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_TGID);
124 detach_pid(p, PIDTYPE_PGID);
125 detach_pid(p, PIDTYPE_SID);
127 list_del_rcu(&p->tasks);
128 list_del_init(&p->sibling);
129 __this_cpu_dec(process_counts);
131 list_del_rcu(&p->thread_group);
132 list_del_rcu(&p->thread_node);
136 * This function expects the tasklist_lock write-locked.
138 static void __exit_signal(struct task_struct *tsk)
140 struct signal_struct *sig = tsk->signal;
141 bool group_dead = thread_group_leader(tsk);
142 struct sighand_struct *sighand;
143 struct tty_struct *uninitialized_var(tty);
146 sighand = rcu_dereference_check(tsk->sighand,
147 lockdep_tasklist_lock_is_held());
148 spin_lock(&sighand->siglock);
150 #ifdef CONFIG_POSIX_TIMERS
151 posix_cpu_timers_exit(tsk);
153 posix_cpu_timers_exit_group(tsk);
156 * This can only happen if the caller is de_thread().
157 * FIXME: this is the temporary hack, we should teach
158 * posix-cpu-timers to handle this case correctly.
160 if (unlikely(has_group_leader_pid(tsk)))
161 posix_cpu_timers_exit_group(tsk);
170 * If there is any task waiting for the group exit
173 if (sig->notify_count > 0 && !--sig->notify_count)
174 wake_up_process(sig->group_exit_task);
176 if (tsk == sig->curr_target)
177 sig->curr_target = next_thread(tsk);
180 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
181 sizeof(unsigned long long));
184 * Accumulate here the counters for all threads as they die. We could
185 * skip the group leader because it is the last user of signal_struct,
186 * but we want to avoid the race with thread_group_cputime() which can
187 * see the empty ->thread_head list.
189 task_cputime(tsk, &utime, &stime);
190 write_seqlock(&sig->stats_lock);
193 sig->gtime += task_gtime(tsk);
194 sig->min_flt += tsk->min_flt;
195 sig->maj_flt += tsk->maj_flt;
196 sig->nvcsw += tsk->nvcsw;
197 sig->nivcsw += tsk->nivcsw;
198 sig->inblock += task_io_get_inblock(tsk);
199 sig->oublock += task_io_get_oublock(tsk);
200 task_io_accounting_add(&sig->ioac, &tsk->ioac);
201 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
203 __unhash_process(tsk, group_dead);
204 write_sequnlock(&sig->stats_lock);
207 * Do this under ->siglock, we can race with another thread
208 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
210 flush_sigqueue(&tsk->pending);
212 spin_unlock(&sighand->siglock);
214 __cleanup_sighand(sighand);
215 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
217 flush_sigqueue(&sig->shared_pending);
222 static void delayed_put_task_struct(struct rcu_head *rhp)
224 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
226 perf_event_delayed_put(tsk);
227 trace_sched_process_free(tsk);
228 put_task_struct(tsk);
232 void release_task(struct task_struct *p)
234 struct task_struct *leader;
237 /* don't need to get the RCU readlock here - the process is dead and
238 * can't be modifying its own credentials. But shut RCU-lockdep up */
240 atomic_dec(&__task_cred(p)->user->processes);
246 write_lock_irq(&tasklist_lock);
247 ptrace_release_task(p);
251 * If we are the last non-leader member of the thread
252 * group, and the leader is zombie, then notify the
253 * group leader's parent process. (if it wants notification.)
256 leader = p->group_leader;
257 if (leader != p && thread_group_empty(leader)
258 && leader->exit_state == EXIT_ZOMBIE) {
260 * If we were the last child thread and the leader has
261 * exited already, and the leader's parent ignores SIGCHLD,
262 * then we are the one who should release the leader.
264 zap_leader = do_notify_parent(leader, leader->exit_signal);
266 leader->exit_state = EXIT_DEAD;
269 write_unlock_irq(&tasklist_lock);
271 call_rcu(&p->rcu, delayed_put_task_struct);
274 if (unlikely(zap_leader))
279 * Note that if this function returns a valid task_struct pointer (!NULL)
280 * task->usage must remain >0 for the duration of the RCU critical section.
282 struct task_struct *task_rcu_dereference(struct task_struct **ptask)
284 struct sighand_struct *sighand;
285 struct task_struct *task;
288 * We need to verify that release_task() was not called and thus
289 * delayed_put_task_struct() can't run and drop the last reference
290 * before rcu_read_unlock(). We check task->sighand != NULL,
291 * but we can read the already freed and reused memory.
294 task = rcu_dereference(*ptask);
298 probe_kernel_address(&task->sighand, sighand);
301 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
302 * was already freed we can not miss the preceding update of this
306 if (unlikely(task != READ_ONCE(*ptask)))
310 * We've re-checked that "task == *ptask", now we have two different
313 * 1. This is actually the same task/task_struct. In this case
314 * sighand != NULL tells us it is still alive.
