moved |
| 2489 | * @from: mem_cgroup which the entry is moved from |
| 2490 | * @to: mem_cgroup which the entry is moved to |
| 2491 | * |
| 2492 | * It succeeds only when the swap_cgroup's record for this entry is the same |
| 2493 | * as the mem_cgroup's id of @from. |
| 2494 | * |
| 2495 | * Returns 0 on success, -EINVAL on failure. |
| 2496 | * |
| 2497 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
| 2498 | * both res and memsw, and called css_get(). |
| 2499 | */ |
| 2500 | static int mem_cgroup_move_swap_account(swp_entry_t entry, |
| 2501 | struct mem_cgroup *from, struct mem_cgroup *to) |
| 2502 | { |
| 2503 | unsigned short old_id, new_id; |
| 2504 | |
| 2505 | old_id = mem_cgroup_id(from); |
| 2506 | new_id = mem_cgroup_id(to); |
| 2507 | |
| 2508 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { |
| 2509 | mem_cgroup_swap_statistics(from, false); |
| 2510 | mem_cgroup_swap_statistics(to, true); |
| 2511 | return 0; |
| 2512 | } |
| 2513 | return -EINVAL; |
| 2514 | } |
| 2515 | #else |
| 2516 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, |
| 2517 | struct mem_cgroup *from, struct mem_cgroup *to) |
| 2518 | { |
| 2519 | return -EINVAL; |
| 2520 | } |
| 2521 | #endif |
| 2522 | |
| 2523 | static DEFINE_MUTEX(memcg_limit_mutex); |
| 2524 | |
| 2525 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
| 2526 | unsigned long limit) |
| 2527 | { |
| 2528 | unsigned long curusage; |
| 2529 | unsigned long oldusage; |
| 2530 | bool enlarge = false; |
| 2531 | int retry_count; |
| 2532 | int ret; |
| 2533 | |
| 2534 | /* |
| 2535 | * For keeping hierarchical_reclaim simple, how long we should retry |
| 2536 | * is depends on callers. We set our retry-count to be function |
| 2537 | * of # of children which we should visit in this loop. |
| 2538 | */ |
| 2539 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
| 2540 | mem_cgroup_count_children(memcg); |
| 2541 | |
| 2542 | oldusage = page_counter_read(&memcg->memory); |
| 2543 | |
| 2544 | do { |
| 2545 | if (signal_pending(current)) { |
| 2546 | ret = -EINTR; |
| 2547 | break; |
| 2548 | } |
| 2549 | |
| 2550 | mutex_lock(&memcg_limit_mutex); |
| 2551 | if (limit > memcg->memsw.limit) { |
| 2552 | mutex_unlock(&memcg_limit_mutex); |
| 2553 | ret = -EINVAL; |
| 2554 | break; |
| 2555 | } |
| 2556 | if (limit > memcg->memory.limit) |
| 2557 | enlarge = true; |
| 2558 | ret = page_counter_limit(&memcg->memory, limit); |
| 2559 | mutex_unlock(&memcg_limit_mutex); |
| 2560 | |
| 2561 | if (!ret) |
| 2562 | break; |
| 2563 | |
| 2564 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
| 2565 | |
| 2566 | curusage = page_counter_read(&memcg->memory); |
| 2567 | /* Usage is reduced ? */ |
| 2568 | if (curusage >= oldusage) |
| 2569 | retry_count--; |
| 2570 | else |
| 2571 | oldusage = curusage; |
| 2572 | } while (retry_count); |
| 2573 | |
| 2574 | if (!ret && enlarge) |
| 2575 | memcg_oom_recover(memcg); |
| 2576 | |
| 2577 | return ret; |
| 2578 | } |
| 2579 | |
| 2580 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
| 2581 | unsigned long limit) |
| 2582 | { |
| 2583 | unsigned long curusage; |
| 2584 | unsigned long oldusage; |
| 2585 | bool enlarge = false; |
| 2586 | int retry_count; |
| 2587 | int ret; |
| 2588 | |
| 2589 | /* see mem_cgroup_resize_res_limit */ |
| 2590 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
| 2591 | mem_cgroup_count_children(memcg); |
| 2592 | |
| 2593 | oldusage = page_counter_read(&memcg->memsw); |
| 2594 | |
| 2595 | do { |
| 2596 | if (signal_pending(current)) { |
| 2597 | ret = -EINTR; |
| 2598 | break; |
| 2599 | } |
| 2600 | |
| 2601 | mutex_lock(&memcg_limit_mutex); |
| 2602 | if (limit < memcg->memory.limit) { |
| 2603 | mutex_unlock(&memcg_limit_mutex); |
| 2604 | ret = -EINVAL; |
| 2605 | break; |
| 2606 | } |
| 2607 | if (limit > memcg->memsw.limit) |
| 2608 | enlarge = true; |
| 2609 | ret = page_counter_limit(&memcg->memsw, limit); |
| 2610 | mutex_unlock(&memcg_limit_mutex); |
| 2611 | |
| 2612 | if (!ret) |
| 2613 | break; |
| 2614 | |
| 2615 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
| 2616 | |
| 2617 | curusage = page_counter_read(&memcg->memsw); |
| 2618 | /* Usage is reduced ? */ |
| 2619 | if (curusage >= oldusage) |
| 2620 | retry_count--; |
| 2621 | else |
| 2622 | oldusage = curusage; |
| 2623 | } while (retry_count); |
| 2624 | |
| 2625 | if (!ret && enlarge) |
| 2626 | memcg_oom_recover(memcg); |
| 2627 | |
| 2628 | return ret; |
| 2629 | } |
| 2630 | |
| 2631 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
| 2632 | gfp_t gfp_mask, |
| 2633 | unsigned long *total_scanned) |
| 2634 | { |
| 2635 | unsigned long nr_reclaimed = 0; |
| 2636 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; |
| 2637 | unsigned long reclaimed; |
| 2638 | int loop = 0; |
| 2639 | struct mem_cgroup_tree_per_zone *mctz; |
| 2640 | unsigned long excess; |
| 2641 | unsigned long nr_scanned; |
| 2642 | |
| 2643 | if (order > 0) |
| 2644 | return 0; |
| 2645 | |
| 2646 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
| 2647 | /* |
| 2648 | * This loop can run a while, specially if mem_cgroup's continuously |
| 2649 | * keep exceeding their soft limit and putting the system under |
| 2650 | * pressure |
| 2651 | */ |
| 2652 | do { |
| 2653 | if (next_mz) |
| 2654 | mz = next_mz; |
| 2655 | else |
| 2656 | mz = mem_cgroup_largest_soft_limit_node(mctz); |
| 2657 | if (!mz) |
| 2658 | break; |
| 2659 | |
| 2660 | nr_scanned = 0; |
| 2661 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
| 2662 | gfp_mask, &nr_scanned); |
| 2663 | nr_reclaimed += reclaimed; |
| 2664 | *total_scanned += nr_scanned; |
| 2665 | spin_lock_irq(&mctz->lock); |
| 2666 | __mem_cgroup_remove_exceeded(mz, mctz); |
| 2667 | |
| 2668 | /* |
| 2669 | * If we failed to reclaim anything from this memory cgroup |
| 2670 | * it is time to move on to the next cgroup |
| 2671 | */ |
| 2672 | next_mz = NULL; |
| 2673 | if (!reclaimed) |
| 2674 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); |
| 2675 | |
| 2676 | excess = soft_limit_excess(mz->memcg); |
| 2677 | /* |
| 2678 | * One school of thought says that we should not add |
| 2679 | * back the node to the tree if reclaim returns 0. |
| 2680 | * But our reclaim could return 0, simply because due |
| 2681 | * to priority we are exposing a smaller subset of |
| 2682 | * memory to reclaim from. Consider this as a longer |
| 2683 | * term TODO. |
| 2684 | */ |
| 2685 | /* If excess == 0, no tree ops */ |
| 2686 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
| 2687 | spin_unlock_irq(&mctz->lock); |
| 2688 | css_put(&mz->memcg->css); |
| 2689 | loop++; |
| 2690 | /* |
| 2691 | * Could not reclaim anything and there are no more |
| 2692 | * mem cgroups to try or we seem to be looping without |
| 2693 | * reclaiming anything. |
| 2694 | */ |
| 2695 | if (!nr_reclaimed && |
| 2696 | (next_mz == NULL || |
| 2697 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) |
| 2698 | break; |
| 2699 | } while (!nr_reclaimed); |
| 2700 | if (next_mz) |
| 2701 | css_put(&next_mz->memcg->css); |
| 2702 | return nr_reclaimed; |
| 2703 | } |
| 2704 | |
| 2705 | /* |
| 2706 | * Test whether @memcg has children, dead or alive. Note that this |
| 2707 | * function doesn't care whether @memcg has use_hierarchy enabled and |
| 2708 | * returns %true if there are child csses according to the cgroup |
| 2709 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. |
| 2710 | */ |
| 2711 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
| 2712 | { |
| 2713 | bool ret; |
| 2714 | |
| 2715 | /* |
| 2716 | * The lock does not prevent addition or deletion of children, but |
| 2717 | * it prevents a new child from being initialized based on this |
| 2718 | * parent in css_online(), so it's enough to decide whether |
| 2719 | * hierarchically inherited attributes can still be changed or not. |
| 2720 | */ |
| 2721 | lockdep_assert_held(&memcg_create_mutex); |
| 2722 | |
| 2723 | rcu_read_lock(); |
| 2724 | ret = css_next_child(NULL, &memcg->css); |
| 2725 | rcu_read_unlock(); |
| 2726 | return ret; |
| 2727 | } |
| 2728 | |
| 2729 | /* |
| 2730 | * Reclaims as many pages from the given memcg as possible and moves |
| 2731 | * the rest to the parent. |
| 2732 | * |
| 2733 | * Caller is responsible for holding css reference for memcg. |
| 2734 | */ |
| 2735 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) |
| 2736 | { |
| 2737 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
| 2738 | |
| 2739 | /* we call try-to-free pages for make this cgroup empty */ |
| 2740 | lru_add_drain_all(); |
| 2741 | /* try to free all pages in this cgroup */ |
| 2742 | while (nr_retries && page_counter_read(&memcg->memory)) { |
| 2743 | int progress; |
| 2744 | |
| 2745 | if (signal_pending(current)) |
| 2746 | return -EINTR; |
| 2747 | |
| 2748 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
| 2749 | GFP_KERNEL, true); |
| 2750 | if (!progress) { |
| 2751 | nr_retries--; |
| 2752 | /* maybe some writeback is necessary */ |
| 2753 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
| 2754 | } |
| 2755 | |
| 2756 | } |
| 2757 | |
| 2758 | return 0; |
| 2759 | } |
| 2760 | |
| 2761 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
| 2762 | char *buf, size_t nbytes, |
| 2763 | loff_t off) |
| 2764 | { |
| 2765 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 2766 | |
| 2767 | if (mem_cgroup_is_root(memcg)) |
| 2768 | return -EINVAL; |
| 2769 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
| 2770 | } |
| 2771 | |
| 2772 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
| 2773 | struct cftype *cft) |
| 2774 | { |
| 2775 | return mem_cgroup_from_css(css)->use_hierarchy; |
| 2776 | } |
| 2777 | |
| 2778 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
| 2779 | struct cftype *cft, u64 val) |
| 2780 | { |
| 2781 | int retval = 0; |
| 2782 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 2783 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
| 2784 | |
| 2785 | mutex_lock(&memcg_create_mutex); |
| 2786 | |
| 2787 | if (memcg->use_hierarchy == val) |
| 2788 | goto out; |
| 2789 | |
| 2790 | /* |
| 2791 | * If parent's use_hierarchy is set, we can't make any modifications |
| 2792 | * in the child subtrees. If it is unset, then the change can |
| 2793 | * occur, provided the current cgroup has no children. |
| 2794 | * |
| 2795 | * For the root cgroup, parent_mem is NULL, we allow value to be |
| 2796 | * set if there are no children. |
| 2797 | */ |
| 2798 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
| 2799 | (val == 1 || val == 0)) { |
| 2800 | if (!memcg_has_children(memcg)) |
| 2801 | memcg->use_hierarchy = val; |
| 2802 | else |
| 2803 | retval = -EBUSY; |
| 2804 | } else |
| 2805 | retval = -EINVAL; |
| 2806 | |
| 2807 | out: |
| 2808 | mutex_unlock(&memcg_create_mutex); |
| 2809 | |
| 2810 | return retval; |
| 2811 | } |
| 2812 | |
| 2813 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
| 2814 | enum mem_cgroup_stat_index idx) |
| 2815 | { |
| 2816 | struct mem_cgroup *iter; |
| 2817 | unsigned long val = 0; |
| 2818 | |
| 2819 | for_each_mem_cgroup_tree(iter, memcg) |
| 2820 | val += mem_cgroup_read_stat(iter, idx); |
| 2821 | |
| 2822 | return val; |
| 2823 | } |
| 2824 | |
| 2825 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
| 2826 | { |
| 2827 | unsigned long val; |
| 2828 | |
| 2829 | if (mem_cgroup_is_root(memcg)) { |
| 2830 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); |
| 2831 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); |
| 2832 | if (swap) |
| 2833 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); |
| 2834 | } else { |
| 2835 | if (!swap) |
| 2836 | val = page_counter_read(&memcg->memory); |
| 2837 | else |
| 2838 | val = page_counter_read(&memcg->memsw); |
| 2839 | } |
| 2840 | return val; |
| 2841 | } |
| 2842 | |
| 2843 | enum { |
| 2844 | RES_USAGE, |
| 2845 | RES_LIMIT, |
| 2846 | RES_MAX_USAGE, |
| 2847 | RES_FAILCNT, |
| 2848 | RES_SOFT_LIMIT, |
| 2849 | }; |
| 2850 | |
| 2851 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
| 2852 | struct cftype *cft) |
| 2853 | { |
| 2854 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 2855 | struct page_counter *counter; |
| 2856 | |
| 2857 | switch (MEMFILE_TYPE(cft->private)) { |
| 2858 | case _MEM: |
| 2859 | counter = &memcg->memory; |
| 2860 | break; |
| 2861 | case _MEMSWAP: |
| 2862 | counter = &memcg->memsw; |
| 2863 | break; |
| 2864 | case _KMEM: |
| 2865 | counter = &memcg->kmem; |
| 2866 | break; |
| 2867 | default: |
| 2868 | BUG(); |
| 2869 | } |
| 2870 | |
| 2871 | switch (MEMFILE_ATTR(cft->private)) { |
| 2872 | case RES_USAGE: |
| 2873 | if (counter == &memcg->memory) |
| 2874 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
| 2875 | if (counter == &memcg->memsw) |
| 2876 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
| 2877 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
| 2878 | case RES_LIMIT: |
| 2879 | return (u64)counter->limit * PAGE_SIZE; |
| 2880 | case RES_MAX_USAGE: |
| 2881 | return (u64)counter->watermark * PAGE_SIZE; |
| 2882 | case RES_FAILCNT: |
| 2883 | return counter->failcnt; |
| 2884 | case RES_SOFT_LIMIT: |
| 2885 | return (u64)memcg->soft_limit * PAGE_SIZE; |
| 2886 | default: |
| 2887 | BUG(); |
| 2888 | } |
| 2889 | } |
| 2890 | |
| 2891 | #ifdef CONFIG_MEMCG_KMEM |
| 2892 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
| 2893 | unsigned long nr_pages) |
| 2894 | { |
| 2895 | int err = 0; |
| 2896 | int memcg_id; |
| 2897 | |
| 2898 | BUG_ON(memcg->kmemcg_id >= 0); |
| 2899 | BUG_ON(memcg->kmem_acct_activated); |
| 2900 | BUG_ON(memcg->kmem_acct_active); |
| 2901 | |
| 2902 | /* |
| 2903 | * For simplicity, we won't allow this to be disabled. It also can't |
| 2904 | * be changed if the cgroup has children already, or if tasks had |
| 2905 | * already joined. |
| 2906 | * |
| 2907 | * If tasks join before we set the limit, a person looking at |
| 2908 | * kmem.usage_in_bytes will have no way to determine when it took |
| 2909 | * place, which makes the value quite meaningless. |
| 2910 | * |
| 2911 | * After it first became limited, changes in the value of the limit are |
| 2912 | * of course permitted. |
| 2913 | */ |
| 2914 | mutex_lock(&memcg_create_mutex); |
| 2915 | if (cgroup_is_populated(memcg->css.cgroup) || |
| 2916 | (memcg->use_hierarchy && memcg_has_children(memcg))) |
| 2917 | err = -EBUSY; |
| 2918 | mutex_unlock(&memcg_create_mutex); |
| 2919 | if (err) |
| 2920 | goto out; |
| 2921 | |
| 2922 | memcg_id = memcg_alloc_cache_id(); |
| 2923 | if (memcg_id < 0) { |
| 2924 | err = memcg_id; |
| 2925 | goto out; |
| 2926 | } |
