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
 291/*
 292 * Simple selection loop. We chose the process with the highest
 293 * number of 'points'. We expect the caller will lock the tasklist.
 294 *
 295 * (not docbooked, we don't want this one cluttering up the manual)
 296 */
 297static struct task_struct *select_bad_process(unsigned int *ppoints,
 298                unsigned long totalpages, struct mem_cgroup *memcg,
 299                const nodemask_t *nodemask, bool force_kill)
 300{
 301        struct task_struct *g, *p;
 302        struct task_struct *chosen = NULL;
 303        unsigned long chosen_points = 0;
 304
 305        do_each_thread(g, p) {
 306                unsigned int points;
 307
 308                if (p->exit_state)
 309                        continue;
 310                if (oom_unkillable_task(p, memcg, nodemask))
 311                        continue;
 312
 313                /*
 314                 * This task already has access to memory reserves and is
 315                 * being killed. Don't allow any other task access to the
 316                 * memory reserve.
 317                 *
 318                 * Note: this may have a chance of deadlock if it gets
 319                 * blocked waiting for another task which itself is waiting
 320                 * for memory. Is there a better alternative?
 321                 */
 322                if (test_tsk_thread_flag(p, TIF_MEMDIE)) {
 323                        if (unlikely(frozen(p)))
 324                                __thaw_task(p);
 325                        if (!force_kill)
 326                                return ERR_PTR(-1UL);
 327                }
 328                if (!p->mm)
 329                        continue;
 330
 331                if (p->flags & PF_EXITING) {
 332                        /*
 333                         * If p is the current task and is in the process of
 334                         * releasing memory, we allow the "kill" to set
 335                         * TIF_MEMDIE, which will allow it to gain access to
 336                         * memory reserves.  Otherwise, it may stall forever.
 337                         *
 338                         * The loop isn't broken here, however, in case other
 339                         * threads are found to have already been oom killed.
 340                         */
 341                        if (p == current) {
 342                                chosen = p;
 343                                chosen_points = ULONG_MAX;
 344                        } else if (!force_kill) {
 345                                /*
 346                                 * If this task is not being ptraced on exit,
 347                                 * then wait for it to finish before killing
 348                                 * some other task unnecessarily.
 349                                 */
 350                                if (!(p->group_leader->ptrace & PT_TRACE_EXIT))
 351                                        return ERR_PTR(-1UL);
 352                        }
 353                }
 354
 355                points = oom_badness(p, memcg, nodemask, totalpages);
 356                if (points > chosen_points) {
 357                        chosen = p;
 358                        chosen_points = points;
 359                }
 360        } while_each_thread(g, p);
 361
 362        *ppoints = chosen_points * 1000 / totalpages;
 363        return chosen;
 364}
 365
 366/**
 367 * dump_tasks - dump current memory state of all system tasks
 368 * @memcg: current's memory controller, if constrained
 369 * @nodemask: nodemask passed to page allocator for mempolicy ooms
 370 *
 371 * Dumps the current memory state of all eligible tasks.  Tasks not in the same
 372 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
 373 * are not shown.
 374 * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj
 375 * value, oom_score_adj value, and name.
 376 *
 377 * Call with tasklist_lock read-locked.
 378 */
 379static void dump_tasks(const struct mem_cgroup *memcg, const nodemask_t *nodemask)
 380{
 381        struct task_struct *p;
 382        struct task_struct *task;
 383
 384        pr_info("[ pid ]   uid  tgid total_vm      rss cpu oom_adj oom_score_adj name\n");
 385        for_each_process(p) {
 386                if (oom_unkillable_task(p, memcg, nodemask))
 387                        continue;
 388
 389                task = find_lock_task_mm(p);
 390                if (!task) {
 391                        /*
 392                         * This is a kthread or all of p's threads have already
 393                         * detached their mm's.  There's no need to report
 394                         * them; they can't be oom killed anyway.