316 * 2. This is another task which got the same memory for task_struct.
317 * We can't know this of course, and we can not trust
320 * In this case we actually return a random value, but this is
323 * If we return NULL - we can pretend that we actually noticed that
324 * *ptask was updated when the previous task has exited. Or pretend
325 * that probe_slab_address(&sighand) reads NULL.
327 * If we return the new task (because sighand is not NULL for any
328 * reason) - this is fine too. This (new) task can't go away before
331 * And note: We could even eliminate the false positive if re-read
332 * task->sighand once again to avoid the falsely NULL. But this case
333 * is very unlikely so we don't care.
341 void rcuwait_wake_up(struct rcuwait *w)
343 struct task_struct *task;
348 * Order condition vs @task, such that everything prior to the load
349 * of @task is visible. This is the condition as to why the user called
350 * rcuwait_trywake() in the first place. Pairs with set_current_state()
351 * barrier (A) in rcuwait_wait_event().
354 * [S] tsk = current [S] cond = true
361 * Avoid using task_rcu_dereference() magic as long as we are careful,
362 * see comment in rcuwait_wait_event() regarding ->exit_state.
364 task = rcu_dereference(w->task);
366 wake_up_process(task);
371 * Determine if a process group is "orphaned", according to the POSIX
372 * definition in 2.2.2.52. Orphaned process groups are not to be affected
373 * by terminal-generated stop signals. Newly orphaned process groups are
374 * to receive a SIGHUP and a SIGCONT.
376 * "I ask you, have you ever known what it is to be an orphan?"
378 static int will_become_orphaned_pgrp(struct pid *pgrp,
379 struct task_struct *ignored_task)
381 struct task_struct *p;
383 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
384 if ((p == ignored_task) ||
385 (p->exit_state && thread_group_empty(p)) ||
386 is_global_init(p->real_parent))
389 if (task_pgrp(p->real_parent) != pgrp &&
390 task_session(p->real_parent) == task_session(p))
392 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
397 int is_current_pgrp_orphaned(void)
401 read_lock(&tasklist_lock);
402 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
403 read_unlock(&tasklist_lock);
408 static bool has_stopped_jobs(struct pid *pgrp)
410 struct task_struct *p;
412 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
413 if (p->signal->flags & SIGNAL_STOP_STOPPED)
415 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
421 * Check to see if any process groups have become orphaned as
422 * a result of our exiting, and if they have any stopped jobs,
423 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
426 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
428 struct pid *pgrp = task_pgrp(tsk);
429 struct task_struct *ignored_task = tsk;
432 /* exit: our father is in a different pgrp than
433 * we are and we were the only connection outside.
435 parent = tsk->real_parent;
437 /* reparent: our child is in a different pgrp than
438 * we are, and it was the only connection outside.
442 if (task_pgrp(parent) != pgrp &&
443 task_session(parent) == task_session(tsk) &&
444 will_become_orphaned_pgrp(pgrp, ignored_task) &&
445 has_stopped_jobs(pgrp)) {
446 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
447 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
453 * A task is exiting. If it owned this mm, find a new owner for the mm.
455 void mm_update_next_owner(struct mm_struct *mm)
457 struct task_struct *c, *g, *p = current;
461 * If the exiting or execing task is not the owner, it's
462 * someone else's problem.
467 * The current owner is exiting/execing and there are no other
468 * candidates. Do not leave the mm pointing to a possibly
469 * freed task structure.
471 if (atomic_read(&mm->mm_users) <= 1) {
476 read_lock(&tasklist_lock);
478 * Search in the children
480 list_for_each_entry(c, &p->children, sibling) {
482 goto assign_new_owner;
486 * Search in the siblings
488 list_for_each_entry(c, &p->real_parent->children, sibling) {
490 goto assign_new_owner;
494 * Search through everything else, we should not get here often.
496 for_each_process(g) {
497 if (g->flags & PF_KTHREAD)
499 for_each_thread(g, c) {
501 goto assign_new_owner;
506 read_unlock(&tasklist_lock);
508 * We found no owner yet mm_users > 1: this implies that we are
509 * most likely racing with swapoff (try_to_unuse()) or /proc or
510 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
519 * The task_lock protects c->mm from changing.
520 * We always want mm->owner->mm == mm
524 * Delay read_unlock() till we have the task_lock()
525 * to ensure that c does not slip away underneath us
527 read_unlock(&tasklist_lock);
537 #endif /* CONFIG_MEMCG */
540 * Turn us into a lazy TLB process if we
543 static void exit_mm(void)
545 struct mm_struct *mm = current->mm;
546 struct core_state *core_state;
548 exit_mm_release(current, mm);
553 * Serialize with any possible pending coredump.
