| 1 | /* |
| 2 | * linux/mm/oom_kill.c |
| 3 | * |
| 4 | * Copyright (C) 1998,2000 Rik van Riel |
| 5 | * Thanks go out to Claus Fischer for some serious inspiration and |
| 6 | * for goading me into coding this file... |
| 7 | * Copyright (C) 2010 Google, Inc. |
| 8 | * Rewritten by David Rientjes |
| 9 | * |
| 10 | * The routines in this file are used to kill a process when |
| 11 | * we're seriously out of memory. This gets called from __alloc_pages() |
| 12 | * in mm/page_alloc.c when we really run out of memory. |
| 13 | * |
| 14 | * Since we won't call these routines often (on a well-configured |
| 15 | * machine) this file will double as a 'coding guide' and a signpost |
| 16 | * for newbie kernel hackers. It features several pointers to major |
| 17 | * kernel subsystems and hints as to where to find out what things do. |
| 18 | */ |
| 19 | |
| 20 | #include <linux/oom.h> |
| 21 | #include <linux/mm.h> |
| 22 | #include <linux/err.h> |
| 23 | #include <linux/gfp.h> |
| 24 | #include <linux/sched.h> |
| 25 | #include <linux/sched/mm.h> |
| 26 | #include <linux/sched/coredump.h> |
| 27 | #include <linux/sched/task.h> |
| 28 | #include <linux/swap.h> |
| 29 | #include <linux/timex.h> |
| 30 | #include <linux/jiffies.h> |
| 31 | #include <linux/cpuset.h> |
| 32 | #include <linux/export.h> |
| 33 | #include <linux/notifier.h> |
| 34 | #include <linux/memcontrol.h> |
| 35 | #include <linux/mempolicy.h> |
| 36 | #include <linux/security.h> |
| 37 | #include <linux/ptrace.h> |
| 38 | #include <linux/freezer.h> |
| 39 | #include <linux/ftrace.h> |
| 40 | #include <linux/ratelimit.h> |
| 41 | #include <linux/kthread.h> |
| 42 | #include <linux/init.h> |
| 43 | #include <linux/mmu_notifier.h> |
| 44 | |
| 45 | #include <asm/tlb.h> |
| 46 | #include "internal.h" |
| 47 | |
| 48 | #define CREATE_TRACE_POINTS |
| 49 | #include <trace/events/oom.h> |
| 50 | |
| 51 | int sysctl_panic_on_oom; |
| 52 | int sysctl_oom_kill_allocating_task; |
| 53 | int sysctl_oom_dump_tasks = 1; |
| 54 | int sysctl_reap_mem_on_sigkill; |
| 55 | |
| 56 | DEFINE_MUTEX(oom_lock); |
| 57 | |
| 58 | #ifdef CONFIG_NUMA |
| 59 | /** |
| 60 | * has_intersects_mems_allowed() - check task eligiblity for kill |
| 61 | * @start: task struct of which task to consider |
| 62 | * @mask: nodemask passed to page allocator for mempolicy ooms |
| 63 | * |
| 64 | * Task eligibility is determined by whether or not a candidate task, @tsk, |
| 65 | * shares the same mempolicy nodes as current if it is bound by such a policy |
| 66 | * and whether or not it has the same set of allowed cpuset nodes. |
| 67 | */ |
| 68 | static bool has_intersects_mems_allowed(struct task_struct *start, |
| 69 | const nodemask_t *mask) |
| 70 | { |
| 71 | struct task_struct *tsk; |
| 72 | bool ret = false; |
| 73 | |
| 74 | rcu_read_lock(); |
| 75 | for_each_thread(start, tsk) { |
| 76 | if (mask) { |
| 77 | /* |
| 78 | * If this is a mempolicy constrained oom, tsk's |
| 79 | * cpuset is irrelevant. Only return true if its |
| 80 | * mempolicy intersects current, otherwise it may be |
| 81 | * needlessly killed. |
| 82 | */ |
| 83 | ret = mempolicy_nodemask_intersects(tsk, mask); |
| 84 | } else { |
| 85 | /* |
| 86 | * This is not a mempolicy constrained oom, so only |
| 87 | * check the mems of tsk's cpuset. |
| 88 | */ |
| 89 | ret = cpuset_mems_allowed_intersects(current, tsk); |
| 90 | } |
| 91 | if (ret) |
| 92 | break; |
| 93 | } |
| 94 | rcu_read_unlock(); |
| 95 | |
| 96 | return ret; |
| 97 | } |
| 98 | #else |
| 99 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
| 100 | const nodemask_t *mask) |
| 101 | { |
| 102 | return true; |
| 103 | } |
| 104 | #endif /* CONFIG_NUMA */ |
| 105 | |
| 106 | /* |
| 107 | * The process p may have detached its own ->mm while exiting or through |
| 108 | * use_mm(), but one or more of its subthreads may still have a valid |
| 109 | * pointer. Return p, or any of its subthreads with a valid ->mm, with |
| 110 | * task_lock() held. |
| 111 | */ |
| 112 | struct task_struct *find_lock_task_mm(struct task_struct *p) |
| 113 | { |
| 114 | struct task_struct *t; |
| 115 | |
| 116 | rcu_read_lock(); |
| 117 | |
| 118 | for_each_thread(p, t) { |
| 119 | task_lock(t); |
| 120 | if (likely(t->mm)) |
| 121 | goto found; |
| 122 | task_unlock(t); |
| 123 | } |
| 124 | t = NULL; |
| 125 | found: |
| 126 | rcu_read_unlock(); |
| 127 | |
| 128 | return t; |
| 129 | } |
| 130 | |
| 131 | /* |
| 132 | * order == -1 means the oom kill is required by sysrq, otherwise only |
| 133 | * for display purposes. |
| 134 | */ |
| 135 | static inline bool is_sysrq_oom(struct oom_control *oc) |
| 136 | { |
| 137 | return oc->order == -1; |
| 138 | } |
| 139 | |
| 140 | static inline bool is_memcg_oom(struct oom_control *oc) |
| 141 | { |
| 142 | return oc->memcg != NULL; |
| 143 | } |
| 144 | |
| 145 | /* return true if the task is not adequate as candidate victim task. */ |
| 146 | static bool oom_unkillable_task(struct task_struct *p, |
| 147 | struct mem_cgroup *memcg, const nodemask_t *nodemask) |
| 148 | { |
| 149 | if (is_global_init(p)) |
