1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 #include <linux/slab.h>
10 #include <linux/sched/autogroup.h>
11 #include <linux/sched/mm.h>
12 #include <linux/sched/stat.h>
13 #include <linux/sched/task.h>
14 #include <linux/sched/task_stack.h>
15 #include <linux/sched/cputime.h>
16 #include <linux/interrupt.h>
17 #include <linux/module.h>
18 #include <linux/capability.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/tty.h>
22 #include <linux/iocontext.h>
23 #include <linux/key.h>
24 #include <linux/cpu.h>
25 #include <linux/acct.h>
26 #include <linux/tsacct_kern.h>
27 #include <linux/file.h>
28 #include <linux/fdtable.h>
29 #include <linux/freezer.h>
30 #include <linux/binfmts.h>
31 #include <linux/nsproxy.h>
32 #include <linux/pid_namespace.h>
33 #include <linux/ptrace.h>
34 #include <linux/profile.h>
35 #include <linux/mount.h>
36 #include <linux/proc_fs.h>
37 #include <linux/kthread.h>
38 #include <linux/mempolicy.h>
39 #include <linux/taskstats_kern.h>
40 #include <linux/delayacct.h>
41 #include <linux/cgroup.h>
42 #include <linux/syscalls.h>
43 #include <linux/signal.h>
44 #include <linux/posix-timers.h>
45 #include <linux/cn_proc.h>
46 #include <linux/mutex.h>
47 #include <linux/futex.h>
48 #include <linux/pipe_fs_i.h>
49 #include <linux/audit.h> /* for audit_free() */
50 #include <linux/resource.h>
51 #include <linux/blkdev.h>
52 #include <linux/task_io_accounting_ops.h>
53 #include <linux/tracehook.h>
54 #include <linux/fs_struct.h>
55 #include <linux/init_task.h>
56 #include <linux/perf_event.h>
57 #include <trace/events/sched.h>
58 #include <linux/hw_breakpoint.h>
59 #include <linux/oom.h>
60 #include <linux/writeback.h>
61 #include <linux/shm.h>
62 #include <linux/kcov.h>
63 #include <linux/random.h>
64 #include <linux/rcuwait.h>
65 #include <linux/compat.h>
66 #include <linux/io_uring.h>
67 #include <linux/sysfs.h>
69 #include <linux/uaccess.h>
70 #include <asm/unistd.h>
71 #include <asm/mmu_context.h>
74 * The default value should be high enough to not crash a system that randomly
75 * crashes its kernel from time to time, but low enough to at least not permit
76 * overflowing 32-bit refcounts or the ldsem writer count.
78 static unsigned int oops_limit = 10000;
81 static struct ctl_table kern_exit_table[] = {
83 .procname = "oops_limit",
85 .maxlen = sizeof(oops_limit),
87 .proc_handler = proc_douintvec,
92 static __init int kernel_exit_sysctls_init(void)
94 register_sysctl_init("kernel", kern_exit_table);
97 late_initcall(kernel_exit_sysctls_init);
100 static atomic_t oops_count = ATOMIC_INIT(0);
103 static ssize_t oops_count_show(struct kobject *kobj, struct kobj_attribute *attr,
106 return sysfs_emit(page, "%d\n", atomic_read(&oops_count));
109 static struct kobj_attribute oops_count_attr = __ATTR_RO(oops_count);
111 static __init int kernel_exit_sysfs_init(void)
113 sysfs_add_file_to_group(kernel_kobj, &oops_count_attr.attr, NULL);
116 late_initcall(kernel_exit_sysfs_init);
119 static void __unhash_process(struct task_struct *p, bool group_dead)
122 detach_pid(p, PIDTYPE_PID);
124 detach_pid(p, PIDTYPE_TGID);
125 detach_pid(p, PIDTYPE_PGID);
126 detach_pid(p, PIDTYPE_SID);
128 list_del_rcu(&p->tasks);
129 list_del_init(&p->sibling);
130 __this_cpu_dec(process_counts);
132 list_del_rcu(&p->thread_group);
133 list_del_rcu(&p->thread_node);
137 * This function expects the tasklist_lock write-locked.
139 static void __exit_signal(struct task_struct *tsk)
141 struct signal_struct *sig = tsk->signal;
142 bool group_dead = thread_group_leader(tsk);
143 struct sighand_struct *sighand;
144 struct tty_struct *tty;
147 sighand = rcu_dereference_check(tsk->sighand,
148 lockdep_tasklist_lock_is_held());
149 spin_lock(&sighand->siglock);
151 #ifdef CONFIG_POSIX_TIMERS
152 posix_cpu_timers_exit(tsk);
154 posix_cpu_timers_exit_group(tsk);
162 * If there is any task waiting for the group exit
165 if (sig->notify_count > 0 && !--sig->notify_count)
166 wake_up_process(sig->group_exit_task);
168 if (tsk == sig->curr_target)
169 sig->curr_target = next_thread(tsk);
172 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
173 sizeof(unsigned long long));
176 * Accumulate here the counters for all threads as they die. We could
177 * skip the group leader because it is the last user of signal_struct,
178 * but we want to avoid the race with thread_group_cputime() which can
179 * see the empty ->thread_head list.
