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/swap.h> 26 #include <linux/timex.h> 27 #include <linux/jiffies.h> 28 #include <linux/cpuset.h> 29 #include <linux/export.h> 30 #include <linux/notifier.h> 31 #include <linux/memcontrol.h> 32 #include <linux/mempolicy.h> 33 #include <linux/security.h> 34 #include <linux/ptrace.h> 35 #include <linux/freezer.h> 36 #include <linux/ftrace.h> 37 #include <linux/ratelimit.h> 38 39 #define CREATE_TRACE_POINTS 40 #include <trace/events/oom.h> 41 42 int sysctl_panic_on_oom; 43 int sysctl_oom_kill_allocating_task; 44 int sysctl_oom_dump_tasks = 1; 45 static DEFINE_SPINLOCK(zone_scan_lock); 46 47 /* 48 * compare_swap_oom_score_adj() - compare and swap current's oom_score_adj 49 * @old_val: old oom_score_adj for compare 50 * @new_val: new oom_score_adj for swap 51 * 52 * Sets the oom_score_adj value for current to @new_val iff its present value is 53 * @old_val. Usually used to reinstate a previous value to prevent racing with 54 * userspacing tuning the value in the interim. 55 */ 56 void compare_swap_oom_score_adj(int old_val, int new_val) 57 { 58 struct sighand_struct *sighand = current->sighand; 59 60 spin_lock_irq(&sighand->siglock); 61 if (current->signal->oom_score_adj == old_val) 62 current->signal->oom_score_adj = new_val; 63 trace_oom_score_adj_update(current); 64 spin_unlock_irq(&sighand->siglock); 65 } 66 67 /** 68 * test_set_oom_score_adj() - set current's oom_score_adj and return old value 69 * @new_val: new oom_score_adj value 70 * 71 * Sets the oom_score_adj value for current to @new_val with proper 72 * synchronization and returns the old value. Usually used to temporarily 73 * set a value, save the old value in the caller, and then reinstate it later. 74 */ 75 int test_set_oom_score_adj(int new_val) 76 { 77 struct sighand_struct *sighand = current->sighand; 78 int old_val; 79 80 spin_lock_irq(&sighand->siglock); 81 old_val = current->signal->oom_score_adj; 82 current->signal->oom_score_adj = new_val; 83 trace_oom_score_adj_update(current); 84 spin_unlock_irq(&sighand->siglock); 85 86 return old_val; 87 } 88 89 #ifdef CONFIG_NUMA 90 /** 91 * has_intersects_mems_allowed() - check task eligiblity for kill 92 * @tsk: task struct of which task to consider 93 * @mask: nodemask passed to page allocator for mempolicy ooms 94 * 95 * Task eligibility is determined by whether or not a candidate task, @tsk, 96 * shares the same mempolicy nodes as current if it is bound by such a policy 97 * and whether or not it has the same set of allowed cpuset nodes. 98 */ 99 static bool has_intersects_mems_allowed(struct task_struct *tsk, 100 const nodemask_t *mask) 101 { 102 struct task_struct *start = tsk; 103 104 do { 105 if (mask) { 106 /* 107 * If this is a mempolicy constrained oom, tsk's 108 * cpuset is irrelevant. Only return true if its 109 * mempolicy intersects current, otherwise it may be 110 * needlessly killed. 111 */ 112 if (mempolicy_nodemask_intersects(tsk, mask)) 113 return true; 114 } else { 115 /* 116 * This is not a mempolicy constrained oom, so only 117 * check the mems of tsk's cpuset. 118 */ 119 if (cpuset_mems_allowed_intersects(current, tsk)) 120 return true; 121 } 122 } while_each_thread(start, tsk); 123 124 return false; 125 } 126 #else 127 static bool has_intersects_mems_allowed(struct task_struct *tsk, 128 const nodemask_t *mask) 129 { 130 return true; 131 } 132 #endif /* CONFIG_NUMA */ 133 134 /* 135 * The process p may have detached its own ->mm while exiting or through 136 * use_mm(), but one or more of its subthreads may still have a valid 137 * pointer. Return p, or any of its subthreads with a valid ->mm, with 138 * task_lock() held. 139 */ 140 struct task_struct *find_lock_task_mm(struct task_struct *p) 141 { 142 struct task_struct *t = p; 143 144 do { 145 task_lock(t); 146 if (likely(t->mm)) 147 return t; 148 task_unlock(t); 149 } while_each_thread(p, t); 150 151 return NULL; 152 } 153 154 /* return true if the task is not adequate as candidate victim task. */ 155 static bool oom_unkillable_task(struct task_struct *p, 156 const struct mem_cgroup *memcg, const nodemask_t *nodemask) 157 { 158 if (is_global_init(p)) 159 return true; 160 if (p->flags & PF_KTHREAD) 161 return true; 162 163 /* When mem_cgroup_out_of_memory() and p is not member of the group */ 164 if (memcg && !task_in_mem_cgroup(p, memcg)) 165 return true; 166 167 /* p may not have freeable memory in nodemask */ 168 if (!has_intersects_mems_allowed(p, nodemask)) 169 return true; 170 171 return false; 172 } 173 174 /** 175 * oom_badness - heuristic function to determine which candidate task to kill 176 * @p: task struct of which task we should calculate 177 * @totalpages: total present RAM allowed for page allocation 178 * 179 * The heuristic for determining which task to kill is made to be as simple and 180 * predictable as possible. The goal is to return the highest value for the 181 * task consuming the most memory to avoid subsequent oom failures. 182 */ 183 unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg, 184 const nodemask_t *nodemask, unsigned long totalpages) 185 { 186 unsigned long points; 187 188 if (oom_unkillable_task(p, memcg, nodemask)) 189 return 0; 190 191 p = find_lock_task_mm(p); 192 if (!p) 193 return 0; 194 195 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) { 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) + p->mm->nr_ptes + 205 get_mm_counter(p->mm, MM_SWAPENTS); 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 -= 30 * totalpages / 1000; 214 215 /* 216 * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may 217 * either completely disable oom killing or always prefer a certain 218 * task. 219 */ 220 points += p->signal->oom_score_adj * totalpages / 1000; 221 222 /* 223 * Never return 0 for an eligible task regardless of the root bonus and 224 * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here). 225 */ 226 return points ? points : 1; 227 } 228 229 /* 230 * Determine the type of allocation constraint. 231 */ 232 #ifdef CONFIG_NUMA 233 static enum oom_constraint constrained_alloc(struct zonelist *zonelist, 234 gfp_t gfp_mask, nodemask_t *nodemask, 235 unsigned long *totalpages) 236 { 237 struct zone *zone; 238 struct zoneref *z; 239 enum zone_type high_zoneidx = gfp_zone(gfp_mask); 240 bool cpuset_limited = false; 241 int nid; 242 243 /* Default to all available memory */ 244 *totalpages = totalram_pages + total_swap_pages; 245 246 if (!zonelist) 247 return CONSTRAINT_NONE; 248 /* 249 * Reach here only when __GFP_NOFAIL is used. So, we should avoid 250 * to kill current.We have to random task kill in this case. 251 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. 252 */ 253 if (gfp_mask & __GFP_THISNODE) 254 return CONSTRAINT_NONE; 255 256 /* 257 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in 258 * the page allocator means a mempolicy is in effect. Cpuset policy 259 * is enforced in get_page_from_freelist(). 260 */ 261 if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) { 262 *totalpages = total_swap_pages; 263 for_each_node_mask(nid, *nodemask) 264 *totalpages += node_spanned_pages(nid); 265 return CONSTRAINT_MEMORY_POLICY; 266 } 267 268 /* Check this allocation failure is caused by cpuset's wall function */ 269 for_each_zone_zonelist_nodemask(zone, z, zonelist, 270 high_zoneidx, nodemask) 271 if (!cpuset_zone_allowed_softwall(zone, gfp_mask)) 272 cpuset_limited = true; 273 274 if (cpuset_limited) { 275 *totalpages = total_swap_pages; 276 for_each_node_mask(nid, cpuset_current_mems_allowed) 277 *totalpages += node_spanned_pages(nid); 278 return CONSTRAINT_CPUSET; 279 } 280 return CONSTRAINT_NONE; 281 } 282 #else 283 static enum oom_constraint constrained_alloc(struct zonelist *zonelist, 284 gfp_t gfp_mask, nodemask_t *nodemask, 285 unsigned long *totalpages) 286 { 287 *totalpages = totalram_pages + total_swap_pages; 288 return CONSTRAINT_NONE; 289 } 290 #endif 291 292 /* 293 * Simple selection loop. We chose the process with the highest 294 * number of 'points'. We expect the caller will lock the tasklist. 