| 150 | return true; |
| 151 | if (p->flags & PF_KTHREAD) |
| 152 | return true; |
| 153 | |
| 154 | /* When mem_cgroup_out_of_memory() and p is not member of the group */ |
| 155 | if (memcg && !task_in_mem_cgroup(p, memcg)) |
| 156 | return true; |
| 157 | |
| 158 | /* p may not have freeable memory in nodemask */ |
| 159 | if (!has_intersects_mems_allowed(p, nodemask)) |
| 160 | return true; |
| 161 | |
| 162 | return false; |
| 163 | } |
| 164 | |
| 165 | /** |
| 166 | * oom_badness - heuristic function to determine which candidate task to kill |
| 167 | * @p: task struct of which task we should calculate |
| 168 | * @totalpages: total present RAM allowed for page allocation |
| 169 | * |
| 170 | * The heuristic for determining which task to kill is made to be as simple and |
| 171 | * predictable as possible. The goal is to return the highest value for the |
| 172 | * task consuming the most memory to avoid subsequent oom failures. |
| 173 | */ |
| 174 | unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg, |
| 175 | const nodemask_t *nodemask, unsigned long totalpages) |
| 176 | { |
| 177 | long points; |
| 178 | long adj; |
| 179 | |
| 180 | if (oom_unkillable_task(p, memcg, nodemask)) |
| 181 | return 0; |
| 182 | |
| 183 | p = find_lock_task_mm(p); |
| 184 | if (!p) |
| 185 | return 0; |
| 186 | |
| 187 | /* |
| 188 | * Do not even consider tasks which are explicitly marked oom |
| 189 | * unkillable or have been already oom reaped or the are in |
| 190 | * the middle of vfork |
| 191 | */ |
| 192 | adj = (long)p->signal->oom_score_adj; |
| 193 | if (adj == OOM_SCORE_ADJ_MIN || |
| 194 | test_bit(MMF_OOM_SKIP, &p->mm->flags) || |
| 195 | in_vfork(p)) { |
| 196 | task_unlock(p); |
| 197 | return 0; |
| 198 | } |
| 199 | |
| 200 | /* |
| 201 | * The baseline for the badness score is the proportion of RAM that each |
| 202 | * task's rss, pagetable and swap space use. |
| 203 | */ |
| 204 | points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) + |
| 205 | atomic_long_read(&p->mm->nr_ptes) + mm_nr_pmds(p->mm); |
| 206 | task_unlock(p); |
| 207 | |
| 208 | /* |
| 209 | * Root processes get 3% bonus, just like the __vm_enough_memory() |
| 210 | * implementation used by LSMs. |
| 211 | */ |
| 212 | if (has_capability_noaudit(p, CAP_SYS_ADMIN)) |
| 213 | points -= (points * 3) / 100; |
| 214 | |
| 215 | /* Normalize to oom_score_adj units */ |
| 216 | adj *= totalpages / 1000; |
| 217 | points += adj; |
| 218 | |
| 219 | /* |
| 220 | * Never return 0 for an eligible task regardless of the root bonus and |
| 221 | * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here). |
| 222 | */ |
| 223 | return points > 0 ? points : 1; |
| 224 | } |
| 225 | |
| 226 | enum oom_constraint { |
| 227 | CONSTRAINT_NONE, |
| 228 | CONSTRAINT_CPUSET, |
| 229 | CONSTRAINT_MEMORY_POLICY, |
| 230 | CONSTRAINT_MEMCG, |
| 231 | }; |
| 232 | |
| 233 | /* |
| 234 | * Determine the type of allocation constraint. |
| 235 | */ |
| 236 | static enum oom_constraint constrained_alloc(struct oom_control *oc) |
| 237 | { |
| 238 | struct zone *zone; |
| 239 | struct zoneref *z; |
| 240 | enum zone_type high_zoneidx = gfp_zone(oc->gfp_mask); |
| 241 | bool cpuset_limited = false; |
| 242 | int nid; |
| 243 | |
| 244 | if (is_memcg_oom(oc)) { |
| 245 | oc->totalpages = mem_cgroup_get_limit(oc->memcg) ?: 1; |
| 246 | return CONSTRAINT_MEMCG; |
| 247 | } |
| 248 | |
| 249 | /* Default to all available memory */ |
| 250 | oc->totalpages = totalram_pages + total_swap_pages; |
| 251 | |
| 252 | if (!IS_ENABLED(CONFIG_NUMA)) |
| 253 | return CONSTRAINT_NONE; |
| 254 | |
| 255 | if (!oc->zonelist) |
| 256 | return CONSTRAINT_NONE; |
| 257 | /* |
| 258 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid |
| 259 | * to kill current.We have to random task kill in this case. |
| 260 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. |
| 261 | */ |
| 262 | if (oc->gfp_mask & __GFP_THISNODE) |
| 263 | return CONSTRAINT_NONE; |
| 264 | |
| 265 | /* |
| 266 | * This is not a __GFP_THISNODE allocation, so a truncated nodemask in |
| 267 | * the page allocator means a mempolicy is in effect. Cpuset policy |
| 268 | * is enforced in get_page_from_freelist(). |
| 269 | */ |
| 270 | if (oc->nodemask && |
| 271 | !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) { |
| 272 | oc->totalpages = total_swap_pages; |
| 273 | for_each_node_mask(nid, *oc->nodemask) |
| 274 | oc->totalpages += node_spanned_pages(nid); |
| 275 | return CONSTRAINT_MEMORY_POLICY; |
| 276 | } |
| 277 | |
| 278 | /* Check this allocation failure is caused by cpuset's wall function */ |
| 279 | for_each_zone_zonelist_nodemask(zone, z, oc->zonelist, |
| 280 | high_zoneidx, oc->nodemask) |
| 281 | if (!cpuset_zone_allowed(zone, oc->gfp_mask)) |
| 282 | cpuset_limited = true; |
| 283 | |
| 284 | if (cpuset_limited) { |
| 285 | oc->totalpages = total_swap_pages; |
| 286 | for_each_node_mask(nid, cpuset_current_mems_allowed) |
| 287 | oc->totalpages += node_spanned_pages(nid); |
| 288 | return CONSTRAINT_CPUSET; |
| 289 | } |
| 290 | return CONSTRAINT_NONE; |
