4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/sched/autogroup.h>
10 #include <linux/sched/mm.h>
11 #include <linux/sched/stat.h>
12 #include <linux/sched/task.h>
13 #include <linux/sched/task_stack.h>
14 #include <linux/sched/cputime.h>
15 #include <linux/interrupt.h>
16 #include <linux/module.h>
17 #include <linux/capability.h>
18 #include <linux/completion.h>
19 #include <linux/personality.h>
20 #include <linux/tty.h>
21 #include <linux/iocontext.h>
22 #include <linux/key.h>
23 #include <linux/cpu.h>
24 #include <linux/acct.h>
25 #include <linux/tsacct_kern.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
28 #include <linux/freezer.h>
29 #include <linux/binfmts.h>
30 #include <linux/nsproxy.h>
31 #include <linux/pid_namespace.h>
32 #include <linux/ptrace.h>
33 #include <linux/profile.h>
34 #include <linux/mount.h>
35 #include <linux/proc_fs.h>
36 #include <linux/kthread.h>
37 #include <linux/mempolicy.h>
38 #include <linux/taskstats_kern.h>
39 #include <linux/delayacct.h>
40 #include <linux/cgroup.h>
41 #include <linux/syscalls.h>
42 #include <linux/signal.h>
43 #include <linux/posix-timers.h>
44 #include <linux/cn_proc.h>
45 #include <linux/mutex.h>
46 #include <linux/futex.h>
47 #include <linux/pipe_fs_i.h>
48 #include <linux/audit.h> /* for audit_free() */
49 #include <linux/resource.h>
50 #include <linux/blkdev.h>
51 #include <linux/task_io_accounting_ops.h>
52 #include <linux/tracehook.h>
53 #include <linux/fs_struct.h>
54 #include <linux/init_task.h>
55 #include <linux/perf_event.h>
56 #include <trace/events/sched.h>
57 #include <linux/hw_breakpoint.h>
58 #include <linux/oom.h>
59 #include <linux/writeback.h>
60 #include <linux/shm.h>
61 #include <linux/kcov.h>
62 #include <linux/random.h>
63 #include <linux/rcuwait.h>
64 #include <linux/compat.h>
65 #include <linux/sysfs.h>
67 #include <linux/uaccess.h>
68 #include <asm/unistd.h>
69 #include <asm/pgtable.h>
70 #include <asm/mmu_context.h>
73 * The default value should be high enough to not crash a system that randomly
74 * crashes its kernel from time to time, but low enough to at least not permit
75 * overflowing 32-bit refcounts or the ldsem writer count.
77 static unsigned int oops_limit = 10000;
80 static struct ctl_table kern_exit_table[] = {
82 .procname = "oops_limit",
84 .maxlen = sizeof(oops_limit),
86 .proc_handler = proc_douintvec,
91 static __init int kernel_exit_sysctls_init(void)
93 register_sysctl_init("kernel", kern_exit_table);
96 late_initcall(kernel_exit_sysctls_init);
99 static atomic_t oops_count = ATOMIC_INIT(0);
102 static ssize_t oops_count_show(struct kobject *kobj, struct kobj_attribute *attr,
105 return sysfs_emit(page, "%d\n", atomic_read(&oops_count));
108 static struct kobj_attribute oops_count_attr = __ATTR_RO(oops_count);
110 static __init int kernel_exit_sysfs_init(void)
112 sysfs_add_file_to_group(kernel_kobj, &oops_count_attr.attr, NULL);
115 late_initcall(kernel_exit_sysfs_init);
118 static void __unhash_process(struct task_struct *p, bool group_dead)
121 detach_pid(p, PIDTYPE_PID);
123 detach_pid(p, PIDTYPE_PGID);
124 detach_pid(p, PIDTYPE_SID);
126 list_del_rcu(&p->tasks);
127 list_del_init(&p->sibling);
128 __this_cpu_dec(process_counts);
130 list_del_rcu(&p->thread_group);
131 list_del_rcu(&p->thread_node);
135 * This function expects the tasklist_lock write-locked.
137 static void __exit_signal(struct task_struct *tsk)
139 struct signal_struct *sig = tsk->signal;
140 bool group_dead = thread_group_leader(tsk);
141 struct sighand_struct *sighand;
142 struct tty_struct *uninitialized_var(tty);
145 sighand = rcu_dereference_check(tsk->sighand,
146 lockdep_tasklist_lock_is_held());
147 spin_lock(&sighand->siglock);
149 #ifdef CONFIG_POSIX_TIMERS
150 posix_cpu_timers_exit(tsk);
152 posix_cpu_timers_exit_group(tsk);
155 * This can only happen if the caller is de_thread().
156 * FIXME: this is the temporary hack, we should teach
157 * posix-cpu-timers to handle this case correctly.
159 if (unlikely(has_group_leader_pid(tsk)))
160 posix_cpu_timers_exit_group(tsk);
169 * If there is any task waiting for the group exit
172 if (sig->notify_count > 0 && !--sig->notify_count)
173 wake_up_process(sig->group_exit_task);
175 if (tsk == sig->curr_target)
176 sig->curr_target = next_thread(tsk);
179 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
180 sizeof(unsigned long long));
183 * Accumulate here the counters for all threads as they die. We could
184 * skip the group leader because it is the last user of signal_struct,
185 * but we want to avoid the race with thread_group_cputime() which can
186 * see the empty ->thread_head list.