295 * 296 * (not docbooked, we don't want this one cluttering up the manual) 297 */ 298 static struct task_struct *select_bad_process(unsigned int *ppoints, 299 unsigned long totalpages, struct mem_cgroup *memcg, 300 const nodemask_t *nodemask, bool force_kill) 301 { 302 struct task_struct *g, *p; 303 struct task_struct *chosen = NULL; 304 unsigned long chosen_points = 0; 305 306 do_each_thread(g, p) { 307 unsigned int points; 308 309 if (p->exit_state) 310 continue; 311 if (oom_unkillable_task(p, memcg, nodemask)) 312 continue; 313 314 /* 315 * This task already has access to memory reserves and is 316 * being killed. Don't allow any other task access to the 317 * memory reserve. 318 * 319 * Note: this may have a chance of deadlock if it gets 320 * blocked waiting for another task which itself is waiting 321 * for memory. Is there a better alternative? 322 */ 323 if (test_tsk_thread_flag(p, TIF_MEMDIE)) { 324 if (unlikely(frozen(p))) 325 __thaw_task(p); 326 if (!force_kill) 327 return ERR_PTR(-1UL); 328 } 329 if (!p->mm) 330 continue; 331 332 if (p->flags & PF_EXITING) { 333 /* 334 * If p is the current task and is in the process of 335 * releasing memory, we allow the "kill" to set 336 * TIF_MEMDIE, which will allow it to gain access to 337 * memory reserves. Otherwise, it may stall forever. 338 * 339 * The loop isn't broken here, however, in case other 340 * threads are found to have already been oom killed. 341 */ 342 if (p == current) { 343 chosen = p; 344 chosen_points = ULONG_MAX; 345 } else if (!force_kill) { 346 /* 347 * If this task is not being ptraced on exit, 348 * then wait for it to finish before killing 349 * some other task unnecessarily. 350 */ 351 if (!(p->group_leader->ptrace & PT_TRACE_EXIT)) 352 return ERR_PTR(-1UL); 353 } 354 } 355 356 points = oom_badness(p, memcg, nodemask, totalpages); 357 if (points > chosen_points) { 358 chosen = p; 359 chosen_points = points; 360 } 361 } while_each_thread(g, p); 362 363 *ppoints = chosen_points * 1000 / totalpages; 364 return chosen; 365 } 366 367 /** 368 * dump_tasks - dump current memory state of all system tasks 369 * @mem: current's memory controller, if constrained 370 * @nodemask: nodemask passed to page allocator for mempolicy ooms 371 * 372 * Dumps the current memory state of all eligible tasks. Tasks not in the same 373 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes 374 * are not shown. 375 * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj 376 * value, oom_score_adj value, and name. 377 * 378 * Call with tasklist_lock read-locked. 379 */ 380 static void dump_tasks(const struct mem_cgroup *memcg, const nodemask_t *nodemask) 381 { 382 struct task_struct *p; 383 struct task_struct *task; 384 385 pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n"); 386 for_each_process(p) { 387 if (oom_unkillable_task(p, memcg, nodemask)) 388 continue; 389 390 task = find_lock_task_mm(p); 391 if (!task) { 392 /* 393 * This is a kthread or all of p's threads have already 394 * detached their mm's. There's no need to report 395 * them; they can't be oom killed anyway. 