| 2927 | |
| 2928 | /* |
| 2929 | * We couldn't have accounted to this cgroup, because it hasn't got |
| 2930 | * activated yet, so this should succeed. |
| 2931 | */ |
| 2932 | err = page_counter_limit(&memcg->kmem, nr_pages); |
| 2933 | VM_BUG_ON(err); |
| 2934 | |
| 2935 | static_key_slow_inc(&memcg_kmem_enabled_key); |
| 2936 | /* |
| 2937 | * A memory cgroup is considered kmem-active as soon as it gets |
| 2938 | * kmemcg_id. Setting the id after enabling static branching will |
| 2939 | * guarantee no one starts accounting before all call sites are |
| 2940 | * patched. |
| 2941 | */ |
| 2942 | memcg->kmemcg_id = memcg_id; |
| 2943 | memcg->kmem_acct_activated = true; |
| 2944 | memcg->kmem_acct_active = true; |
| 2945 | out: |
| 2946 | return err; |
| 2947 | } |
| 2948 | |
| 2949 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
| 2950 | unsigned long limit) |
| 2951 | { |
| 2952 | int ret; |
| 2953 | |
| 2954 | mutex_lock(&memcg_limit_mutex); |
| 2955 | if (!memcg_kmem_is_active(memcg)) |
| 2956 | ret = memcg_activate_kmem(memcg, limit); |
| 2957 | else |
| 2958 | ret = page_counter_limit(&memcg->kmem, limit); |
| 2959 | mutex_unlock(&memcg_limit_mutex); |
| 2960 | return ret; |
| 2961 | } |
| 2962 | |
| 2963 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
| 2964 | { |
| 2965 | int ret = 0; |
| 2966 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
| 2967 | |
| 2968 | if (!parent) |
| 2969 | return 0; |
| 2970 | |
| 2971 | mutex_lock(&memcg_limit_mutex); |
| 2972 | /* |
| 2973 | * If the parent cgroup is not kmem-active now, it cannot be activated |
| 2974 | * after this point, because it has at least one child already. |
| 2975 | */ |
| 2976 | if (memcg_kmem_is_active(parent)) |
| 2977 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
| 2978 | mutex_unlock(&memcg_limit_mutex); |
| 2979 | return ret; |
| 2980 | } |
| 2981 | #else |
| 2982 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
| 2983 | unsigned long limit) |
| 2984 | { |
| 2985 | return -EINVAL; |
| 2986 | } |
| 2987 | #endif /* CONFIG_MEMCG_KMEM */ |
| 2988 | |
| 2989 | /* |
| 2990 | * The user of this function is... |
| 2991 | * RES_LIMIT. |
| 2992 | */ |
| 2993 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
| 2994 | char *buf, size_t nbytes, loff_t off) |
| 2995 | { |
| 2996 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 2997 | unsigned long nr_pages; |
| 2998 | int ret; |
| 2999 | |
| 3000 | buf = strstrip(buf); |
| 3001 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
| 3002 | if (ret) |
| 3003 | return ret; |
| 3004 | |
| 3005 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
| 3006 | case RES_LIMIT: |
| 3007 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
| 3008 | ret = -EINVAL; |
| 3009 | break; |
| 3010 | } |
| 3011 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
| 3012 | case _MEM: |
| 3013 | ret = mem_cgroup_resize_limit(memcg, nr_pages); |
| 3014 | break; |
| 3015 | case _MEMSWAP: |
| 3016 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); |
| 3017 | break; |
| 3018 | case _KMEM: |
| 3019 | ret = memcg_update_kmem_limit(memcg, nr_pages); |
| 3020 | break; |
| 3021 | } |
| 3022 | break; |
| 3023 | case RES_SOFT_LIMIT: |
| 3024 | memcg->soft_limit = nr_pages; |
| 3025 | ret = 0; |
| 3026 | break; |
| 3027 | } |
| 3028 | return ret ?: nbytes; |
| 3029 | } |
| 3030 | |
| 3031 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
| 3032 | size_t nbytes, loff_t off) |
| 3033 | { |
| 3034 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 3035 | struct page_counter *counter; |
| 3036 | |
| 3037 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
| 3038 | case _MEM: |
| 3039 | counter = &memcg->memory; |
| 3040 | break; |
| 3041 | case _MEMSWAP: |
| 3042 | counter = &memcg->memsw; |
| 3043 | break; |
| 3044 | case _KMEM: |
| 3045 | counter = &memcg->kmem; |
| 3046 | break; |
| 3047 | default: |
| 3048 | BUG(); |
| 3049 | } |
| 3050 | |
| 3051 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
| 3052 | case RES_MAX_USAGE: |
| 3053 | page_counter_reset_watermark(counter); |
| 3054 | break; |
| 3055 | case RES_FAILCNT: |
| 3056 | counter->failcnt = 0; |
| 3057 | break; |
| 3058 | default: |
| 3059 | BUG(); |
| 3060 | } |
| 3061 | |
| 3062 | return nbytes; |
| 3063 | } |
| 3064 | |
| 3065 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
| 3066 | struct cftype *cft) |
| 3067 | { |
| 3068 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
| 3069 | } |
| 3070 | |
| 3071 | #ifdef CONFIG_MMU |
| 3072 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
| 3073 | struct cftype *cft, u64 val) |
| 3074 | { |
| 3075 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 3076 | |
| 3077 | if (val & ~MOVE_MASK) |
| 3078 | return -EINVAL; |
| 3079 | |
| 3080 | /* |
| 3081 | * No kind of locking is needed in here, because ->can_attach() will |
| 3082 | * check this value once in the beginning of the process, and then carry |
| 3083 | * on with stale data. This means that changes to this value will only |
| 3084 | * affect task migrations starting after the change. |
| 3085 | */ |
| 3086 | memcg->move_charge_at_immigrate = val; |
| 3087 | return 0; |
| 3088 | } |
| 3089 | #else |
| 3090 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
| 3091 | struct cftype *cft, u64 val) |
| 3092 | { |
| 3093 | return -ENOSYS; |
| 3094 | } |
| 3095 | #endif |
| 3096 | |
| 3097 | #ifdef CONFIG_NUMA |
| 3098 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
| 3099 | { |
| 3100 | struct numa_stat { |
| 3101 | const char *name; |
| 3102 | unsigned int lru_mask; |
| 3103 | }; |
| 3104 | |
| 3105 | static const struct numa_stat stats[] = { |
| 3106 | { "total", LRU_ALL }, |
| 3107 | { "file", LRU_ALL_FILE }, |
| 3108 | { "anon", LRU_ALL_ANON }, |
| 3109 | { "unevictable", BIT(LRU_UNEVICTABLE) }, |
| 3110 | }; |
| 3111 | const struct numa_stat *stat; |
| 3112 | int nid; |
| 3113 | unsigned long nr; |
| 3114 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 3115 | |
| 3116 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
| 3117 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); |
| 3118 | seq_printf(m, "%s=%lu", stat->name, nr); |
| 3119 | for_each_node_state(nid, N_MEMORY) { |
| 3120 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
| 3121 | stat->lru_mask); |
| 3122 | seq_printf(m, " N%d=%lu", nid, nr); |
| 3123 | } |
| 3124 | seq_putc(m, '\n'); |
| 3125 | } |
| 3126 | |
| 3127 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
| 3128 | struct mem_cgroup *iter; |
| 3129 | |
| 3130 | nr = 0; |
| 3131 | for_each_mem_cgroup_tree(iter, memcg) |
| 3132 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); |
| 3133 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); |
| 3134 | for_each_node_state(nid, N_MEMORY) { |
| 3135 | nr = 0; |
| 3136 | for_each_mem_cgroup_tree(iter, memcg) |
| 3137 | nr += mem_cgroup_node_nr_lru_pages( |
| 3138 | iter, nid, stat->lru_mask); |
| 3139 | seq_printf(m, " N%d=%lu", nid, nr); |
| 3140 | } |
| 3141 | seq_putc(m, '\n'); |
| 3142 | } |
| 3143 | |
| 3144 | return 0; |
| 3145 | } |
| 3146 | #endif /* CONFIG_NUMA */ |
| 3147 | |
| 3148 | static int memcg_stat_show(struct seq_file *m, void *v) |
| 3149 | { |
| 3150 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 3151 | unsigned long memory, memsw; |
| 3152 | struct mem_cgroup *mi; |
| 3153 | unsigned int i; |
| 3154 | |
| 3155 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
| 3156 | MEM_CGROUP_STAT_NSTATS); |
| 3157 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != |
| 3158 | MEM_CGROUP_EVENTS_NSTATS); |
| 3159 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
| 3160 | |
| 3161 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
| 3162 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
| 3163 | continue; |
| 3164 | seq_printf(m, "%s %lu\n", mem_cgroup_stat_names[i], |
| 3165 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
| 3166 | } |
| 3167 | |
| 3168 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
| 3169 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], |
| 3170 | mem_cgroup_read_events(memcg, i)); |
| 3171 | |
| 3172 | for (i = 0; i < NR_LRU_LISTS; i++) |
| 3173 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], |
| 3174 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); |
| 3175 | |
| 3176 | /* Hierarchical information */ |
| 3177 | memory = memsw = PAGE_COUNTER_MAX; |
| 3178 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { |
| 3179 | memory = min(memory, mi->memory.limit); |
| 3180 | memsw = min(memsw, mi->memsw.limit); |
| 3181 | } |
| 3182 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
| 3183 | (u64)memory * PAGE_SIZE); |
| 3184 | if (do_swap_account) |
| 3185 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
| 3186 | (u64)memsw * PAGE_SIZE); |
| 3187 | |
| 3188 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
| 3189 | unsigned long long val = 0; |
| 3190 | |
| 3191 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
| 3192 | continue; |
| 3193 | for_each_mem_cgroup_tree(mi, memcg) |
| 3194 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; |
| 3195 | seq_printf(m, "total_%s %llu\n", mem_cgroup_stat_names[i], val); |
| 3196 | } |
| 3197 | |
| 3198 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
| 3199 | unsigned long long val = 0; |
| 3200 | |
| 3201 | for_each_mem_cgroup_tree(mi, memcg) |
| 3202 | val += mem_cgroup_read_events(mi, i); |
| 3203 | seq_printf(m, "total_%s %llu\n", |
| 3204 | mem_cgroup_events_names[i], val); |
| 3205 | } |
| 3206 | |
| 3207 | for (i = 0; i < NR_LRU_LISTS; i++) { |
| 3208 | unsigned long long val = 0; |
| 3209 | |
| 3210 | for_each_mem_cgroup_tree(mi, memcg) |
| 3211 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; |
| 3212 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); |
| 3213 | } |
| 3214 | |
| 3215 | #ifdef CONFIG_DEBUG_VM |
| 3216 | { |
| 3217 | int nid, zid; |
| 3218 | struct mem_cgroup_per_zone *mz; |
| 3219 | struct zone_reclaim_stat *rstat; |
| 3220 | unsigned long recent_rotated[2] = {0, 0}; |
| 3221 | unsigned long recent_scanned[2] = {0, 0}; |
| 3222 | |
| 3223 | for_each_online_node(nid) |
| 3224 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 3225 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
| 3226 | rstat = &mz->lruvec.reclaim_stat; |
| 3227 | |
| 3228 | recent_rotated[0] += rstat->recent_rotated[0]; |
| 3229 | recent_rotated[1] += rstat->recent_rotated[1]; |
| 3230 | recent_scanned[0] += rstat->recent_scanned[0]; |
| 3231 | recent_scanned[1] += rstat->recent_scanned[1]; |
| 3232 | } |
| 3233 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
| 3234 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); |
| 3235 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); |
| 3236 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); |
| 3237 | } |
| 3238 | #endif |
| 3239 | |
| 3240 | return 0; |
| 3241 | } |
| 3242 | static u64 mem_cgroup_vmpressure_read(struct cgroup_subsys_state *css, |
| 3243 | struct cftype *cft) |
| 3244 | { |
| 3245 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 3246 | struct vmpressure *vmpr = memcg_to_vmpressure(memcg); |
| 3247 | unsigned long vmpressure; |
| 3248 | |
| 3249 | vmpressure = vmpr->pressure; |
| 3250 | |
| 3251 | return vmpressure; |
| 3252 | } |
| 3253 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
| 3254 | struct cftype *cft) |
| 3255 | { |
| 3256 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 3257 | |
| 3258 | return mem_cgroup_swappiness(memcg); |
| 3259 | } |
| 3260 | |
| 3261 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
| 3262 | struct cftype *cft, u64 val) |
| 3263 | { |
| 3264 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 3265 | |
| 3266 | if (val > 100) |
| 3267 | return -EINVAL; |
| 3268 | |
| 3269 | if (css->parent) |
| 3270 | memcg->swappiness = val; |
| 3271 | else |
| 3272 | vm_swappiness = val; |
| 3273 | |
| 3274 | return 0; |
| 3275 | } |
| 3276 | |
| 3277 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
| 3278 | { |
| 3279 | struct mem_cgroup_threshold_ary *t; |
| 3280 | unsigned long usage; |
| 3281 | int i; |
| 3282 | |
| 3283 | rcu_read_lock(); |
| 3284 | if (!swap) |
| 3285 | t = rcu_dereference(memcg->thresholds.primary); |
| 3286 | else |
| 3287 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
| 3288 | |
| 3289 | if (!t) |
| 3290 | goto unlock; |
| 3291 | |
| 3292 | usage = mem_cgroup_usage(memcg, swap); |
| 3293 | |
| 3294 | /* |
| 3295 | * current_threshold points to threshold just below or equal to usage. |
| 3296 | * If it's not true, a threshold was crossed after last |
| 3297 | * call of __mem_cgroup_threshold(). |
| 3298 | */ |
| 3299 | i = t->current_threshold; |
| 3300 | |
| 3301 | /* |
| 3302 | * Iterate backward over array of thresholds starting from |
| 3303 | * current_threshold and check if a threshold is crossed. |
| 3304 | * If none of thresholds below usage is crossed, we read |
| 3305 | * only one element of the array here. |
| 3306 | */ |
| 3307 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) |
| 3308 | eventfd_signal(t->entries[i].eventfd, 1); |
| 3309 | |
| 3310 | /* i = current_threshold + 1 */ |
| 3311 | i++; |
| 3312 | |
| 3313 | /* |
| 3314 | * Iterate forward over array of thresholds starting from |
| 3315 | * current_threshold+1 and check if a threshold is crossed. |
| 3316 | * If none of thresholds above usage is crossed, we read |
| 3317 | * only one element of the array here. |
| 3318 | */ |
| 3319 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) |
| 3320 | eventfd_signal(t->entries[i].eventfd, 1); |
| 3321 | |
| 3322 | /* Update current_threshold */ |
| 3323 | t->current_threshold = i - 1; |
| 3324 | unlock: |
| 3325 | rcu_read_unlock(); |
| 3326 | } |
| 3327 | |
| 3328 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) |
| 3329 | { |
| 3330 | while (memcg) { |
| 3331 | __mem_cgroup_threshold(memcg, false); |
| 3332 | if (do_swap_account) |
| 3333 | __mem_cgroup_threshold(memcg, true); |
| 3334 | |
| 3335 | memcg = parent_mem_cgroup(memcg); |
| 3336 | } |
| 3337 | } |
| 3338 | |
| 3339 | static int compare_thresholds(const void *a, const void *b) |
| 3340 | { |
| 3341 | const struct mem_cgroup_threshold *_a = a; |
| 3342 | const struct mem_cgroup_threshold *_b = b; |
| 3343 | |
| 3344 | if (_a->threshold > _b->threshold) |
| 3345 | return 1; |
| 3346 | |
| 3347 | if (_a->threshold < _b->threshold) |
| 3348 | return -1; |
| 3349 | |
| 3350 | return 0; |
| 3351 | } |
| 3352 | |
| 3353 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
| 3354 | { |
| 3355 | struct mem_cgroup_eventfd_list *ev; |
| 3356 | |