188 task_cputime(tsk, &utime, &stime);
189 write_seqlock(&sig->stats_lock);
192 sig->gtime += task_gtime(tsk);
193 sig->min_flt += tsk->min_flt;
194 sig->maj_flt += tsk->maj_flt;
195 sig->nvcsw += tsk->nvcsw;
196 sig->nivcsw += tsk->nivcsw;
197 sig->inblock += task_io_get_inblock(tsk);
198 sig->oublock += task_io_get_oublock(tsk);
199 task_io_accounting_add(&sig->ioac, &tsk->ioac);
200 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
202 __unhash_process(tsk, group_dead);
203 write_sequnlock(&sig->stats_lock);
206 * Do this under ->siglock, we can race with another thread
207 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
209 flush_sigqueue(&tsk->pending);
211 spin_unlock(&sighand->siglock);
213 __cleanup_sighand(sighand);
214 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
216 flush_sigqueue(&sig->shared_pending);
221 static void delayed_put_task_struct(struct rcu_head *rhp)
223 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
225 perf_event_delayed_put(tsk);
226 trace_sched_process_free(tsk);
227 put_task_struct(tsk);
231 void release_task(struct task_struct *p)
233 struct task_struct *leader;
236 /* don't need to get the RCU readlock here - the process is dead and
237 * can't be modifying its own credentials. But shut RCU-lockdep up */
239 atomic_dec(&__task_cred(p)->user->processes);
245 write_lock_irq(&tasklist_lock);
246 ptrace_release_task(p);
250 * If we are the last non-leader member of the thread
251 * group, and the leader is zombie, then notify the
252 * group leader's parent process. (if it wants notification.)
255 leader = p->group_leader;
256 if (leader != p && thread_group_empty(leader)
257 && leader->exit_state == EXIT_ZOMBIE) {
259 * If we were the last child thread and the leader has
260 * exited already, and the leader's parent ignores SIGCHLD,
261 * then we are the one who should release the leader.
263 zap_leader = do_notify_parent(leader, leader->exit_signal);
265 leader->exit_state = EXIT_DEAD;
268 write_unlock_irq(&tasklist_lock);
270 call_rcu(&p->rcu, delayed_put_task_struct);
273 if (unlikely(zap_leader))
278 * Note that if this function returns a valid task_struct pointer (!NULL)
279 * task->usage must remain >0 for the duration of the RCU critical section.
281 struct task_struct *task_rcu_dereference(struct task_struct **ptask)
283 struct sighand_struct *sighand;
284 struct task_struct *task;
287 * We need to verify that release_task() was not called and thus
288 * delayed_put_task_struct() can't run and drop the last reference
289 * before rcu_read_unlock(). We check task->sighand != NULL,
290 * but we can read the already freed and reused memory.
293 task = rcu_dereference(*ptask);
297 probe_kernel_address(&task->sighand, sighand);
300 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
301 * was already freed we can not miss the preceding update of this
305 if (unlikely(task != READ_ONCE(*ptask)))
309 * We've re-checked that "task == *ptask", now we have two different
312 * 1. This is actually the same task/task_struct. In this case
313 * sighand != NULL tells us it is still alive.
315 * 2. This is another task which got the same memory for task_struct.
316 * We can't know this of course, and we can not trust
319 * In this case we actually return a random value, but this is
322 * If we return NULL - we can pretend that we actually noticed that
323 * *ptask was updated when the previous task has exited. Or pretend
324 * that probe_slab_address(&sighand) reads NULL.
326 * If we return the new task (because sighand is not NULL for any
327 * reason) - this is fine too. This (new) task can't go away before
330 * And note: We could even eliminate the false positive if re-read
331 * task->sighand once again to avoid the falsely NULL. But this case
332 * is very unlikely so we don't care.
340 void rcuwait_wake_up(struct rcuwait *w)
342 struct task_struct *task;
347 * Order condition vs @task, such that everything prior to the load
348 * of @task is visible. This is the condition as to why the user called
349 * rcuwait_trywake() in the first place. Pairs with set_current_state()
350 * barrier (A) in rcuwait_wait_event().
353 * [S] tsk = current [S] cond = true
360 * Avoid using task_rcu_dereference() magic as long as we are careful,
361 * see comment in rcuwait_wait_event() regarding ->exit_state.
363 task = rcu_dereference(w->task);
365 wake_up_process(task);
370 * Determine if a process group is "orphaned", according to the POSIX
371 * definition in 2.2.2.52. Orphaned process groups are not to be affected
372 * by terminal-generated stop signals. Newly orphaned process groups are
373 * to receive a SIGHUP and a SIGCONT.
375 * "I ask you, have you ever known what it is to be an orphan?"