396 */ 397 continue; 398 } 399 400 pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n", 401 task->pid, from_kuid(&init_user_ns, task_uid(task)), 402 task->tgid, task->mm->total_vm, get_mm_rss(task->mm), 403 task_cpu(task), task->signal->oom_adj, 404 task->signal->oom_score_adj, task->comm); 405 task_unlock(task); 406 } 407 } 408 409 static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, 410 struct mem_cgroup *memcg, const nodemask_t *nodemask) 411 { 412 task_lock(current); 413 pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " 414 "oom_adj=%d, oom_score_adj=%d\n", 415 current->comm, gfp_mask, order, current->signal->oom_adj, 416 current->signal->oom_score_adj); 417 cpuset_print_task_mems_allowed(current); 418 task_unlock(current); 419 dump_stack(); 420 mem_cgroup_print_oom_info(memcg, p); 421 show_mem(SHOW_MEM_FILTER_NODES); 422 if (sysctl_oom_dump_tasks) 423 dump_tasks(memcg, nodemask); 424 } 425 426 #define K(x) ((x) << (PAGE_SHIFT-10)) 427 static void oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, 428 unsigned int points, unsigned long totalpages, 429 struct mem_cgroup *memcg, nodemask_t *nodemask, 430 const char *message) 431 { 432 struct task_struct *victim = p; 433 struct task_struct *child; 434 struct task_struct *t = p; 435 struct mm_struct *mm; 436 unsigned int victim_points = 0; 437 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL, 438 DEFAULT_RATELIMIT_BURST); 439 440 /* 441 * If the task is already exiting, don't alarm the sysadmin or kill 442 * its children or threads, just set TIF_MEMDIE so it can die quickly 443 */ 444 if (p->flags & PF_EXITING) { 445 set_tsk_thread_flag(p, TIF_MEMDIE); 446 return; 447 } 448 449 if (__ratelimit(&oom_rs)) 450 dump_header(p, gfp_mask, order, memcg, nodemask); 451 452 task_lock(p); 453 pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n", 454 message, task_pid_nr(p), p->comm, points); 455 task_unlock(p); 456 457 /* 458 * If any of p's children has a different mm and is eligible for kill, 459 * the one with the highest oom_badness() score is sacrificed for its 460 * parent. This attempts to lose the minimal amount of work done while 461 * still freeing memory. 462 */ 463 do { 464 list_for_each_entry(child, &t->children, sibling) { 465 unsigned int child_points; 466 467 if (child->mm == p->mm) 468 continue; 469 /* 470 * oom_badness() returns 0 if the thread is unkillable 471 */ 472 child_points = oom_badness(child, memcg, nodemask, 473 totalpages); 474 if (child_points > victim_points) { 475 victim = child; 476 victim_points = child_points; 477 } 478 } 479 } while_each_thread(p, t); 480 481 victim = find_lock_task_mm(victim); 482 if (!victim) 483 return; 484 485 /* mm cannot safely be dereferenced after task_unlock(victim) */ 486 mm = victim->mm; 487 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n", 488 task_pid_nr(victim), victim->comm, K(victim->mm->total_vm), 489 K(get_mm_counter(victim->mm, MM_ANONPAGES)), 490 K(get_mm_counter(victim->mm, MM_FILEPAGES))); 491 task_unlock(victim); 492 493 /* 494 * Kill all user processes sharing victim->mm in other thread groups, if 495 * any. They don't get access to memory reserves, though, to avoid 496 * depletion of all memory. This prevents mm->mmap_sem livelock when an 497 * oom killed thread cannot exit because it requires the semaphore and 498 * its contended by another thread trying to allocate memory itself. 499 * That thread will now get access to memory reserves since it has a 500 * pending fatal signal. 501 */ 502 for_each_process(p) 503 if (p->mm == mm && !same_thread_group(p, victim) && 504 !(p->flags & PF_KTHREAD)) { 505 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) 506 continue; 507 508 task_lock(p); /* Protect ->comm from prctl() */ 509 pr_err("Kill process %d (%s) sharing same memory\n", 510 task_pid_nr(p), p->comm); 511 task_unlock(p); 512 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true); 513 } 514 515 set_tsk_thread_flag(victim, TIF_MEMDIE); 516 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true); 517 } 518 #undef K 519 520 /* 521 * Determines whether the kernel must panic because of the panic_on_oom sysctl. 