| 3357 | spin_lock(&memcg_oom_lock); |
| 3358 | |
| 3359 | list_for_each_entry(ev, &memcg->oom_notify, list) |
| 3360 | eventfd_signal(ev->eventfd, 1); |
| 3361 | |
| 3362 | spin_unlock(&memcg_oom_lock); |
| 3363 | return 0; |
| 3364 | } |
| 3365 | |
| 3366 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
| 3367 | { |
| 3368 | struct mem_cgroup *iter; |
| 3369 | |
| 3370 | for_each_mem_cgroup_tree(iter, memcg) |
| 3371 | mem_cgroup_oom_notify_cb(iter); |
| 3372 | } |
| 3373 | |
| 3374 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
| 3375 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
| 3376 | { |
| 3377 | struct mem_cgroup_thresholds *thresholds; |
| 3378 | struct mem_cgroup_threshold_ary *new; |
| 3379 | unsigned long threshold; |
| 3380 | unsigned long usage; |
| 3381 | int i, size, ret; |
| 3382 | |
| 3383 | ret = page_counter_memparse(args, "-1", &threshold); |
| 3384 | if (ret) |
| 3385 | return ret; |
| 3386 | |
| 3387 | mutex_lock(&memcg->thresholds_lock); |
| 3388 | |
| 3389 | if (type == _MEM) { |
| 3390 | thresholds = &memcg->thresholds; |
| 3391 | usage = mem_cgroup_usage(memcg, false); |
| 3392 | } else if (type == _MEMSWAP) { |
| 3393 | thresholds = &memcg->memsw_thresholds; |
| 3394 | usage = mem_cgroup_usage(memcg, true); |
| 3395 | } else |
| 3396 | BUG(); |
| 3397 | |
| 3398 | /* Check if a threshold crossed before adding a new one */ |
| 3399 | if (thresholds->primary) |
| 3400 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 3401 | |
| 3402 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
| 3403 | |
| 3404 | /* Allocate memory for new array of thresholds */ |
| 3405 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
| 3406 | GFP_KERNEL); |
| 3407 | if (!new) { |
| 3408 | ret = -ENOMEM; |
| 3409 | goto unlock; |
| 3410 | } |
| 3411 | new->size = size; |
| 3412 | |
| 3413 | /* Copy thresholds (if any) to new array */ |
| 3414 | if (thresholds->primary) { |
| 3415 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * |
| 3416 | sizeof(struct mem_cgroup_threshold)); |
| 3417 | } |
| 3418 | |
| 3419 | /* Add new threshold */ |
| 3420 | new->entries[size - 1].eventfd = eventfd; |
| 3421 | new->entries[size - 1].threshold = threshold; |
| 3422 | |
| 3423 | /* Sort thresholds. Registering of new threshold isn't time-critical */ |
| 3424 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
| 3425 | compare_thresholds, NULL); |
| 3426 | |
| 3427 | /* Find current threshold */ |
| 3428 | new->current_threshold = -1; |
| 3429 | for (i = 0; i < size; i++) { |
| 3430 | if (new->entries[i].threshold <= usage) { |
| 3431 | /* |
| 3432 | * new->current_threshold will not be used until |
| 3433 | * rcu_assign_pointer(), so it's safe to increment |
| 3434 | * it here. |
| 3435 | */ |
| 3436 | ++new->current_threshold; |
| 3437 | } else |
| 3438 | break; |
| 3439 | } |
| 3440 | |
| 3441 | /* Free old spare buffer and save old primary buffer as spare */ |
| 3442 | kfree(thresholds->spare); |
| 3443 | thresholds->spare = thresholds->primary; |
| 3444 | |
| 3445 | rcu_assign_pointer(thresholds->primary, new); |
| 3446 | |
| 3447 | /* To be sure that nobody uses thresholds */ |
| 3448 | synchronize_rcu(); |
| 3449 | |
| 3450 | unlock: |
| 3451 | mutex_unlock(&memcg->thresholds_lock); |
| 3452 | |
| 3453 | return ret; |
| 3454 | } |
| 3455 | |
| 3456 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
| 3457 | struct eventfd_ctx *eventfd, const char *args) |
| 3458 | { |
| 3459 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
| 3460 | } |
| 3461 | |
| 3462 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
| 3463 | struct eventfd_ctx *eventfd, const char *args) |
| 3464 | { |
| 3465 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
| 3466 | } |
| 3467 | |
| 3468 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
| 3469 | struct eventfd_ctx *eventfd, enum res_type type) |
| 3470 | { |
| 3471 | struct mem_cgroup_thresholds *thresholds; |
| 3472 | struct mem_cgroup_threshold_ary *new; |
| 3473 | unsigned long usage; |
| 3474 | int i, j, size; |
| 3475 | |
| 3476 | mutex_lock(&memcg->thresholds_lock); |
| 3477 | |
| 3478 | if (type == _MEM) { |
| 3479 | thresholds = &memcg->thresholds; |
| 3480 | usage = mem_cgroup_usage(memcg, false); |
| 3481 | } else if (type == _MEMSWAP) { |
| 3482 | thresholds = &memcg->memsw_thresholds; |
| 3483 | usage = mem_cgroup_usage(memcg, true); |
| 3484 | } else |
| 3485 | BUG(); |
| 3486 | |
| 3487 | if (!thresholds->primary) |
| 3488 | goto unlock; |
| 3489 | |
| 3490 | /* Check if a threshold crossed before removing */ |
| 3491 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 3492 | |
| 3493 | /* Calculate new number of threshold */ |
| 3494 | size = 0; |
| 3495 | for (i = 0; i < thresholds->primary->size; i++) { |
| 3496 | if (thresholds->primary->entries[i].eventfd != eventfd) |
| 3497 | size++; |
| 3498 | } |
| 3499 | |
| 3500 | new = thresholds->spare; |
| 3501 | |
| 3502 | /* Set thresholds array to NULL if we don't have thresholds */ |
| 3503 | if (!size) { |
| 3504 | kfree(new); |
| 3505 | new = NULL; |
| 3506 | goto swap_buffers; |
| 3507 | } |
| 3508 | |
| 3509 | new->size = size; |
| 3510 | |
| 3511 | /* Copy thresholds and find current threshold */ |
| 3512 | new->current_threshold = -1; |
| 3513 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { |
| 3514 | if (thresholds->primary->entries[i].eventfd == eventfd) |
| 3515 | continue; |
| 3516 | |
| 3517 | new->entries[j] = thresholds->primary->entries[i]; |
| 3518 | if (new->entries[j].threshold <= usage) { |
| 3519 | /* |
| 3520 | * new->current_threshold will not be used |
| 3521 | * until rcu_assign_pointer(), so it's safe to increment |
| 3522 | * it here. |
| 3523 | */ |
| 3524 | ++new->current_threshold; |
| 3525 | } |
| 3526 | j++; |
| 3527 | } |
| 3528 | |
| 3529 | swap_buffers: |
| 3530 | /* Swap primary and spare array */ |
| 3531 | thresholds->spare = thresholds->primary; |
| 3532 | |
| 3533 | rcu_assign_pointer(thresholds->primary, new); |
| 3534 | |
| 3535 | /* To be sure that nobody uses thresholds */ |
| 3536 | synchronize_rcu(); |
| 3537 | |
| 3538 | /* If all events are unregistered, free the spare array */ |
| 3539 | if (!new) { |
| 3540 | kfree(thresholds->spare); |
| 3541 | thresholds->spare = NULL; |
| 3542 | } |
| 3543 | unlock: |
| 3544 | mutex_unlock(&memcg->thresholds_lock); |
| 3545 | } |
| 3546 | |
| 3547 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
| 3548 | struct eventfd_ctx *eventfd) |
| 3549 | { |
| 3550 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
| 3551 | } |
| 3552 | |
| 3553 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
| 3554 | struct eventfd_ctx *eventfd) |
| 3555 | { |
| 3556 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
| 3557 | } |
| 3558 | |
| 3559 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
| 3560 | struct eventfd_ctx *eventfd, const char *args) |
| 3561 | { |
| 3562 | struct mem_cgroup_eventfd_list *event; |
| 3563 | |
| 3564 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
| 3565 | if (!event) |
| 3566 | return -ENOMEM; |
| 3567 | |
| 3568 | spin_lock(&memcg_oom_lock); |
| 3569 | |
| 3570 | event->eventfd = eventfd; |
| 3571 | list_add(&event->list, &memcg->oom_notify); |
| 3572 | |
| 3573 | /* already in OOM ? */ |
| 3574 | if (memcg->under_oom) |
| 3575 | eventfd_signal(eventfd, 1); |
| 3576 | spin_unlock(&memcg_oom_lock); |
| 3577 | |
| 3578 | return 0; |
| 3579 | } |
| 3580 | |
| 3581 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
| 3582 | struct eventfd_ctx *eventfd) |
| 3583 | { |
| 3584 | struct mem_cgroup_eventfd_list *ev, *tmp; |
| 3585 | |
| 3586 | spin_lock(&memcg_oom_lock); |
| 3587 | |
| 3588 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
| 3589 | if (ev->eventfd == eventfd) { |
| 3590 | list_del(&ev->list); |
| 3591 | kfree(ev); |
| 3592 | } |
| 3593 | } |
| 3594 | |
| 3595 | spin_unlock(&memcg_oom_lock); |
| 3596 | } |
| 3597 | |
| 3598 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
| 3599 | { |
| 3600 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
| 3601 | |
| 3602 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
| 3603 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
| 3604 | return 0; |
| 3605 | } |
| 3606 | |
| 3607 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
| 3608 | struct cftype *cft, u64 val) |
| 3609 | { |
| 3610 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 3611 | |
| 3612 | /* cannot set to root cgroup and only 0 and 1 are allowed */ |
| 3613 | if (!css->parent || !((val == 0) || (val == 1))) |
| 3614 | return -EINVAL; |
| 3615 | |
| 3616 | memcg->oom_kill_disable = val; |
| 3617 | if (!val) |
| 3618 | memcg_oom_recover(memcg); |
| 3619 | |
| 3620 | return 0; |
| 3621 | } |
| 3622 | |
| 3623 | #ifdef CONFIG_MEMCG_KMEM |
| 3624 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
| 3625 | { |
| 3626 | int ret; |
| 3627 | |
| 3628 | ret = memcg_propagate_kmem(memcg); |
| 3629 | if (ret) |
| 3630 | return ret; |
| 3631 | |
| 3632 | return mem_cgroup_sockets_init(memcg, ss); |
| 3633 | } |
| 3634 | |
| 3635 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
| 3636 | { |
| 3637 | struct cgroup_subsys_state *css; |
| 3638 | struct mem_cgroup *parent, *child; |
| 3639 | int kmemcg_id; |
| 3640 | |
| 3641 | if (!memcg->kmem_acct_active) |
| 3642 | return; |
| 3643 | |
| 3644 | /* |
| 3645 | * Clear the 'active' flag before clearing memcg_caches arrays entries. |
| 3646 | * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it |
| 3647 | * guarantees no cache will be created for this cgroup after we are |
| 3648 | * done (see memcg_create_kmem_cache()). |
| 3649 | */ |
| 3650 | memcg->kmem_acct_active = false; |
| 3651 | |
| 3652 | memcg_deactivate_kmem_caches(memcg); |
| 3653 | |
| 3654 | kmemcg_id = memcg->kmemcg_id; |
| 3655 | BUG_ON(kmemcg_id < 0); |
| 3656 | |
| 3657 | parent = parent_mem_cgroup(memcg); |
| 3658 | if (!parent) |
| 3659 | parent = root_mem_cgroup; |
| 3660 | |
| 3661 | /* |
| 3662 | * Change kmemcg_id of this cgroup and all its descendants to the |
| 3663 | * parent's id, and then move all entries from this cgroup's list_lrus |
| 3664 | * to ones of the parent. After we have finished, all list_lrus |
| 3665 | * corresponding to this cgroup are guaranteed to remain empty. The |
| 3666 | * ordering is imposed by list_lru_node->lock taken by |
| 3667 | * memcg_drain_all_list_lrus(). |
| 3668 | */ |
| 3669 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
| 3670 | css_for_each_descendant_pre(css, &memcg->css) { |
| 3671 | child = mem_cgroup_from_css(css); |
| 3672 | BUG_ON(child->kmemcg_id != kmemcg_id); |
| 3673 | child->kmemcg_id = parent->kmemcg_id; |
| 3674 | if (!memcg->use_hierarchy) |
| 3675 | break; |
| 3676 | } |
| 3677 | rcu_read_unlock(); |
| 3678 | |
| 3679 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); |
| 3680 | |
| 3681 | memcg_free_cache_id(kmemcg_id); |
| 3682 | } |
| 3683 | |
| 3684 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
| 3685 | { |
| 3686 | if (memcg->kmem_acct_activated) { |
| 3687 | memcg_destroy_kmem_caches(memcg); |
| 3688 | static_key_slow_dec(&memcg_kmem_enabled_key); |
| 3689 | WARN_ON(page_counter_read(&memcg->kmem)); |
| 3690 | } |
| 3691 | mem_cgroup_sockets_destroy(memcg); |
| 3692 | } |
| 3693 | #else |
| 3694 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
| 3695 | { |
| 3696 | return 0; |
| 3697 | } |
| 3698 | |
| 3699 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
| 3700 | { |
| 3701 | } |
| 3702 | |
| 3703 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
| 3704 | { |
| 3705 | } |
| 3706 | #endif |
| 3707 | |
| 3708 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 3709 | |
| 3710 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) |
| 3711 | { |
| 3712 | return &memcg->cgwb_list; |
| 3713 | } |
| 3714 | |
| 3715 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
| 3716 | { |
| 3717 | return wb_domain_init(&memcg->cgwb_domain, gfp); |
| 3718 | } |
| 3719 | |
| 3720 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) |
| 3721 | { |
| 3722 | wb_domain_exit(&memcg->cgwb_domain); |
| 3723 | } |
| 3724 | |
| 3725 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
| 3726 | { |
| 3727 | wb_domain_size_changed(&memcg->cgwb_domain); |
| 3728 | } |
| 3729 | |
| 3730 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
| 3731 | { |
| 3732 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); |
| 3733 | |
| 3734 | if (!memcg->css.parent) |
| 3735 | return NULL; |
| 3736 | |
| 3737 | return &memcg->cgwb_domain; |
| 3738 | } |
| 3739 | |
| 3740 | /** |
| 3741 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg |
| 3742 | * @wb: bdi_writeback in question |
| 3743 | * @pfilepages: out parameter for number of file pages |
| 3744 | * @pheadroom: out parameter for number of allocatable pages according to memcg |
| 3745 | * @pdirty: out parameter for number of dirty pages |
| 3746 | * @pwriteback: out parameter for number of pages under writeback |
| 3747 | * |
| 3748 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
| 3749 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom |
| 3750 | * is a bit more involved. |
| 3751 | * |
| 3752 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
| 3753 | * headroom is calculated as the lowest headroom of itself and the |
| 3754 | * ancestors. Note that this doesn't consider the actual amount of |
| 3755 | * available memory in the system. The caller should further cap |
| 3756 | * *@pheadroom accordingly. |
| 3757 | */ |
| 3758 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
| 3759 | unsigned long *pheadroom, unsigned long *pdirty, |
| 3760 | unsigned long *pwriteback) |
| 3761 | { |
| 3762 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); |
| 3763 | struct mem_cgroup *parent; |
| 3764 | |
| 3765 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); |
| 3766 | |
| 3767 | /* this should eventually include NR_UNSTABLE_NFS */ |
| 3768 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); |
| 3769 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
| 3770 | (1 << LRU_ACTIVE_FILE)); |
| 3771 | *pheadroom = PAGE_COUNTER_MAX; |
| 3772 | |
| 3773 | while ((parent = parent_mem_cgroup(memcg))) { |
| 3774 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); |
| 3775 | unsigned long used = page_counter_read(&memcg->memory); |
| 3776 | |
| 3777 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
| 3778 | memcg = parent; |
| 3779 | } |
| 3780 | } |
| 3781 | |
| 3782 | #else /* CONFIG_CGROUP_WRITEBACK */ |
| 3783 | |
| 3784 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
| 3785 | { |
| 3786 | return 0; |