554 * We must hold mmap_sem around checking core_state
555 * and clearing tsk->mm. The core-inducing thread
556 * will increment ->nr_threads for each thread in the
557 * group with ->mm != NULL.
559 down_read(&mm->mmap_sem);
560 core_state = mm->core_state;
562 struct core_thread self;
564 up_read(&mm->mmap_sem);
567 if (self.task->flags & PF_SIGNALED)
568 self.next = xchg(&core_state->dumper.next, &self);
572 * Implies mb(), the result of xchg() must be visible
573 * to core_state->dumper.
575 if (atomic_dec_and_test(&core_state->nr_threads))
576 complete(&core_state->startup);
579 set_current_state(TASK_UNINTERRUPTIBLE);
580 if (!self.task) /* see coredump_finish() */
582 freezable_schedule();
584 __set_current_state(TASK_RUNNING);
585 down_read(&mm->mmap_sem);
588 BUG_ON(mm != current->active_mm);
589 /* more a memory barrier than a real lock */
592 up_read(&mm->mmap_sem);
593 enter_lazy_tlb(mm, current);
594 task_unlock(current);
595 mm_update_next_owner(mm);
597 if (test_thread_flag(TIF_MEMDIE))
601 static struct task_struct *find_alive_thread(struct task_struct *p)
603 struct task_struct *t;
605 for_each_thread(p, t) {
606 if (!(t->flags & PF_EXITING))
612 static struct task_struct *find_child_reaper(struct task_struct *father,
613 struct list_head *dead)
614 __releases(&tasklist_lock)
615 __acquires(&tasklist_lock)
617 struct pid_namespace *pid_ns = task_active_pid_ns(father);
618 struct task_struct *reaper = pid_ns->child_reaper;
619 struct task_struct *p, *n;
621 if (likely(reaper != father))
624 reaper = find_alive_thread(father);
626 pid_ns->child_reaper = reaper;
630 write_unlock_irq(&tasklist_lock);
632 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
633 list_del_init(&p->ptrace_entry);
637 zap_pid_ns_processes(pid_ns);
638 write_lock_irq(&tasklist_lock);
644 * When we die, we re-parent all our children, and try to:
645 * 1. give them to another thread in our thread group, if such a member exists
646 * 2. give it to the first ancestor process which prctl'd itself as a
647 * child_subreaper for its children (like a service manager)
648 * 3. give it to the init process (PID 1) in our pid namespace
650 static struct task_struct *find_new_reaper(struct task_struct *father,
651 struct task_struct *child_reaper)
653 struct task_struct *thread, *reaper;
655 thread = find_alive_thread(father);
659 if (father->signal->has_child_subreaper) {
660 unsigned int ns_level = task_pid(father)->level;
662 * Find the first ->is_child_subreaper ancestor in our pid_ns.
663 * We can't check reaper != child_reaper to ensure we do not
664 * cross the namespaces, the exiting parent could be injected
665 * by setns() + fork().
666 * We check pid->level, this is slightly more efficient than
667 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
669 for (reaper = father->real_parent;
670 task_pid(reaper)->level == ns_level;
671 reaper = reaper->real_parent) {
672 if (reaper == &init_task)
674 if (!reaper->signal->is_child_subreaper)
676 thread = find_alive_thread(reaper);
686 * Any that need to be release_task'd are put on the @dead list.
688 static void reparent_leader(struct task_struct *father, struct task_struct *p,
689 struct list_head *dead)
691 if (unlikely(p->exit_state == EXIT_DEAD))
694 /* We don't want people slaying init. */
695 p->exit_signal = SIGCHLD;
697 /* If it has exited notify the new parent about this child's death. */
699 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
700 if (do_notify_parent(p, p->exit_signal)) {
701 p->exit_state = EXIT_DEAD;
702 list_add(&p->ptrace_entry, dead);
706 kill_orphaned_pgrp(p, father);
710 * This does two things:
712 * A. Make init inherit all the child processes
713 * B. Check to see if any process groups have become orphaned
714 * as a result of our exiting, and if they have any stopped
715 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
717 static void forget_original_parent(struct task_struct *father,
718 struct list_head *dead)
720 struct task_struct *p, *t, *reaper;
722 if (unlikely(!list_empty(&father->ptraced)))
723 exit_ptrace(father, dead);
725 /* Can drop and reacquire tasklist_lock */
726 reaper = find_child_reaper(father, dead);
727 if (list_empty(&father->children))
730 reaper = find_new_reaper(father, reaper);
731 list_for_each_entry(p, &father->children, sibling) {
732 for_each_thread(p, t) {
733 t->real_parent = reaper;
734 BUG_ON((!t->ptrace) != (t->parent == father));
735 if (likely(!t->ptrace))
736 t->parent = t->real_parent;
737 if (t->pdeath_signal)
738 group_send_sig_info(t->pdeath_signal,
743 * If this is a threaded reparent there is no need to
744 * notify anyone anything has happened.