| 291 | } |
| 292 | |
| 293 | static int oom_evaluate_task(struct task_struct *task, void *arg) |
| 294 | { |
| 295 | struct oom_control *oc = arg; |
| 296 | unsigned long points; |
| 297 | |
| 298 | if (oom_unkillable_task(task, NULL, oc->nodemask)) |
| 299 | goto next; |
| 300 | |
| 301 | /* |
| 302 | * This task already has access to memory reserves and is being killed. |
| 303 | * Don't allow any other task to have access to the reserves unless |
| 304 | * the task has MMF_OOM_SKIP because chances that it would release |
| 305 | * any memory is quite low. |
| 306 | */ |
| 307 | if (!is_sysrq_oom(oc) && tsk_is_oom_victim(task)) { |
| 308 | if (test_bit(MMF_OOM_SKIP, &task->signal->oom_mm->flags)) |
| 309 | goto next; |
| 310 | goto abort; |
| 311 | } |
| 312 | |
| 313 | /* |
| 314 | * If task is allocating a lot of memory and has been marked to be |
| 315 | * killed first if it triggers an oom, then select it. |
| 316 | */ |
| 317 | if (oom_task_origin(task)) { |
| 318 | points = ULONG_MAX; |
| 319 | goto select; |
| 320 | } |
| 321 | |
| 322 | points = oom_badness(task, NULL, oc->nodemask, oc->totalpages); |
| 323 | if (!points || points < oc->chosen_points) |
| 324 | goto next; |
| 325 | |
| 326 | /* Prefer thread group leaders for display purposes */ |
| 327 | if (points == oc->chosen_points && thread_group_leader(oc->chosen)) |
| 328 | goto next; |
| 329 | select: |
| 330 | if (oc->chosen) |
| 331 | put_task_struct(oc->chosen); |
| 332 | get_task_struct(task); |
| 333 | oc->chosen = task; |
| 334 | oc->chosen_points = points; |
| 335 | next: |
| 336 | return 0; |
| 337 | abort: |
| 338 | if (oc->chosen) |
| 339 | put_task_struct(oc->chosen); |
| 340 | oc->chosen = (void *)-1UL; |
| 341 | return 1; |
| 342 | } |
| 343 | |
| 344 | /* |
| 345 | * Simple selection loop. We choose the process with the highest number of |
| 346 | * 'points'. In case scan was aborted, oc->chosen is set to -1. |
| 347 | */ |
| 348 | static void select_bad_process(struct oom_control *oc) |
| 349 | { |
| 350 | if (is_memcg_oom(oc)) |
| 351 | mem_cgroup_scan_tasks(oc->memcg, oom_evaluate_task, oc); |
| 352 | else { |
| 353 | struct task_struct *p; |
| 354 | |
| 355 | rcu_read_lock(); |
| 356 | for_each_process(p) |
| 357 | if (oom_evaluate_task(p, oc)) |
| 358 | break; |
| 359 | rcu_read_unlock(); |
| 360 | } |
| 361 | |
| 362 | oc->chosen_points = oc->chosen_points * 1000 / oc->totalpages; |
| 363 | } |
| 364 | |
| 365 | /** |
| 366 | * dump_tasks - dump current memory state of all system tasks |
| 367 | * @memcg: current's memory controller, if constrained |
| 368 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
| 369 | * |
| 370 | * Dumps the current memory state of all eligible tasks. Tasks not in the same |
| 371 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes |
| 372 | * are not shown. |
| 373 | * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes, |
| 374 | * swapents, oom_score_adj value, and name. |
| 375 | */ |
| 376 | static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask) |
| 377 | { |
| 378 | struct task_struct *p; |
| 379 | struct task_struct *task; |
| 380 | |
| 381 | pr_info("[ pid ] uid tgid total_vm rss nr_ptes nr_pmds swapents oom_score_adj name\n"); |
| 382 | rcu_read_lock(); |
| 383 | for_each_process(p) { |
| 384 | if (oom_unkillable_task(p, memcg, nodemask)) |
| 385 | continue; |
| 386 | |
| 387 | task = find_lock_task_mm(p); |
| 388 | if (!task) { |
| 389 | /* |
| 390 | * This is a kthread or all of p's threads have already |
| 391 | * detached their mm's. There's no need to report |
| 392 | * them; they can't be oom killed anyway. |
| 393 | */ |
| 394 | continue; |
| 395 | } |
| 396 | |
| 397 | pr_info("[%5d] %5d %5d %8lu %8lu %7ld %7ld %8lu %5hd %s\n", |
| 398 | task->pid, from_kuid(&init_user_ns, task_uid(task)), |
| 399 | task->tgid, task->mm->total_vm, get_mm_rss(task->mm), |
| 400 | atomic_long_read(&task->mm->nr_ptes), |
| 401 | mm_nr_pmds(task->mm), |
| 402 | get_mm_counter(task->mm, MM_SWAPENTS), |
| 403 | task->signal->oom_score_adj, task->comm); |
| 404 | task_unlock(task); |
| 405 | } |
| 406 | rcu_read_unlock(); |
| 407 | } |
| 408 | |
| 409 | static void dump_header(struct oom_control *oc, struct task_struct *p) |
| 410 | { |
| 411 | pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), nodemask=", |
| 412 | current->comm, oc->gfp_mask, &oc->gfp_mask); |
| 413 | if (oc->nodemask) |
| 414 | pr_cont("%*pbl", nodemask_pr_args(oc->nodemask)); |
| 415 | else |
| 416 | pr_cont("(null)"); |
| 417 | pr_cont(", order=%d, oom_score_adj=%hd\n", |
| 418 | oc->order, current->signal->oom_score_adj); |
| 419 | if (!IS_ENABLED(CONFIG_COMPACTION) && oc->order) |
| 420 | pr_warn("COMPACTION is disabled!!!\n"); |
| 421 | |
| 422 | cpuset_print_current_mems_allowed(); |
| 423 | dump_stack(); |
| 424 | if (oc->memcg) |
| 425 | mem_cgroup_print_oom_info(oc->memcg, p); |
| 426 | else |
| 427 | show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask); |
| 428 | if (sysctl_oom_dump_tasks) |
| 429 | dump_tasks(oc->memcg, oc->nodemask); |
| 430 | } |
| 431 | |