| 3787 | } |
| 3788 | |
| 3789 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) |
| 3790 | { |
| 3791 | } |
| 3792 | |
| 3793 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
| 3794 | { |
| 3795 | } |
| 3796 | |
| 3797 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
| 3798 | |
| 3799 | /* |
| 3800 | * DO NOT USE IN NEW FILES. |
| 3801 | * |
| 3802 | * "cgroup.event_control" implementation. |
| 3803 | * |
| 3804 | * This is way over-engineered. It tries to support fully configurable |
| 3805 | * events for each user. Such level of flexibility is completely |
| 3806 | * unnecessary especially in the light of the planned unified hierarchy. |
| 3807 | * |
| 3808 | * Please deprecate this and replace with something simpler if at all |
| 3809 | * possible. |
| 3810 | */ |
| 3811 | |
| 3812 | /* |
| 3813 | * Unregister event and free resources. |
| 3814 | * |
| 3815 | * Gets called from workqueue. |
| 3816 | */ |
| 3817 | static void memcg_event_remove(struct work_struct *work) |
| 3818 | { |
| 3819 | struct mem_cgroup_event *event = |
| 3820 | container_of(work, struct mem_cgroup_event, remove); |
| 3821 | struct mem_cgroup *memcg = event->memcg; |
| 3822 | |
| 3823 | remove_wait_queue(event->wqh, &event->wait); |
| 3824 | |
| 3825 | event->unregister_event(memcg, event->eventfd); |
| 3826 | |
| 3827 | /* Notify userspace the event is going away. */ |
| 3828 | eventfd_signal(event->eventfd, 1); |
| 3829 | |
| 3830 | eventfd_ctx_put(event->eventfd); |
| 3831 | kfree(event); |
| 3832 | css_put(&memcg->css); |
| 3833 | } |
| 3834 | |
| 3835 | /* |
| 3836 | * Gets called on POLLHUP on eventfd when user closes it. |
| 3837 | * |
| 3838 | * Called with wqh->lock held and interrupts disabled. |
| 3839 | */ |
| 3840 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
| 3841 | int sync, void *key) |
| 3842 | { |
| 3843 | struct mem_cgroup_event *event = |
| 3844 | container_of(wait, struct mem_cgroup_event, wait); |
| 3845 | struct mem_cgroup *memcg = event->memcg; |
| 3846 | unsigned long flags = (unsigned long)key; |
| 3847 | |
| 3848 | if (flags & POLLHUP) { |
| 3849 | /* |
| 3850 | * If the event has been detached at cgroup removal, we |
| 3851 | * can simply return knowing the other side will cleanup |
| 3852 | * for us. |
| 3853 | * |
| 3854 | * We can't race against event freeing since the other |
| 3855 | * side will require wqh->lock via remove_wait_queue(), |
| 3856 | * which we hold. |
| 3857 | */ |
| 3858 | spin_lock(&memcg->event_list_lock); |
| 3859 | if (!list_empty(&event->list)) { |
| 3860 | list_del_init(&event->list); |
| 3861 | /* |
| 3862 | * We are in atomic context, but cgroup_event_remove() |
| 3863 | * may sleep, so we have to call it in workqueue. |
| 3864 | */ |
| 3865 | schedule_work(&event->remove); |
| 3866 | } |
| 3867 | spin_unlock(&memcg->event_list_lock); |
| 3868 | } |
| 3869 | |
| 3870 | return 0; |
| 3871 | } |
| 3872 | |
| 3873 | static void memcg_event_ptable_queue_proc(struct file *file, |
| 3874 | wait_queue_head_t *wqh, poll_table *pt) |
| 3875 | { |
| 3876 | struct mem_cgroup_event *event = |
| 3877 | container_of(pt, struct mem_cgroup_event, pt); |
| 3878 | |
| 3879 | event->wqh = wqh; |
| 3880 | add_wait_queue(wqh, &event->wait); |
| 3881 | } |
| 3882 | |
| 3883 | /* |
| 3884 | * DO NOT USE IN NEW FILES. |
| 3885 | * |
| 3886 | * Parse input and register new cgroup event handler. |
| 3887 | * |
| 3888 | * Input must be in format '<event_fd> <control_fd> <args>'. |
| 3889 | * Interpretation of args is defined by control file implementation. |
| 3890 | */ |
| 3891 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
| 3892 | char *buf, size_t nbytes, loff_t off) |
| 3893 | { |
| 3894 | struct cgroup_subsys_state *css = of_css(of); |
| 3895 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 3896 | struct mem_cgroup_event *event; |
| 3897 | struct cgroup_subsys_state *cfile_css; |
| 3898 | unsigned int efd, cfd; |
| 3899 | struct fd efile; |
| 3900 | struct fd cfile; |
| 3901 | const char *name; |
| 3902 | char *endp; |
| 3903 | int ret; |
| 3904 | |
| 3905 | buf = strstrip(buf); |
| 3906 | |
| 3907 | efd = simple_strtoul(buf, &endp, 10); |
| 3908 | if (*endp != ' ') |
| 3909 | return -EINVAL; |
| 3910 | buf = endp + 1; |
| 3911 | |
| 3912 | cfd = simple_strtoul(buf, &endp, 10); |
| 3913 | if ((*endp != ' ') && (*endp != '\0')) |
| 3914 | return -EINVAL; |
| 3915 | buf = endp + 1; |
| 3916 | |
| 3917 | event = kzalloc(sizeof(*event), GFP_KERNEL); |
| 3918 | if (!event) |
| 3919 | return -ENOMEM; |
| 3920 | |
| 3921 | event->memcg = memcg; |
| 3922 | INIT_LIST_HEAD(&event->list); |
| 3923 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
| 3924 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); |
| 3925 | INIT_WORK(&event->remove, memcg_event_remove); |
| 3926 | |
| 3927 | efile = fdget(efd); |
| 3928 | if (!efile.file) { |
| 3929 | ret = -EBADF; |
| 3930 | goto out_kfree; |
| 3931 | } |
| 3932 | |
| 3933 | event->eventfd = eventfd_ctx_fileget(efile.file); |
| 3934 | if (IS_ERR(event->eventfd)) { |
| 3935 | ret = PTR_ERR(event->eventfd); |
| 3936 | goto out_put_efile; |
| 3937 | } |
| 3938 | |
| 3939 | cfile = fdget(cfd); |
| 3940 | if (!cfile.file) { |
| 3941 | ret = -EBADF; |
| 3942 | goto out_put_eventfd; |
| 3943 | } |
| 3944 | |
| 3945 | /* the process need read permission on control file */ |
| 3946 | /* AV: shouldn't we check that it's been opened for read instead? */ |
| 3947 | ret = inode_permission(file_inode(cfile.file), MAY_READ); |
| 3948 | if (ret < 0) |
| 3949 | goto out_put_cfile; |
| 3950 | |
| 3951 | /* |
| 3952 | * Determine the event callbacks and set them in @event. This used |
| 3953 | * to be done via struct cftype but cgroup core no longer knows |
| 3954 | * about these events. The following is crude but the whole thing |
| 3955 | * is for compatibility anyway. |
| 3956 | * |
| 3957 | * DO NOT ADD NEW FILES. |
| 3958 | */ |
| 3959 | name = cfile.file->f_path.dentry->d_name.name; |
| 3960 | |
| 3961 | if (!strcmp(name, "memory.usage_in_bytes")) { |
| 3962 | event->register_event = mem_cgroup_usage_register_event; |
| 3963 | event->unregister_event = mem_cgroup_usage_unregister_event; |
| 3964 | } else if (!strcmp(name, "memory.oom_control")) { |
| 3965 | event->register_event = mem_cgroup_oom_register_event; |
| 3966 | event->unregister_event = mem_cgroup_oom_unregister_event; |
| 3967 | } else if (!strcmp(name, "memory.pressure_level")) { |
| 3968 | event->register_event = vmpressure_register_event; |
| 3969 | event->unregister_event = vmpressure_unregister_event; |
| 3970 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { |
| 3971 | event->register_event = memsw_cgroup_usage_register_event; |
| 3972 | event->unregister_event = memsw_cgroup_usage_unregister_event; |
| 3973 | } else { |
| 3974 | ret = -EINVAL; |
| 3975 | goto out_put_cfile; |
| 3976 | } |
| 3977 | |
| 3978 | /* |
| 3979 | * Verify @cfile should belong to @css. Also, remaining events are |
| 3980 | * automatically removed on cgroup destruction but the removal is |
| 3981 | * asynchronous, so take an extra ref on @css. |
| 3982 | */ |
| 3983 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
| 3984 | &memory_cgrp_subsys); |
| 3985 | ret = -EINVAL; |
| 3986 | if (IS_ERR(cfile_css)) |
| 3987 | goto out_put_cfile; |
| 3988 | if (cfile_css != css) { |
| 3989 | css_put(cfile_css); |
| 3990 | goto out_put_cfile; |
| 3991 | } |
| 3992 | |
| 3993 | ret = event->register_event(memcg, event->eventfd, buf); |
| 3994 | if (ret) |
| 3995 | goto out_put_css; |
| 3996 | |
| 3997 | efile.file->f_op->poll(efile.file, &event->pt); |
| 3998 | |
| 3999 | spin_lock(&memcg->event_list_lock); |
| 4000 | list_add(&event->list, &memcg->event_list); |
| 4001 | spin_unlock(&memcg->event_list_lock); |
| 4002 | |
| 4003 | fdput(cfile); |
| 4004 | fdput(efile); |
| 4005 | |
| 4006 | return nbytes; |
| 4007 | |
| 4008 | out_put_css: |
| 4009 | css_put(css); |
| 4010 | out_put_cfile: |
| 4011 | fdput(cfile); |
| 4012 | out_put_eventfd: |
| 4013 | eventfd_ctx_put(event->eventfd); |
| 4014 | out_put_efile: |
| 4015 | fdput(efile); |
| 4016 | out_kfree: |
| 4017 | kfree(event); |
| 4018 | |
| 4019 | return ret; |
| 4020 | } |
| 4021 | |
| 4022 | static struct cftype mem_cgroup_legacy_files[] = { |
| 4023 | { |
| 4024 | .name = "usage_in_bytes", |
| 4025 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
| 4026 | .read_u64 = mem_cgroup_read_u64, |
| 4027 | }, |
| 4028 | { |
| 4029 | .name = "max_usage_in_bytes", |
| 4030 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
| 4031 | .write = mem_cgroup_reset, |
| 4032 | .read_u64 = mem_cgroup_read_u64, |
| 4033 | }, |
| 4034 | { |
| 4035 | .name = "limit_in_bytes", |
| 4036 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
| 4037 | .write = mem_cgroup_write, |
| 4038 | .read_u64 = mem_cgroup_read_u64, |
| 4039 | }, |
| 4040 | { |
| 4041 | .name = "soft_limit_in_bytes", |
| 4042 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), |
| 4043 | .write = mem_cgroup_write, |
| 4044 | .read_u64 = mem_cgroup_read_u64, |
| 4045 | }, |
| 4046 | { |
| 4047 | .name = "failcnt", |
| 4048 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
| 4049 | .write = mem_cgroup_reset, |
| 4050 | .read_u64 = mem_cgroup_read_u64, |
| 4051 | }, |
| 4052 | { |
| 4053 | .name = "stat", |
| 4054 | .seq_show = memcg_stat_show, |
| 4055 | }, |
| 4056 | { |
| 4057 | .name = "force_empty", |
| 4058 | .write = mem_cgroup_force_empty_write, |
| 4059 | }, |
| 4060 | { |
| 4061 | .name = "use_hierarchy", |
| 4062 | .write_u64 = mem_cgroup_hierarchy_write, |
| 4063 | .read_u64 = mem_cgroup_hierarchy_read, |
| 4064 | }, |
| 4065 | { |
| 4066 | .name = "cgroup.event_control", /* XXX: for compat */ |
| 4067 | .write = memcg_write_event_control, |
| 4068 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
| 4069 | }, |
| 4070 | { |
| 4071 | .name = "swappiness", |
| 4072 | .read_u64 = mem_cgroup_swappiness_read, |
| 4073 | .write_u64 = mem_cgroup_swappiness_write, |
| 4074 | }, |
| 4075 | { |
| 4076 | .name = "move_charge_at_immigrate", |
| 4077 | .read_u64 = mem_cgroup_move_charge_read, |
| 4078 | .write_u64 = mem_cgroup_move_charge_write, |
| 4079 | }, |
| 4080 | { |
| 4081 | .name = "oom_control", |
| 4082 | .seq_show = mem_cgroup_oom_control_read, |
| 4083 | .write_u64 = mem_cgroup_oom_control_write, |
| 4084 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
| 4085 | }, |
| 4086 | { |
| 4087 | .name = "pressure_level", |
| 4088 | }, |
| 4089 | { |
| 4090 | .name = "vmpressure", |
| 4091 | .read_u64 = mem_cgroup_vmpressure_read, |
| 4092 | }, |
| 4093 | #ifdef CONFIG_NUMA |
| 4094 | { |
| 4095 | .name = "numa_stat", |
| 4096 | .seq_show = memcg_numa_stat_show, |
| 4097 | }, |
| 4098 | #endif |
| 4099 | #ifdef CONFIG_MEMCG_KMEM |
| 4100 | { |
| 4101 | .name = "kmem.limit_in_bytes", |
| 4102 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), |
| 4103 | .write = mem_cgroup_write, |
| 4104 | .read_u64 = mem_cgroup_read_u64, |
| 4105 | }, |
| 4106 | { |
| 4107 | .name = "kmem.usage_in_bytes", |
| 4108 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), |
| 4109 | .read_u64 = mem_cgroup_read_u64, |
| 4110 | }, |
| 4111 | { |
| 4112 | .name = "kmem.failcnt", |
| 4113 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), |
| 4114 | .write = mem_cgroup_reset, |
| 4115 | .read_u64 = mem_cgroup_read_u64, |
| 4116 | }, |
| 4117 | { |
| 4118 | .name = "kmem.max_usage_in_bytes", |
| 4119 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), |
| 4120 | .write = mem_cgroup_reset, |
| 4121 | .read_u64 = mem_cgroup_read_u64, |
| 4122 | }, |
| 4123 | #ifdef CONFIG_SLABINFO |
| 4124 | { |
| 4125 | .name = "kmem.slabinfo", |
| 4126 | .seq_start = slab_start, |
| 4127 | .seq_next = slab_next, |
| 4128 | .seq_stop = slab_stop, |
| 4129 | .seq_show = memcg_slab_show, |
| 4130 | }, |
| 4131 | #endif |
| 4132 | #endif |
| 4133 | { }, /* terminate */ |
| 4134 | }; |
| 4135 | |
| 4136 | /* |
| 4137 | * Private memory cgroup IDR |
| 4138 | * |
| 4139 | * Swap-out records and page cache shadow entries need to store memcg |
| 4140 | * references in constrained space, so we maintain an ID space that is |
| 4141 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of |
| 4142 | * memory-controlled cgroups to 64k. |
| 4143 | * |
| 4144 | * However, there usually are many references to the oflline CSS after |
| 4145 | * the cgroup has been destroyed, such as page cache or reclaimable |
| 4146 | * slab objects, that don't need to hang on to the ID. We want to keep |
| 4147 | * those dead CSS from occupying IDs, or we might quickly exhaust the |
| 4148 | * relatively small ID space and prevent the creation of new cgroups |
| 4149 | * even when there are much fewer than 64k cgroups - possibly none. |
| 4150 | * |
| 4151 | * Maintain a private 16-bit ID space for memcg, and allow the ID to |
| 4152 | * be freed and recycled when it's no longer needed, which is usually |
| 4153 | * when the CSS is offlined. |
| 4154 | * |
| 4155 | * The only exception to that are records of swapped out tmpfs/shmem |
| 4156 | * pages that need to be attributed to live ancestors on swapin. But |
| 4157 | * those references are manageable from userspace. |
| 4158 | */ |
| 4159 | |
| 4160 | static DEFINE_IDR(mem_cgroup_idr); |
| 4161 | |
| 4162 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
| 4163 | { |
| 4164 | atomic_add(n, &memcg->id.ref); |
| 4165 | } |
| 4166 | |
| 4167 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
| 4168 | { |
| 4169 | if (atomic_sub_and_test(n, &memcg->id.ref)) { |
| 4170 | idr_remove(&mem_cgroup_idr, memcg->id.id); |
| 4171 | memcg->id.id = 0; |
| 4172 | |
| 4173 | /* Memcg ID pins CSS */ |
| 4174 | css_put(&memcg->css); |
| 4175 | } |
| 4176 | } |
| 4177 | |
| 4178 | static inline void mem_cgroup_id_get(struct mem_cgroup *memcg) |
| 4179 | { |
| 4180 | mem_cgroup_id_get_many(memcg, 1); |
| 4181 | } |
| 4182 | |
| 4183 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
| 4184 | { |
| 4185 | mem_cgroup_id_put_many(memcg, 1); |
| 4186 | } |
| 4187 | |
| 4188 | /** |
| 4189 | * mem_cgroup_from_id - look up a memcg from a memcg id |
| 4190 | * @id: the memcg id to look up |
| 4191 | * |
| 4192 | * Caller must hold rcu_read_lock(). |
| 4193 | */ |
| 4194 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) |
| 4195 | { |
| 4196 | WARN_ON_ONCE(!rcu_read_lock_held()); |
| 4197 | return idr_find(&mem_cgroup_idr, id); |
| 4198 | } |
| 4199 | |
| 4200 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
| 4201 | { |
| 4202 | struct mem_cgroup_per_node *pn; |
| 4203 | struct mem_cgroup_per_zone *mz; |
| 4204 | int zone, tmp = node; |
