linux/mm/oom_kill.c
<<
>>
Prefs
   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
  42int sysctl_panic_on_oom;
  43int sysctl_oom_kill_allocating_task;
  44int sysctl_oom_dump_tasks = 1;
  45static 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 */
  56void 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 */
  75int 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 */
  99static 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
 127static 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 */
 140struct 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. */
 155static 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 */
 183unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
 184                          const nodemask_t *nodemask, unsigned long totalpages)
 185{
 186        long points;
 187        long adj;
 188
 189        if (oom_unkillable_task(p, memcg, nodemask))
 190                return 0;
 191
 192        p = find_lock_task_mm(p);
 193        if (!p)
 194                return 0;
 195
 196        adj = p->signal->oom_score_adj;
 197        if (adj == OOM_SCORE_ADJ_MIN) {
 198                task_unlock(p);
 199                return 0;
 200        }
 201
 202        /*
 203         * The baseline for the badness score is the proportion of RAM that each
 204         * task's rss, pagetable and swap space use.
 205         */
 206        points = get_mm_rss(p->mm) + p->mm->nr_ptes +
 207                 get_mm_counter(p->mm, MM_SWAPENTS);
 208        task_unlock(p);
 209
 210        /*
 211         * Root processes get 3% bonus, just like the __vm_enough_memory()
 212         * implementation used by LSMs.
 213         */
 214        if (has_capability_noaudit(p, CAP_SYS_ADMIN))
 215                adj -= 30;
 216
 217        /* Normalize to oom_score_adj units */
 218        adj *= totalpages / 1000;
 219        points += adj;
 220
 221        /*
 222         * Never return 0 for an eligible task regardless of the root bonus and
 223         * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here).
 224         */
 225        return points > 0 ? points : 1;
 226}
 227
 228/*
 229 * Determine the type of allocation constraint.
 230 */
 231#ifdef CONFIG_NUMA
 232static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
 233                                gfp_t gfp_mask, nodemask_t *nodemask,
 234                                unsigned long *totalpages)
 235{
 236        struct zone *zone;
 237        struct zoneref *z;
 238        enum zone_type high_zoneidx = gfp_zone(gfp_mask);
 239        bool cpuset_limited = false;
 240        int nid;
 241
 242        /* Default to all available memory */
 243        *totalpages = totalram_pages + total_swap_pages;
 244
 245        if (!zonelist)
 246                return CONSTRAINT_NONE;
 247        /*
 248         * Reach here only when __GFP_NOFAIL is used. So, we should avoid
 249         * to kill current.We have to random task kill in this case.
 250         * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
 251         */
 252        if (gfp_mask & __GFP_THISNODE)
 253                return CONSTRAINT_NONE;
 254
 255        /*
 256         * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
 257         * the page allocator means a mempolicy is in effect.  Cpuset policy
 258         * is enforced in get_page_from_freelist().
 259         */
 260        if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) {
 261                *totalpages = total_swap_pages;
 262                for_each_node_mask(nid, *nodemask)
 263                        *totalpages += node_spanned_pages(nid);
 264                return CONSTRAINT_MEMORY_POLICY;
 265        }
 266
 267        /* Check this allocation failure is caused by cpuset's wall function */
 268        for_each_zone_zonelist_nodemask(zone, z, zonelist,
 269                        high_zoneidx, nodemask)
 270                if (!cpuset_zone_allowed_softwall(zone, gfp_mask))
 271                        cpuset_limited = true;
 272
 273        if (cpuset_limited) {
 274                *totalpages = total_swap_pages;
 275                for_each_node_mask(nid, cpuset_current_mems_allowed)
 276                        *totalpages += node_spanned_pages(nid);
 277                return CONSTRAINT_CPUSET;
 278        }
 279        return CONSTRAINT_NONE;
 280}
 281#else
 282static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
 283                                gfp_t gfp_mask, nodemask_t *nodemask,
 284                                unsigned long *totalpages)
 285{
 286        *totalpages = totalram_pages + total_swap_pages;
 287        return CONSTRAINT_NONE;
 288}
 289#endif
 290
 291enum oom_scan_t oom_scan_process_thread(struct task_struct *task,
 292                unsigned long totalpages, const nodemask_t *nodemask,
 293                bool force_kill)
 294{
 295        if (task->exit_state)
 296                return OOM_SCAN_CONTINUE;
 297        if (oom_unkillable_task(task, NULL, nodemask))
 298                return OOM_SCAN_CONTINUE;
 299
 300        /*
 301         * This task already has access to memory reserves and is being killed.