| 432 | /* |
| 433 | * Number of OOM victims in flight |
| 434 | */ |
| 435 | static atomic_t oom_victims = ATOMIC_INIT(0); |
| 436 | static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait); |
| 437 | |
| 438 | static bool oom_killer_disabled __read_mostly; |
| 439 | |
| 440 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
| 441 | |
| 442 | /* |
| 443 | * task->mm can be NULL if the task is the exited group leader. So to |
| 444 | * determine whether the task is using a particular mm, we examine all the |
| 445 | * task's threads: if one of those is using this mm then this task was also |
| 446 | * using it. |
| 447 | */ |
| 448 | bool process_shares_mm(struct task_struct *p, struct mm_struct *mm) |
| 449 | { |
| 450 | struct task_struct *t; |
| 451 | |
| 452 | for_each_thread(p, t) { |
| 453 | struct mm_struct *t_mm = READ_ONCE(t->mm); |
| 454 | if (t_mm) |
| 455 | return t_mm == mm; |
| 456 | } |
| 457 | return false; |
| 458 | } |
| 459 | |
| 460 | #ifdef CONFIG_MMU |
| 461 | /* |
| 462 | * OOM Reaper kernel thread which tries to reap the memory used by the OOM |
| 463 | * victim (if that is possible) to help the OOM killer to move on. |
| 464 | */ |
| 465 | static struct task_struct *oom_reaper_th; |
| 466 | static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait); |
| 467 | static struct task_struct *oom_reaper_list; |
| 468 | static DEFINE_SPINLOCK(oom_reaper_lock); |
| 469 | |
| 470 | void __oom_reap_task_mm(struct mm_struct *mm) |
| 471 | { |
| 472 | struct vm_area_struct *vma; |
| 473 | |
| 474 | /* |
| 475 | * Tell all users of get_user/copy_from_user etc... that the content |
| 476 | * is no longer stable. No barriers really needed because unmapping |
| 477 | * should imply barriers already and the reader would hit a page fault |
| 478 | * if it stumbled over a reaped memory. |
| 479 | */ |
| 480 | set_bit(MMF_UNSTABLE, &mm->flags); |
| 481 | |
| 482 | for (vma = mm->mmap ; vma; vma = vma->vm_next) { |
| 483 | if (!can_madv_dontneed_vma(vma)) |
| 484 | continue; |
| 485 | |
| 486 | /* |
| 487 | * Only anonymous pages have a good chance to be dropped |
| 488 | * without additional steps which we cannot afford as we |
| 489 | * are OOM already. |
| 490 | * |
| 491 | * We do not even care about fs backed pages because all |
| 492 | * which are reclaimable have already been reclaimed and |
| 493 | * we do not want to block exit_mmap by keeping mm ref |
| 494 | * count elevated without a good reason. |
| 495 | */ |
| 496 | if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED)) { |
| 497 | struct mmu_gather tlb; |
| 498 | |
| 499 | tlb_gather_mmu(&tlb, mm, vma->vm_start, vma->vm_end); |
| 500 | unmap_page_range(&tlb, vma, vma->vm_start, vma->vm_end, |
| 501 | NULL); |
| 502 | tlb_finish_mmu(&tlb, vma->vm_start, vma->vm_end); |
| 503 | } |
| 504 | } |
| 505 | } |
| 506 | |
| 507 | static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm) |
| 508 | { |
| 509 | bool ret = true; |
| 510 | |
| 511 | /* |
| 512 | * We have to make sure to not race with the victim exit path |
| 513 | * and cause premature new oom victim selection: |
| 514 | * oom_reap_task_mm exit_mm |
| 515 | * mmget_not_zero |
| 516 | * mmput |
| 517 | * atomic_dec_and_test |
| 518 | * exit_oom_victim |
| 519 | * [...] |
| 520 | * out_of_memory |
| 521 | * select_bad_process |
| 522 | * # no TIF_MEMDIE task selects new victim |
| 523 | * unmap_page_range # frees some memory |
| 524 | */ |
| 525 | mutex_lock(&oom_lock); |
| 526 | |
| 527 | if (!down_read_trylock(&mm->mmap_sem)) { |
| 528 | ret = false; |
| 529 | trace_skip_task_reaping(tsk->pid); |
| 530 | goto unlock_oom; |
| 531 | } |
| 532 | |
| 533 | /* |
| 534 | * If the mm has notifiers then we would need to invalidate them around |
| 535 | * unmap_page_range and that is risky because notifiers can sleep and |
| 536 | * what they do is basically undeterministic. So let's have a short |
| 537 | * sleep to give the oom victim some more time. |
| 538 | * TODO: we really want to get rid of this ugly hack and make sure that |
| 539 | * notifiers cannot block for unbounded amount of time and add |
| 540 | * mmu_notifier_invalidate_range_{start,end} around unmap_page_range |
| 541 | */ |
| 542 | if (mm_has_notifiers(mm)) { |
| 543 | up_read(&mm->mmap_sem); |
| 544 | schedule_timeout_idle(HZ); |
| 545 | goto unlock_oom; |
| 546 | } |
| 547 | |
| 548 | /* |
| 549 | * MMF_OOM_SKIP is set by exit_mmap when the OOM reaper can't |
| 550 | * work on the mm anymore. The check for MMF_OOM_SKIP must run |
| 551 | * under mmap_sem for reading because it serializes against the |
| 552 | * down_write();up_write() cycle in exit_mmap(). |
| 553 | */ |
| 554 | if (test_bit(MMF_OOM_SKIP, &mm->flags)) { |
| 555 | up_read(&mm->mmap_sem); |
| 556 | trace_skip_task_reaping(tsk->pid); |
| 557 | goto unlock_oom; |
| 558 | } |
| 559 | |
| 560 | trace_start_task_reaping(tsk->pid); |
| 561 | |
| 562 | __oom_reap_task_mm(mm); |
| 563 | |
| 564 | pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n", |
| 565 | task_pid_nr(tsk), tsk->comm, |
| 566 | K(get_mm_counter(mm, MM_ANONPAGES)), |