181 task_cputime(tsk, &utime, &stime);
182 write_seqlock(&sig->stats_lock);
185 sig->gtime += task_gtime(tsk);
186 sig->min_flt += tsk->min_flt;
187 sig->maj_flt += tsk->maj_flt;
188 sig->nvcsw += tsk->nvcsw;
189 sig->nivcsw += tsk->nivcsw;
190 sig->inblock += task_io_get_inblock(tsk);
191 sig->oublock += task_io_get_oublock(tsk);
192 task_io_accounting_add(&sig->ioac, &tsk->ioac);
193 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
195 __unhash_process(tsk, group_dead);
196 write_sequnlock(&sig->stats_lock);
199 * Do this under ->siglock, we can race with another thread
200 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
202 flush_sigqueue(&tsk->pending);
204 spin_unlock(&sighand->siglock);
206 __cleanup_sighand(sighand);
207 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
209 flush_sigqueue(&sig->shared_pending);
214 static void delayed_put_task_struct(struct rcu_head *rhp)
216 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
218 perf_event_delayed_put(tsk);
219 trace_sched_process_free(tsk);
220 put_task_struct(tsk);
223 void put_task_struct_rcu_user(struct task_struct *task)
225 if (refcount_dec_and_test(&task->rcu_users))
226 call_rcu(&task->rcu, delayed_put_task_struct);
229 void release_task(struct task_struct *p)
231 struct task_struct *leader;
232 struct pid *thread_pid;
235 /* don't need to get the RCU readlock here - the process is dead and
236 * can't be modifying its own credentials. But shut RCU-lockdep up */
238 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
243 write_lock_irq(&tasklist_lock);
244 ptrace_release_task(p);
245 thread_pid = get_pid(p->thread_pid);
249 * If we are the last non-leader member of the thread
250 * group, and the leader is zombie, then notify the
251 * group leader's parent process. (if it wants notification.)
254 leader = p->group_leader;
255 if (leader != p && thread_group_empty(leader)
256 && leader->exit_state == EXIT_ZOMBIE) {
258 * If we were the last child thread and the leader has
259 * exited already, and the leader's parent ignores SIGCHLD,
260 * then we are the one who should release the leader.
262 zap_leader = do_notify_parent(leader, leader->exit_signal);
264 leader->exit_state = EXIT_DEAD;
267 write_unlock_irq(&tasklist_lock);
268 seccomp_filter_release(p);
269 proc_flush_pid(thread_pid);
272 put_task_struct_rcu_user(p);
275 if (unlikely(zap_leader))
279 int rcuwait_wake_up(struct rcuwait *w)
282 struct task_struct *task;
287 * Order condition vs @task, such that everything prior to the load
288 * of @task is visible. This is the condition as to why the user called
289 * rcuwait_wake() in the first place. Pairs with set_current_state()
290 * barrier (A) in rcuwait_wait_event().
293 * [S] tsk = current [S] cond = true
299 task = rcu_dereference(w->task);
301 ret = wake_up_process(task);
306 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
309 * Determine if a process group is "orphaned", according to the POSIX
310 * definition in 2.2.2.52. Orphaned process groups are not to be affected
311 * by terminal-generated stop signals. Newly orphaned process groups are
312 * to receive a SIGHUP and a SIGCONT.
314 * "I ask you, have you ever known what it is to be an orphan?"
316 static int will_become_orphaned_pgrp(struct pid *pgrp,
317 struct task_struct *ignored_task)
319 struct task_struct *p;
321 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
322 if ((p == ignored_task) ||
323 (p->exit_state && thread_group_empty(p)) ||
324 is_global_init(p->real_parent))
327 if (task_pgrp(p->real_parent) != pgrp &&
328 task_session(p->real_parent) == task_session(p))
330 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
335 int is_current_pgrp_orphaned(void)
339 read_lock(&tasklist_lock);
340 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
341 read_unlock(&tasklist_lock);
346 static bool has_stopped_jobs(struct pid *pgrp)
348 struct task_struct *p;
350 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
351 if (p->signal->flags & SIGNAL_STOP_STOPPED)
353 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
359 * Check to see if any process groups have become orphaned as
360 * a result of our exiting, and if they have any stopped jobs,
361 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
364 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
366 struct pid *pgrp = task_pgrp(tsk);
367 struct task_struct *ignored_task = tsk;
370 /* exit: our father is in a different pgrp than
371 * we are and we were the only connection outside.
373 parent = tsk->real_parent;
375 /* reparent: our child is in a different pgrp than
376 * we are, and it was the only connection outside.
380 if (task_pgrp(parent) != pgrp &&
381 task_session(parent) == task_session(tsk) &&
382 will_become_orphaned_pgrp(pgrp, ignored_task) &&
383 has_stopped_jobs(pgrp)) {
384 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
385 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
391 * A task is exiting. If it owned this mm, find a new owner for the mm.
393 void mm_update_next_owner(struct mm_struct *mm)
395 struct task_struct *c, *g, *p = current;
399 * If the exiting or execing task is not the owner, it's
400 * someone else's problem.
405 * The current owner is exiting/execing and there are no other
406 * candidates. Do not leave the mm pointing to a possibly
407 * freed task structure.