 302         * Don't allow any other task to have access to the reserves.
 303         */
 304        if (test_tsk_thread_flag(task, TIF_MEMDIE)) {
 305                if (unlikely(frozen(task)))
 306                        __thaw_task(task);
 307                if (!force_kill)
 308                        return OOM_SCAN_ABORT;
 309        }
 310        if (!task->mm)
 311                return OOM_SCAN_CONTINUE;
 312
 313        if (task->flags & PF_EXITING) {
 314                /*
 315                 * If task is current and is in the process of releasing memory,
 316                 * allow the "kill" to set TIF_MEMDIE, which will allow it to
 317                 * access memory reserves.  Otherwise, it may stall forever.
 318                 *
 319                 * The iteration isn't broken here, however, in case other
 320                 * threads are found to have already been oom killed.
 321                 */
 322                if (task == current)
 323                        return OOM_SCAN_SELECT;
 324                else if (!force_kill) {
 325                        /*
 326                         * If this task is not being ptraced on exit, then wait
 327                         * for it to finish before killing some other task
 328                         * unnecessarily.
 329                         */
 330                        if (!(task->group_leader->ptrace & PT_TRACE_EXIT))
 331                                return OOM_SCAN_ABORT;
 332                }
 333        }
 334        return OOM_SCAN_OK;
 335}
 336
 337/*
 338 * Simple selection loop. We chose the process with the highest
 339 * number of 'points'.
 340 *
 341 * (not docbooked, we don't want this one cluttering up the manual)
 342 */
 343static struct task_struct *select_bad_process(unsigned int *ppoints,
 344                unsigned long totalpages, const nodemask_t *nodemask,
 345                bool force_kill)
 346{
 347        struct task_struct *g, *p;
 348        struct task_struct *chosen = NULL;
 349        unsigned long chosen_points = 0;
 350
 351        rcu_read_lock();
 352        do_each_thread(g, p) {
 353                unsigned int points;
 354
 355                switch (oom_scan_process_thread(p, totalpages, nodemask,
 356                                                force_kill)) {
 357                case OOM_SCAN_SELECT:
 358                        chosen = p;
 359                        chosen_points = ULONG_MAX;
 360                        /* fall through */
 361                case OOM_SCAN_CONTINUE:
 362                        continue;
 363                case OOM_SCAN_ABORT:
 364                        rcu_read_unlock();
 365                        return ERR_PTR(-1UL);
 366                case OOM_SCAN_OK:
 367                        break;
 368                };
 369                points = oom_badness(p, NULL, nodemask, totalpages);
 370                if (points > chosen_points) {
 371                        chosen = p;
 372                        chosen_points = points;
 373                }
 374        } while_each_thread(g, p);
 375        if (chosen)
 376                get_task_struct(chosen);
 377        rcu_read_unlock();
 378
 379        *ppoints = chosen_points * 1000 / totalpages;
 380        return chosen;
 381}
 382
 383/**
 384 * dump_tasks - dump current memory state of all system tasks
 385 * @memcg: current's memory controller, if constrained
 386 * @nodemask: nodemask passed to page allocator for mempolicy ooms
 387 *
 388 * Dumps the current memory state of all eligible tasks.  Tasks not in the same
 389 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
 390 * are not shown.
 391 * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes,
 392 * swapents, oom_score_adj value, and name.
 393 */
 394static void dump_tasks(const struct mem_cgroup *memcg, const nodemask_t *nodemask)
 395{
 396        struct task_struct *p;
 397        struct task_struct *task;
 398
 399        pr_info("[ pid ]   uid  tgid total_vm      rss nr_ptes swapents oom_score_adj name\n");
 400        rcu_read_lock();
 401        for_each_process(p) {
 402                if (oom_unkillable_task(p, memcg, nodemask))
 403                        continue;
 404
 405                task = find_lock_task_mm(p);
 406                if (!task) {
 407                        /*
 408                         * This is a kthread or all of p's threads have already
 409                         * detached their mm's.  There's no need to report
 410                         * them; they can't be oom killed anyway.