377 static int will_become_orphaned_pgrp(struct pid *pgrp,
378 struct task_struct *ignored_task)
380 struct task_struct *p;
382 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
383 if ((p == ignored_task) ||
384 (p->exit_state && thread_group_empty(p)) ||
385 is_global_init(p->real_parent))
388 if (task_pgrp(p->real_parent) != pgrp &&
389 task_session(p->real_parent) == task_session(p))
391 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
396 int is_current_pgrp_orphaned(void)
400 read_lock(&tasklist_lock);
401 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
402 read_unlock(&tasklist_lock);
407 static bool has_stopped_jobs(struct pid *pgrp)
409 struct task_struct *p;
411 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
412 if (p->signal->flags & SIGNAL_STOP_STOPPED)
414 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
420 * Check to see if any process groups have become orphaned as
421 * a result of our exiting, and if they have any stopped jobs,
422 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
425 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
427 struct pid *pgrp = task_pgrp(tsk);
428 struct task_struct *ignored_task = tsk;
431 /* exit: our father is in a different pgrp than
432 * we are and we were the only connection outside.
434 parent = tsk->real_parent;
436 /* reparent: our child is in a different pgrp than
437 * we are, and it was the only connection outside.
441 if (task_pgrp(parent) != pgrp &&
442 task_session(parent) == task_session(tsk) &&
443 will_become_orphaned_pgrp(pgrp, ignored_task) &&
444 has_stopped_jobs(pgrp)) {
445 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
446 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
452 * A task is exiting. If it owned this mm, find a new owner for the mm.
454 void mm_update_next_owner(struct mm_struct *mm)
456 struct task_struct *c, *g, *p = current;
460 * If the exiting or execing task is not the owner, it's
461 * someone else's problem.
466 * The current owner is exiting/execing and there are no other
467 * candidates. Do not leave the mm pointing to a possibly
468 * freed task structure.
470 if (atomic_read(&mm->mm_users) <= 1) {
475 read_lock(&tasklist_lock);
477 * Search in the children
479 list_for_each_entry(c, &p->children, sibling) {
481 goto assign_new_owner;
485 * Search in the siblings
487 list_for_each_entry(c, &p->real_parent->children, sibling) {
489 goto assign_new_owner;
493 * Search through everything else, we should not get here often.
495 for_each_process(g) {
496 if (g->flags & PF_KTHREAD)
498 for_each_thread(g, c) {
500 goto assign_new_owner;
505 read_unlock(&tasklist_lock);
507 * We found no owner yet mm_users > 1: this implies that we are
508 * most likely racing with swapoff (try_to_unuse()) or /proc or
509 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
518 * The task_lock protects c->mm from changing.
519 * We always want mm->owner->mm == mm
523 * Delay read_unlock() till we have the task_lock()
524 * to ensure that c does not slip away underneath us
526 read_unlock(&tasklist_lock);
536 #endif /* CONFIG_MEMCG */
539 * Turn us into a lazy TLB process if we
542 static void exit_mm(void)
544 struct mm_struct *mm = current->mm;
545 struct core_state *core_state;
547 exit_mm_release(current, mm);
552 * Serialize with any possible pending coredump.
553 * We must hold mmap_sem around checking core_state
554 * and clearing tsk->mm. The core-inducing thread
555 * will increment ->nr_threads for each thread in the
556 * group with ->mm != NULL.
558 down_read(&mm->mmap_sem);
559 core_state = mm->core_state;
561 struct core_thread self;
563 up_read(&mm->mmap_sem);
566 if (self.task->flags & PF_SIGNALED)
567 self.next = xchg(&core_state->dumper.next, &self);
571 * Implies mb(), the result of xchg() must be visible
572 * to core_state->dumper.
574 if (atomic_dec_and_test(&core_state->nr_threads))
575 complete(&core_state->startup);
578 set_current_state(TASK_UNINTERRUPTIBLE);
579 if (!self.task) /* see coredump_finish() */
581 freezable_schedule();
583 __set_current_state(TASK_RUNNING);
584 down_read(&mm->mmap_sem);
587 BUG_ON(mm != current->active_mm);
588 /* more a memory barrier than a real lock */
591 up_read(&mm->mmap_sem);
592 enter_lazy_tlb(mm, current);
593 task_unlock(current);
594 mm_update_next_owner(mm);
596 if (test_thread_flag(TIF_MEMDIE))
600 static struct task_struct *find_alive_thread(struct task_struct *p)
602 struct task_struct *t;
604 for_each_thread(p, t) {
605 if (!(t->flags & PF_EXITING))
611 static struct task_struct *find_child_reaper(struct task_struct *father,
612 struct list_head *dead)
613 __releases(&tasklist_lock)
614 __acquires(&tasklist_lock)
616 struct pid_namespace *pid_ns = task_active_pid_ns(father);
617 struct task_struct *reaper = pid_ns->child_reaper;
618 struct task_struct *p, *n;
620 if (likely(reaper != father))
623 reaper = find_alive_thread(father);
625 pid_ns->child_reaper = reaper;
629 write_unlock_irq(&tasklist_lock);
631 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
632 list_del_init(&p->ptrace_entry);
636 zap_pid_ns_processes(pid_ns);
637 write_lock_irq(&tasklist_lock);
643 * When we die, we re-parent all our children, and try to:
644 * 1. give them to another thread in our thread group, if such a member exists
645 * 2. give it to the first ancestor process which prctl'd itself as a
646 * child_subreaper for its children (like a service manager)
647 * 3. give it to the init process (PID 1) in our pid namespace
649 static struct task_struct *find_new_reaper(struct task_struct *father,
650 struct task_struct *child_reaper)
652 struct task_struct *thread, *reaper;
654 thread = find_alive_thread(father);
658 if (father->signal->has_child_subreaper) {
659 unsigned int ns_level = task_pid(father)->level;
661 * Find the first ->is_child_subreaper ancestor in our pid_ns.