| 567 | K(get_mm_counter(mm, MM_FILEPAGES)), |
| 568 | K(get_mm_counter(mm, MM_SHMEMPAGES))); |
| 569 | up_read(&mm->mmap_sem); |
| 570 | |
| 571 | trace_finish_task_reaping(tsk->pid); |
| 572 | unlock_oom: |
| 573 | mutex_unlock(&oom_lock); |
| 574 | return ret; |
| 575 | } |
| 576 | |
| 577 | #define MAX_OOM_REAP_RETRIES 10 |
| 578 | static void oom_reap_task(struct task_struct *tsk) |
| 579 | { |
| 580 | int attempts = 0; |
| 581 | struct mm_struct *mm = tsk->signal->oom_mm; |
| 582 | |
| 583 | /* Retry the down_read_trylock(mmap_sem) a few times */ |
| 584 | while (attempts++ < MAX_OOM_REAP_RETRIES && !oom_reap_task_mm(tsk, mm)) |
| 585 | schedule_timeout_idle(HZ/10); |
| 586 | |
| 587 | if (attempts <= MAX_OOM_REAP_RETRIES) |
| 588 | goto done; |
| 589 | |
| 590 | pr_info("oom_reaper: unable to reap pid:%d (%s)\n", |
| 591 | task_pid_nr(tsk), tsk->comm); |
| 592 | debug_show_all_locks(); |
| 593 | |
| 594 | done: |
| 595 | tsk->oom_reaper_list = NULL; |
| 596 | |
| 597 | /* |
| 598 | * Hide this mm from OOM killer because it has been either reaped or |
| 599 | * somebody can't call up_write(mmap_sem). |
| 600 | */ |
| 601 | set_bit(MMF_OOM_SKIP, &mm->flags); |
| 602 | |
| 603 | /* Drop a reference taken by wake_oom_reaper */ |
| 604 | put_task_struct(tsk); |
| 605 | } |
| 606 | |
| 607 | static int oom_reaper(void *unused) |
| 608 | { |
| 609 | while (true) { |
| 610 | struct task_struct *tsk = NULL; |
| 611 | |
| 612 | wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL); |
| 613 | spin_lock(&oom_reaper_lock); |
| 614 | if (oom_reaper_list != NULL) { |
| 615 | tsk = oom_reaper_list; |
| 616 | oom_reaper_list = tsk->oom_reaper_list; |
| 617 | } |
| 618 | spin_unlock(&oom_reaper_lock); |
| 619 | |
| 620 | if (tsk) |
| 621 | oom_reap_task(tsk); |
| 622 | } |
| 623 | |
| 624 | return 0; |
| 625 | } |
| 626 | |
| 627 | static void wake_oom_reaper(struct task_struct *tsk) |
| 628 | { |
| 629 | if (!oom_reaper_th) |
| 630 | return; |
| 631 | /* |
| 632 | * Move the lock here to avoid scenario of queuing |
| 633 | * the same task by both OOM killer and any other SIGKILL |
| 634 | * path. |
| 635 | */ |
| 636 | |
| 637 | spin_lock(&oom_reaper_lock); |
| 638 | |
| 639 | /* mm is already queued? */ |
| 640 | if (test_and_set_bit(MMF_OOM_REAP_QUEUED, &tsk->signal->oom_mm->flags)) { |
| 641 | spin_unlock(&oom_reaper_lock); |
| 642 | return; |
| 643 | } |
| 644 | |
| 645 | get_task_struct(tsk); |
| 646 | |
| 647 | tsk->oom_reaper_list = oom_reaper_list; |
| 648 | oom_reaper_list = tsk; |
| 649 | spin_unlock(&oom_reaper_lock); |
| 650 | trace_wake_reaper(tsk->pid); |
| 651 | wake_up(&oom_reaper_wait); |
| 652 | } |
| 653 | |
| 654 | static int __init oom_init(void) |
| 655 | { |
| 656 | oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper"); |
| 657 | if (IS_ERR(oom_reaper_th)) { |
| 658 | pr_err("Unable to start OOM reaper %ld. Continuing regardless\n", |
| 659 | PTR_ERR(oom_reaper_th)); |
| 660 | oom_reaper_th = NULL; |
| 661 | } |
| 662 | return 0; |
| 663 | } |
| 664 | subsys_initcall(oom_init) |
| 665 | #else |
| 666 | static inline void wake_oom_reaper(struct task_struct *tsk) |
| 667 | { |
| 668 | } |
| 669 | #endif /* CONFIG_MMU */ |
| 670 | |
| 671 | static void __mark_oom_victim(struct task_struct *tsk) |
| 672 | { |
| 673 | struct mm_struct *mm = tsk->mm; |
| 674 | |
| 675 | if (!cmpxchg(&tsk->signal->oom_mm, NULL, mm)) { |
| 676 | mmgrab(tsk->signal->oom_mm); |
| 677 | set_bit(MMF_OOM_VICTIM, &mm->flags); |
| 678 | } |
| 679 | } |
| 680 | |
| 681 | /** |
| 682 | * mark_oom_victim - mark the given task as OOM victim |
| 683 | * @tsk: task to mark |
| 684 | * |
| 685 | * Has to be called with oom_lock held and never after |
| 686 | * oom has been disabled already. |
| 687 | * |
| 688 | * tsk->mm has to be non NULL and caller has to guarantee it is stable (either |
| 689 | * under task_lock or operate on the current). |
| 690 | */ |
| 691 | static void mark_oom_victim(struct task_struct *tsk) |
| 692 | { |
| 693 | WARN_ON(oom_killer_disabled); |
| 694 | /* OOM killer might race with memcg OOM */ |
| 695 | if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE)) |
| 696 | return; |
| 697 | |
| 698 | /* oom_mm is bound to the signal struct life time. */ |
| 699 | __mark_oom_victim(tsk); |
| 700 | |
| 701 | /* |
| 702 | * Make sure that the task is woken up from uninterruptible sleep |
| 703 | * if it is frozen because OOM killer wouldn't be able to free |
| 704 | * any memory and livelock. freezing_slow_path will tell the freezer |
| 705 | * that TIF_MEMDIE tasks should be ignored. |
| 706 | */ |
| 707 | __thaw_task(tsk); |
| 708 | atomic_inc(&oom_victims); |
| 709 | trace_mark_victim(tsk->pid); |
| 710 | } |
| 711 | |
| 712 | /** |
| 713 | * exit_oom_victim - note the exit of an OOM victim |
| 714 | */ |
| 715 | void exit_oom_victim(void) |
| 716 | { |
| 717 | clear_thread_flag(TIF_MEMDIE); |
| 718 | |
| 719 | if (!atomic_dec_return(&oom_victims)) |
| 720 | wake_up_all(&oom_victims_wait); |
| 721 | } |
| 722 | |
| 723 | /** |
| 724 | * oom_killer_enable - enable OOM killer |
