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