662 * We can't check reaper != child_reaper to ensure we do not
663 * cross the namespaces, the exiting parent could be injected
664 * by setns() + fork().
665 * We check pid->level, this is slightly more efficient than
666 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
668 for (reaper = father->real_parent;
669 task_pid(reaper)->level == ns_level;
670 reaper = reaper->real_parent) {
671 if (reaper == &init_task)
673 if (!reaper->signal->is_child_subreaper)
675 thread = find_alive_thread(reaper);
685 * Any that need to be release_task'd are put on the @dead list.
687 static void reparent_leader(struct task_struct *father, struct task_struct *p,
688 struct list_head *dead)
690 if (unlikely(p->exit_state == EXIT_DEAD))
693 /* We don't want people slaying init. */
694 p->exit_signal = SIGCHLD;
696 /* If it has exited notify the new parent about this child's death. */
698 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
699 if (do_notify_parent(p, p->exit_signal)) {
700 p->exit_state = EXIT_DEAD;
701 list_add(&p->ptrace_entry, dead);
705 kill_orphaned_pgrp(p, father);
709 * This does two things:
711 * A. Make init inherit all the child processes
712 * B. Check to see if any process groups have become orphaned
713 * as a result of our exiting, and if they have any stopped
714 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
716 static void forget_original_parent(struct task_struct *father,
717 struct list_head *dead)
719 struct task_struct *p, *t, *reaper;
721 if (unlikely(!list_empty(&father->ptraced)))
722 exit_ptrace(father, dead);
724 /* Can drop and reacquire tasklist_lock */
725 reaper = find_child_reaper(father, dead);
726 if (list_empty(&father->children))
729 reaper = find_new_reaper(father, reaper);
730 list_for_each_entry(p, &father->children, sibling) {
731 for_each_thread(p, t) {
732 t->real_parent = reaper;
733 BUG_ON((!t->ptrace) != (t->parent == father));
734 if (likely(!t->ptrace))
735 t->parent = t->real_parent;
736 if (t->pdeath_signal)
737 group_send_sig_info(t->pdeath_signal,
741 * If this is a threaded reparent there is no need to
742 * notify anyone anything has happened.
744 if (!same_thread_group(reaper, father))
745 reparent_leader(father, p, dead);
747 list_splice_tail_init(&father->children, &reaper->children);
751 * Send signals to all our closest relatives so that they know
752 * to properly mourn us..
754 static void exit_notify(struct task_struct *tsk, int group_dead)
757 struct task_struct *p, *n;
760 write_lock_irq(&tasklist_lock);
761 forget_original_parent(tsk, &dead);
764 kill_orphaned_pgrp(tsk->group_leader, NULL);
766 if (unlikely(tsk->ptrace)) {
767 int sig = thread_group_leader(tsk) &&
768 thread_group_empty(tsk) &&
769 !ptrace_reparented(tsk) ?
770 tsk->exit_signal : SIGCHLD;
771 autoreap = do_notify_parent(tsk, sig);
772 } else if (thread_group_leader(tsk)) {
773 autoreap = thread_group_empty(tsk) &&
774 do_notify_parent(tsk, tsk->exit_signal);
779 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
780 if (tsk->exit_state == EXIT_DEAD)
781 list_add(&tsk->ptrace_entry, &dead);
783 /* mt-exec, de_thread() is waiting for group leader */
784 if (unlikely(tsk->signal->notify_count < 0))
785 wake_up_process(tsk->signal->group_exit_task);
786 write_unlock_irq(&tasklist_lock);
788 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
789 list_del_init(&p->ptrace_entry);
794 #ifdef CONFIG_DEBUG_STACK_USAGE
795 static void check_stack_usage(void)
797 static DEFINE_SPINLOCK(low_water_lock);
798 static int lowest_to_date = THREAD_SIZE;
801 free = stack_not_used(current);
803 if (free >= lowest_to_date)
806 spin_lock(&low_water_lock);
807 if (free < lowest_to_date) {
808 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
809 current->comm, task_pid_nr(current), free);
810 lowest_to_date = free;
812 spin_unlock(&low_water_lock);
815 static inline void check_stack_usage(void) {}
818 void __noreturn do_exit(long code)
820 struct task_struct *tsk = current;
824 * We can get here from a kernel oops, sometimes with preemption off.
825 * Start by checking for critical errors.
826 * Then fix up important state like USER_DS and preemption.
827 * Then do everything else.