746 if (!same_thread_group(reaper, father))
747 reparent_leader(father, p, dead);
749 list_splice_tail_init(&father->children, &reaper->children);
753 * Send signals to all our closest relatives so that they know
754 * to properly mourn us..
756 static void exit_notify(struct task_struct *tsk, int group_dead)
759 struct task_struct *p, *n;
762 write_lock_irq(&tasklist_lock);
763 forget_original_parent(tsk, &dead);
766 kill_orphaned_pgrp(tsk->group_leader, NULL);
768 if (unlikely(tsk->ptrace)) {
769 int sig = thread_group_leader(tsk) &&
770 thread_group_empty(tsk) &&
771 !ptrace_reparented(tsk) ?
772 tsk->exit_signal : SIGCHLD;
773 autoreap = do_notify_parent(tsk, sig);
774 } else if (thread_group_leader(tsk)) {
775 autoreap = thread_group_empty(tsk) &&
776 do_notify_parent(tsk, tsk->exit_signal);
781 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
782 if (tsk->exit_state == EXIT_DEAD)
783 list_add(&tsk->ptrace_entry, &dead);
785 /* mt-exec, de_thread() is waiting for group leader */
786 if (unlikely(tsk->signal->notify_count < 0))
787 wake_up_process(tsk->signal->group_exit_task);
788 write_unlock_irq(&tasklist_lock);
790 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
791 list_del_init(&p->ptrace_entry);
796 #ifdef CONFIG_DEBUG_STACK_USAGE
797 static void check_stack_usage(void)
799 static DEFINE_SPINLOCK(low_water_lock);
800 static int lowest_to_date = THREAD_SIZE;
803 free = stack_not_used(current);
805 if (free >= lowest_to_date)
808 spin_lock(&low_water_lock);
809 if (free < lowest_to_date) {
810 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
811 current->comm, task_pid_nr(current), free);
812 lowest_to_date = free;
814 spin_unlock(&low_water_lock);
817 static inline void check_stack_usage(void) {}
820 void __noreturn do_exit(long code)
822 struct task_struct *tsk = current;
826 * We can get here from a kernel oops, sometimes with preemption off.
827 * Start by checking for critical errors.
828 * Then fix up important state like USER_DS and preemption.
829 * Then do everything else.
832 WARN_ON(blk_needs_flush_plug(tsk));
834 if (unlikely(in_interrupt()))
835 panic("Aiee, killing interrupt handler!");
836 if (unlikely(!tsk->pid))
837 panic("Attempted to kill the idle task!");
840 * If do_exit is called because this processes oopsed, it's possible
841 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
842 * continuing. Amongst other possible reasons, this is to prevent
843 * mm_release()->clear_child_tid() from writing to a user-controlled
848 if (unlikely(in_atomic())) {
849 pr_info("note: %s[%d] exited with preempt_count %d\n",
850 current->comm, task_pid_nr(current),
852 preempt_count_set(PREEMPT_ENABLED);
855 profile_task_exit(tsk);
858 ptrace_event(PTRACE_EVENT_EXIT, code);
860 validate_creds_for_do_exit(tsk);
863 * We're taking recursive faults here in do_exit. Safest is to just
864 * leave this task alone and wait for reboot.
866 if (unlikely(tsk->flags & PF_EXITING)) {
867 pr_alert("Fixing recursive fault but reboot is needed!\n");
868 futex_exit_recursive(tsk);
869 set_current_state(TASK_UNINTERRUPTIBLE);
873 exit_signals(tsk); /* sets PF_EXITING */
875 /* sync mm's RSS info before statistics gathering */
877 sync_mm_rss(tsk->mm);
878 acct_update_integrals(tsk);
879 group_dead = atomic_dec_and_test(&tsk->signal->live);
882 * If the last thread of global init has exited, panic
883 * immediately to get a useable coredump.
885 if (unlikely(is_global_init(tsk)))
886 panic("Attempted to kill init! exitcode=0x%08x\n",
887 tsk->signal->group_exit_code ?: (int)code);
889 #ifdef CONFIG_POSIX_TIMERS
890 hrtimer_cancel(&tsk->signal->real_timer);
891 exit_itimers(tsk->signal);
894 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
896 acct_collect(code, group_dead);
901 tsk->exit_code = code;
902 taskstats_exit(tsk, group_dead);
908 trace_sched_process_exit(tsk);
915 disassociate_ctty(1);
916 exit_task_namespaces(tsk);
921 * Flush inherited counters to the parent - before the parent
922 * gets woken up by child-exit notifications.