409 if (atomic_read(&mm->mm_users) <= 1) {
410 WRITE_ONCE(mm->owner, NULL);
414 read_lock(&tasklist_lock);
416 * Search in the children
418 list_for_each_entry(c, &p->children, sibling) {
420 goto assign_new_owner;
424 * Search in the siblings
426 list_for_each_entry(c, &p->real_parent->children, sibling) {
428 goto assign_new_owner;
432 * Search through everything else, we should not get here often.
434 for_each_process(g) {
435 if (g->flags & PF_KTHREAD)
437 for_each_thread(g, c) {
439 goto assign_new_owner;
444 read_unlock(&tasklist_lock);
446 * We found no owner yet mm_users > 1: this implies that we are
447 * most likely racing with swapoff (try_to_unuse()) or /proc or
448 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
450 WRITE_ONCE(mm->owner, NULL);
457 * The task_lock protects c->mm from changing.
458 * We always want mm->owner->mm == mm
462 * Delay read_unlock() till we have the task_lock()
463 * to ensure that c does not slip away underneath us
465 read_unlock(&tasklist_lock);
471 WRITE_ONCE(mm->owner, c);
475 #endif /* CONFIG_MEMCG */
478 * Turn us into a lazy TLB process if we
481 static void exit_mm(void)
483 struct mm_struct *mm = current->mm;
484 struct core_state *core_state;
486 exit_mm_release(current, mm);
491 * Serialize with any possible pending coredump.
492 * We must hold mmap_lock around checking core_state
493 * and clearing tsk->mm. The core-inducing thread
494 * will increment ->nr_threads for each thread in the
495 * group with ->mm != NULL.
498 core_state = mm->core_state;
500 struct core_thread self;
502 mmap_read_unlock(mm);
505 if (self.task->flags & PF_SIGNALED)
506 self.next = xchg(&core_state->dumper.next, &self);
510 * Implies mb(), the result of xchg() must be visible
511 * to core_state->dumper.
513 if (atomic_dec_and_test(&core_state->nr_threads))
514 complete(&core_state->startup);
517 set_current_state(TASK_UNINTERRUPTIBLE);
518 if (!self.task) /* see coredump_finish() */
520 freezable_schedule();
522 __set_current_state(TASK_RUNNING);
526 BUG_ON(mm != current->active_mm);
527 /* more a memory barrier than a real lock */
530 * When a thread stops operating on an address space, the loop
531 * in membarrier_private_expedited() may not observe that
532 * tsk->mm, and the loop in membarrier_global_expedited() may
533 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
534 * rq->membarrier_state, so those would not issue an IPI.
535 * Membarrier requires a memory barrier after accessing
536 * user-space memory, before clearing tsk->mm or the
537 * rq->membarrier_state.
539 smp_mb__after_spinlock();
542 membarrier_update_current_mm(NULL);
543 enter_lazy_tlb(mm, current);
545 task_unlock(current);
546 mmap_read_unlock(mm);
547 mm_update_next_owner(mm);
549 if (test_thread_flag(TIF_MEMDIE))
553 static struct task_struct *find_alive_thread(struct task_struct *p)
555 struct task_struct *t;
557 for_each_thread(p, t) {
558 if (!(t->flags & PF_EXITING))
564 static struct task_struct *find_child_reaper(struct task_struct *father,
565 struct list_head *dead)
566 __releases(&tasklist_lock)
567 __acquires(&tasklist_lock)
569 struct pid_namespace *pid_ns = task_active_pid_ns(father);
570 struct task_struct *reaper = pid_ns->child_reaper;
571 struct task_struct *p, *n;
573 if (likely(reaper != father))
576 reaper = find_alive_thread(father);
578 pid_ns->child_reaper = reaper;
582 write_unlock_irq(&tasklist_lock);
584 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
585 list_del_init(&p->ptrace_entry);
589 zap_pid_ns_processes(pid_ns);
590 write_lock_irq(&tasklist_lock);
596 * When we die, we re-parent all our children, and try to:
597 * 1. give them to another thread in our thread group, if such a member exists
598 * 2. give it to the first ancestor process which prctl'd itself as a
599 * child_subreaper for its children (like a service manager)
600 * 3. give it to the init process (PID 1) in our pid namespace
602 static struct task_struct *find_new_reaper(struct task_struct *father,
603 struct task_struct *child_reaper)
605 struct task_struct *thread, *reaper;
607 thread = find_alive_thread(father);
611 if (father->signal->has_child_subreaper) {
612 unsigned int ns_level = task_pid(father)->level;
614 * Find the first ->is_child_subreaper ancestor in our pid_ns.
615 * We can't check reaper != child_reaper to ensure we do not
616 * cross the namespaces, the exiting parent could be injected
617 * by setns() + fork().
618 * We check pid->level, this is slightly more efficient than
619 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
621 for (reaper = father->real_parent;
622 task_pid(reaper)->level == ns_level;
623 reaper = reaper->real_parent) {
624 if (reaper == &init_task)
626 if (!reaper->signal->is_child_subreaper)
628 thread = find_alive_thread(reaper);
638 * Any that need to be release_task'd are put on the @dead list.