830 WARN_ON(blk_needs_flush_plug(tsk));
832 if (unlikely(in_interrupt()))
833 panic("Aiee, killing interrupt handler!");
834 if (unlikely(!tsk->pid))
835 panic("Attempted to kill the idle task!");
838 * If do_exit is called because this processes oopsed, it's possible
839 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
840 * continuing. Amongst other possible reasons, this is to prevent
841 * mm_release()->clear_child_tid() from writing to a user-controlled
846 if (unlikely(in_atomic())) {
847 pr_info("note: %s[%d] exited with preempt_count %d\n",
848 current->comm, task_pid_nr(current),
850 preempt_count_set(PREEMPT_ENABLED);
853 profile_task_exit(tsk);
856 ptrace_event(PTRACE_EVENT_EXIT, code);
858 validate_creds_for_do_exit(tsk);
861 * We're taking recursive faults here in do_exit. Safest is to just
862 * leave this task alone and wait for reboot.
864 if (unlikely(tsk->flags & PF_EXITING)) {
865 pr_alert("Fixing recursive fault but reboot is needed!\n");
866 futex_exit_recursive(tsk);
867 set_current_state(TASK_UNINTERRUPTIBLE);
871 exit_signals(tsk); /* sets PF_EXITING */
873 /* sync mm's RSS info before statistics gathering */
875 sync_mm_rss(tsk->mm);
876 acct_update_integrals(tsk);
877 group_dead = atomic_dec_and_test(&tsk->signal->live);
880 * If the last thread of global init has exited, panic
881 * immediately to get a useable coredump.
883 if (unlikely(is_global_init(tsk)))
884 panic("Attempted to kill init! exitcode=0x%08x\n",
885 tsk->signal->group_exit_code ?: (int)code);
887 #ifdef CONFIG_POSIX_TIMERS
888 hrtimer_cancel(&tsk->signal->real_timer);
889 exit_itimers(tsk->signal);
892 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
894 acct_collect(code, group_dead);
899 tsk->exit_code = code;
900 taskstats_exit(tsk, group_dead);
906 trace_sched_process_exit(tsk);
913 disassociate_ctty(1);
914 exit_task_namespaces(tsk);
919 * Flush inherited counters to the parent - before the parent
920 * gets woken up by child-exit notifications.
922 * because of cgroup mode, must be called before cgroup_exit()
924 perf_event_exit_task(tsk);
926 sched_autogroup_exit_task(tsk);
930 * FIXME: do that only when needed, using sched_exit tracepoint
932 flush_ptrace_hw_breakpoint(tsk);
934 exit_tasks_rcu_start();
935 exit_notify(tsk, group_dead);
936 proc_exit_connector(tsk);
937 mpol_put_task_policy(tsk);
939 if (unlikely(current->pi_state_cache))
940 kfree(current->pi_state_cache);
943 * Make sure we are holding no locks:
945 debug_check_no_locks_held();
948 exit_io_context(tsk);
950 if (tsk->splice_pipe)
951 free_pipe_info(tsk->splice_pipe);
953 if (tsk->task_frag.page)
954 put_page(tsk->task_frag.page);
956 validate_creds_for_do_exit(tsk);
961 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
963 exit_tasks_rcu_finish();
965 lockdep_free_task(tsk);
968 EXPORT_SYMBOL_GPL(do_exit);
970 void __noreturn make_task_dead(int signr)
973 * Take the task off the cpu after something catastrophic has
979 * Every time the system oopses, if the oops happens while a reference
980 * to an object was held, the reference leaks.
981 * If the oops doesn't also leak memory, repeated oopsing can cause
982 * reference counters to wrap around (if they're not using refcount_t).
983 * This means that repeated oopsing can make unexploitable-looking bugs
984 * exploitable through repeated oopsing.
985 * To make sure this can't happen, place an upper bound on how often the
986 * kernel may oops without panic().
988 limit = READ_ONCE(oops_limit);
989 if (atomic_inc_return(&oops_count) >= limit && limit)
990 panic("Oopsed too often (kernel.oops_limit is %d)", limit);
995 void complete_and_exit(struct completion *comp, long code)
1002 EXPORT_SYMBOL(complete_and_exit);
1004 SYSCALL_DEFINE1(exit, int, error_code)
1006 do_exit((error_code&0xff)<<8);
1010 * Take down every thread in the group. This is called by fatal signals
1011 * as well as by sys_exit_group (below).
1014 do_group_exit(int exit_code)
1016 struct signal_struct *sig = current->signal;
1018 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1020 if (signal_group_exit(sig))
1021 exit_code = sig->group_exit_code;
1022 else if (!thread_group_empty(current)) {
1023 struct sighand_struct *const sighand = current->sighand;
1025 spin_lock_irq(&sighand->siglock);
1026 if (signal_group_exit(sig))
1027 /* Another thread got here before we took the lock. */
1028 exit_code = sig->group_exit_code;
1030 sig->group_exit_code = exit_code;
1031 sig->flags = SIGNAL_GROUP_EXIT;
1032 zap_other_threads(current);
1034 spin_unlock_irq(&sighand->siglock);
1042 * this kills every thread in the thread group. Note that any externally
1043 * wait4()-ing process will get the correct exit code - even if this
1044 * thread is not the thread group leader.
1046 SYSCALL_DEFINE1(exit_group, int, error_code)
1048 do_group_exit((error_code & 0xff) << 8);
1053 struct waitid_info {
1061 enum pid_type wo_type;
1065 struct waitid_info *wo_info;
1067 struct rusage *wo_rusage;
1069 wait_queue_entry_t child_wait;
1074 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1076 if (type != PIDTYPE_PID)
1077 task = task->group_leader;
1078 return task->pids[type].pid;
1081 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1083 return wo->wo_type == PIDTYPE_MAX ||
1084 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1088 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1090 if (!eligible_pid(wo, p))
1094 * Wait for all children (clone and not) if __WALL is set or
1095 * if it is traced by us.