 395                         */
 396                        continue;
 397                }
 398
 399                pr_info("[%5d] %5d %5d %8lu %8lu %3u     %3d         %5d %s\n",
 400                        task->pid, from_kuid(&init_user_ns, task_uid(task)),
 401                        task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
 402                        task_cpu(task), task->signal->oom_adj,
 403                        task->signal->oom_score_adj, task->comm);
 404                task_unlock(task);
 405        }
 406}
 407
 408static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
 409                        struct mem_cgroup *memcg, const nodemask_t *nodemask)
 410{
 411        task_lock(current);
 412        pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
 413                "oom_adj=%d, oom_score_adj=%d\n",
 414                current->comm, gfp_mask, order, current->signal->oom_adj,
 415                current->signal->oom_score_adj);
 416        cpuset_print_task_mems_allowed(current);
 417        task_unlock(current);
 418        dump_stack();
 419        mem_cgroup_print_oom_info(memcg, p);
 420        show_mem(SHOW_MEM_FILTER_NODES);
 421        if (sysctl_oom_dump_tasks)
 422                dump_tasks(memcg, nodemask);
 423}
 424
 425#define K(x) ((x) << (PAGE_SHIFT-10))
 426static void oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
 427                             unsigned int points, unsigned long totalpages,
 428                             struct mem_cgroup *memcg, nodemask_t *nodemask,
 429                             const char *message)
 430{
 431        struct task_struct *victim = p;
 432        struct task_struct *child;
 433        struct task_struct *t = p;
 434        struct mm_struct *mm;
 435        unsigned int victim_points = 0;
 436        static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
 437                                              DEFAULT_RATELIMIT_BURST);
 438
 439        /*
 440         * If the task is already exiting, don't alarm the sysadmin or kill
 441         * its children or threads, just set TIF_MEMDIE so it can die quickly
 442         */
 443        if (p->flags & PF_EXITING) {
 444                set_tsk_thread_flag(p, TIF_MEMDIE);
 445                return;
 446        }
 447
 448        if (__ratelimit(&oom_rs))
 449                dump_header(p, gfp_mask, order, memcg, nodemask);
 450
 451        task_lock(p);
 452        pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
 453                message, task_pid_nr(p), p->comm, points);
 454        task_unlock(p);
 455
 456        /*
 457         * If any of p's children has a different mm and is eligible for kill,
 458         * the one with the highest oom_badness() score is sacrificed for its
 459         * parent.  This attempts to lose the minimal amount of work done while
 460         * still freeing memory.
 461         */
 462        do {
 463                list_for_each_entry(child, &t->children, sibling) {
 464                        unsigned int child_points;
 465
 466                        if (child->mm == p->mm)
 467                                continue;
 468                        /*
 469                         * oom_badness() returns 0 if the thread is unkillable
 470                         */
 471                        child_points = oom_badness(child, memcg, nodemask,
 472                                                                totalpages);
 473                        if (child_points > victim_points) {
 474                                victim = child;
 475                                victim_points = child_points;
 476                        }
 477                }
 478        } while_each_thread(p, t);
 479
 480        victim = find_lock_task_mm(victim);
 481        if (!victim)
 482                return;
 483
 484        /* mm cannot safely be dereferenced after task_unlock(victim) */
 485        mm = victim->mm;
 486        pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
 487                task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
 488                K(get_mm_counter(victim->mm, MM_ANONPAGES)),
 489                K(get_mm_counter(victim->mm, MM_FILEPAGES)));
 490        task_unlock(victim);
 491
 492        /*
 493         * Kill all user processes sharing victim->mm in other thread groups, if
 494         * any.  They don't get access to memory reserves, though, to avoid
 495         * depletion of all memory.  This prevents mm->mmap_sem livelock when an
 496         * oom killed thread cannot exit because it requires the semaphore and
 497         * its contended by another thread trying to allocate memory itself.
 498         * That thread will now get access to memory reserves since it has a
 499         * pending fatal signal.
 500         */
 501        for_each_process(p)
 502                if (p->mm == mm && !same_thread_group(p, victim) &&
 503                    !(p->flags & PF_KTHREAD)) {
 504                        if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
 505                                continue;
 506
 507                        task_lock(p);   /* Protect ->comm from prctl() */
 508                        pr_err("Kill process %d (%s) sharing same memory\n",
 509                                task_pid_nr(p), p->comm);
 510                        task_unlock(p);
 511                        do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
 512                }
 513
 514        set_tsk_thread_flag(victim, TIF_MEMDIE);
 515        do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
 516}
 517#undef K
 518
 519/*
 520 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
 521 */
 522static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
 523                                int order, const nodemask_t *nodemask)
 524{
 525        if (likely(!sysctl_panic_on_oom))
 526                return;
 527        if (sysctl_panic_on_oom != 2) {
 528                /*
 529                 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
 530                 * does not panic for cpuset, mempolicy, or memcg allocation
 531                 * failures.