924 * because of cgroup mode, must be called before cgroup_exit()
926 perf_event_exit_task(tsk);
928 sched_autogroup_exit_task(tsk);
932 * FIXME: do that only when needed, using sched_exit tracepoint
934 flush_ptrace_hw_breakpoint(tsk);
936 exit_tasks_rcu_start();
937 exit_notify(tsk, group_dead);
938 proc_exit_connector(tsk);
939 mpol_put_task_policy(tsk);
941 if (unlikely(current->pi_state_cache))
942 kfree(current->pi_state_cache);
945 * Make sure we are holding no locks:
947 debug_check_no_locks_held();
950 exit_io_context(tsk);
952 if (tsk->splice_pipe)
953 free_pipe_info(tsk->splice_pipe);
955 if (tsk->task_frag.page)
956 put_page(tsk->task_frag.page);
958 validate_creds_for_do_exit(tsk);
963 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
965 exit_tasks_rcu_finish();
967 lockdep_free_task(tsk);
970 EXPORT_SYMBOL_GPL(do_exit);
972 void __noreturn make_task_dead(int signr)
975 * Take the task off the cpu after something catastrophic has
981 * Every time the system oopses, if the oops happens while a reference
982 * to an object was held, the reference leaks.
983 * If the oops doesn't also leak memory, repeated oopsing can cause
984 * reference counters to wrap around (if they're not using refcount_t).
985 * This means that repeated oopsing can make unexploitable-looking bugs
986 * exploitable through repeated oopsing.
987 * To make sure this can't happen, place an upper bound on how often the
988 * kernel may oops without panic().
990 limit = READ_ONCE(oops_limit);
991 if (atomic_inc_return(&oops_count) >= limit && limit)
992 panic("Oopsed too often (kernel.oops_limit is %d)", limit);
997 void complete_and_exit(struct completion *comp, long code)
1004 EXPORT_SYMBOL(complete_and_exit);
1006 SYSCALL_DEFINE1(exit, int, error_code)
1008 do_exit((error_code&0xff)<<8);
1012 * Take down every thread in the group. This is called by fatal signals
1013 * as well as by sys_exit_group (below).
1016 do_group_exit(int exit_code)
1018 struct signal_struct *sig = current->signal;
1020 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1022 if (signal_group_exit(sig))
1023 exit_code = sig->group_exit_code;
1024 else if (!thread_group_empty(current)) {
1025 struct sighand_struct *const sighand = current->sighand;
1027 spin_lock_irq(&sighand->siglock);
1028 if (signal_group_exit(sig))
1029 /* Another thread got here before we took the lock. */
1030 exit_code = sig->group_exit_code;
1032 sig->group_exit_code = exit_code;
1033 sig->flags = SIGNAL_GROUP_EXIT;
1034 zap_other_threads(current);
1036 spin_unlock_irq(&sighand->siglock);
1044 * this kills every thread in the thread group. Note that any externally
1045 * wait4()-ing process will get the correct exit code - even if this
1046 * thread is not the thread group leader.
1048 SYSCALL_DEFINE1(exit_group, int, error_code)
1050 do_group_exit((error_code & 0xff) << 8);
1055 struct waitid_info {
1063 enum pid_type wo_type;
1067 struct waitid_info *wo_info;
1069 struct rusage *wo_rusage;
1071 wait_queue_entry_t child_wait;
1075 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1077 return wo->wo_type == PIDTYPE_MAX ||
1078 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1082 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1084 if (!eligible_pid(wo, p))
1088 * Wait for all children (clone and not) if __WALL is set or
1089 * if it is traced by us.
1091 if (ptrace || (wo->wo_flags & __WALL))
1095 * Otherwise, wait for clone children *only* if __WCLONE is set;
1096 * otherwise, wait for non-clone children *only*.
1098 * Note: a "clone" child here is one that reports to its parent
1099 * using a signal other than SIGCHLD, or a non-leader thread which
1100 * we can only see if it is traced by us.
1102 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1109 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1110 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1111 * the lock and this task is uninteresting. If we return nonzero, we have
1112 * released the lock and the system call should return.
1114 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1117 pid_t pid = task_pid_vnr(p);
1118 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1119 struct waitid_info *infop;
1121 if (!likely(wo->wo_flags & WEXITED))
1124 if (unlikely(wo->wo_flags & WNOWAIT)) {
1125 status = p->exit_code;
1127 read_unlock(&tasklist_lock);
1128 sched_annotate_sleep();
1130 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1135 * Move the task's state to DEAD/TRACE, only one thread can do this.
1137 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1138 EXIT_TRACE : EXIT_DEAD;
1139 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1142 * We own this thread, nobody else can reap it.