640 static void reparent_leader(struct task_struct *father, struct task_struct *p,
641 struct list_head *dead)
643 if (unlikely(p->exit_state == EXIT_DEAD))
646 /* We don't want people slaying init. */
647 p->exit_signal = SIGCHLD;
649 /* If it has exited notify the new parent about this child's death. */
651 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
652 if (do_notify_parent(p, p->exit_signal)) {
653 p->exit_state = EXIT_DEAD;
654 list_add(&p->ptrace_entry, dead);
658 kill_orphaned_pgrp(p, father);
662 * This does two things:
664 * A. Make init inherit all the child processes
665 * B. Check to see if any process groups have become orphaned
666 * as a result of our exiting, and if they have any stopped
667 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
669 static void forget_original_parent(struct task_struct *father,
670 struct list_head *dead)
672 struct task_struct *p, *t, *reaper;
674 if (unlikely(!list_empty(&father->ptraced)))
675 exit_ptrace(father, dead);
677 /* Can drop and reacquire tasklist_lock */
678 reaper = find_child_reaper(father, dead);
679 if (list_empty(&father->children))
682 reaper = find_new_reaper(father, reaper);
683 list_for_each_entry(p, &father->children, sibling) {
684 for_each_thread(p, t) {
685 RCU_INIT_POINTER(t->real_parent, reaper);
686 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
687 if (likely(!t->ptrace))
688 t->parent = t->real_parent;
689 if (t->pdeath_signal)
690 group_send_sig_info(t->pdeath_signal,
695 * If this is a threaded reparent there is no need to
696 * notify anyone anything has happened.
698 if (!same_thread_group(reaper, father))
699 reparent_leader(father, p, dead);
701 list_splice_tail_init(&father->children, &reaper->children);
705 * Send signals to all our closest relatives so that they know
706 * to properly mourn us..
708 static void exit_notify(struct task_struct *tsk, int group_dead)
711 struct task_struct *p, *n;
714 write_lock_irq(&tasklist_lock);
715 forget_original_parent(tsk, &dead);
718 kill_orphaned_pgrp(tsk->group_leader, NULL);
720 tsk->exit_state = EXIT_ZOMBIE;
721 if (unlikely(tsk->ptrace)) {
722 int sig = thread_group_leader(tsk) &&
723 thread_group_empty(tsk) &&
724 !ptrace_reparented(tsk) ?
725 tsk->exit_signal : SIGCHLD;
726 autoreap = do_notify_parent(tsk, sig);
727 } else if (thread_group_leader(tsk)) {
728 autoreap = thread_group_empty(tsk) &&
729 do_notify_parent(tsk, tsk->exit_signal);
735 tsk->exit_state = EXIT_DEAD;
736 list_add(&tsk->ptrace_entry, &dead);
739 /* mt-exec, de_thread() is waiting for group leader */
740 if (unlikely(tsk->signal->notify_count < 0))
741 wake_up_process(tsk->signal->group_exit_task);
742 write_unlock_irq(&tasklist_lock);
744 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
745 list_del_init(&p->ptrace_entry);
750 #ifdef CONFIG_DEBUG_STACK_USAGE
751 static void check_stack_usage(void)
753 static DEFINE_SPINLOCK(low_water_lock);
754 static int lowest_to_date = THREAD_SIZE;
757 free = stack_not_used(current);
759 if (free >= lowest_to_date)
762 spin_lock(&low_water_lock);
763 if (free < lowest_to_date) {
764 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
765 current->comm, task_pid_nr(current), free);
766 lowest_to_date = free;
768 spin_unlock(&low_water_lock);
771 static inline void check_stack_usage(void) {}
774 void __noreturn do_exit(long code)
776 struct task_struct *tsk = current;
780 * We can get here from a kernel oops, sometimes with preemption off.
781 * Start by checking for critical errors.
782 * Then fix up important state like USER_DS and preemption.
783 * Then do everything else.
786 WARN_ON(blk_needs_flush_plug(tsk));
788 if (unlikely(in_interrupt()))
789 panic("Aiee, killing interrupt handler!");
790 if (unlikely(!tsk->pid))
791 panic("Attempted to kill the idle task!");
794 * If do_exit is called because this processes oopsed, it's possible
795 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
796 * continuing. Amongst other possible reasons, this is to prevent
797 * mm_release()->clear_child_tid() from writing to a user-controlled
800 force_uaccess_begin();
802 if (unlikely(in_atomic())) {
803 pr_info("note: %s[%d] exited with preempt_count %d\n",
804 current->comm, task_pid_nr(current),
806 preempt_count_set(PREEMPT_ENABLED);
809 profile_task_exit(tsk);
812 ptrace_event(PTRACE_EVENT_EXIT, code);
814 validate_creds_for_do_exit(tsk);
817 * We're taking recursive faults here in do_exit. Safest is to just
818 * leave this task alone and wait for reboot.
820 if (unlikely(tsk->flags & PF_EXITING)) {
821 pr_alert("Fixing recursive fault but reboot is needed!\n");
822 futex_exit_recursive(tsk);
823 set_current_state(TASK_UNINTERRUPTIBLE);
827 io_uring_files_cancel();
828 exit_signals(tsk); /* sets PF_EXITING */
830 /* sync mm's RSS info before statistics gathering */
832 sync_mm_rss(tsk->mm);
833 acct_update_integrals(tsk);
834 group_dead = atomic_dec_and_test(&tsk->signal->live);
837 * If the last thread of global init has exited, panic
838 * immediately to get a useable coredump.