 532                 */
 533                if (constraint != CONSTRAINT_NONE)
 534                        return;
 535        }
 536        read_lock(&tasklist_lock);
 537        dump_header(NULL, gfp_mask, order, NULL, nodemask);
 538        read_unlock(&tasklist_lock);
 539        panic("Out of memory: %s panic_on_oom is enabled\n",
 540                sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
 541}
 542
 543#ifdef CONFIG_CGROUP_MEM_RES_CTLR
 544void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
 545                              int order)
 546{
 547        unsigned long limit;
 548        unsigned int points = 0;
 549        struct task_struct *p;
 550
 551        /*
 552         * If current has a pending SIGKILL, then automatically select it.  The
 553         * goal is to allow it to allocate so that it may quickly exit and free
 554         * its memory.
 555         */
 556        if (fatal_signal_pending(current)) {
 557                set_thread_flag(TIF_MEMDIE);
 558                return;
 559        }
 560
 561        check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
 562        limit = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1;
 563        read_lock(&tasklist_lock);
 564        p = select_bad_process(&points, limit, memcg, NULL, false);
 565        if (p && PTR_ERR(p) != -1UL)
 566                oom_kill_process(p, gfp_mask, order, points, limit, memcg, NULL,
 567                                 "Memory cgroup out of memory");
 568        read_unlock(&tasklist_lock);
 569}
 570#endif
 571
 572static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
 573
 574int register_oom_notifier(struct notifier_block *nb)
 575{
 576        return blocking_notifier_chain_register(&oom_notify_list, nb);
 577}
 578EXPORT_SYMBOL_GPL(register_oom_notifier);
 579
 580int unregister_oom_notifier(struct notifier_block *nb)
 581{
 582        return blocking_notifier_chain_unregister(&oom_notify_list, nb);
 583}
 584EXPORT_SYMBOL_GPL(unregister_oom_notifier);
 585
 586/*
 587 * Try to acquire the OOM killer lock for the zones in zonelist.  Returns zero
 588 * if a parallel OOM killing is already taking place that includes a zone in
 589 * the zonelist.  Otherwise, locks all zones in the zonelist and returns 1.
 590 */
 591int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
 592{
 593        struct zoneref *z;
 594        struct zone *zone;
 595        int ret = 1;
 596
 597        spin_lock(&zone_scan_lock);
 598        for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
 599                if (zone_is_oom_locked(zone)) {
 600                        ret = 0;
 601                        goto out;
 602                }
 603        }
 604
 605        for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
 606                /*
 607                 * Lock each zone in the zonelist under zone_scan_lock so a
 608                 * parallel invocation of try_set_zonelist_oom() doesn't succeed
 609                 * when it shouldn't.
 610                 */
 611                zone_set_flag(zone, ZONE_OOM_LOCKED);
 612        }
 613
 614out:
 615        spin_unlock(&zone_scan_lock);
 616        return ret;
 617}
 618
 619/*
 620 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
 621 * allocation attempts with zonelists containing them may now recall the OOM
 622 * killer, if necessary.
 623 */
 624void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
 625{
 626        struct zoneref *z;
 627        struct zone *zone;
 628
 629        spin_lock(&zone_scan_lock);
 630        for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
 631                zone_clear_flag(zone, ZONE_OOM_LOCKED);
 632        }
 633        spin_unlock(&zone_scan_lock);
 634}
 635
 636/*
 637 * Try to acquire the oom killer lock for all system zones.  Returns zero if a
 638 * parallel oom killing is taking place, otherwise locks all zones and returns
 639 * non-zero.