1144 read_unlock(&tasklist_lock);
1145 sched_annotate_sleep();
1148 * Check thread_group_leader() to exclude the traced sub-threads.
1150 if (state == EXIT_DEAD && thread_group_leader(p)) {
1151 struct signal_struct *sig = p->signal;
1152 struct signal_struct *psig = current->signal;
1153 unsigned long maxrss;
1154 u64 tgutime, tgstime;
1157 * The resource counters for the group leader are in its
1158 * own task_struct. Those for dead threads in the group
1159 * are in its signal_struct, as are those for the child
1160 * processes it has previously reaped. All these
1161 * accumulate in the parent's signal_struct c* fields.
1163 * We don't bother to take a lock here to protect these
1164 * p->signal fields because the whole thread group is dead
1165 * and nobody can change them.
1167 * psig->stats_lock also protects us from our sub-theads
1168 * which can reap other children at the same time. Until
1169 * we change k_getrusage()-like users to rely on this lock
1170 * we have to take ->siglock as well.
1172 * We use thread_group_cputime_adjusted() to get times for
1173 * the thread group, which consolidates times for all threads
1174 * in the group including the group leader.
1176 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1177 spin_lock_irq(¤t->sighand->siglock);
1178 write_seqlock(&psig->stats_lock);
1179 psig->cutime += tgutime + sig->cutime;
1180 psig->cstime += tgstime + sig->cstime;
1181 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1183 p->min_flt + sig->min_flt + sig->cmin_flt;
1185 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1187 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1189 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1191 task_io_get_inblock(p) +
1192 sig->inblock + sig->cinblock;
1194 task_io_get_oublock(p) +
1195 sig->oublock + sig->coublock;
1196 maxrss = max(sig->maxrss, sig->cmaxrss);
1197 if (psig->cmaxrss < maxrss)
1198 psig->cmaxrss = maxrss;
1199 task_io_accounting_add(&psig->ioac, &p->ioac);
1200 task_io_accounting_add(&psig->ioac, &sig->ioac);
1201 write_sequnlock(&psig->stats_lock);
1202 spin_unlock_irq(¤t->sighand->siglock);
1206 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1207 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1208 ? p->signal->group_exit_code : p->exit_code;
1209 wo->wo_stat = status;
1211 if (state == EXIT_TRACE) {
1212 write_lock_irq(&tasklist_lock);
1213 /* We dropped tasklist, ptracer could die and untrace */
1216 /* If parent wants a zombie, don't release it now */
1217 state = EXIT_ZOMBIE;
1218 if (do_notify_parent(p, p->exit_signal))
1220 p->exit_state = state;
1221 write_unlock_irq(&tasklist_lock);
1223 if (state == EXIT_DEAD)
1227 infop = wo->wo_info;
1229 if ((status & 0x7f) == 0) {
1230 infop->cause = CLD_EXITED;
1231 infop->status = status >> 8;
1233 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1234 infop->status = status & 0x7f;
1243 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1246 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1247 return &p->exit_code;
1249 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1250 return &p->signal->group_exit_code;
1256 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1258 * @ptrace: is the wait for ptrace
1259 * @p: task to wait for
1261 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1264 * read_lock(&tasklist_lock), which is released if return value is
1265 * non-zero. Also, grabs and releases @p->sighand->siglock.
1268 * 0 if wait condition didn't exist and search for other wait conditions
1269 * should continue. Non-zero return, -errno on failure and @p's pid on
1270 * success, implies that tasklist_lock is released and wait condition
1271 * search should terminate.
1273 static int wait_task_stopped(struct wait_opts *wo,
1274 int ptrace, struct task_struct *p)
1276 struct waitid_info *infop;
1277 int exit_code, *p_code, why;
1278 uid_t uid = 0; /* unneeded, required by compiler */
1282 * Traditionally we see ptrace'd stopped tasks regardless of options.
1284 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1287 if (!task_stopped_code(p, ptrace))
1291 spin_lock_irq(&p->sighand->siglock);
1293 p_code = task_stopped_code(p, ptrace);
1294 if (unlikely(!p_code))
1297 exit_code = *p_code;
1301 if (!unlikely(wo->wo_flags & WNOWAIT))
1304 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1306 spin_unlock_irq(&p->sighand->siglock);
1311 * Now we are pretty sure this task is interesting.
1312 * Make sure it doesn't get reaped out from under us while we
1313 * give up the lock and then examine it below. We don't want to
1314 * keep holding onto the tasklist_lock while we call getrusage and
1315 * possibly take page faults for user memory.
1318 pid = task_pid_vnr(p);
1319 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1320 read_unlock(&tasklist_lock);
1321 sched_annotate_sleep();
1323 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1326 if (likely(!(wo->wo_flags & WNOWAIT)))
1327 wo->wo_stat = (exit_code << 8) | 0x7f;
1329 infop = wo->wo_info;
1332 infop->status = exit_code;
1340 * Handle do_wait work for one task in a live, non-stopped state.