840 if (unlikely(is_global_init(tsk)))
841 panic("Attempted to kill init! exitcode=0x%08x\n",
842 tsk->signal->group_exit_code ?: (int)code);
844 #ifdef CONFIG_POSIX_TIMERS
845 hrtimer_cancel(&tsk->signal->real_timer);
849 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
851 acct_collect(code, group_dead);
856 tsk->exit_code = code;
857 taskstats_exit(tsk, group_dead);
863 trace_sched_process_exit(tsk);
870 disassociate_ctty(1);
871 exit_task_namespaces(tsk);
876 * Flush inherited counters to the parent - before the parent
877 * gets woken up by child-exit notifications.
879 * because of cgroup mode, must be called before cgroup_exit()
881 perf_event_exit_task(tsk);
883 sched_autogroup_exit_task(tsk);
887 * FIXME: do that only when needed, using sched_exit tracepoint
889 flush_ptrace_hw_breakpoint(tsk);
891 exit_tasks_rcu_start();
892 exit_notify(tsk, group_dead);
893 proc_exit_connector(tsk);
894 mpol_put_task_policy(tsk);
896 if (unlikely(current->pi_state_cache))
897 kfree(current->pi_state_cache);
900 * Make sure we are holding no locks:
902 debug_check_no_locks_held();
905 exit_io_context(tsk);
907 if (tsk->splice_pipe)
908 free_pipe_info(tsk->splice_pipe);
910 if (tsk->task_frag.page)
911 put_page(tsk->task_frag.page);
913 validate_creds_for_do_exit(tsk);
918 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
920 exit_tasks_rcu_finish();
922 lockdep_free_task(tsk);
925 EXPORT_SYMBOL_GPL(do_exit);
927 void __noreturn make_task_dead(int signr)
930 * Take the task off the cpu after something catastrophic has
935 * Every time the system oopses, if the oops happens while a reference
936 * to an object was held, the reference leaks.
937 * If the oops doesn't also leak memory, repeated oopsing can cause
938 * reference counters to wrap around (if they're not using refcount_t).
939 * This means that repeated oopsing can make unexploitable-looking bugs
940 * exploitable through repeated oopsing.
941 * To make sure this can't happen, place an upper bound on how often the
942 * kernel may oops without panic().
944 if (atomic_inc_return(&oops_count) >= READ_ONCE(oops_limit))
945 panic("Oopsed too often (kernel.oops_limit is %d)", oops_limit);
950 void complete_and_exit(struct completion *comp, long code)
957 EXPORT_SYMBOL(complete_and_exit);
959 SYSCALL_DEFINE1(exit, int, error_code)
961 do_exit((error_code&0xff)<<8);
965 * Take down every thread in the group. This is called by fatal signals
966 * as well as by sys_exit_group (below).
969 do_group_exit(int exit_code)
971 struct signal_struct *sig = current->signal;
973 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
975 if (signal_group_exit(sig))
976 exit_code = sig->group_exit_code;
977 else if (!thread_group_empty(current)) {
978 struct sighand_struct *const sighand = current->sighand;
980 spin_lock_irq(&sighand->siglock);
981 if (signal_group_exit(sig))
982 /* Another thread got here before we took the lock. */
983 exit_code = sig->group_exit_code;
985 sig->group_exit_code = exit_code;
986 sig->flags = SIGNAL_GROUP_EXIT;
987 zap_other_threads(current);
989 spin_unlock_irq(&sighand->siglock);
997 * this kills every thread in the thread group. Note that any externally
998 * wait4()-ing process will get the correct exit code - even if this
999 * thread is not the thread group leader.
1001 SYSCALL_DEFINE1(exit_group, int, error_code)
1003 do_group_exit((error_code & 0xff) << 8);
1008 struct waitid_info {
1016 enum pid_type wo_type;
1020 struct waitid_info *wo_info;
1022 struct rusage *wo_rusage;
1024 wait_queue_entry_t child_wait;
1028 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1030 return wo->wo_type == PIDTYPE_MAX ||
1031 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1035 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1037 if (!eligible_pid(wo, p))
1041 * Wait for all children (clone and not) if __WALL is set or
1042 * if it is traced by us.
1044 if (ptrace || (wo->wo_flags & __WALL))
1048 * Otherwise, wait for clone children *only* if __WCLONE is set;
1049 * otherwise, wait for non-clone children *only*.
1051 * Note: a "clone" child here is one that reports to its parent
1052 * using a signal other than SIGCHLD, or a non-leader thread which
1053 * we can only see if it is traced by us.
1055 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1062 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1063 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1064 * the lock and this task is uninteresting. If we return nonzero, we have
1065 * released the lock and the system call should return.
1067 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1070 pid_t pid = task_pid_vnr(p);
1071 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1072 struct waitid_info *infop;
1074 if (!likely(wo->wo_flags & WEXITED))
1077 if (unlikely(wo->wo_flags & WNOWAIT)) {
1078 status = p->exit_code;
1080 read_unlock(&tasklist_lock);
1081 sched_annotate_sleep();
1083 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1088 * Move the task's state to DEAD/TRACE, only one thread can do this.
1090 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1091 EXIT_TRACE : EXIT_DEAD;
1092 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1095 * We own this thread, nobody else can reap it.