| 725 | */ |
| 726 | void oom_killer_enable(void) |
| 727 | { |
| 728 | oom_killer_disabled = false; |
| 729 | pr_info("OOM killer enabled.\n"); |
| 730 | } |
| 731 | |
| 732 | /** |
| 733 | * oom_killer_disable - disable OOM killer |
| 734 | * @timeout: maximum timeout to wait for oom victims in jiffies |
| 735 | * |
| 736 | * Forces all page allocations to fail rather than trigger OOM killer. |
| 737 | * Will block and wait until all OOM victims are killed or the given |
| 738 | * timeout expires. |
| 739 | * |
| 740 | * The function cannot be called when there are runnable user tasks because |
| 741 | * the userspace would see unexpected allocation failures as a result. Any |
| 742 | * new usage of this function should be consulted with MM people. |
| 743 | * |
| 744 | * Returns true if successful and false if the OOM killer cannot be |
| 745 | * disabled. |
| 746 | */ |
| 747 | bool oom_killer_disable(signed long timeout) |
| 748 | { |
| 749 | signed long ret; |
| 750 | |
| 751 | /* |
| 752 | * Make sure to not race with an ongoing OOM killer. Check that the |
| 753 | * current is not killed (possibly due to sharing the victim's memory). |
| 754 | */ |
| 755 | if (mutex_lock_killable(&oom_lock)) |
| 756 | return false; |
| 757 | oom_killer_disabled = true; |
| 758 | mutex_unlock(&oom_lock); |
| 759 | |
| 760 | ret = wait_event_interruptible_timeout(oom_victims_wait, |
| 761 | !atomic_read(&oom_victims), timeout); |
| 762 | if (ret <= 0) { |
| 763 | oom_killer_enable(); |
| 764 | return false; |
| 765 | } |
| 766 | pr_info("OOM killer disabled.\n"); |
| 767 | |
| 768 | return true; |
| 769 | } |
| 770 | |
| 771 | static inline bool __task_will_free_mem(struct task_struct *task) |
| 772 | { |
| 773 | struct signal_struct *sig = task->signal; |
| 774 | |
| 775 | /* |
| 776 | * A coredumping process may sleep for an extended period in exit_mm(), |
| 777 | * so the oom killer cannot assume that the process will promptly exit |
| 778 | * and release memory. |
| 779 | */ |
| 780 | if (sig->flags & SIGNAL_GROUP_COREDUMP) |
| 781 | return false; |
| 782 | |
| 783 | if (sig->flags & SIGNAL_GROUP_EXIT) |
| 784 | return true; |
| 785 | |
| 786 | if (thread_group_empty(task) && (task->flags & PF_EXITING)) |
| 787 | return true; |
| 788 | |
| 789 | return false; |
| 790 | } |
| 791 | |
| 792 | /* |
| 793 | * Checks whether the given task is dying or exiting and likely to |
| 794 | * release its address space. This means that all threads and processes |
| 795 | * sharing the same mm have to be killed or exiting. |
| 796 | * Caller has to make sure that task->mm is stable (hold task_lock or |
| 797 | * it operates on the current). |
| 798 | */ |
| 799 | static bool task_will_free_mem(struct task_struct *task) |
| 800 | { |
| 801 | struct mm_struct *mm = task->mm; |
| 802 | struct task_struct *p; |
| 803 | bool ret = true; |
| 804 | |
| 805 | /* |
| 806 | * Skip tasks without mm because it might have passed its exit_mm and |
| 807 | * exit_oom_victim. oom_reaper could have rescued that but do not rely |
| 808 | * on that for now. We can consider find_lock_task_mm in future. |
| 809 | */ |
| 810 | if (!mm) |
| 811 | return false; |
| 812 | |
| 813 | if (!__task_will_free_mem(task)) |
| 814 | return false; |
| 815 | |
| 816 | /* |
| 817 | * This task has already been drained by the oom reaper so there are |
| 818 | * only small chances it will free some more |
| 819 | */ |
| 820 | if (test_bit(MMF_OOM_SKIP, &mm->flags)) |
| 821 | return false; |
| 822 | |
| 823 | if (atomic_read(&mm->mm_users) <= 1) |
| 824 | return true; |
| 825 | |
| 826 | /* |
| 827 | * Make sure that all tasks which share the mm with the given tasks |
| 828 | * are dying as well to make sure that a) nobody pins its mm and |
| 829 | * b) the task is also reapable by the oom reaper. |
| 830 | */ |
| 831 | rcu_read_lock(); |
| 832 | for_each_process(p) { |
| 833 | if (!process_shares_mm(p, mm)) |
| 834 | continue; |
| 835 | if (same_thread_group(task, p)) |
| 836 | continue; |
| 837 | ret = __task_will_free_mem(p); |
| 838 | if (!ret) |
| 839 | break; |
| 840 | } |
| 841 | rcu_read_unlock(); |
| 842 | |
| 843 | return ret; |
| 844 | } |
| 845 | |
| 846 | static void oom_kill_process(struct oom_control *oc, const char *message) |
| 847 | { |
| 848 | struct task_struct *p = oc->chosen; |
| 849 | unsigned int points = oc->chosen_points; |
| 850 | struct task_struct *victim = p; |
| 851 | struct task_struct *child; |
| 852 | struct task_struct *t; |
| 853 | struct mm_struct *mm; |
| 854 | unsigned int victim_points = 0; |
| 855 | static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
| 856 | DEFAULT_RATELIMIT_BURST); |
| 857 | bool can_oom_reap = true; |
| 858 | |
| 859 | /* |
| 860 | * If the task is already exiting, don't alarm the sysadmin or kill |
| 861 | * its children or threads, just give it access to memory reserves |
| 862 | * so it can die quickly |
| 863 | */ |
| 864 | task_lock(p); |
| 865 | if (task_will_free_mem(p)) { |
| 866 | mark_oom_victim(p); |
| 867 | wake_oom_reaper(p); |
| 868 | task_unlock(p); |