| 4205 | /* |
| 4206 | * This routine is called against possible nodes. |
| 4207 | * But it's BUG to call kmalloc() against offline node. |
| 4208 | * |
| 4209 | * TODO: this routine can waste much memory for nodes which will |
| 4210 | * never be onlined. It's better to use memory hotplug callback |
| 4211 | * function. |
| 4212 | */ |
| 4213 | if (!node_state(node, N_NORMAL_MEMORY)) |
| 4214 | tmp = -1; |
| 4215 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
| 4216 | if (!pn) |
| 4217 | return 1; |
| 4218 | |
| 4219 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 4220 | mz = &pn->zoneinfo[zone]; |
| 4221 | lruvec_init(&mz->lruvec); |
| 4222 | mz->usage_in_excess = 0; |
| 4223 | mz->on_tree = false; |
| 4224 | mz->memcg = memcg; |
| 4225 | } |
| 4226 | memcg->nodeinfo[node] = pn; |
| 4227 | return 0; |
| 4228 | } |
| 4229 | |
| 4230 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
| 4231 | { |
| 4232 | kfree(memcg->nodeinfo[node]); |
| 4233 | } |
| 4234 | |
| 4235 | static struct mem_cgroup *mem_cgroup_alloc(void) |
| 4236 | { |
| 4237 | struct mem_cgroup *memcg; |
| 4238 | size_t size; |
| 4239 | |
| 4240 | size = sizeof(struct mem_cgroup); |
| 4241 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); |
| 4242 | |
| 4243 | memcg = kzalloc(size, GFP_KERNEL); |
| 4244 | if (!memcg) |
| 4245 | return NULL; |
| 4246 | |
| 4247 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
| 4248 | if (!memcg->stat) |
| 4249 | goto out_free; |
| 4250 | |
| 4251 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
| 4252 | goto out_free_stat; |
| 4253 | |
| 4254 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
| 4255 | 1, MEM_CGROUP_ID_MAX, |
| 4256 | GFP_KERNEL); |
| 4257 | if (memcg->id.id < 0) |
| 4258 | goto out_free_stat; |
| 4259 | |
| 4260 | return memcg; |
| 4261 | |
| 4262 | out_free_stat: |
| 4263 | free_percpu(memcg->stat); |
| 4264 | out_free: |
| 4265 | kfree(memcg); |
| 4266 | return NULL; |
| 4267 | } |
| 4268 | |
| 4269 | /* |
| 4270 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
| 4271 | * (scanning all at force_empty is too costly...) |
| 4272 | * |
| 4273 | * Instead of clearing all references at force_empty, we remember |
| 4274 | * the number of reference from swap_cgroup and free mem_cgroup when |
| 4275 | * it goes down to 0. |
| 4276 | * |
| 4277 | * Removal of cgroup itself succeeds regardless of refs from swap. |
| 4278 | */ |
| 4279 | |
| 4280 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
| 4281 | { |
| 4282 | int node; |
| 4283 | |
| 4284 | mem_cgroup_remove_from_trees(memcg); |
| 4285 | |
| 4286 | for_each_node(node) |
| 4287 | free_mem_cgroup_per_zone_info(memcg, node); |
| 4288 | |
| 4289 | free_percpu(memcg->stat); |
| 4290 | memcg_wb_domain_exit(memcg); |
| 4291 | kfree(memcg); |
| 4292 | } |
| 4293 | |
| 4294 | /* |
| 4295 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. |
| 4296 | */ |
| 4297 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
| 4298 | { |
| 4299 | if (!memcg->memory.parent) |
| 4300 | return NULL; |
| 4301 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
| 4302 | } |
| 4303 | EXPORT_SYMBOL(parent_mem_cgroup); |
| 4304 | |
| 4305 | static struct cgroup_subsys_state * __ref |
| 4306 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
| 4307 | { |
| 4308 | struct mem_cgroup *memcg; |
| 4309 | long error = -ENOMEM; |
| 4310 | int node; |
| 4311 | |
| 4312 | memcg = mem_cgroup_alloc(); |
| 4313 | if (!memcg) |
| 4314 | return ERR_PTR(error); |
| 4315 | |
| 4316 | for_each_node(node) |
| 4317 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
| 4318 | goto free_out; |
| 4319 | |
| 4320 | /* root ? */ |
| 4321 | if (parent_css == NULL) { |
| 4322 | root_mem_cgroup = memcg; |
| 4323 | mem_cgroup_root_css = &memcg->css; |
| 4324 | page_counter_init(&memcg->memory, NULL); |
| 4325 | memcg->high = PAGE_COUNTER_MAX; |
| 4326 | memcg->soft_limit = PAGE_COUNTER_MAX; |
| 4327 | page_counter_init(&memcg->memsw, NULL); |
| 4328 | page_counter_init(&memcg->kmem, NULL); |
| 4329 | } |
| 4330 | |
| 4331 | memcg->last_scanned_node = MAX_NUMNODES; |
| 4332 | INIT_LIST_HEAD(&memcg->oom_notify); |
| 4333 | memcg->move_charge_at_immigrate = 0; |
| 4334 | mutex_init(&memcg->thresholds_lock); |
| 4335 | spin_lock_init(&memcg->move_lock); |
| 4336 | vmpressure_init(&memcg->vmpressure); |
| 4337 | INIT_LIST_HEAD(&memcg->event_list); |
| 4338 | spin_lock_init(&memcg->event_list_lock); |
| 4339 | #ifdef CONFIG_MEMCG_KMEM |
| 4340 | memcg->kmemcg_id = -1; |
| 4341 | #endif |
| 4342 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 4343 | INIT_LIST_HEAD(&memcg->cgwb_list); |
| 4344 | #endif |
| 4345 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
| 4346 | return &memcg->css; |
| 4347 | |
| 4348 | free_out: |
| 4349 | idr_remove(&mem_cgroup_idr, memcg->id.id); |
| 4350 | __mem_cgroup_free(memcg); |
| 4351 | return ERR_PTR(error); |
| 4352 | } |
| 4353 | |
| 4354 | static int |
| 4355 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
| 4356 | { |
| 4357 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 4358 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
| 4359 | int ret; |
| 4360 | |
| 4361 | /* Online state pins memcg ID, memcg ID pins CSS */ |
| 4362 | mem_cgroup_id_get(mem_cgroup_from_css(css)); |
| 4363 | css_get(css); |
| 4364 | |
| 4365 | if (!parent) |
| 4366 | return 0; |
| 4367 | |
| 4368 | mutex_lock(&memcg_create_mutex); |
| 4369 | |
| 4370 | memcg->use_hierarchy = parent->use_hierarchy; |
| 4371 | memcg->oom_kill_disable = parent->oom_kill_disable; |
| 4372 | memcg->swappiness = mem_cgroup_swappiness(parent); |
| 4373 | |
| 4374 | if (parent->use_hierarchy) { |
| 4375 | page_counter_init(&memcg->memory, &parent->memory); |
| 4376 | memcg->high = PAGE_COUNTER_MAX; |
| 4377 | memcg->soft_limit = PAGE_COUNTER_MAX; |
| 4378 | page_counter_init(&memcg->memsw, &parent->memsw); |
| 4379 | page_counter_init(&memcg->kmem, &parent->kmem); |
| 4380 | |
| 4381 | /* |
| 4382 | * No need to take a reference to the parent because cgroup |
| 4383 | * core guarantees its existence. |
| 4384 | */ |
| 4385 | } else { |
| 4386 | page_counter_init(&memcg->memory, NULL); |
| 4387 | memcg->high = PAGE_COUNTER_MAX; |
| 4388 | memcg->soft_limit = PAGE_COUNTER_MAX; |
| 4389 | page_counter_init(&memcg->memsw, NULL); |
| 4390 | page_counter_init(&memcg->kmem, NULL); |
| 4391 | /* |
| 4392 | * Deeper hierachy with use_hierarchy == false doesn't make |
| 4393 | * much sense so let cgroup subsystem know about this |
| 4394 | * unfortunate state in our controller. |
| 4395 | */ |
| 4396 | if (parent != root_mem_cgroup) |
| 4397 | memory_cgrp_subsys.broken_hierarchy = true; |
| 4398 | } |
| 4399 | mutex_unlock(&memcg_create_mutex); |
| 4400 | |
| 4401 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
| 4402 | if (ret) |
| 4403 | return ret; |
| 4404 | |
| 4405 | /* |
| 4406 | * Make sure the memcg is initialized: mem_cgroup_iter() |
| 4407 | * orders reading memcg->initialized against its callers |
| 4408 | * reading the memcg members. |
| 4409 | */ |
| 4410 | smp_store_release(&memcg->initialized, 1); |
| 4411 | |
| 4412 | return 0; |
| 4413 | } |
| 4414 | |
| 4415 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
| 4416 | { |
| 4417 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 4418 | struct mem_cgroup_event *event, *tmp; |
| 4419 | |
| 4420 | /* |
| 4421 | * Unregister events and notify userspace. |
| 4422 | * Notify userspace about cgroup removing only after rmdir of cgroup |
| 4423 | * directory to avoid race between userspace and kernelspace. |
| 4424 | */ |
| 4425 | spin_lock(&memcg->event_list_lock); |
| 4426 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { |
| 4427 | list_del_init(&event->list); |
| 4428 | schedule_work(&event->remove); |
| 4429 | } |
| 4430 | spin_unlock(&memcg->event_list_lock); |
| 4431 | |
| 4432 | vmpressure_cleanup(&memcg->vmpressure); |
| 4433 | |
| 4434 | memcg_deactivate_kmem(memcg); |
| 4435 | |
| 4436 | wb_memcg_offline(memcg); |
| 4437 | |
| 4438 | mem_cgroup_id_put(memcg); |
| 4439 | } |
| 4440 | |
| 4441 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
| 4442 | { |
| 4443 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 4444 | |
| 4445 | invalidate_reclaim_iterators(memcg); |
| 4446 | } |
| 4447 | |
| 4448 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
| 4449 | { |
| 4450 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 4451 | |
| 4452 | memcg_destroy_kmem(memcg); |
| 4453 | __mem_cgroup_free(memcg); |
| 4454 | } |
| 4455 | |
| 4456 | /** |
| 4457 | * mem_cgroup_css_reset - reset the states of a mem_cgroup |
| 4458 | * @css: the target css |
| 4459 | * |
| 4460 | * Reset the states of the mem_cgroup associated with @css. This is |
| 4461 | * invoked when the userland requests disabling on the default hierarchy |
| 4462 | * but the memcg is pinned through dependency. The memcg should stop |
| 4463 | * applying policies and should revert to the vanilla state as it may be |
| 4464 | * made visible again. |
| 4465 | * |
| 4466 | * The current implementation only resets the essential configurations. |
| 4467 | * This needs to be expanded to cover all the visible parts. |
| 4468 | */ |
| 4469 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) |
| 4470 | { |
| 4471 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 4472 | |
| 4473 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
| 4474 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); |
| 4475 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); |
| 4476 | memcg->low = 0; |
| 4477 | memcg->high = PAGE_COUNTER_MAX; |
| 4478 | memcg->soft_limit = PAGE_COUNTER_MAX; |
| 4479 | memcg_wb_domain_size_changed(memcg); |
| 4480 | } |
| 4481 | |
| 4482 | #ifdef CONFIG_MMU |
| 4483 | /* Handlers for move charge at task migration. */ |
| 4484 | static int mem_cgroup_do_precharge(unsigned long count) |
| 4485 | { |
| 4486 | int ret; |
| 4487 | |
| 4488 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
| 4489 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); |
| 4490 | if (!ret) { |
| 4491 | mc.precharge += count; |
| 4492 | return ret; |
| 4493 | } |
| 4494 | |
| 4495 | /* Try charges one by one with reclaim, but do not retry */ |
| 4496 | while (count--) { |
| 4497 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
| 4498 | if (ret) |
| 4499 | return ret; |
| 4500 | mc.precharge++; |
| 4501 | cond_resched(); |
| 4502 | } |
| 4503 | return 0; |
| 4504 | } |
| 4505 | |
| 4506 | /** |
| 4507 | * get_mctgt_type - get target type of moving charge |
| 4508 | * @vma: the vma the pte to be checked belongs |
| 4509 | * @addr: the address corresponding to the pte to be checked |
| 4510 | * @ptent: the pte to be checked |
| 4511 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
| 4512 | * |
| 4513 | * Returns |
| 4514 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. |
| 4515 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for |
| 4516 | * move charge. if @target is not NULL, the page is stored in target->page |
| 4517 | * with extra refcnt got(Callers should handle it). |
| 4518 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
| 4519 | * target for charge migration. if @target is not NULL, the entry is stored |
| 4520 | * in target->ent. |
| 4521 | * |
| 4522 | * Called with pte lock held. |
| 4523 | */ |
| 4524 | union mc_target { |
| 4525 | struct page *page; |
| 4526 | swp_entry_t ent; |
| 4527 | }; |
| 4528 | |
| 4529 | enum mc_target_type { |
| 4530 | MC_TARGET_NONE = 0, |
| 4531 | MC_TARGET_PAGE, |
| 4532 | MC_TARGET_SWAP, |
| 4533 | }; |
| 4534 | |
| 4535 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
| 4536 | unsigned long addr, pte_t ptent) |
| 4537 | { |
| 4538 | struct page *page = vm_normal_page(vma, addr, ptent); |
| 4539 | |
| 4540 | if (!page || !page_mapped(page)) |
| 4541 | return NULL; |
| 4542 | if (PageAnon(page)) { |
| 4543 | if (!(mc.flags & MOVE_ANON)) |
| 4544 | return NULL; |
| 4545 | } else { |
| 4546 | if (!(mc.flags & MOVE_FILE)) |
| 4547 | return NULL; |
| 4548 | } |
| 4549 | if (!get_page_unless_zero(page)) |
| 4550 | return NULL; |
| 4551 | |
| 4552 | return page; |
| 4553 | } |
| 4554 | |
| 4555 | #ifdef CONFIG_SWAP |
| 4556 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
| 4557 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4558 | { |
| 4559 | struct page *page = NULL; |
| 4560 | swp_entry_t ent = pte_to_swp_entry(ptent); |
| 4561 | |
| 4562 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
| 4563 | return NULL; |
| 4564 | /* |
| 4565 | * Because lookup_swap_cache() updates some statistics counter, |
| 4566 | * we call find_get_page() with swapper_space directly. |
| 4567 | */ |
| 4568 | page = find_get_page(swap_address_space(ent), ent.val); |
| 4569 | if (do_swap_account) |
| 4570 | entry->val = ent.val; |
| 4571 | |
| 4572 | return page; |
| 4573 | } |
| 4574 | #else |
| 4575 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
| 4576 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4577 | { |
| 4578 | return NULL; |
| 4579 | } |
| 4580 | #endif |
| 4581 | |
| 4582 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
| 4583 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4584 | { |
| 4585 | struct page *page = NULL; |
| 4586 | struct address_space *mapping; |
| 4587 | pgoff_t pgoff; |
| 4588 | |
| 4589 | if (!vma->vm_file) /* anonymous vma */ |
| 4590 | return NULL; |
| 4591 | if (!(mc.flags & MOVE_FILE)) |
| 4592 | return NULL; |
| 4593 | |
| 4594 | mapping = vma->vm_file->f_mapping; |
| 4595 | pgoff = linear_page_index(vma, addr); |
| 4596 | |
| 4597 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
| 4598 | #ifdef CONFIG_SWAP |
| 4599 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
| 4600 | if (shmem_mapping(mapping)) { |
| 4601 | page = find_get_entry(mapping, pgoff); |
| 4602 | if (radix_tree_exceptional_entry(page)) { |
| 4603 | swp_entry_t swp = radix_to_swp_entry(page); |
| 4604 | if (do_swap_account) |
| 4605 | *entry = swp; |
| 4606 | page = find_get_page(swap_address_space(swp), swp.val); |
| 4607 | } |
| 4608 | } else |
| 4609 | page = find_get_page(mapping, pgoff); |
| 4610 | #else |
| 4611 | page = find_get_page(mapping, pgoff); |
| 4612 | #endif |
| 4613 | return page; |
| 4614 | } |
| 4615 | |
| 4616 | /** |
| 4617 | * mem_cgroup_move_account - move account of the page |
| 4618 | * @page: the page |
| 4619 | * @nr_pages: number of regular pages (>1 for huge pages) |
| 4620 | * @from: mem_cgroup which the page is moved from. |
| 4621 | * @to: mem_cgroup which the page is moved to. @from != @to. |
| 4622 | * |
| 4623 | * The caller must confirm following. |