1097 if (ptrace || (wo->wo_flags & __WALL))
1101 * Otherwise, wait for clone children *only* if __WCLONE is set;
1102 * otherwise, wait for non-clone children *only*.
1104 * Note: a "clone" child here is one that reports to its parent
1105 * using a signal other than SIGCHLD, or a non-leader thread which
1106 * we can only see if it is traced by us.
1108 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1115 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1116 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1117 * the lock and this task is uninteresting. If we return nonzero, we have
1118 * released the lock and the system call should return.
1120 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1123 pid_t pid = task_pid_vnr(p);
1124 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1125 struct waitid_info *infop;
1127 if (!likely(wo->wo_flags & WEXITED))
1130 if (unlikely(wo->wo_flags & WNOWAIT)) {
1131 status = p->exit_code;
1133 read_unlock(&tasklist_lock);
1134 sched_annotate_sleep();
1136 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1141 * Move the task's state to DEAD/TRACE, only one thread can do this.
1143 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1144 EXIT_TRACE : EXIT_DEAD;
1145 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1148 * We own this thread, nobody else can reap it.
1150 read_unlock(&tasklist_lock);
1151 sched_annotate_sleep();
1154 * Check thread_group_leader() to exclude the traced sub-threads.
1156 if (state == EXIT_DEAD && thread_group_leader(p)) {
1157 struct signal_struct *sig = p->signal;
1158 struct signal_struct *psig = current->signal;
1159 unsigned long maxrss;
1160 u64 tgutime, tgstime;
1163 * The resource counters for the group leader are in its
1164 * own task_struct. Those for dead threads in the group
1165 * are in its signal_struct, as are those for the child
1166 * processes it has previously reaped. All these
1167 * accumulate in the parent's signal_struct c* fields.
1169 * We don't bother to take a lock here to protect these
1170 * p->signal fields because the whole thread group is dead
1171 * and nobody can change them.
1173 * psig->stats_lock also protects us from our sub-theads
1174 * which can reap other children at the same time. Until
1175 * we change k_getrusage()-like users to rely on this lock
1176 * we have to take ->siglock as well.
1178 * We use thread_group_cputime_adjusted() to get times for
1179 * the thread group, which consolidates times for all threads
1180 * in the group including the group leader.
1182 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1183 spin_lock_irq(¤t->sighand->siglock);
1184 write_seqlock(&psig->stats_lock);
1185 psig->cutime += tgutime + sig->cutime;
1186 psig->cstime += tgstime + sig->cstime;
1187 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1189 p->min_flt + sig->min_flt + sig->cmin_flt;
1191 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1193 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1195 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1197 task_io_get_inblock(p) +
1198 sig->inblock + sig->cinblock;
1200 task_io_get_oublock(p) +
1201 sig->oublock + sig->coublock;
1202 maxrss = max(sig->maxrss, sig->cmaxrss);
1203 if (psig->cmaxrss < maxrss)
1204 psig->cmaxrss = maxrss;
1205 task_io_accounting_add(&psig->ioac, &p->ioac);
1206 task_io_accounting_add(&psig->ioac, &sig->ioac);
1207 write_sequnlock(&psig->stats_lock);
1208 spin_unlock_irq(¤t->sighand->siglock);
1212 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1213 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1214 ? p->signal->group_exit_code : p->exit_code;
1215 wo->wo_stat = status;
1217 if (state == EXIT_TRACE) {
1218 write_lock_irq(&tasklist_lock);
1219 /* We dropped tasklist, ptracer could die and untrace */
1222 /* If parent wants a zombie, don't release it now */
1223 state = EXIT_ZOMBIE;
1224 if (do_notify_parent(p, p->exit_signal))
1226 p->exit_state = state;
1227 write_unlock_irq(&tasklist_lock);
1229 if (state == EXIT_DEAD)
1233 infop = wo->wo_info;
1235 if ((status & 0x7f) == 0) {
1236 infop->cause = CLD_EXITED;
1237 infop->status = status >> 8;
1239 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1240 infop->status = status & 0x7f;
1249 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1252 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1253 return &p->exit_code;
1255 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1256 return &p->signal->group_exit_code;
1262 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1264 * @ptrace: is the wait for ptrace
1265 * @p: task to wait for
1267 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1270 * read_lock(&tasklist_lock), which is released if return value is
1271 * non-zero. Also, grabs and releases @p->sighand->siglock.
1274 * 0 if wait condition didn't exist and search for other wait conditions
1275 * should continue. Non-zero return, -errno on failure and @p's pid on
1276 * success, implies that tasklist_lock is released and wait condition
1277 * search should terminate.
1279 static int wait_task_stopped(struct wait_opts *wo,
1280 int ptrace, struct task_struct *p)
1282 struct waitid_info *infop;
1283 int exit_code, *p_code, why;
1284 uid_t uid = 0; /* unneeded, required by compiler */
1288 * Traditionally we see ptrace'd stopped tasks regardless of options.