| 869 | put_task_struct(p); |
| 870 | return; |
| 871 | } |
| 872 | task_unlock(p); |
| 873 | |
| 874 | if (__ratelimit(&oom_rs)) |
| 875 | dump_header(oc, p); |
| 876 | |
| 877 | pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n", |
| 878 | message, task_pid_nr(p), p->comm, points); |
| 879 | |
| 880 | /* |
| 881 | * If any of p's children has a different mm and is eligible for kill, |
| 882 | * the one with the highest oom_badness() score is sacrificed for its |
| 883 | * parent. This attempts to lose the minimal amount of work done while |
| 884 | * still freeing memory. |
| 885 | */ |
| 886 | read_lock(&tasklist_lock); |
| 887 | |
| 888 | /* |
| 889 | * The task 'p' might have already exited before reaching here. The |
| 890 | * put_task_struct() will free task_struct 'p' while the loop still try |
| 891 | * to access the field of 'p', so, get an extra reference. |
| 892 | */ |
| 893 | get_task_struct(p); |
| 894 | for_each_thread(p, t) { |
| 895 | list_for_each_entry(child, &t->children, sibling) { |
| 896 | unsigned int child_points; |
| 897 | |
| 898 | if (process_shares_mm(child, p->mm)) |
| 899 | continue; |
| 900 | /* |
| 901 | * oom_badness() returns 0 if the thread is unkillable |
| 902 | */ |
| 903 | child_points = oom_badness(child, |
| 904 | oc->memcg, oc->nodemask, oc->totalpages); |
| 905 | if (child_points > victim_points) { |
| 906 | put_task_struct(victim); |
| 907 | victim = child; |
| 908 | victim_points = child_points; |
| 909 | get_task_struct(victim); |
| 910 | } |
| 911 | } |
| 912 | } |
| 913 | put_task_struct(p); |
| 914 | read_unlock(&tasklist_lock); |
| 915 | |
| 916 | p = find_lock_task_mm(victim); |
| 917 | if (!p) { |
| 918 | put_task_struct(victim); |
| 919 | return; |
| 920 | } else if (victim != p) { |
| 921 | get_task_struct(p); |
| 922 | put_task_struct(victim); |
| 923 | victim = p; |
| 924 | } |
| 925 | |
| 926 | /* Get a reference to safely compare mm after task_unlock(victim) */ |
| 927 | mm = victim->mm; |
| 928 | mmgrab(mm); |
| 929 | |
| 930 | /* Raise event before sending signal: task reaper must see this */ |
| 931 | count_vm_event(OOM_KILL); |
| 932 | count_memcg_event_mm(mm, OOM_KILL); |
| 933 | |
| 934 | /* |
| 935 | * We should send SIGKILL before granting access to memory reserves |
| 936 | * in order to prevent the OOM victim from depleting the memory |
| 937 | * reserves from the user space under its control. |
| 938 | */ |
| 939 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true); |
| 940 | mark_oom_victim(victim); |
| 941 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n", |
| 942 | task_pid_nr(victim), victim->comm, K(victim->mm->total_vm), |
| 943 | K(get_mm_counter(victim->mm, MM_ANONPAGES)), |
| 944 | K(get_mm_counter(victim->mm, MM_FILEPAGES)), |
| 945 | K(get_mm_counter(victim->mm, MM_SHMEMPAGES))); |
| 946 | task_unlock(victim); |
| 947 | |
| 948 | /* |
| 949 | * Kill all user processes sharing victim->mm in other thread groups, if |
| 950 | * any. They don't get access to memory reserves, though, to avoid |
| 951 | * depletion of all memory. This prevents mm->mmap_sem livelock when an |
| 952 | * oom killed thread cannot exit because it requires the semaphore and |
| 953 | * its contended by another thread trying to allocate memory itself. |
| 954 | * That thread will now get access to memory reserves since it has a |
| 955 | * pending fatal signal. |
| 956 | */ |
| 957 | rcu_read_lock(); |
| 958 | for_each_process(p) { |
| 959 | if (!process_shares_mm(p, mm)) |
| 960 | continue; |
| 961 | if (same_thread_group(p, victim)) |
| 962 | continue; |
| 963 | if (is_global_init(p)) { |
| 964 | can_oom_reap = false; |
| 965 | set_bit(MMF_OOM_SKIP, &mm->flags); |
| 966 | pr_info("oom killer %d (%s) has mm pinned by %d (%s)\n", |
| 967 | task_pid_nr(victim), victim->comm, |
| 968 | task_pid_nr(p), p->comm); |
| 969 | continue; |
| 970 | } |
| 971 | /* |
| 972 | * No use_mm() user needs to read from the userspace so we are |
| 973 | * ok to reap it. |
| 974 | */ |
| 975 | if (unlikely(p->flags & PF_KTHREAD)) |
| 976 | continue; |
| 977 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true); |
| 978 | } |
| 979 | rcu_read_unlock(); |
| 980 | |
| 981 | if (can_oom_reap) |
| 982 | wake_oom_reaper(victim); |
| 983 | |
| 984 | mmdrop(mm); |
| 985 | put_task_struct(victim); |
| 986 | } |
| 987 | #undef K |
| 988 | |
| 989 | /* |
| 990 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. |
| 991 | */ |
| 992 | static void check_panic_on_oom(struct oom_control *oc, |
| 993 | enum oom_constraint constraint) |
| 994 | { |
| 995 | if (likely(!sysctl_panic_on_oom)) |
| 996 | return; |
| 997 | if (sysctl_panic_on_oom != 2) { |
| 998 | /* |
| 999 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel |
| 1000 | * does not panic for cpuset, mempolicy, or memcg allocation |
| 1001 | * failures. |
| 1002 | */ |
| 1003 | if (constraint != CONSTRAINT_NONE) |
| 1004 | return; |
| 1005 | } |
| 1006 | /* Do not panic for oom kills triggered by sysrq */ |
| 1007 | if (is_sysrq_oom(oc)) |
| 1008 | return; |
| 1009 | dump_header(oc, NULL); |