| 4624 | * - page is not on LRU (isolate_page() is useful.) |
| 4625 | * - compound_lock is held when nr_pages > 1 |
| 4626 | * |
| 4627 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
| 4628 | * from old cgroup. |
| 4629 | */ |
| 4630 | static int mem_cgroup_move_account(struct page *page, |
| 4631 | unsigned int nr_pages, |
| 4632 | struct mem_cgroup *from, |
| 4633 | struct mem_cgroup *to) |
| 4634 | { |
| 4635 | unsigned long flags; |
| 4636 | int ret; |
| 4637 | bool anon; |
| 4638 | |
| 4639 | VM_BUG_ON(from == to); |
| 4640 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 4641 | /* |
| 4642 | * The page is isolated from LRU. So, collapse function |
| 4643 | * will not handle this page. But page splitting can happen. |
| 4644 | * Do this check under compound_page_lock(). The caller should |
| 4645 | * hold it. |
| 4646 | */ |
| 4647 | ret = -EBUSY; |
| 4648 | if (nr_pages > 1 && !PageTransHuge(page)) |
| 4649 | goto out; |
| 4650 | |
| 4651 | /* |
| 4652 | * Prevent mem_cgroup_replace_page() from looking at |
| 4653 | * page->mem_cgroup of its source page while we change it. |
| 4654 | */ |
| 4655 | if (!trylock_page(page)) |
| 4656 | goto out; |
| 4657 | |
| 4658 | ret = -EINVAL; |
| 4659 | if (page->mem_cgroup != from) |
| 4660 | goto out_unlock; |
| 4661 | |
| 4662 | anon = PageAnon(page); |
| 4663 | |
| 4664 | spin_lock_irqsave(&from->move_lock, flags); |
| 4665 | |
| 4666 | if (!anon && page_mapped(page)) { |
| 4667 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
| 4668 | nr_pages); |
| 4669 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
| 4670 | nr_pages); |
| 4671 | } |
| 4672 | |
| 4673 | /* |
| 4674 | * move_lock grabbed above and caller set from->moving_account, so |
| 4675 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. |
| 4676 | * So mapping should be stable for dirty pages. |
| 4677 | */ |
| 4678 | if (!anon && PageDirty(page)) { |
| 4679 | struct address_space *mapping = page_mapping(page); |
| 4680 | |
| 4681 | if (mapping_cap_account_dirty(mapping)) { |
| 4682 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], |
| 4683 | nr_pages); |
| 4684 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], |
| 4685 | nr_pages); |
| 4686 | } |
| 4687 | } |
| 4688 | |
| 4689 | if (PageWriteback(page)) { |
| 4690 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], |
| 4691 | nr_pages); |
| 4692 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], |
| 4693 | nr_pages); |
| 4694 | } |
| 4695 | |
| 4696 | /* |
| 4697 | * It is safe to change page->mem_cgroup here because the page |
| 4698 | * is referenced, charged, and isolated - we can't race with |
| 4699 | * uncharging, charging, migration, or LRU putback. |
| 4700 | */ |
| 4701 | |
| 4702 | /* caller should have done css_get */ |
| 4703 | page->mem_cgroup = to; |
| 4704 | spin_unlock_irqrestore(&from->move_lock, flags); |
| 4705 | |
| 4706 | ret = 0; |
| 4707 | |
| 4708 | local_irq_disable(); |
| 4709 | mem_cgroup_charge_statistics(to, page, nr_pages); |
| 4710 | memcg_check_events(to, page); |
| 4711 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
| 4712 | memcg_check_events(from, page); |
| 4713 | local_irq_enable(); |
| 4714 | out_unlock: |
| 4715 | unlock_page(page); |
| 4716 | out: |
| 4717 | return ret; |
| 4718 | } |
| 4719 | |
| 4720 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
| 4721 | unsigned long addr, pte_t ptent, union mc_target *target) |
| 4722 | { |
| 4723 | struct page *page = NULL; |
| 4724 | enum mc_target_type ret = MC_TARGET_NONE; |
| 4725 | swp_entry_t ent = { .val = 0 }; |
| 4726 | |
| 4727 | if (pte_present(ptent)) |
| 4728 | page = mc_handle_present_pte(vma, addr, ptent); |
| 4729 | else if (is_swap_pte(ptent)) |
| 4730 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); |
| 4731 | else if (pte_none(ptent)) |
| 4732 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
| 4733 | |
| 4734 | if (!page && !ent.val) |
| 4735 | return ret; |
| 4736 | if (page) { |
| 4737 | /* |
| 4738 | * Do only loose check w/o serialization. |
| 4739 | * mem_cgroup_move_account() checks the page is valid or |
| 4740 | * not under LRU exclusion. |
| 4741 | */ |
| 4742 | if (page->mem_cgroup == mc.from) { |
| 4743 | ret = MC_TARGET_PAGE; |
| 4744 | if (target) |
| 4745 | target->page = page; |
| 4746 | } |
| 4747 | if (!ret || !target) |
| 4748 | put_page(page); |
| 4749 | } |
| 4750 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
| 4751 | if (ent.val && !ret && |
| 4752 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
| 4753 | ret = MC_TARGET_SWAP; |
| 4754 | if (target) |
| 4755 | target->ent = ent; |
| 4756 | } |
| 4757 | return ret; |
| 4758 | } |
| 4759 | |
| 4760 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 4761 | /* |
| 4762 | * We don't consider swapping or file mapped pages because THP does not |
| 4763 | * support them for now. |
| 4764 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
| 4765 | */ |
| 4766 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 4767 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 4768 | { |
| 4769 | struct page *page = NULL; |
| 4770 | enum mc_target_type ret = MC_TARGET_NONE; |
| 4771 | |
| 4772 | page = pmd_page(pmd); |
| 4773 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
| 4774 | if (!(mc.flags & MOVE_ANON)) |
| 4775 | return ret; |
| 4776 | if (page->mem_cgroup == mc.from) { |
| 4777 | ret = MC_TARGET_PAGE; |
| 4778 | if (target) { |
| 4779 | get_page(page); |
| 4780 | target->page = page; |
| 4781 | } |
| 4782 | } |
| 4783 | return ret; |
| 4784 | } |
| 4785 | #else |
| 4786 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 4787 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 4788 | { |
| 4789 | return MC_TARGET_NONE; |
| 4790 | } |
| 4791 | #endif |
| 4792 | |
| 4793 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
| 4794 | unsigned long addr, unsigned long end, |
| 4795 | struct mm_walk *walk) |
| 4796 | { |
| 4797 | struct vm_area_struct *vma = walk->vma; |
| 4798 | pte_t *pte; |
| 4799 | spinlock_t *ptl; |
| 4800 | |
| 4801 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
| 4802 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
| 4803 | mc.precharge += HPAGE_PMD_NR; |
| 4804 | spin_unlock(ptl); |
| 4805 | return 0; |
| 4806 | } |
| 4807 | |
| 4808 | if (pmd_trans_unstable(pmd)) |
| 4809 | return 0; |
| 4810 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 4811 | for (; addr != end; pte++, addr += PAGE_SIZE) |
| 4812 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
| 4813 | mc.precharge++; /* increment precharge temporarily */ |
| 4814 | pte_unmap_unlock(pte - 1, ptl); |
| 4815 | cond_resched(); |
| 4816 | |
| 4817 | return 0; |
| 4818 | } |
| 4819 | |
| 4820 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
| 4821 | { |
| 4822 | unsigned long precharge; |
| 4823 | |
| 4824 | struct mm_walk mem_cgroup_count_precharge_walk = { |
| 4825 | .pmd_entry = mem_cgroup_count_precharge_pte_range, |
| 4826 | .mm = mm, |
| 4827 | }; |
| 4828 | down_read(&mm->mmap_sem); |
| 4829 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
| 4830 | up_read(&mm->mmap_sem); |
| 4831 | |
| 4832 | precharge = mc.precharge; |
| 4833 | mc.precharge = 0; |
| 4834 | |
| 4835 | return precharge; |
| 4836 | } |
| 4837 | |
| 4838 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
| 4839 | { |
| 4840 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
| 4841 | |
| 4842 | VM_BUG_ON(mc.moving_task); |
| 4843 | mc.moving_task = current; |
| 4844 | return mem_cgroup_do_precharge(precharge); |
| 4845 | } |
| 4846 | |
| 4847 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
| 4848 | static void __mem_cgroup_clear_mc(void) |
| 4849 | { |
| 4850 | struct mem_cgroup *from = mc.from; |
| 4851 | struct mem_cgroup *to = mc.to; |
| 4852 | |
| 4853 | /* we must uncharge all the leftover precharges from mc.to */ |
| 4854 | if (mc.precharge) { |
| 4855 | cancel_charge(mc.to, mc.precharge); |
| 4856 | mc.precharge = 0; |
| 4857 | } |
| 4858 | /* |
| 4859 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so |
| 4860 | * we must uncharge here. |
| 4861 | */ |
| 4862 | if (mc.moved_charge) { |
| 4863 | cancel_charge(mc.from, mc.moved_charge); |
| 4864 | mc.moved_charge = 0; |
| 4865 | } |
| 4866 | /* we must fixup refcnts and charges */ |
| 4867 | if (mc.moved_swap) { |
| 4868 | /* uncharge swap account from the old cgroup */ |
| 4869 | if (!mem_cgroup_is_root(mc.from)) |
| 4870 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
| 4871 | |
| 4872 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
| 4873 | |
| 4874 | /* |
| 4875 | * we charged both to->memory and to->memsw, so we |
| 4876 | * should uncharge to->memory. |
| 4877 | */ |
| 4878 | if (!mem_cgroup_is_root(mc.to)) |
| 4879 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
| 4880 | |
| 4881 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
| 4882 | css_put_many(&mc.to->css, mc.moved_swap); |
| 4883 | |
| 4884 | mc.moved_swap = 0; |
| 4885 | } |
| 4886 | memcg_oom_recover(from); |
| 4887 | memcg_oom_recover(to); |
| 4888 | wake_up_all(&mc.waitq); |
| 4889 | } |
| 4890 | |
| 4891 | static void mem_cgroup_clear_mc(void) |
| 4892 | { |
| 4893 | struct mm_struct *mm = mc.mm; |
| 4894 | |
| 4895 | /* |
| 4896 | * we must clear moving_task before waking up waiters at the end of |
| 4897 | * task migration. |
| 4898 | */ |
| 4899 | mc.moving_task = NULL; |
| 4900 | __mem_cgroup_clear_mc(); |
| 4901 | spin_lock(&mc.lock); |
| 4902 | mc.from = NULL; |
| 4903 | mc.to = NULL; |
| 4904 | mc.mm = NULL; |
| 4905 | spin_unlock(&mc.lock); |
| 4906 | |
| 4907 | mmput(mm); |
| 4908 | } |
| 4909 | |
| 4910 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
| 4911 | { |
| 4912 | struct cgroup_subsys_state *css; |
| 4913 | struct mem_cgroup *memcg; |
| 4914 | struct mem_cgroup *from; |
| 4915 | struct task_struct *leader, *p; |
| 4916 | struct mm_struct *mm; |
| 4917 | unsigned long move_flags; |
| 4918 | int ret = 0; |
| 4919 | |
| 4920 | /* charge immigration isn't supported on the default hierarchy */ |
| 4921 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
| 4922 | return 0; |
| 4923 | |
| 4924 | /* |
| 4925 | * Multi-process migrations only happen on the default hierarchy |
| 4926 | * where charge immigration is not used. Perform charge |
| 4927 | * immigration if @tset contains a leader and whine if there are |
| 4928 | * multiple. |
| 4929 | */ |
| 4930 | p = NULL; |
| 4931 | cgroup_taskset_for_each_leader(leader, css, tset) { |
| 4932 | WARN_ON_ONCE(p); |
| 4933 | p = leader; |
| 4934 | memcg = mem_cgroup_from_css(css); |
| 4935 | } |
| 4936 | if (!p) |
| 4937 | return 0; |
| 4938 | |
| 4939 | /* |
| 4940 | * We are now commited to this value whatever it is. Changes in this |
| 4941 | * tunable will only affect upcoming migrations, not the current one. |
| 4942 | * So we need to save it, and keep it going. |
| 4943 | */ |
| 4944 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); |
| 4945 | if (!move_flags) |
| 4946 | return 0; |
| 4947 | |
| 4948 | from = mem_cgroup_from_task(p); |
| 4949 | |
| 4950 | VM_BUG_ON(from == memcg); |
| 4951 | |
| 4952 | mm = get_task_mm(p); |
| 4953 | if (!mm) |
| 4954 | return 0; |
| 4955 | /* We move charges only when we move a owner of the mm */ |
| 4956 | if (mm->owner == p) { |
| 4957 | VM_BUG_ON(mc.from); |
| 4958 | VM_BUG_ON(mc.to); |
| 4959 | VM_BUG_ON(mc.precharge); |
| 4960 | VM_BUG_ON(mc.moved_charge); |
| 4961 | VM_BUG_ON(mc.moved_swap); |
| 4962 | |
| 4963 | spin_lock(&mc.lock); |
| 4964 | mc.mm = mm; |
| 4965 | mc.from = from; |
| 4966 | mc.to = memcg; |
| 4967 | mc.flags = move_flags; |
| 4968 | spin_unlock(&mc.lock); |
| 4969 | /* We set mc.moving_task later */ |
| 4970 | |
| 4971 | ret = mem_cgroup_precharge_mc(mm); |
| 4972 | if (ret) |
| 4973 | mem_cgroup_clear_mc(); |
| 4974 | } else { |
| 4975 | mmput(mm); |
| 4976 | } |
| 4977 | return ret; |
| 4978 | } |
| 4979 | |
| 4980 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
| 4981 | { |
| 4982 | if (mc.to) |
| 4983 | mem_cgroup_clear_mc(); |
| 4984 | } |
| 4985 | |
| 4986 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
| 4987 | unsigned long addr, unsigned long end, |
| 4988 | struct mm_walk *walk) |
| 4989 | { |
| 4990 | int ret = 0; |
| 4991 | struct vm_area_struct *vma = walk->vma; |
| 4992 | pte_t *pte; |
| 4993 | spinlock_t *ptl; |
| 4994 | enum mc_target_type target_type; |
| 4995 | union mc_target target; |
| 4996 | struct page *page; |
| 4997 | |
| 4998 | /* |
| 4999 | * We don't take compound_lock() here but no race with splitting thp |
| 5000 | * happens because: |
| 5001 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not |
| 5002 | * under splitting, which means there's no concurrent thp split, |
| 5003 | * - if another thread runs into split_huge_page() just after we |
| 5004 | * entered this if-block, the thread must wait for page table lock |
| 5005 | * to be unlocked in __split_huge_page_splitting(), where the main |
| 5006 | * part of thp split is not executed yet. |
| 5007 | */ |
| 5008 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
| 5009 | if (mc.precharge < HPAGE_PMD_NR) { |
| 5010 | spin_unlock(ptl); |
| 5011 | return 0; |
| 5012 | } |
| 5013 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); |
| 5014 | if (target_type == MC_TARGET_PAGE) { |
| 5015 | page = target.page; |
| 5016 | if (!isolate_lru_page(page)) { |
| 5017 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, |
| 5018 | mc.from, mc.to)) { |
| 5019 | mc.precharge -= HPAGE_PMD_NR; |
| 5020 | mc.moved_charge += HPAGE_PMD_NR; |
| 5021 | } |
| 5022 | putback_lru_page(page); |
| 5023 | } |
| 5024 | put_page(page); |
| 5025 | } |
| 5026 | spin_unlock(ptl); |
| 5027 | return 0; |
| 5028 | } |
| 5029 | |
| 5030 | if (pmd_trans_unstable(pmd)) |
| 5031 | return 0; |
| 5032 | retry: |
| 5033 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 5034 | for (; addr != end; addr += PAGE_SIZE) { |
| 5035 | pte_t ptent = *(pte++); |
| 5036 | swp_entry_t ent; |
| 5037 | |
| 5038 | if (!mc.precharge) |
| 5039 | break; |
| 5040 | |
| 5041 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
| 5042 | case MC_TARGET_PAGE: |
| 5043 | page = target.page; |
| 5044 | if (isolate_lru_page(page)) |
| 5045 | goto put; |
| 5046 | if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) { |
| 5047 | mc.precharge--; |
| 5048 | /* we uncharge from mc.from later. */ |
| 5049 | mc.moved_charge++; |
| 5050 | } |
| 5051 | putback_lru_page(page); |
| 5052 | put: /* get_mctgt_type() gets the page */ |
| 5053 | put_page(page); |
| 5054 | break; |
| 5055 | case MC_TARGET_SWAP: |
| 5056 | ent = target.ent; |