522 */ 523 static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask, 524 int order, const nodemask_t *nodemask) 525 { 526 if (likely(!sysctl_panic_on_oom)) 527 return; 528 if (sysctl_panic_on_oom != 2) { 529 /* 530 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel 531 * does not panic for cpuset, mempolicy, or memcg allocation 532 * failures. 533 */ 534 if (constraint != CONSTRAINT_NONE) 535 return; 536 } 537 read_lock(&tasklist_lock); 538 dump_header(NULL, gfp_mask, order, NULL, nodemask); 539 read_unlock(&tasklist_lock); 540 panic("Out of memory: %s panic_on_oom is enabled\n", 541 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); 542 } 543 544 #ifdef CONFIG_CGROUP_MEM_RES_CTLR 545 void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, 546 int order) 547 { 548 unsigned long limit; 549 unsigned int points = 0; 550 struct task_struct *p; 551 552 /* 553 * If current has a pending SIGKILL, then automatically select it. The 554 * goal is to allow it to allocate so that it may quickly exit and free 555 * its memory. 556 */ 557 if (fatal_signal_pending(current)) { 558 set_thread_flag(TIF_MEMDIE); 559 return; 560 } 561 562 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); 563 limit = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; 564 read_lock(&tasklist_lock); 565 p = select_bad_process(&points, limit, memcg, NULL, false); 566 if (p && PTR_ERR(p) != -1UL) 567 oom_kill_process(p, gfp_mask, order, points, limit, memcg, NULL, 568 "Memory cgroup out of memory"); 569 read_unlock(&tasklist_lock); 570 } 571 #endif 572 573 static BLOCKING_NOTIFIER_HEAD(oom_notify_list); 574 575 int register_oom_notifier(struct notifier_block *nb) 576 { 577 return blocking_notifier_chain_register(&oom_notify_list, nb); 578 } 579 EXPORT_SYMBOL_GPL(register_oom_notifier); 580 581 int unregister_oom_notifier(struct notifier_block *nb) 582 { 583 return blocking_notifier_chain_unregister(&oom_notify_list, nb); 584 } 585 EXPORT_SYMBOL_GPL(unregister_oom_notifier); 586 587 /* 588 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero 589 * if a parallel OOM killing is already taking place that includes a zone in 590 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. 591 */ 592 int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) 593 { 594 struct zoneref *z; 595 struct zone *zone; 596 int ret = 1; 597 598 spin_lock(&zone_scan_lock); 599 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { 600 if (zone_is_oom_locked(zone)) { 601 ret = 0; 602 goto out; 603 } 604 } 605 606 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { 607 /* 608 * Lock each zone in the zonelist under zone_scan_lock so a 609 * parallel invocation of try_set_zonelist_oom() doesn't succeed 610 * when it shouldn't. 611 */ 612 zone_set_flag(zone, ZONE_OOM_LOCKED); 613 } 614 615 out: 616 spin_unlock(&zone_scan_lock); 617 return ret; 618 } 619 620 /* 621 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed 622 * allocation attempts with zonelists containing them may now recall the OOM 623 * killer, if necessary. 624 */ 625 void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) 626 { 627 struct zoneref *z; 628 struct zone *zone; 629 630 spin_lock(&zone_scan_lock); 631 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { 632 zone_clear_flag(zone, ZONE_OOM_LOCKED); 633 } 634 spin_unlock(&zone_scan_lock); 635 } 636 637 /* 638 * Try to acquire the oom killer lock for all system zones. Returns zero if a 639 * parallel oom killing is taking place, otherwise locks all zones and returns 640 * non-zero. 641 */ 642 static int try_set_system_oom(void) 643 { 644 struct zone *zone; 645 int ret = 1; 646 647 spin_lock(&zone_scan_lock); 648 for_each_populated_zone(zone) 649 if (zone_is_oom_locked(zone)) { 650 ret = 0; 651 goto out; 652 } 653 for_each_populated_zone(zone) 654 zone_set_flag(zone, ZONE_OOM_LOCKED); 655 out: 656 spin_unlock(&zone_scan_lock); 657 return ret; 658 } 659 660 /* 661 * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation 662 * attempts or page faults may now recall the oom killer, if necessary. 