1341 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1342 * the lock and this task is uninteresting. If we return nonzero, we have
1343 * released the lock and the system call should return.
1345 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1347 struct waitid_info *infop;
1351 if (!unlikely(wo->wo_flags & WCONTINUED))
1354 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1357 spin_lock_irq(&p->sighand->siglock);
1358 /* Re-check with the lock held. */
1359 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1360 spin_unlock_irq(&p->sighand->siglock);
1363 if (!unlikely(wo->wo_flags & WNOWAIT))
1364 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1365 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1366 spin_unlock_irq(&p->sighand->siglock);
1368 pid = task_pid_vnr(p);
1370 read_unlock(&tasklist_lock);
1371 sched_annotate_sleep();
1373 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1376 infop = wo->wo_info;
1378 wo->wo_stat = 0xffff;
1380 infop->cause = CLD_CONTINUED;
1383 infop->status = SIGCONT;
1389 * Consider @p for a wait by @parent.
1391 * -ECHILD should be in ->notask_error before the first call.
1392 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1393 * Returns zero if the search for a child should continue;
1394 * then ->notask_error is 0 if @p is an eligible child,
1397 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1398 struct task_struct *p)
1401 * We can race with wait_task_zombie() from another thread.
1402 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1403 * can't confuse the checks below.
1405 int exit_state = READ_ONCE(p->exit_state);
1408 if (unlikely(exit_state == EXIT_DEAD))
1411 ret = eligible_child(wo, ptrace, p);
1415 if (unlikely(exit_state == EXIT_TRACE)) {
1417 * ptrace == 0 means we are the natural parent. In this case
1418 * we should clear notask_error, debugger will notify us.
1420 if (likely(!ptrace))
1421 wo->notask_error = 0;
1425 if (likely(!ptrace) && unlikely(p->ptrace)) {
1427 * If it is traced by its real parent's group, just pretend
1428 * the caller is ptrace_do_wait() and reap this child if it
1431 * This also hides group stop state from real parent; otherwise
1432 * a single stop can be reported twice as group and ptrace stop.
1433 * If a ptracer wants to distinguish these two events for its
1434 * own children it should create a separate process which takes
1435 * the role of real parent.
1437 if (!ptrace_reparented(p))
1442 if (exit_state == EXIT_ZOMBIE) {
1443 /* we don't reap group leaders with subthreads */
1444 if (!delay_group_leader(p)) {
1446 * A zombie ptracee is only visible to its ptracer.
1447 * Notification and reaping will be cascaded to the
1448 * real parent when the ptracer detaches.
1450 if (unlikely(ptrace) || likely(!p->ptrace))
1451 return wait_task_zombie(wo, p);
1455 * Allow access to stopped/continued state via zombie by
1456 * falling through. Clearing of notask_error is complex.
1460 * If WEXITED is set, notask_error should naturally be
1461 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1462 * so, if there are live subthreads, there are events to
1463 * wait for. If all subthreads are dead, it's still safe
1464 * to clear - this function will be called again in finite
1465 * amount time once all the subthreads are released and
1466 * will then return without clearing.
1470 * Stopped state is per-task and thus can't change once the
1471 * target task dies. Only continued and exited can happen.
1472 * Clear notask_error if WCONTINUED | WEXITED.
1474 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1475 wo->notask_error = 0;
1478 * @p is alive and it's gonna stop, continue or exit, so
1479 * there always is something to wait for.
1481 wo->notask_error = 0;
1485 * Wait for stopped. Depending on @ptrace, different stopped state
1486 * is used and the two don't interact with each other.
1488 ret = wait_task_stopped(wo, ptrace, p);
1493 * Wait for continued. There's only one continued state and the
1494 * ptracer can consume it which can confuse the real parent. Don't
1495 * use WCONTINUED from ptracer. You don't need or want it.
1497 return wait_task_continued(wo, p);
1501 * Do the work of do_wait() for one thread in the group, @tsk.
1503 * -ECHILD should be in ->notask_error before the first call.
1504 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1505 * Returns zero if the search for a child should continue; then
1506 * ->notask_error is 0 if there were any eligible children,
1509 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1511 struct task_struct *p;
1513 list_for_each_entry(p, &tsk->children, sibling) {
1514 int ret = wait_consider_task(wo, 0, p);
1523 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1525 struct task_struct *p;
1527 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1528 int ret = wait_consider_task(wo, 1, p);
1537 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1538 int sync, void *key)
1540 struct wait_opts *wo = container_of(wait, struct wait_opts,
1542 struct task_struct *p = key;
1544 if (!eligible_pid(wo, p))
1547 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1550 return default_wake_function(wait, mode, sync, key);
1553 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1555 __wake_up_sync_key(&parent->signal->wait_chldexit,
1556 TASK_INTERRUPTIBLE, 1, p);
1559 static long do_wait(struct wait_opts *wo)
1561 struct task_struct *tsk;
1564 trace_sched_process_wait(wo->wo_pid);
1566 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1567 wo->child_wait.private = current;
1568 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1571 * If there is nothing that can match our criteria, just get out.