 411                         */
 412                        continue;
 413                }
 414
 415                pr_info("[%5d] %5d %5d %8lu %8lu %7lu %8lu         %5d %s\n",
 416                        task->pid, from_kuid(&init_user_ns, task_uid(task)),
 417                        task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
 418                        task->mm->nr_ptes,
 419                        get_mm_counter(task->mm, MM_SWAPENTS),
 420                        task->signal->oom_score_adj, task->comm);
 421                task_unlock(task);
 422        }
 423        rcu_read_unlock();
 424}
 425
 426static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
 427                        struct mem_cgroup *memcg, const nodemask_t *nodemask)
 428{
 429        task_lock(current);
 430        pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
 431                "oom_adj=%d, oom_score_adj=%d\n",
 432                current->comm, gfp_mask, order, current->signal->oom_adj,
 433                current->signal->oom_score_adj);
 434        cpuset_print_task_mems_allowed(current);
 435        task_unlock(current);
 436        dump_stack();
 437        mem_cgroup_print_oom_info(memcg, p);
 438        show_mem(SHOW_MEM_FILTER_NODES);
 439        if (sysctl_oom_dump_tasks)
 440                dump_tasks(memcg, nodemask);
 441}
 442
 443#define K(x) ((x) << (PAGE_SHIFT-10))
 444/*
 445 * Must be called while holding a reference to p, which will be released upon
 446 * returning.
 447 */
 448void oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
 449                      unsigned int points, unsigned long totalpages,
 450                      struct mem_cgroup *memcg, nodemask_t *nodemask,
 451                      const char *message)
 452{
 453        struct task_struct *victim = p;
 454        struct task_struct *child;
 455        struct task_struct *t = p;
 456        struct mm_struct *mm;
 457        unsigned int victim_points = 0;
 458        static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
 459                                              DEFAULT_RATELIMIT_BURST);
 460
 461        /*
 462         * If the task is already exiting, don't alarm the sysadmin or kill
 463         * its children or threads, just set TIF_MEMDIE so it can die quickly
 464         */
 465        if (p->flags & PF_EXITING) {
 466                set_tsk_thread_flag(p, TIF_MEMDIE);
 467                put_task_struct(p);
 468                return;
 469        }
 470
 471        if (__ratelimit(&oom_rs))
 472                dump_header(p, gfp_mask, order, memcg, nodemask);
 473
 474        task_lock(p);
 475        pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
 476                message, task_pid_nr(p), p->comm, points);
 477        task_unlock(p);
 478
 479        /*
 480         * If any of p's children has a different mm and is eligible for kill,
 481         * the one with the highest oom_badness() score is sacrificed for its
 482         * parent.  This attempts to lose the minimal amount of work done while
 483         * still freeing memory.
 484         */
 485        read_lock(&tasklist_lock);
 486        do {
 487                list_for_each_entry(child, &t->children, sibling) {
 488                        unsigned int child_points;
 489
 490                        if (child->mm == p->mm)
 491                                continue;
 492                        /*
 493                         * oom_badness() returns 0 if the thread is unkillable
 494                         */
 495                        child_points = oom_badness(child, memcg, nodemask,
 496                                                                totalpages);
 497                        if (child_points > victim_points) {
 498                                put_task_struct(victim);
 499                                victim = child;
 500                                victim_points = child_points;
 501                                get_task_struct(victim);
 502                        }
 503                }
 504        } while_each_thread(p, t);
 505        read_unlock(&tasklist_lock);
 506
 507        rcu_read_lock();
 508        p = find_lock_task_mm(victim);
 509        if (!p) {
 510                rcu_read_unlock();
 511                put_task_struct(victim);
 512                return;
 513        } else if (victim != p) {
 514                get_task_struct(p);
 515                put_task_struct(victim);
 516                victim = p;
 517        }
 518
 519        /* mm cannot safely be dereferenced after task_unlock(victim) */
 520        mm = victim->mm;
 521        pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
 522                task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
 523                K(get_mm_counter(victim->mm, MM_ANONPAGES)),
 524                K(get_mm_counter(victim->mm, MM_FILEPAGES)));
 525        task_unlock(victim);
 526
 527        /*
 528         * Kill all user processes sharing victim->mm in other thread groups, if
 529         * any.  They don't get access to memory reserves, though, to avoid
 530         * depletion of all memory.  This prevents mm->mmap_sem livelock when an
 531         * oom killed thread cannot exit because it requires the semaphore and
 532         * its contended by another thread trying to allocate memory itself.