663 */ 664 static void clear_system_oom(void) 665 { 666 struct zone *zone; 667 668 spin_lock(&zone_scan_lock); 669 for_each_populated_zone(zone) 670 zone_clear_flag(zone, ZONE_OOM_LOCKED); 671 spin_unlock(&zone_scan_lock); 672 } 673 674 /** 675 * out_of_memory - kill the "best" process when we run out of memory 676 * @zonelist: zonelist pointer 677 * @gfp_mask: memory allocation flags 678 * @order: amount of memory being requested as a power of 2 679 * @nodemask: nodemask passed to page allocator 680 * @force_kill: true if a task must be killed, even if others are exiting 681 * 682 * If we run out of memory, we have the choice between either 683 * killing a random task (bad), letting the system crash (worse) 684 * OR try to be smart about which process to kill. Note that we 685 * don't have to be perfect here, we just have to be good. 686 */ 687 void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, 688 int order, nodemask_t *nodemask, bool force_kill) 689 { 690 const nodemask_t *mpol_mask; 691 struct task_struct *p; 692 unsigned long totalpages; 693 unsigned long freed = 0; 694 unsigned int points; 695 enum oom_constraint constraint = CONSTRAINT_NONE; 696 int killed = 0; 697 698 blocking_notifier_call_chain(&oom_notify_list, 0, &freed); 699 if (freed > 0) 700 /* Got some memory back in the last second. */ 701 return; 702 703 /* 704 * If current has a pending SIGKILL, then automatically select it. The 705 * goal is to allow it to allocate so that it may quickly exit and free 706 * its memory. 707 */ 708 if (fatal_signal_pending(current)) { 709 set_thread_flag(TIF_MEMDIE); 710 return; 711 } 712 713 /* 714 * Check if there were limitations on the allocation (only relevant for 715 * NUMA) that may require different handling. 716 */ 717 constraint = constrained_alloc(zonelist, gfp_mask, nodemask, 718 &totalpages); 719 mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL; 720 check_panic_on_oom(constraint, gfp_mask, order, mpol_mask); 721 722 read_lock(&tasklist_lock); 723 if (sysctl_oom_kill_allocating_task && 724 !oom_unkillable_task(current, NULL, nodemask) && 725 current->mm) { 726 oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL, 727 nodemask, 728 "Out of memory (oom_kill_allocating_task)"); 729 goto out; 730 } 731 732 p = select_bad_process(&points, totalpages, NULL, mpol_mask, 733 force_kill); 734 /* Found nothing?!?! Either we hang forever, or we panic. */ 735 if (!p) { 736 dump_header(NULL, gfp_mask, order, NULL, mpol_mask); 737 read_unlock(&tasklist_lock); 738 panic("Out of memory and no killable processes...\n"); 739 } 740 if (PTR_ERR(p) != -1UL) { 741 oom_kill_process(p, gfp_mask, order, points, totalpages, NULL, 742 nodemask, "Out of memory"); 743 killed = 1; 744 } 745 out: 746 read_unlock(&tasklist_lock); 747 748 /* 749 * Give "p" a good chance of killing itself before we 750 * retry to allocate memory unless "p" is current 751 */ 752 if (killed && !test_thread_flag(TIF_MEMDIE)) 753 schedule_timeout_uninterruptible(1); 754 } 755 756 /* 757 * The pagefault handler calls here because it is out of memory, so kill a 758 * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel 759 * oom killing is already in progress so do nothing. If a task is found with 760 * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit. 761 */ 762 void pagefault_out_of_memory(void) 763 { 764 if (try_set_system_oom()) { 765 out_of_memory(NULL, 0, 0, NULL, false); 766 clear_system_oom(); 767 } 768 if (!test_thread_flag(TIF_MEMDIE)) 769 schedule_timeout_uninterruptible(1); 770 } 771