| 1010 | panic("Out of memory: %s panic_on_oom is enabled\n", |
| 1011 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); |
| 1012 | } |
| 1013 | |
| 1014 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
| 1015 | |
| 1016 | int register_oom_notifier(struct notifier_block *nb) |
| 1017 | { |
| 1018 | return blocking_notifier_chain_register(&oom_notify_list, nb); |
| 1019 | } |
| 1020 | EXPORT_SYMBOL_GPL(register_oom_notifier); |
| 1021 | |
| 1022 | int unregister_oom_notifier(struct notifier_block *nb) |
| 1023 | { |
| 1024 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); |
| 1025 | } |
| 1026 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); |
| 1027 | |
| 1028 | /** |
| 1029 | * out_of_memory - kill the "best" process when we run out of memory |
| 1030 | * @oc: pointer to struct oom_control |
| 1031 | * |
| 1032 | * If we run out of memory, we have the choice between either |
| 1033 | * killing a random task (bad), letting the system crash (worse) |
| 1034 | * OR try to be smart about which process to kill. Note that we |
| 1035 | * don't have to be perfect here, we just have to be good. |
| 1036 | */ |
| 1037 | bool out_of_memory(struct oom_control *oc) |
| 1038 | { |
| 1039 | unsigned long freed = 0; |
| 1040 | enum oom_constraint constraint = CONSTRAINT_NONE; |
| 1041 | |
| 1042 | if (oom_killer_disabled) |
| 1043 | return false; |
| 1044 | |
| 1045 | if (!is_memcg_oom(oc)) { |
| 1046 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); |
| 1047 | if (freed > 0) |
| 1048 | /* Got some memory back in the last second. */ |
| 1049 | return true; |
| 1050 | } |
| 1051 | |
| 1052 | /* |
| 1053 | * If current has a pending SIGKILL or is exiting, then automatically |
| 1054 | * select it. The goal is to allow it to allocate so that it may |
| 1055 | * quickly exit and free its memory. |
| 1056 | */ |
| 1057 | if (task_will_free_mem(current)) { |
| 1058 | mark_oom_victim(current); |
| 1059 | wake_oom_reaper(current); |
| 1060 | return true; |
| 1061 | } |
| 1062 | |
| 1063 | /* |
| 1064 | * The OOM killer does not compensate for IO-less reclaim. |
| 1065 | * pagefault_out_of_memory lost its gfp context so we have to |
| 1066 | * make sure exclude 0 mask - all other users should have at least |
| 1067 | * ___GFP_DIRECT_RECLAIM to get here. |
| 1068 | */ |
| 1069 | if (oc->gfp_mask && !(oc->gfp_mask & __GFP_FS)) |
| 1070 | return true; |
| 1071 | |
| 1072 | /* |
| 1073 | * Check if there were limitations on the allocation (only relevant for |
| 1074 | * NUMA and memcg) that may require different handling. |
| 1075 | */ |
| 1076 | constraint = constrained_alloc(oc); |
| 1077 | if (constraint != CONSTRAINT_MEMORY_POLICY) |
| 1078 | oc->nodemask = NULL; |
| 1079 | check_panic_on_oom(oc, constraint); |
| 1080 | |
| 1081 | if (!is_memcg_oom(oc) && sysctl_oom_kill_allocating_task && |
| 1082 | current->mm && !oom_unkillable_task(current, NULL, oc->nodemask) && |
| 1083 | current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) { |
| 1084 | get_task_struct(current); |
| 1085 | oc->chosen = current; |
| 1086 | oom_kill_process(oc, "Out of memory (oom_kill_allocating_task)"); |
| 1087 | return true; |
| 1088 | } |
| 1089 | |
| 1090 | select_bad_process(oc); |
| 1091 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
| 1092 | if (!oc->chosen && !is_sysrq_oom(oc) && !is_memcg_oom(oc)) { |
| 1093 | dump_header(oc, NULL); |
| 1094 | panic("Out of memory and no killable processes...\n"); |
| 1095 | } |
| 1096 | if (oc->chosen && oc->chosen != (void *)-1UL) { |
| 1097 | oom_kill_process(oc, !is_memcg_oom(oc) ? "Out of memory" : |
| 1098 | "Memory cgroup out of memory"); |
| 1099 | /* |
| 1100 | * Give the killed process a good chance to exit before trying |
| 1101 | * to allocate memory again. |
| 1102 | */ |
| 1103 | schedule_timeout_killable(1); |
| 1104 | } |
| 1105 | return !!oc->chosen; |
| 1106 | } |
| 1107 | |
| 1108 | /* |
| 1109 | * The pagefault handler calls here because it is out of memory, so kill a |
| 1110 | * memory-hogging task. If oom_lock is held by somebody else, a parallel oom |
| 1111 | * killing is already in progress so do nothing. |
| 1112 | */ |
| 1113 | void pagefault_out_of_memory(void) |
| 1114 | { |
| 1115 | struct oom_control oc = { |
| 1116 | .zonelist = NULL, |
| 1117 | .nodemask = NULL, |
| 1118 | .memcg = NULL, |
| 1119 | .gfp_mask = 0, |
| 1120 | .order = 0, |
| 1121 | }; |
| 1122 | |
| 1123 | if (mem_cgroup_oom_synchronize(true)) |
| 1124 | return; |
| 1125 | |
| 1126 | if (!mutex_trylock(&oom_lock)) |
| 1127 | return; |
| 1128 | out_of_memory(&oc); |
| 1129 | mutex_unlock(&oom_lock); |
| 1130 | } |
| 1131 | |
| 1132 | void add_to_oom_reaper(struct task_struct *p) |
| 1133 | { |
| 1134 | if (!sysctl_reap_mem_on_sigkill) |
| 1135 | return; |
| 1136 | |
| 1137 | p = find_lock_task_mm(p); |
| 1138 | if (!p) |
| 1139 | return; |
| 1140 | |
| 1141 | get_task_struct(p); |
| 1142 | if (task_will_free_mem(p)) { |
| 1143 | __mark_oom_victim(p); |
| 1144 | wake_oom_reaper(p); |
| 1145 | } |
| 1146 | task_unlock(p); |
| 1147 | put_task_struct(p); |
| 1148 | } |