1097 read_unlock(&tasklist_lock);
1098 sched_annotate_sleep();
1101 * Check thread_group_leader() to exclude the traced sub-threads.
1103 if (state == EXIT_DEAD && thread_group_leader(p)) {
1104 struct signal_struct *sig = p->signal;
1105 struct signal_struct *psig = current->signal;
1106 unsigned long maxrss;
1107 u64 tgutime, tgstime;
1110 * The resource counters for the group leader are in its
1111 * own task_struct. Those for dead threads in the group
1112 * are in its signal_struct, as are those for the child
1113 * processes it has previously reaped. All these
1114 * accumulate in the parent's signal_struct c* fields.
1116 * We don't bother to take a lock here to protect these
1117 * p->signal fields because the whole thread group is dead
1118 * and nobody can change them.
1120 * psig->stats_lock also protects us from our sub-theads
1121 * which can reap other children at the same time. Until
1122 * we change k_getrusage()-like users to rely on this lock
1123 * we have to take ->siglock as well.
1125 * We use thread_group_cputime_adjusted() to get times for
1126 * the thread group, which consolidates times for all threads
1127 * in the group including the group leader.
1129 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1130 spin_lock_irq(¤t->sighand->siglock);
1131 write_seqlock(&psig->stats_lock);
1132 psig->cutime += tgutime + sig->cutime;
1133 psig->cstime += tgstime + sig->cstime;
1134 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1136 p->min_flt + sig->min_flt + sig->cmin_flt;
1138 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1140 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1142 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1144 task_io_get_inblock(p) +
1145 sig->inblock + sig->cinblock;
1147 task_io_get_oublock(p) +
1148 sig->oublock + sig->coublock;
1149 maxrss = max(sig->maxrss, sig->cmaxrss);
1150 if (psig->cmaxrss < maxrss)
1151 psig->cmaxrss = maxrss;
1152 task_io_accounting_add(&psig->ioac, &p->ioac);
1153 task_io_accounting_add(&psig->ioac, &sig->ioac);
1154 write_sequnlock(&psig->stats_lock);
1155 spin_unlock_irq(¤t->sighand->siglock);
1159 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1160 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1161 ? p->signal->group_exit_code : p->exit_code;
1162 wo->wo_stat = status;
1164 if (state == EXIT_TRACE) {
1165 write_lock_irq(&tasklist_lock);
1166 /* We dropped tasklist, ptracer could die and untrace */
1169 /* If parent wants a zombie, don't release it now */
1170 state = EXIT_ZOMBIE;
1171 if (do_notify_parent(p, p->exit_signal))
1173 p->exit_state = state;
1174 write_unlock_irq(&tasklist_lock);
1176 if (state == EXIT_DEAD)
1180 infop = wo->wo_info;
1182 if ((status & 0x7f) == 0) {
1183 infop->cause = CLD_EXITED;
1184 infop->status = status >> 8;
1186 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1187 infop->status = status & 0x7f;
1196 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1199 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1200 return &p->exit_code;
1202 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1203 return &p->signal->group_exit_code;
1209 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1211 * @ptrace: is the wait for ptrace
1212 * @p: task to wait for
1214 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1217 * read_lock(&tasklist_lock), which is released if return value is
1218 * non-zero. Also, grabs and releases @p->sighand->siglock.
1221 * 0 if wait condition didn't exist and search for other wait conditions
1222 * should continue. Non-zero return, -errno on failure and @p's pid on
1223 * success, implies that tasklist_lock is released and wait condition
1224 * search should terminate.
1226 static int wait_task_stopped(struct wait_opts *wo,
1227 int ptrace, struct task_struct *p)
1229 struct waitid_info *infop;
1230 int exit_code, *p_code, why;
1231 uid_t uid = 0; /* unneeded, required by compiler */
1235 * Traditionally we see ptrace'd stopped tasks regardless of options.
1237 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1240 if (!task_stopped_code(p, ptrace))
1244 spin_lock_irq(&p->sighand->siglock);
1246 p_code = task_stopped_code(p, ptrace);
1247 if (unlikely(!p_code))
1250 exit_code = *p_code;
1254 if (!unlikely(wo->wo_flags & WNOWAIT))
1257 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1259 spin_unlock_irq(&p->sighand->siglock);
1264 * Now we are pretty sure this task is interesting.
1265 * Make sure it doesn't get reaped out from under us while we
1266 * give up the lock and then examine it below. We don't want to
1267 * keep holding onto the tasklist_lock while we call getrusage and
1268 * possibly take page faults for user memory.
1271 pid = task_pid_vnr(p);
1272 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1273 read_unlock(&tasklist_lock);
1274 sched_annotate_sleep();
1276 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1279 if (likely(!(wo->wo_flags & WNOWAIT)))
1280 wo->wo_stat = (exit_code << 8) | 0x7f;
1282 infop = wo->wo_info;
1285 infop->status = exit_code;
1293 * Handle do_wait work for one task in a live, non-stopped state.
1294 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1295 * the lock and this task is uninteresting. If we return nonzero, we have
1296 * released the lock and the system call should return.