1290 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1293 if (!task_stopped_code(p, ptrace))
1297 spin_lock_irq(&p->sighand->siglock);
1299 p_code = task_stopped_code(p, ptrace);
1300 if (unlikely(!p_code))
1303 exit_code = *p_code;
1307 if (!unlikely(wo->wo_flags & WNOWAIT))
1310 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1312 spin_unlock_irq(&p->sighand->siglock);
1317 * Now we are pretty sure this task is interesting.
1318 * Make sure it doesn't get reaped out from under us while we
1319 * give up the lock and then examine it below. We don't want to
1320 * keep holding onto the tasklist_lock while we call getrusage and
1321 * possibly take page faults for user memory.
1324 pid = task_pid_vnr(p);
1325 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1326 read_unlock(&tasklist_lock);
1327 sched_annotate_sleep();
1329 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1332 if (likely(!(wo->wo_flags & WNOWAIT)))
1333 wo->wo_stat = (exit_code << 8) | 0x7f;
1335 infop = wo->wo_info;
1338 infop->status = exit_code;
1346 * Handle do_wait work for one task in a live, non-stopped state.
1347 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1348 * the lock and this task is uninteresting. If we return nonzero, we have
1349 * released the lock and the system call should return.
1351 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1353 struct waitid_info *infop;
1357 if (!unlikely(wo->wo_flags & WCONTINUED))
1360 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1363 spin_lock_irq(&p->sighand->siglock);
1364 /* Re-check with the lock held. */
1365 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1366 spin_unlock_irq(&p->sighand->siglock);
1369 if (!unlikely(wo->wo_flags & WNOWAIT))
1370 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1371 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1372 spin_unlock_irq(&p->sighand->siglock);
1374 pid = task_pid_vnr(p);
1376 read_unlock(&tasklist_lock);
1377 sched_annotate_sleep();
1379 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1382 infop = wo->wo_info;
1384 wo->wo_stat = 0xffff;
1386 infop->cause = CLD_CONTINUED;
1389 infop->status = SIGCONT;
1395 * Consider @p for a wait by @parent.
1397 * -ECHILD should be in ->notask_error before the first call.
1398 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1399 * Returns zero if the search for a child should continue;
1400 * then ->notask_error is 0 if @p is an eligible child,
1403 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1404 struct task_struct *p)
1407 * We can race with wait_task_zombie() from another thread.
1408 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1409 * can't confuse the checks below.
1411 int exit_state = ACCESS_ONCE(p->exit_state);
1414 if (unlikely(exit_state == EXIT_DEAD))
1417 ret = eligible_child(wo, ptrace, p);
1421 if (unlikely(exit_state == EXIT_TRACE)) {
1423 * ptrace == 0 means we are the natural parent. In this case
1424 * we should clear notask_error, debugger will notify us.
1426 if (likely(!ptrace))
1427 wo->notask_error = 0;
1431 if (likely(!ptrace) && unlikely(p->ptrace)) {
1433 * If it is traced by its real parent's group, just pretend
1434 * the caller is ptrace_do_wait() and reap this child if it
1437 * This also hides group stop state from real parent; otherwise
1438 * a single stop can be reported twice as group and ptrace stop.
1439 * If a ptracer wants to distinguish these two events for its
1440 * own children it should create a separate process which takes
1441 * the role of real parent.
1443 if (!ptrace_reparented(p))
1448 if (exit_state == EXIT_ZOMBIE) {
1449 /* we don't reap group leaders with subthreads */
1450 if (!delay_group_leader(p)) {
1452 * A zombie ptracee is only visible to its ptracer.
1453 * Notification and reaping will be cascaded to the
1454 * real parent when the ptracer detaches.
1456 if (unlikely(ptrace) || likely(!p->ptrace))
1457 return wait_task_zombie(wo, p);
1461 * Allow access to stopped/continued state via zombie by
1462 * falling through. Clearing of notask_error is complex.
1466 * If WEXITED is set, notask_error should naturally be
1467 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1468 * so, if there are live subthreads, there are events to
1469 * wait for. If all subthreads are dead, it's still safe
1470 * to clear - this function will be called again in finite
1471 * amount time once all the subthreads are released and
1472 * will then return without clearing.
1476 * Stopped state is per-task and thus can't change once the
1477 * target task dies. Only continued and exited can happen.
1478 * Clear notask_error if WCONTINUED | WEXITED.
1480 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1481 wo->notask_error = 0;
1484 * @p is alive and it's gonna stop, continue or exit, so
1485 * there always is something to wait for.
1487 wo->notask_error = 0;
1491 * Wait for stopped. Depending on @ptrace, different stopped state
1492 * is used and the two don't interact with each other.
1494 ret = wait_task_stopped(wo, ptrace, p);
1499 * Wait for continued. There's only one continued state and the
1500 * ptracer can consume it which can confuse the real parent. Don't
1501 * use WCONTINUED from ptracer. You don't need or want it.
1503 return wait_task_continued(wo, p);
1507 * Do the work of do_wait() for one thread in the group, @tsk.
1509 * -ECHILD should be in ->notask_error before the first call.