| 5057 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
| 5058 | mc.precharge--; |
| 5059 | /* we fixup refcnts and charges later. */ |
| 5060 | mc.moved_swap++; |
| 5061 | } |
| 5062 | break; |
| 5063 | default: |
| 5064 | break; |
| 5065 | } |
| 5066 | } |
| 5067 | pte_unmap_unlock(pte - 1, ptl); |
| 5068 | cond_resched(); |
| 5069 | |
| 5070 | if (addr != end) { |
| 5071 | /* |
| 5072 | * We have consumed all precharges we got in can_attach(). |
| 5073 | * We try charge one by one, but don't do any additional |
| 5074 | * charges to mc.to if we have failed in charge once in attach() |
| 5075 | * phase. |
| 5076 | */ |
| 5077 | ret = mem_cgroup_do_precharge(1); |
| 5078 | if (!ret) |
| 5079 | goto retry; |
| 5080 | } |
| 5081 | |
| 5082 | return ret; |
| 5083 | } |
| 5084 | |
| 5085 | static void mem_cgroup_move_charge(void) |
| 5086 | { |
| 5087 | struct mm_walk mem_cgroup_move_charge_walk = { |
| 5088 | .pmd_entry = mem_cgroup_move_charge_pte_range, |
| 5089 | .mm = mc.mm, |
| 5090 | }; |
| 5091 | |
| 5092 | lru_add_drain_all(); |
| 5093 | /* |
| 5094 | * Signal mem_cgroup_begin_page_stat() to take the memcg's |
| 5095 | * move_lock while we're moving its pages to another memcg. |
| 5096 | * Then wait for already started RCU-only updates to finish. |
| 5097 | */ |
| 5098 | atomic_inc(&mc.from->moving_account); |
| 5099 | synchronize_rcu(); |
| 5100 | retry: |
| 5101 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
| 5102 | /* |
| 5103 | * Someone who are holding the mmap_sem might be waiting in |
| 5104 | * waitq. So we cancel all extra charges, wake up all waiters, |
| 5105 | * and retry. Because we cancel precharges, we might not be able |
| 5106 | * to move enough charges, but moving charge is a best-effort |
| 5107 | * feature anyway, so it wouldn't be a big problem. |
| 5108 | */ |
| 5109 | __mem_cgroup_clear_mc(); |
| 5110 | cond_resched(); |
| 5111 | goto retry; |
| 5112 | } |
| 5113 | /* |
| 5114 | * When we have consumed all precharges and failed in doing |
| 5115 | * additional charge, the page walk just aborts. |
| 5116 | */ |
| 5117 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); |
| 5118 | up_read(&mc.mm->mmap_sem); |
| 5119 | atomic_dec(&mc.from->moving_account); |
| 5120 | } |
| 5121 | |
| 5122 | static void mem_cgroup_move_task(void) |
| 5123 | { |
| 5124 | if (mc.to) { |
| 5125 | mem_cgroup_move_charge(); |
| 5126 | mem_cgroup_clear_mc(); |
| 5127 | } |
| 5128 | } |
| 5129 | #else /* !CONFIG_MMU */ |
| 5130 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
| 5131 | { |
| 5132 | return 0; |
| 5133 | } |
| 5134 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
| 5135 | { |
| 5136 | } |
| 5137 | static void mem_cgroup_move_task(void) |
| 5138 | { |
| 5139 | } |
| 5140 | #endif |
| 5141 | |
| 5142 | /* |
| 5143 | * Cgroup retains root cgroups across [un]mount cycles making it necessary |
| 5144 | * to verify whether we're attached to the default hierarchy on each mount |
| 5145 | * attempt. |
| 5146 | */ |
| 5147 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
| 5148 | { |
| 5149 | /* |
| 5150 | * use_hierarchy is forced on the default hierarchy. cgroup core |
| 5151 | * guarantees that @root doesn't have any children, so turning it |
| 5152 | * on for the root memcg is enough. |
| 5153 | */ |
| 5154 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
| 5155 | root_mem_cgroup->use_hierarchy = true; |
| 5156 | else |
| 5157 | root_mem_cgroup->use_hierarchy = false; |
| 5158 | } |
| 5159 | |
| 5160 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
| 5161 | struct cftype *cft) |
| 5162 | { |
| 5163 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 5164 | |
| 5165 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; |
| 5166 | } |
| 5167 | |
| 5168 | static int memory_low_show(struct seq_file *m, void *v) |
| 5169 | { |
| 5170 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 5171 | unsigned long low = READ_ONCE(memcg->low); |
| 5172 | |
| 5173 | if (low == PAGE_COUNTER_MAX) |
| 5174 | seq_puts(m, "max\n"); |
| 5175 | else |
| 5176 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); |
| 5177 | |
| 5178 | return 0; |
| 5179 | } |
| 5180 | |
| 5181 | static ssize_t memory_low_write(struct kernfs_open_file *of, |
| 5182 | char *buf, size_t nbytes, loff_t off) |
| 5183 | { |
| 5184 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 5185 | unsigned long low; |
| 5186 | int err; |
| 5187 | |
| 5188 | buf = strstrip(buf); |
| 5189 | err = page_counter_memparse(buf, "max", &low); |
| 5190 | if (err) |
| 5191 | return err; |
| 5192 | |
| 5193 | memcg->low = low; |
| 5194 | |
| 5195 | return nbytes; |
| 5196 | } |
| 5197 | |
| 5198 | static int memory_high_show(struct seq_file *m, void *v) |
| 5199 | { |
| 5200 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 5201 | unsigned long high = READ_ONCE(memcg->high); |
| 5202 | |
| 5203 | if (high == PAGE_COUNTER_MAX) |
| 5204 | seq_puts(m, "max\n"); |
| 5205 | else |
| 5206 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); |
| 5207 | |
| 5208 | return 0; |
| 5209 | } |
| 5210 | |
| 5211 | static ssize_t memory_high_write(struct kernfs_open_file *of, |
| 5212 | char *buf, size_t nbytes, loff_t off) |
| 5213 | { |
| 5214 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 5215 | unsigned long nr_pages; |
| 5216 | unsigned long high; |
| 5217 | int err; |
| 5218 | |
| 5219 | buf = strstrip(buf); |
| 5220 | err = page_counter_memparse(buf, "max", &high); |
| 5221 | if (err) |
| 5222 | return err; |
| 5223 | |
| 5224 | memcg->high = high; |
| 5225 | |
| 5226 | nr_pages = page_counter_read(&memcg->memory); |
| 5227 | if (nr_pages > high) |
| 5228 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, |
| 5229 | GFP_KERNEL, true); |
| 5230 | |
| 5231 | memcg_wb_domain_size_changed(memcg); |
| 5232 | return nbytes; |
| 5233 | } |
| 5234 | |
| 5235 | static int memory_max_show(struct seq_file *m, void *v) |
| 5236 | { |
| 5237 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 5238 | unsigned long max = READ_ONCE(memcg->memory.limit); |
| 5239 | |
| 5240 | if (max == PAGE_COUNTER_MAX) |
| 5241 | seq_puts(m, "max\n"); |
| 5242 | else |
| 5243 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); |
| 5244 | |
| 5245 | return 0; |
| 5246 | } |
| 5247 | |
| 5248 | static ssize_t memory_max_write(struct kernfs_open_file *of, |
| 5249 | char *buf, size_t nbytes, loff_t off) |
| 5250 | { |
| 5251 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 5252 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
| 5253 | bool drained = false; |
| 5254 | unsigned long max; |
| 5255 | int err; |
| 5256 | |
| 5257 | buf = strstrip(buf); |
| 5258 | err = page_counter_memparse(buf, "max", &max); |
| 5259 | if (err) |
| 5260 | return err; |
| 5261 | |
| 5262 | xchg(&memcg->memory.limit, max); |
| 5263 | |
| 5264 | for (;;) { |
| 5265 | unsigned long nr_pages = page_counter_read(&memcg->memory); |
| 5266 | |
| 5267 | if (nr_pages <= max) |
| 5268 | break; |
| 5269 | |
| 5270 | if (signal_pending(current)) { |
| 5271 | err = -EINTR; |
| 5272 | break; |
| 5273 | } |
| 5274 | |
| 5275 | if (!drained) { |
| 5276 | drain_all_stock(memcg); |
| 5277 | drained = true; |
| 5278 | continue; |
| 5279 | } |
| 5280 | |
| 5281 | if (nr_reclaims) { |
| 5282 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, |
| 5283 | GFP_KERNEL, true)) |
| 5284 | nr_reclaims--; |
| 5285 | continue; |
| 5286 | } |
| 5287 | |
| 5288 | mem_cgroup_events(memcg, MEMCG_OOM, 1); |
| 5289 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
| 5290 | break; |
| 5291 | } |
| 5292 | |
| 5293 | memcg_wb_domain_size_changed(memcg); |
| 5294 | return nbytes; |
| 5295 | } |
| 5296 | |
| 5297 | static int memory_events_show(struct seq_file *m, void *v) |
| 5298 | { |
| 5299 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 5300 | |
| 5301 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); |
| 5302 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); |
| 5303 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); |
| 5304 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); |
| 5305 | |
| 5306 | return 0; |
| 5307 | } |
| 5308 | |
| 5309 | static struct cftype memory_files[] = { |
| 5310 | { |
| 5311 | .name = "current", |
| 5312 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5313 | .read_u64 = memory_current_read, |
| 5314 | }, |
| 5315 | { |
| 5316 | .name = "low", |
| 5317 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5318 | .seq_show = memory_low_show, |
| 5319 | .write = memory_low_write, |
| 5320 | }, |
| 5321 | { |
| 5322 | .name = "high", |
| 5323 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5324 | .seq_show = memory_high_show, |
| 5325 | .write = memory_high_write, |
| 5326 | }, |
| 5327 | { |
| 5328 | .name = "max", |
| 5329 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5330 | .seq_show = memory_max_show, |
| 5331 | .write = memory_max_write, |
| 5332 | }, |
| 5333 | { |
| 5334 | .name = "events", |
| 5335 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5336 | .file_offset = offsetof(struct mem_cgroup, events_file), |
| 5337 | .seq_show = memory_events_show, |
| 5338 | }, |
| 5339 | { } /* terminate */ |
| 5340 | }; |
| 5341 | |
| 5342 | struct cgroup_subsys memory_cgrp_subsys = { |
| 5343 | .css_alloc = mem_cgroup_css_alloc, |
| 5344 | .css_online = mem_cgroup_css_online, |
| 5345 | .css_offline = mem_cgroup_css_offline, |
| 5346 | .css_released = mem_cgroup_css_released, |
| 5347 | .css_free = mem_cgroup_css_free, |
| 5348 | .css_reset = mem_cgroup_css_reset, |
| 5349 | .can_attach = mem_cgroup_can_attach, |
| 5350 | .cancel_attach = mem_cgroup_cancel_attach, |
| 5351 | .post_attach = mem_cgroup_move_task, |
| 5352 | .bind = mem_cgroup_bind, |
| 5353 | .dfl_cftypes = memory_files, |
| 5354 | .legacy_cftypes = mem_cgroup_legacy_files, |
| 5355 | .early_init = 0, |
| 5356 | }; |
| 5357 | |
| 5358 | /** |
| 5359 | * mem_cgroup_low - check if memory consumption is below the normal range |
| 5360 | * @root: the highest ancestor to consider |
| 5361 | * @memcg: the memory cgroup to check |
| 5362 | * |
| 5363 | * Returns %true if memory consumption of @memcg, and that of all |
| 5364 | * configurable ancestors up to @root, is below the normal range. |
| 5365 | */ |
| 5366 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) |
| 5367 | { |
| 5368 | if (mem_cgroup_disabled()) |
| 5369 | return false; |
| 5370 | |
| 5371 | /* |
| 5372 | * The toplevel group doesn't have a configurable range, so |
| 5373 | * it's never low when looked at directly, and it is not |
| 5374 | * considered an ancestor when assessing the hierarchy. |
| 5375 | */ |
| 5376 | |
| 5377 | if (memcg == root_mem_cgroup) |
| 5378 | return false; |
| 5379 | |
| 5380 | if (page_counter_read(&memcg->memory) >= memcg->low) |
| 5381 | return false; |
| 5382 | |
| 5383 | while (memcg != root) { |
| 5384 | memcg = parent_mem_cgroup(memcg); |
| 5385 | |
| 5386 | if (memcg == root_mem_cgroup) |
| 5387 | break; |
| 5388 | |
| 5389 | if (page_counter_read(&memcg->memory) >= memcg->low) |
| 5390 | return false; |
| 5391 | } |
| 5392 | return true; |
| 5393 | } |
| 5394 | |
| 5395 | /** |
| 5396 | * mem_cgroup_try_charge - try charging a page |
| 5397 | * @page: page to charge |
| 5398 | * @mm: mm context of the victim |
| 5399 | * @gfp_mask: reclaim mode |
| 5400 | * @memcgp: charged memcg return |
| 5401 | * |
| 5402 | * Try to charge @page to the memcg that @mm belongs to, reclaiming |
| 5403 | * pages according to @gfp_mask if necessary. |
| 5404 | * |
| 5405 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. |
| 5406 | * Otherwise, an error code is returned. |
| 5407 | * |
| 5408 | * After page->mapping has been set up, the caller must finalize the |
| 5409 | * charge with mem_cgroup_commit_charge(). Or abort the transaction |
| 5410 | * with mem_cgroup_cancel_charge() in case page instantiation fails. |
| 5411 | */ |
| 5412 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, |
| 5413 | gfp_t gfp_mask, struct mem_cgroup **memcgp) |
| 5414 | { |
| 5415 | struct mem_cgroup *memcg = NULL; |
| 5416 | unsigned int nr_pages = 1; |
| 5417 | int ret = 0; |
| 5418 | |
| 5419 | if (mem_cgroup_disabled()) |
| 5420 | goto out; |
| 5421 | |
| 5422 | if (PageSwapCache(page)) { |
| 5423 | /* |
| 5424 | * Every swap fault against a single page tries to charge the |
| 5425 | * page, bail as early as possible. shmem_unuse() encounters |
| 5426 | * already charged pages, too. The USED bit is protected by |
| 5427 | * the page lock, which serializes swap cache removal, which |
| 5428 | * in turn serializes uncharging. |
| 5429 | */ |
| 5430 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
| 5431 | if (page->mem_cgroup) |
| 5432 | goto out; |
| 5433 | |
| 5434 | if (do_swap_account) { |
| 5435 | swp_entry_t ent = { .val = page_private(page), }; |
| 5436 | unsigned short id = lookup_swap_cgroup_id(ent); |
| 5437 | |
| 5438 | rcu_read_lock(); |
| 5439 | memcg = mem_cgroup_from_id(id); |
| 5440 | if (memcg && !css_tryget_online(&memcg->css)) |
| 5441 | memcg = NULL; |
| 5442 | rcu_read_unlock(); |
| 5443 | } |
| 5444 | } |
| 5445 | |
| 5446 | if (PageTransHuge(page)) { |
| 5447 | nr_pages <<= compound_order(page); |
| 5448 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
| 5449 | } |
| 5450 | |
| 5451 | if (!memcg) |
| 5452 | memcg = get_mem_cgroup_from_mm(mm); |
| 5453 | |
| 5454 | ret = try_charge(memcg, gfp_mask, nr_pages); |
| 5455 | |
| 5456 | css_put(&memcg->css); |
| 5457 | out: |
| 5458 | *memcgp = memcg; |
| 5459 | return ret; |
| 5460 | } |
| 5461 | |
| 5462 | /** |
| 5463 | * mem_cgroup_commit_charge - commit a page charge |
| 5464 | * @page: page to charge |
| 5465 | * @memcg: memcg to charge the page to |
| 5466 | * @lrucare: page might be on LRU already |
| 5467 | * |
| 5468 | * Finalize a charge transaction started by mem_cgroup_try_charge(), |
| 5469 | * after page->mapping has been set up. This must happen atomically |
| 5470 | * as part of the page instantiation, i.e. under the page table lock |
| 5471 | * for anonymous pages, under the page lock for page and swap cache. |
| 5472 | * |
| 5473 | * In addition, the page must not be on the LRU during the commit, to |
| 5474 | * prevent racing with task migration. If it might be, use @lrucare. |
| 5475 | * |
| 5476 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. |
| 5477 | */ |
| 5478 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, |
| 5479 | bool lrucare) |
| 5480 | { |
| 5481 | unsigned int nr_pages = 1; |
| 5482 | |
| 5483 | VM_BUG_ON_PAGE(!page->mapping, page); |
| 5484 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); |