1298 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1300 struct waitid_info *infop;
1304 if (!unlikely(wo->wo_flags & WCONTINUED))
1307 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1310 spin_lock_irq(&p->sighand->siglock);
1311 /* Re-check with the lock held. */
1312 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1313 spin_unlock_irq(&p->sighand->siglock);
1316 if (!unlikely(wo->wo_flags & WNOWAIT))
1317 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1318 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1319 spin_unlock_irq(&p->sighand->siglock);
1321 pid = task_pid_vnr(p);
1323 read_unlock(&tasklist_lock);
1324 sched_annotate_sleep();
1326 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1329 infop = wo->wo_info;
1331 wo->wo_stat = 0xffff;
1333 infop->cause = CLD_CONTINUED;
1336 infop->status = SIGCONT;
1342 * Consider @p for a wait by @parent.
1344 * -ECHILD should be in ->notask_error before the first call.
1345 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1346 * Returns zero if the search for a child should continue;
1347 * then ->notask_error is 0 if @p is an eligible child,
1350 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1351 struct task_struct *p)
1354 * We can race with wait_task_zombie() from another thread.
1355 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1356 * can't confuse the checks below.
1358 int exit_state = READ_ONCE(p->exit_state);
1361 if (unlikely(exit_state == EXIT_DEAD))
1364 ret = eligible_child(wo, ptrace, p);
1368 if (unlikely(exit_state == EXIT_TRACE)) {
1370 * ptrace == 0 means we are the natural parent. In this case
1371 * we should clear notask_error, debugger will notify us.
1373 if (likely(!ptrace))
1374 wo->notask_error = 0;
1378 if (likely(!ptrace) && unlikely(p->ptrace)) {
1380 * If it is traced by its real parent's group, just pretend
1381 * the caller is ptrace_do_wait() and reap this child if it
1384 * This also hides group stop state from real parent; otherwise
1385 * a single stop can be reported twice as group and ptrace stop.
1386 * If a ptracer wants to distinguish these two events for its
1387 * own children it should create a separate process which takes
1388 * the role of real parent.
1390 if (!ptrace_reparented(p))
1395 if (exit_state == EXIT_ZOMBIE) {
1396 /* we don't reap group leaders with subthreads */
1397 if (!delay_group_leader(p)) {
1399 * A zombie ptracee is only visible to its ptracer.
1400 * Notification and reaping will be cascaded to the
1401 * real parent when the ptracer detaches.
1403 if (unlikely(ptrace) || likely(!p->ptrace))
1404 return wait_task_zombie(wo, p);
1408 * Allow access to stopped/continued state via zombie by
1409 * falling through. Clearing of notask_error is complex.
1413 * If WEXITED is set, notask_error should naturally be
1414 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1415 * so, if there are live subthreads, there are events to
1416 * wait for. If all subthreads are dead, it's still safe
1417 * to clear - this function will be called again in finite
1418 * amount time once all the subthreads are released and
1419 * will then return without clearing.
1423 * Stopped state is per-task and thus can't change once the
1424 * target task dies. Only continued and exited can happen.
1425 * Clear notask_error if WCONTINUED | WEXITED.
1427 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1428 wo->notask_error = 0;
1431 * @p is alive and it's gonna stop, continue or exit, so
1432 * there always is something to wait for.
1434 wo->notask_error = 0;
1438 * Wait for stopped. Depending on @ptrace, different stopped state
1439 * is used and the two don't interact with each other.
1441 ret = wait_task_stopped(wo, ptrace, p);
1446 * Wait for continued. There's only one continued state and the
1447 * ptracer can consume it which can confuse the real parent. Don't
1448 * use WCONTINUED from ptracer. You don't need or want it.
1450 return wait_task_continued(wo, p);
1454 * Do the work of do_wait() for one thread in the group, @tsk.
1456 * -ECHILD should be in ->notask_error before the first call.
1457 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1458 * Returns zero if the search for a child should continue; then
1459 * ->notask_error is 0 if there were any eligible children,
1462 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1464 struct task_struct *p;
1466 list_for_each_entry(p, &tsk->children, sibling) {
1467 int ret = wait_consider_task(wo, 0, p);
1476 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1478 struct task_struct *p;
1480 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1481 int ret = wait_consider_task(wo, 1, p);
1490 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1491 int sync, void *key)
1493 struct wait_opts *wo = container_of(wait, struct wait_opts,
1495 struct task_struct *p = key;
1497 if (!eligible_pid(wo, p))
1500 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1503 return default_wake_function(wait, mode, sync, key);
1506 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1508 __wake_up_sync_key(&parent->signal->wait_chldexit,
1509 TASK_INTERRUPTIBLE, p);
1512 static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1513 struct task_struct *target)
1515 struct task_struct *parent =
1516 !ptrace ? target->real_parent : target->parent;
1518 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1519 same_thread_group(current, parent));
1523 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1524 * and tracee lists to find the target task.
1526 static int do_wait_pid(struct wait_opts *wo)
1529 struct task_struct *target;
1533 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1534 if (target && is_effectively_child(wo, ptrace, target)) {
1535 retval = wait_consider_task(wo, ptrace, target);
1541 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1542 if (target && target->ptrace &&
1543 is_effectively_child(wo, ptrace, target)) {
1544 retval = wait_consider_task(wo, ptrace, target);
1552 static long do_wait(struct wait_opts *wo)
1556 trace_sched_process_wait(wo->wo_pid);
1558 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1559 wo->child_wait.private = current;
1560 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1563 * If there is nothing that can match our criteria, just get out.