 533         * That thread will now get access to memory reserves since it has a
 534         * pending fatal signal.
 535         */
 536        for_each_process(p)
 537                if (p->mm == mm && !same_thread_group(p, victim) &&
 538                    !(p->flags & PF_KTHREAD)) {
 539                        if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
 540                                continue;
 541
 542                        task_lock(p);   /* Protect ->comm from prctl() */
 543                        pr_err("Kill process %d (%s) sharing same memory\n",
 544                                task_pid_nr(p), p->comm);
 545                        task_unlock(p);
 546                        do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
 547                }
 548        rcu_read_unlock();
 549
 550        set_tsk_thread_flag(victim, TIF_MEMDIE);
 551        do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
 552        put_task_struct(victim);
 553}
 554#undef K
 555
 556/*
 557 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
 558 */
 559void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
 560                        int order, const nodemask_t *nodemask)
 561{
 562        if (likely(!sysctl_panic_on_oom))
 563                return;
 564        if (sysctl_panic_on_oom != 2) {
 565                /*
 566                 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
 567                 * does not panic for cpuset, mempolicy, or memcg allocation
 568                 * failures.
 569                 */
 570                if (constraint != CONSTRAINT_NONE)
 571                        return;
 572        }
 573        dump_header(NULL, gfp_mask, order, NULL, nodemask);
 574        panic("Out of memory: %s panic_on_oom is enabled\n",
 575                sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
 576}
 577
 578static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
 579
 580int register_oom_notifier(struct notifier_block *nb)
 581{
 582        return blocking_notifier_chain_register(&oom_notify_list, nb);
 583}
 584EXPORT_SYMBOL_GPL(register_oom_notifier);
 585
 586int unregister_oom_notifier(struct notifier_block *nb)
 587{
 588        return blocking_notifier_chain_unregister(&oom_notify_list, nb);
 589}
 590EXPORT_SYMBOL_GPL(unregister_oom_notifier);
 591
 592/*
 593 * Try to acquire the OOM killer lock for the zones in zonelist.  Returns zero
 594 * if a parallel OOM killing is already taking place that includes a zone in
 595 * the zonelist.  Otherwise, locks all zones in the zonelist and returns 1.
 596 */
 597int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
 598{
 599        struct zoneref *z;
 600        struct zone *zone;
 601        int ret = 1;
 602
 603        spin_lock(&zone_scan_lock);
 604        for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
 605                if (zone_is_oom_locked(zone)) {
 606                        ret = 0;
 607                        goto out;
 608                }
 609        }
 610
 611        for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
 612                /*
 613                 * Lock each zone in the zonelist under zone_scan_lock so a
 614                 * parallel invocation of try_set_zonelist_oom() doesn't succeed
 615                 * when it shouldn't.
 616                 */
 617                zone_set_flag(zone, ZONE_OOM_LOCKED);
 618        }
 619
 620out:
 621        spin_unlock(&zone_scan_lock);
 622        return ret;
 623}
 624
 625/*
 626 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
 627 * allocation attempts with zonelists containing them may now recall the OOM
 628 * killer, if necessary.
 629 */
 630void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
 631{
 632        struct zoneref *z;
 633        struct zone *zone;
 634
 635        spin_lock(&zone_scan_lock);
 636        for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
 637                zone_clear_flag(zone, ZONE_OOM_LOCKED);
 638        }
 639        spin_unlock(&zone_scan_lock);
 640}
 641
 642/*
 643 * Try to acquire the oom killer lock for all system zones.  Returns zero if a
 644 * parallel oom killing is taking place, otherwise locks all zones and returns
 645 * non-zero.