 640 */
 641static int try_set_system_oom(void)
 642{
 643        struct zone *zone;
 644        int ret = 1;
 645
 646        spin_lock(&zone_scan_lock);
 647        for_each_populated_zone(zone)
 648                if (zone_is_oom_locked(zone)) {
 649                        ret = 0;
 650                        goto out;
 651                }
 652        for_each_populated_zone(zone)
 653                zone_set_flag(zone, ZONE_OOM_LOCKED);
 654out:
 655        spin_unlock(&zone_scan_lock);
 656        return ret;
 657}
 658
 659/*
 660 * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation
 661 * attempts or page faults may now recall the oom killer, if necessary.
 662 */
 663static void clear_system_oom(void)
 664{
 665        struct zone *zone;
 666
 667        spin_lock(&zone_scan_lock);
 668        for_each_populated_zone(zone)
 669                zone_clear_flag(zone, ZONE_OOM_LOCKED);
 670        spin_unlock(&zone_scan_lock);
 671}
 672
 673/**
 674 * out_of_memory - kill the "best" process when we run out of memory
 675 * @zonelist: zonelist pointer
 676 * @gfp_mask: memory allocation flags
 677 * @order: amount of memory being requested as a power of 2
 678 * @nodemask: nodemask passed to page allocator
 679 * @force_kill: true if a task must be killed, even if others are exiting
 680 *
 681 * If we run out of memory, we have the choice between either
 682 * killing a random task (bad), letting the system crash (worse)
 683 * OR try to be smart about which process to kill. Note that we
 684 * don't have to be perfect here, we just have to be good.
 685 */
 686void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
 687                int order, nodemask_t *nodemask, bool force_kill)
 688{
 689        const nodemask_t *mpol_mask;
 690        struct task_struct *p;
 691        unsigned long totalpages;
 692        unsigned long freed = 0;
 693        unsigned int points;
 694        enum oom_constraint constraint = CONSTRAINT_NONE;
 695        int killed = 0;
 696
 697        blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
 698        if (freed > 0)
 699                /* Got some memory back in the last second. */
 700                return;
 701
 702        /*
 703         * If current has a pending SIGKILL, then automatically select it.  The
 704         * goal is to allow it to allocate so that it may quickly exit and free
 705         * its memory.
 706         */
 707        if (fatal_signal_pending(current)) {
 708                set_thread_flag(TIF_MEMDIE);
 709                return;
 710        }
 711
 712        /*
 713         * Check if there were limitations on the allocation (only relevant for
 714         * NUMA) that may require different handling.
 715         */
 716        constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
 717                                                &totalpages);
 718        mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
 719        check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
 720
 721        read_lock(&tasklist_lock);
 722        if (sysctl_oom_kill_allocating_task &&
 723            !oom_unkillable_task(current, NULL, nodemask) &&
 724            current->mm) {
 725                oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL,
 726                                 nodemask,
 727                                 "Out of memory (oom_kill_allocating_task)");
 728                goto out;
 729        }
 730
 731        p = select_bad_process(&points, totalpages, NULL, mpol_mask,
 732                               force_kill);
 733        /* Found nothing?!?! Either we hang forever, or we panic. */
 734        if (!p) {
 735                dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
 736                read_unlock(&tasklist_lock);
 737                panic("Out of memory and no killable processes...\n");
 738        }
 739        if (PTR_ERR(p) != -1UL) {
 740                oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
 741                                 nodemask, "Out of memory");
 742                killed = 1;
 743        }
 744out:
 745        read_unlock(&tasklist_lock);
 746
 747        /*
 748         * Give "p" a good chance of killing itself before we
 749         * retry to allocate memory unless "p" is current
 750         */
 751        if (killed && !test_thread_flag(TIF_MEMDIE))
 752                schedule_timeout_uninterruptible(1);
 753}
 754
 755/*
 756 * The pagefault handler calls here because it is out of memory, so kill a
 757 * memory-hogging task.  If a populated zone has ZONE_OOM_LOCKED set, a parallel
 758 * oom killing is already in progress so do nothing.  If a task is found with
 759 * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit.
 760 */
 761void pagefault_out_of_memory(void)
 762{
 763        if (try_set_system_oom()) {
 764                out_of_memory(NULL, 0, 0, NULL, false);
 765                clear_system_oom();
 766        }
 767        if (!test_thread_flag(TIF_MEMDIE))
 768                schedule_timeout_uninterruptible(1);
 769}
 770
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.