| 5485 | |
| 5486 | if (mem_cgroup_disabled()) |
| 5487 | return; |
| 5488 | /* |
| 5489 | * Swap faults will attempt to charge the same page multiple |
| 5490 | * times. But reuse_swap_page() might have removed the page |
| 5491 | * from swapcache already, so we can't check PageSwapCache(). |
| 5492 | */ |
| 5493 | if (!memcg) |
| 5494 | return; |
| 5495 | |
| 5496 | commit_charge(page, memcg, lrucare); |
| 5497 | |
| 5498 | if (PageTransHuge(page)) { |
| 5499 | nr_pages <<= compound_order(page); |
| 5500 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
| 5501 | } |
| 5502 | |
| 5503 | local_irq_disable(); |
| 5504 | mem_cgroup_charge_statistics(memcg, page, nr_pages); |
| 5505 | memcg_check_events(memcg, page); |
| 5506 | local_irq_enable(); |
| 5507 | |
| 5508 | if (do_swap_account && PageSwapCache(page)) { |
| 5509 | swp_entry_t entry = { .val = page_private(page) }; |
| 5510 | /* |
| 5511 | * The swap entry might not get freed for a long time, |
| 5512 | * let's not wait for it. The page already received a |
| 5513 | * memory+swap charge, drop the swap entry duplicate. |
| 5514 | */ |
| 5515 | mem_cgroup_uncharge_swap(entry); |
| 5516 | } |
| 5517 | } |
| 5518 | |
| 5519 | /** |
| 5520 | * mem_cgroup_cancel_charge - cancel a page charge |
| 5521 | * @page: page to charge |
| 5522 | * @memcg: memcg to charge the page to |
| 5523 | * |
| 5524 | * Cancel a charge transaction started by mem_cgroup_try_charge(). |
| 5525 | */ |
| 5526 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) |
| 5527 | { |
| 5528 | unsigned int nr_pages = 1; |
| 5529 | |
| 5530 | if (mem_cgroup_disabled()) |
| 5531 | return; |
| 5532 | /* |
| 5533 | * Swap faults will attempt to charge the same page multiple |
| 5534 | * times. But reuse_swap_page() might have removed the page |
| 5535 | * from swapcache already, so we can't check PageSwapCache(). |
| 5536 | */ |
| 5537 | if (!memcg) |
| 5538 | return; |
| 5539 | |
| 5540 | if (PageTransHuge(page)) { |
| 5541 | nr_pages <<= compound_order(page); |
| 5542 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
| 5543 | } |
| 5544 | |
| 5545 | cancel_charge(memcg, nr_pages); |
| 5546 | } |
| 5547 | |
| 5548 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
| 5549 | unsigned long nr_anon, unsigned long nr_file, |
| 5550 | unsigned long nr_huge, struct page *dummy_page) |
| 5551 | { |
| 5552 | unsigned long nr_pages = nr_anon + nr_file; |
| 5553 | unsigned long flags; |
| 5554 | |
| 5555 | if (!mem_cgroup_is_root(memcg)) { |
| 5556 | page_counter_uncharge(&memcg->memory, nr_pages); |
| 5557 | if (do_swap_account) |
| 5558 | page_counter_uncharge(&memcg->memsw, nr_pages); |
| 5559 | memcg_oom_recover(memcg); |
| 5560 | } |
| 5561 | |
| 5562 | local_irq_save(flags); |
| 5563 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); |
| 5564 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); |
| 5565 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); |
| 5566 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); |
| 5567 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
| 5568 | memcg_check_events(memcg, dummy_page); |
| 5569 | local_irq_restore(flags); |
| 5570 | |
| 5571 | if (!mem_cgroup_is_root(memcg)) |
| 5572 | css_put_many(&memcg->css, nr_pages); |
| 5573 | } |
| 5574 | |
| 5575 | static void uncharge_list(struct list_head *page_list) |
| 5576 | { |
| 5577 | struct mem_cgroup *memcg = NULL; |
| 5578 | unsigned long nr_anon = 0; |
| 5579 | unsigned long nr_file = 0; |
| 5580 | unsigned long nr_huge = 0; |
| 5581 | unsigned long pgpgout = 0; |
| 5582 | struct list_head *next; |
| 5583 | struct page *page; |
| 5584 | |
| 5585 | next = page_list->next; |
| 5586 | do { |
| 5587 | unsigned int nr_pages = 1; |
| 5588 | |
| 5589 | page = list_entry(next, struct page, lru); |
| 5590 | next = page->lru.next; |
| 5591 | |
| 5592 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 5593 | VM_BUG_ON_PAGE(page_count(page), page); |
| 5594 | |
| 5595 | if (!page->mem_cgroup) |
| 5596 | continue; |
| 5597 | |
| 5598 | /* |
| 5599 | * Nobody should be changing or seriously looking at |
| 5600 | * page->mem_cgroup at this point, we have fully |
| 5601 | * exclusive access to the page. |
| 5602 | */ |
| 5603 | |
| 5604 | if (memcg != page->mem_cgroup) { |
| 5605 | if (memcg) { |
| 5606 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
| 5607 | nr_huge, page); |
| 5608 | pgpgout = nr_anon = nr_file = nr_huge = 0; |
| 5609 | } |
| 5610 | memcg = page->mem_cgroup; |
| 5611 | } |
| 5612 | |
| 5613 | if (PageTransHuge(page)) { |
| 5614 | nr_pages <<= compound_order(page); |
| 5615 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
| 5616 | nr_huge += nr_pages; |
| 5617 | } |
| 5618 | |
| 5619 | if (PageAnon(page)) |
| 5620 | nr_anon += nr_pages; |
| 5621 | else |
| 5622 | nr_file += nr_pages; |
| 5623 | |
| 5624 | page->mem_cgroup = NULL; |
| 5625 | |
| 5626 | pgpgout++; |
| 5627 | } while (next != page_list); |
| 5628 | |
| 5629 | if (memcg) |
| 5630 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
| 5631 | nr_huge, page); |
| 5632 | } |
| 5633 | |
| 5634 | /** |
| 5635 | * mem_cgroup_uncharge - uncharge a page |
| 5636 | * @page: page to uncharge |
| 5637 | * |
| 5638 | * Uncharge a page previously charged with mem_cgroup_try_charge() and |
| 5639 | * mem_cgroup_commit_charge(). |
| 5640 | */ |
| 5641 | void mem_cgroup_uncharge(struct page *page) |
| 5642 | { |
| 5643 | if (mem_cgroup_disabled()) |
| 5644 | return; |
| 5645 | |
| 5646 | /* Don't touch page->lru of any random page, pre-check: */ |
| 5647 | if (!page->mem_cgroup) |
| 5648 | return; |
| 5649 | |
| 5650 | INIT_LIST_HEAD(&page->lru); |
| 5651 | uncharge_list(&page->lru); |
| 5652 | } |
| 5653 | |
| 5654 | /** |
| 5655 | * mem_cgroup_uncharge_list - uncharge a list of page |
| 5656 | * @page_list: list of pages to uncharge |
| 5657 | * |
| 5658 | * Uncharge a list of pages previously charged with |
| 5659 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). |
| 5660 | */ |
| 5661 | void mem_cgroup_uncharge_list(struct list_head *page_list) |
| 5662 | { |
| 5663 | if (mem_cgroup_disabled()) |
| 5664 | return; |
| 5665 | |
| 5666 | if (!list_empty(page_list)) |
| 5667 | uncharge_list(page_list); |
| 5668 | } |
| 5669 | |
| 5670 | /** |
| 5671 | * mem_cgroup_replace_page - migrate a charge to another page |
| 5672 | * @oldpage: currently charged page |
| 5673 | * @newpage: page to transfer the charge to |
| 5674 | * |
| 5675 | * Migrate the charge from @oldpage to @newpage. |
| 5676 | * |
| 5677 | * Both pages must be locked, @newpage->mapping must be set up. |
| 5678 | * Either or both pages might be on the LRU already. |
| 5679 | */ |
| 5680 | void mem_cgroup_replace_page(struct page *oldpage, struct page *newpage) |
| 5681 | { |
| 5682 | struct mem_cgroup *memcg; |
| 5683 | int isolated; |
| 5684 | |
| 5685 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); |
| 5686 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); |
| 5687 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
| 5688 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
| 5689 | newpage); |
| 5690 | |
| 5691 | if (mem_cgroup_disabled()) |
| 5692 | return; |
| 5693 | |
| 5694 | /* Page cache replacement: new page already charged? */ |
| 5695 | if (newpage->mem_cgroup) |
| 5696 | return; |
| 5697 | |
| 5698 | /* Swapcache readahead pages can get replaced before being charged */ |
| 5699 | memcg = oldpage->mem_cgroup; |
| 5700 | if (!memcg) |
| 5701 | return; |
| 5702 | |
| 5703 | lock_page_lru(oldpage, &isolated); |
| 5704 | oldpage->mem_cgroup = NULL; |
| 5705 | unlock_page_lru(oldpage, isolated); |
| 5706 | |
| 5707 | commit_charge(newpage, memcg, true); |
| 5708 | } |
| 5709 | |
| 5710 | /* |
| 5711 | * subsys_initcall() for memory controller. |
| 5712 | * |
| 5713 | * Some parts like hotcpu_notifier() have to be initialized from this context |
| 5714 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically |
| 5715 | * everything that doesn't depend on a specific mem_cgroup structure should |
| 5716 | * be initialized from here. |
| 5717 | */ |
| 5718 | static int __init mem_cgroup_init(void) |
| 5719 | { |
| 5720 | int cpu, node; |
| 5721 | |
| 5722 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
| 5723 | |
| 5724 | for_each_possible_cpu(cpu) |
| 5725 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, |
| 5726 | drain_local_stock); |
| 5727 | |
| 5728 | for_each_node(node) { |
| 5729 | struct mem_cgroup_tree_per_node *rtpn; |
| 5730 | int zone; |
| 5731 | |
| 5732 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, |
| 5733 | node_online(node) ? node : NUMA_NO_NODE); |
| 5734 | |
| 5735 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 5736 | struct mem_cgroup_tree_per_zone *rtpz; |
| 5737 | |
| 5738 | rtpz = &rtpn->rb_tree_per_zone[zone]; |
| 5739 | rtpz->rb_root = RB_ROOT; |
| 5740 | spin_lock_init(&rtpz->lock); |
| 5741 | } |
| 5742 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
| 5743 | } |
| 5744 | |
| 5745 | return 0; |
| 5746 | } |
| 5747 | subsys_initcall(mem_cgroup_init); |
| 5748 | |
| 5749 | #ifdef CONFIG_MEMCG_SWAP |
| 5750 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
| 5751 | { |
| 5752 | while (!atomic_inc_not_zero(&memcg->id.ref)) { |
| 5753 | /* |
| 5754 | * The root cgroup cannot be destroyed, so it's refcount must |
| 5755 | * always be >= 1. |
| 5756 | */ |
| 5757 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { |
| 5758 | VM_BUG_ON(1); |
| 5759 | break; |
| 5760 | } |
| 5761 | memcg = parent_mem_cgroup(memcg); |
| 5762 | if (!memcg) |
| 5763 | memcg = root_mem_cgroup; |
| 5764 | } |
| 5765 | return memcg; |
| 5766 | } |
| 5767 | |
| 5768 | /** |
| 5769 | * mem_cgroup_swapout - transfer a memsw charge to swap |
| 5770 | * @page: page whose memsw charge to transfer |
| 5771 | * @entry: swap entry to move the charge to |
| 5772 | * |
| 5773 | * Transfer the memsw charge of @page to @entry. |
| 5774 | */ |
| 5775 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) |
| 5776 | { |
| 5777 | struct mem_cgroup *memcg, *swap_memcg; |
| 5778 | unsigned short oldid; |
| 5779 | |
| 5780 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 5781 | VM_BUG_ON_PAGE(page_count(page), page); |
| 5782 | |
| 5783 | if (!do_swap_account) |
| 5784 | return; |
| 5785 | |
| 5786 | memcg = page->mem_cgroup; |
| 5787 | |
| 5788 | /* Readahead page, never charged */ |
| 5789 | if (!memcg) |
| 5790 | return; |
| 5791 | |
| 5792 | /* |
| 5793 | * In case the memcg owning these pages has been offlined and doesn't |
| 5794 | * have an ID allocated to it anymore, charge the closest online |
| 5795 | * ancestor for the swap instead and transfer the memory+swap charge. |
| 5796 | */ |
| 5797 | swap_memcg = mem_cgroup_id_get_online(memcg); |
| 5798 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg)); |
| 5799 | VM_BUG_ON_PAGE(oldid, page); |
| 5800 | mem_cgroup_swap_statistics(swap_memcg, true); |
| 5801 | |
| 5802 | page->mem_cgroup = NULL; |
| 5803 | |
| 5804 | if (!mem_cgroup_is_root(memcg)) |
| 5805 | page_counter_uncharge(&memcg->memory, 1); |
| 5806 | |
| 5807 | if (memcg != swap_memcg) { |
| 5808 | if (!mem_cgroup_is_root(swap_memcg)) |
| 5809 | page_counter_charge(&swap_memcg->memsw, 1); |
| 5810 | page_counter_uncharge(&memcg->memsw, 1); |
| 5811 | } |
| 5812 | |
| 5813 | /* |
| 5814 | * Interrupts should be disabled here because the caller holds the |
| 5815 | * mapping->tree_lock lock which is taken with interrupts-off. It is |
| 5816 | * important here to have the interrupts disabled because it is the |
| 5817 | * only synchronisation we have for udpating the per-CPU variables. |
| 5818 | */ |
| 5819 | VM_BUG_ON(!irqs_disabled()); |
| 5820 | mem_cgroup_charge_statistics(memcg, page, -1); |
| 5821 | memcg_check_events(memcg, page); |
| 5822 | |
| 5823 | if (!mem_cgroup_is_root(memcg)) |
| 5824 | css_put(&memcg->css); |
| 5825 | } |
| 5826 | |
| 5827 | /** |
| 5828 | * mem_cgroup_uncharge_swap - uncharge a swap entry |
| 5829 | * @entry: swap entry to uncharge |
| 5830 | * |
| 5831 | * Drop the memsw charge associated with @entry. |
| 5832 | */ |
| 5833 | void mem_cgroup_uncharge_swap(swp_entry_t entry) |
| 5834 | { |
| 5835 | struct mem_cgroup *memcg; |
| 5836 | unsigned short id; |
| 5837 | |
| 5838 | if (!do_swap_account) |
| 5839 | return; |
| 5840 | |
| 5841 | id = swap_cgroup_record(entry, 0); |
| 5842 | rcu_read_lock(); |
| 5843 | memcg = mem_cgroup_from_id(id); |
| 5844 | if (memcg) { |
| 5845 | if (!mem_cgroup_is_root(memcg)) |
| 5846 | page_counter_uncharge(&memcg->memsw, 1); |
| 5847 | mem_cgroup_swap_statistics(memcg, false); |
| 5848 | mem_cgroup_id_put(memcg); |
| 5849 | } |
| 5850 | rcu_read_unlock(); |
| 5851 | } |
| 5852 | |
| 5853 | /* for remember boot option*/ |
| 5854 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
| 5855 | static int really_do_swap_account __initdata = 1; |
| 5856 | #else |
| 5857 | static int really_do_swap_account __initdata; |
| 5858 | #endif |
| 5859 | |
| 5860 | static int __init enable_swap_account(char *s) |
| 5861 | { |
| 5862 | if (!strcmp(s, "1")) |
| 5863 | really_do_swap_account = 1; |
| 5864 | else if (!strcmp(s, "0")) |
| 5865 | really_do_swap_account = 0; |
| 5866 | return 1; |
| 5867 | } |
| 5868 | __setup("swapaccount=", enable_swap_account); |
| 5869 | |
| 5870 | static struct cftype memsw_cgroup_files[] = { |
| 5871 | { |
| 5872 | .name = "memsw.usage_in_bytes", |
| 5873 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), |
| 5874 | .read_u64 = mem_cgroup_read_u64, |
| 5875 | }, |
| 5876 | { |
| 5877 | .name = "memsw.max_usage_in_bytes", |
| 5878 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), |
| 5879 | .write = mem_cgroup_reset, |
| 5880 | .read_u64 = mem_cgroup_read_u64, |
| 5881 | }, |
| 5882 | { |
| 5883 | .name = "memsw.limit_in_bytes", |
| 5884 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), |
| 5885 | .write = mem_cgroup_write, |
| 5886 | .read_u64 = mem_cgroup_read_u64, |
| 5887 | }, |
| 5888 | { |
| 5889 | .name = "memsw.failcnt", |
| 5890 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), |
| 5891 | .write = mem_cgroup_reset, |
| 5892 | .read_u64 = mem_cgroup_read_u64, |
| 5893 | }, |
| 5894 | { }, /* terminate */ |
| 5895 | }; |
| 5896 | |
| 5897 | static int __init mem_cgroup_swap_init(void) |
| 5898 | { |
| 5899 | if (!mem_cgroup_disabled() && really_do_swap_account) { |
| 5900 | do_swap_account = 1; |
| 5901 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
| 5902 | memsw_cgroup_files)); |
| 5903 | } |
| 5904 | return 0; |
| 5905 | } |
| 5906 | subsys_initcall(mem_cgroup_swap_init); |
| 5907 | |
| 5908 | #endif /* CONFIG_MEMCG_SWAP */ |