 646 */
 647static int try_set_system_oom(void)
 648{
 649        struct zone *zone;
 650        int ret = 1;
 651
 652        spin_lock(&zone_scan_lock);
 653        for_each_populated_zone(zone)
 654                if (zone_is_oom_locked(zone)) {
 655                        ret = 0;
 656                        goto out;
 657                }
 658        for_each_populated_zone(zone)
 659                zone_set_flag(zone, ZONE_OOM_LOCKED);
 660out:
 661        spin_unlock(&zone_scan_lock);
 662        return ret;
 663}
 664
 665/*
 666 * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation
 667 * attempts or page faults may now recall the oom killer, if necessary.
 668 */
 669static void clear_system_oom(void)
 670{
 671        struct zone *zone;
 672
 673        spin_lock(&zone_scan_lock);
 674        for_each_populated_zone(zone)
 675                zone_clear_flag(zone, ZONE_OOM_LOCKED);
 676        spin_unlock(&zone_scan_lock);
 677}
 678
 679/**
 680 * out_of_memory - kill the "best" process when we run out of memory
 681 * @zonelist: zonelist pointer
 682 * @gfp_mask: memory allocation flags
 683 * @order: amount of memory being requested as a power of 2
 684 * @nodemask: nodemask passed to page allocator
 685 * @force_kill: true if a task must be killed, even if others are exiting
 686 *
 687 * If we run out of memory, we have the choice between either
 688 * killing a random task (bad), letting the system crash (worse)
 689 * OR try to be smart about which process to kill. Note that we
 690 * don't have to be perfect here, we just have to be good.
 691 */
 692void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
 693                int order, nodemask_t *nodemask, bool force_kill)
 694{
 695        const nodemask_t *mpol_mask;
 696        struct task_struct *p;
 697        unsigned long totalpages;
 698        unsigned long freed = 0;
 699        unsigned int uninitialized_var(points);
 700        enum oom_constraint constraint = CONSTRAINT_NONE;
 701        int killed = 0;
 702
 703        blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
 704        if (freed > 0)
 705                /* Got some memory back in the last second. */
 706                return;
 707
 708        /*
 709         * If current has a pending SIGKILL, then automatically select it.  The
 710         * goal is to allow it to allocate so that it may quickly exit and free
 711         * its memory.
 712         */
 713        if (fatal_signal_pending(current)) {
 714                set_thread_flag(TIF_MEMDIE);
 715                return;
 716        }
 717
 718        /*
 719         * Check if there were limitations on the allocation (only relevant for
 720         * NUMA) that may require different handling.
 721         */
 722        constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
 723                                                &totalpages);
 724        mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
 725        check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
 726
 727        if (sysctl_oom_kill_allocating_task && current->mm &&
 728            !oom_unkillable_task(current, NULL, nodemask) &&
 729            current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
 730                get_task_struct(current);
 731                oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL,
 732                                 nodemask,
 733                                 "Out of memory (oom_kill_allocating_task)");
 734                goto out;
 735        }
 736
 737        p = select_bad_process(&points, totalpages, mpol_mask, force_kill);
 738        /* Found nothing?!?! Either we hang forever, or we panic. */
 739        if (!p) {
 740                dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
 741                panic("Out of memory and no killable processes...\n");
 742        }
 743        if (PTR_ERR(p) != -1UL) {
 744                oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
 745                                 nodemask, "Out of memory");
 746                killed = 1;
 747        }
 748out:
 749        /*
 750         * Give the killed threads a good chance of exiting before trying to
 751         * allocate memory again.
 752         */
 753        if (killed)
 754                schedule_timeout_killable(1);
 755}
 756
 757/*
 758 * The pagefault handler calls here because it is out of memory, so kill a
 759 * memory-hogging task.  If a populated zone has ZONE_OOM_LOCKED set, a parallel
 760 * oom killing is already in progress so do nothing.  If a task is found with
 761 * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit.
 762 */
 763void pagefault_out_of_memory(void)
 764{
 765        if (try_set_system_oom()) {
 766                out_of_memory(NULL, 0, 0, NULL, false);
 767                clear_system_oom();
 768        }
 769        schedule_timeout_killable(1);
 770}
 771
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.