1572 * We will clear ->notask_error to zero if we see any child that
1573 * might later match our criteria, even if we are not able to reap
1576 wo->notask_error = -ECHILD;
1577 if ((wo->wo_type < PIDTYPE_MAX) &&
1578 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1581 set_current_state(TASK_INTERRUPTIBLE);
1582 read_lock(&tasklist_lock);
1585 retval = do_wait_thread(wo, tsk);
1589 retval = ptrace_do_wait(wo, tsk);
1593 if (wo->wo_flags & __WNOTHREAD)
1595 } while_each_thread(current, tsk);
1596 read_unlock(&tasklist_lock);
1599 retval = wo->notask_error;
1600 if (!retval && !(wo->wo_flags & WNOHANG)) {
1601 retval = -ERESTARTSYS;
1602 if (!signal_pending(current)) {
1608 __set_current_state(TASK_RUNNING);
1609 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1613 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1614 int options, struct rusage *ru)
1616 struct wait_opts wo;
1617 struct pid *pid = NULL;
1621 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1622 __WNOTHREAD|__WCLONE|__WALL))
1624 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1637 type = PIDTYPE_PGID;
1645 if (type < PIDTYPE_MAX)
1646 pid = find_get_pid(upid);
1650 wo.wo_flags = options;
1659 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1660 infop, int, options, struct rusage __user *, ru)
1663 struct waitid_info info = {.status = 0};
1664 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1670 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1676 if (!user_access_begin(VERIFY_WRITE, infop, sizeof(*infop)))
1679 unsafe_put_user(signo, &infop->si_signo, Efault);
1680 unsafe_put_user(0, &infop->si_errno, Efault);
1681 unsafe_put_user(info.cause, &infop->si_code, Efault);
1682 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1683 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1684 unsafe_put_user(info.status, &infop->si_status, Efault);
1692 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1695 struct wait_opts wo;
1696 struct pid *pid = NULL;
1700 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1701 __WNOTHREAD|__WCLONE|__WALL))
1704 /* -INT_MIN is not defined */
1705 if (upid == INT_MIN)
1710 else if (upid < 0) {
1711 type = PIDTYPE_PGID;
1712 pid = find_get_pid(-upid);
1713 } else if (upid == 0) {
1714 type = PIDTYPE_PGID;
1715 pid = get_task_pid(current, PIDTYPE_PGID);
1716 } else /* upid > 0 */ {
1718 pid = find_get_pid(upid);
1723 wo.wo_flags = options | WEXITED;
1729 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1735 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1736 int, options, struct rusage __user *, ru)
1739 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1742 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1748 #ifdef __ARCH_WANT_SYS_WAITPID
1751 * sys_waitpid() remains for compatibility. waitpid() should be
1752 * implemented by calling sys_wait4() from libc.a.
1754 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1756 return kernel_wait4(pid, stat_addr, options, NULL);
1761 #ifdef CONFIG_COMPAT
1762 COMPAT_SYSCALL_DEFINE4(wait4,
1764 compat_uint_t __user *, stat_addr,
1766 struct compat_rusage __user *, ru)
1769 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1771 if (ru && put_compat_rusage(&r, ru))
1777 COMPAT_SYSCALL_DEFINE5(waitid,
1778 int, which, compat_pid_t, pid,
1779 struct compat_siginfo __user *, infop, int, options,
1780 struct compat_rusage __user *, uru)
1783 struct waitid_info info = {.status = 0};
1784 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1790 /* kernel_waitid() overwrites everything in ru */
1791 if (COMPAT_USE_64BIT_TIME)
1792 err = copy_to_user(uru, &ru, sizeof(ru));
1794 err = put_compat_rusage(&ru, uru);
1803 if (!user_access_begin(VERIFY_WRITE, infop, sizeof(*infop)))
1806 unsafe_put_user(signo, &infop->si_signo, Efault);
1807 unsafe_put_user(0, &infop->si_errno, Efault);
1808 unsafe_put_user(info.cause, &infop->si_code, Efault);
1809 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1810 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1811 unsafe_put_user(info.status, &infop->si_status, Efault);
1820 __weak void abort(void)
1824 /* if that doesn't kill us, halt */
1825 panic("Oops failed to kill thread");
1827 EXPORT_SYMBOL(abort);