1564 * We will clear ->notask_error to zero if we see any child that
1565 * might later match our criteria, even if we are not able to reap
1568 wo->notask_error = -ECHILD;
1569 if ((wo->wo_type < PIDTYPE_MAX) &&
1570 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1573 set_current_state(TASK_INTERRUPTIBLE);
1574 read_lock(&tasklist_lock);
1576 if (wo->wo_type == PIDTYPE_PID) {
1577 retval = do_wait_pid(wo);
1581 struct task_struct *tsk = current;
1584 retval = do_wait_thread(wo, tsk);
1588 retval = ptrace_do_wait(wo, tsk);
1592 if (wo->wo_flags & __WNOTHREAD)
1594 } 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;
1620 unsigned int f_flags = 0;
1622 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1623 __WNOTHREAD|__WCLONE|__WALL))
1625 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1637 pid = find_get_pid(upid);
1640 type = PIDTYPE_PGID;
1645 pid = find_get_pid(upid);
1647 pid = get_task_pid(current, PIDTYPE_PGID);
1654 pid = pidfd_get_pid(upid, &f_flags);
1656 return PTR_ERR(pid);
1665 wo.wo_flags = options;
1668 if (f_flags & O_NONBLOCK)
1669 wo.wo_flags |= WNOHANG;
1672 if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1679 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1680 infop, int, options, struct rusage __user *, ru)
1683 struct waitid_info info = {.status = 0};
1684 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1690 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1696 if (!user_write_access_begin(infop, sizeof(*infop)))
1699 unsafe_put_user(signo, &infop->si_signo, Efault);
1700 unsafe_put_user(0, &infop->si_errno, Efault);
1701 unsafe_put_user(info.cause, &infop->si_code, Efault);
1702 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1703 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1704 unsafe_put_user(info.status, &infop->si_status, Efault);
1705 user_write_access_end();
1708 user_write_access_end();
1712 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1715 struct wait_opts wo;
1716 struct pid *pid = NULL;
1720 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1721 __WNOTHREAD|__WCLONE|__WALL))
1724 /* -INT_MIN is not defined */
1725 if (upid == INT_MIN)
1730 else if (upid < 0) {
1731 type = PIDTYPE_PGID;
1732 pid = find_get_pid(-upid);
1733 } else if (upid == 0) {
1734 type = PIDTYPE_PGID;
1735 pid = get_task_pid(current, PIDTYPE_PGID);
1736 } else /* upid > 0 */ {
1738 pid = find_get_pid(upid);
1743 wo.wo_flags = options | WEXITED;
1749 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1755 int kernel_wait(pid_t pid, int *stat)
1757 struct wait_opts wo = {
1758 .wo_type = PIDTYPE_PID,
1759 .wo_pid = find_get_pid(pid),
1760 .wo_flags = WEXITED,
1765 if (ret > 0 && wo.wo_stat)
1771 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1772 int, options, struct rusage __user *, ru)
1775 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1778 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1784 #ifdef __ARCH_WANT_SYS_WAITPID
1787 * sys_waitpid() remains for compatibility. waitpid() should be
1788 * implemented by calling sys_wait4() from libc.a.
1790 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1792 return kernel_wait4(pid, stat_addr, options, NULL);
1797 #ifdef CONFIG_COMPAT
1798 COMPAT_SYSCALL_DEFINE4(wait4,
1800 compat_uint_t __user *, stat_addr,
1802 struct compat_rusage __user *, ru)
1805 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1807 if (ru && put_compat_rusage(&r, ru))
1813 COMPAT_SYSCALL_DEFINE5(waitid,
1814 int, which, compat_pid_t, pid,
1815 struct compat_siginfo __user *, infop, int, options,
1816 struct compat_rusage __user *, uru)
1819 struct waitid_info info = {.status = 0};
1820 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1826 /* kernel_waitid() overwrites everything in ru */
1827 if (COMPAT_USE_64BIT_TIME)
1828 err = copy_to_user(uru, &ru, sizeof(ru));
1830 err = put_compat_rusage(&ru, uru);
1839 if (!user_write_access_begin(infop, sizeof(*infop)))
1842 unsafe_put_user(signo, &infop->si_signo, Efault);
1843 unsafe_put_user(0, &infop->si_errno, Efault);
1844 unsafe_put_user(info.cause, &infop->si_code, Efault);
1845 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1846 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1847 unsafe_put_user(info.status, &infop->si_status, Efault);
1848 user_write_access_end();
1851 user_write_access_end();
1857 * thread_group_exited - check that a thread group has exited
1858 * @pid: tgid of thread group to be checked.
1860 * Test if the thread group represented by tgid has exited (all
1861 * threads are zombies, dead or completely gone).
1863 * Return: true if the thread group has exited. false otherwise.
1865 bool thread_group_exited(struct pid *pid)
1867 struct task_struct *task;
1871 task = pid_task(pid, PIDTYPE_PID);
1873 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1878 EXPORT_SYMBOL(thread_group_exited);
1880 __weak void abort(void)
1884 /* if that doesn't kill us, halt */
1885 panic("Oops failed to kill thread");
1887 EXPORT_SYMBOL(abort);