1510 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1511 * Returns zero if the search for a child should continue; then
1512 * ->notask_error is 0 if there were any eligible children,
1515 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1517 struct task_struct *p;
1519 list_for_each_entry(p, &tsk->children, sibling) {
1520 int ret = wait_consider_task(wo, 0, p);
1529 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1531 struct task_struct *p;
1533 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1534 int ret = wait_consider_task(wo, 1, p);
1543 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1544 int sync, void *key)
1546 struct wait_opts *wo = container_of(wait, struct wait_opts,
1548 struct task_struct *p = key;
1550 if (!eligible_pid(wo, p))
1553 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1556 return default_wake_function(wait, mode, sync, key);
1559 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1561 __wake_up_sync_key(&parent->signal->wait_chldexit,
1562 TASK_INTERRUPTIBLE, 1, p);
1565 static long do_wait(struct wait_opts *wo)
1567 struct task_struct *tsk;
1570 trace_sched_process_wait(wo->wo_pid);
1572 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1573 wo->child_wait.private = current;
1574 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1577 * If there is nothing that can match our criteria, just get out.
1578 * We will clear ->notask_error to zero if we see any child that
1579 * might later match our criteria, even if we are not able to reap
1582 wo->notask_error = -ECHILD;
1583 if ((wo->wo_type < PIDTYPE_MAX) &&
1584 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1587 set_current_state(TASK_INTERRUPTIBLE);
1588 read_lock(&tasklist_lock);
1591 retval = do_wait_thread(wo, tsk);
1595 retval = ptrace_do_wait(wo, tsk);
1599 if (wo->wo_flags & __WNOTHREAD)
1601 } while_each_thread(current, tsk);
1602 read_unlock(&tasklist_lock);
1605 retval = wo->notask_error;
1606 if (!retval && !(wo->wo_flags & WNOHANG)) {
1607 retval = -ERESTARTSYS;
1608 if (!signal_pending(current)) {
1614 __set_current_state(TASK_RUNNING);
1615 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1619 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1620 int options, struct rusage *ru)
1622 struct wait_opts wo;
1623 struct pid *pid = NULL;
1627 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1628 __WNOTHREAD|__WCLONE|__WALL))
1630 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1643 type = PIDTYPE_PGID;
1651 if (type < PIDTYPE_MAX)
1652 pid = find_get_pid(upid);
1656 wo.wo_flags = options;
1665 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1666 infop, int, options, struct rusage __user *, ru)
1669 struct waitid_info info = {.status = 0};
1670 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1676 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1682 if (!user_access_begin(VERIFY_WRITE, infop, sizeof(*infop)))
1685 unsafe_put_user(signo, &infop->si_signo, Efault);
1686 unsafe_put_user(0, &infop->si_errno, Efault);
1687 unsafe_put_user(info.cause, &infop->si_code, Efault);
1688 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1689 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1690 unsafe_put_user(info.status, &infop->si_status, Efault);
1698 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1701 struct wait_opts wo;
1702 struct pid *pid = NULL;
1706 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1707 __WNOTHREAD|__WCLONE|__WALL))
1710 /* -INT_MIN is not defined */
1711 if (upid == INT_MIN)
1716 else if (upid < 0) {
1717 type = PIDTYPE_PGID;
1718 pid = find_get_pid(-upid);
1719 } else if (upid == 0) {
1720 type = PIDTYPE_PGID;
1721 pid = get_task_pid(current, PIDTYPE_PGID);
1722 } else /* upid > 0 */ {
1724 pid = find_get_pid(upid);
1729 wo.wo_flags = options | WEXITED;
1735 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1741 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1742 int, options, struct rusage __user *, ru)
1745 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1748 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1754 #ifdef __ARCH_WANT_SYS_WAITPID
1757 * sys_waitpid() remains for compatibility. waitpid() should be
1758 * implemented by calling sys_wait4() from libc.a.
1760 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1762 return sys_wait4(pid, stat_addr, options, NULL);
1767 #ifdef CONFIG_COMPAT
1768 COMPAT_SYSCALL_DEFINE4(wait4,
1770 compat_uint_t __user *, stat_addr,
1772 struct compat_rusage __user *, ru)
1775 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1777 if (ru && put_compat_rusage(&r, ru))
1783 COMPAT_SYSCALL_DEFINE5(waitid,
1784 int, which, compat_pid_t, pid,
1785 struct compat_siginfo __user *, infop, int, options,
1786 struct compat_rusage __user *, uru)
1789 struct waitid_info info = {.status = 0};
1790 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1796 /* kernel_waitid() overwrites everything in ru */
1797 if (COMPAT_USE_64BIT_TIME)
1798 err = copy_to_user(uru, &ru, sizeof(ru));
1800 err = put_compat_rusage(&ru, uru);
1809 if (!user_access_begin(VERIFY_WRITE, infop, sizeof(*infop)))
1812 unsafe_put_user(signo, &infop->si_signo, Efault);
1813 unsafe_put_user(0, &infop->si_errno, Efault);
1814 unsafe_put_user(info.cause, &infop->si_code, Efault);
1815 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1816 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1817 unsafe_put_user(info.status, &infop->si_status, Efault);
1826 __weak void abort(void)
1830 /* if that doesn't kill us, halt */
1831 panic("Oops failed to kill thread");
1833 EXPORT_SYMBOL(abort);