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/sysfs.h>
68 #include <linux/uaccess.h>
69 #include <asm/unistd.h>
70 #include <asm/pgtable.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);
157 * This can only happen if the caller is de_thread().
158 * FIXME: this is the temporary hack, we should teach
159 * posix-cpu-timers to handle this case correctly.
161 if (unlikely(has_group_leader_pid(tsk)))
162 posix_cpu_timers_exit_group(tsk);
171 * If there is any task waiting for the group exit
174 if (sig->notify_count > 0 && !--sig->notify_count)
175 wake_up_process(sig->group_exit_task);
177 if (tsk == sig->curr_target)
178 sig->curr_target = next_thread(tsk);
181 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
182 sizeof(unsigned long long));
185 * Accumulate here the counters for all threads as they die. We could
186 * skip the group leader because it is the last user of signal_struct,
187 * but we want to avoid the race with thread_group_cputime() which can
188 * see the empty ->thread_head list.
190 task_cputime(tsk, &utime, &stime);
191 write_seqlock(&sig->stats_lock);
194 sig->gtime += task_gtime(tsk);
195 sig->min_flt += tsk->min_flt;
196 sig->maj_flt += tsk->maj_flt;
197 sig->nvcsw += tsk->nvcsw;
198 sig->nivcsw += tsk->nivcsw;
199 sig->inblock += task_io_get_inblock(tsk);
200 sig->oublock += task_io_get_oublock(tsk);
201 task_io_accounting_add(&sig->ioac, &tsk->ioac);
202 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
204 __unhash_process(tsk, group_dead);
205 write_sequnlock(&sig->stats_lock);
208 * Do this under ->siglock, we can race with another thread
209 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
211 flush_sigqueue(&tsk->pending);
213 spin_unlock(&sighand->siglock);
215 __cleanup_sighand(sighand);
216 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
218 flush_sigqueue(&sig->shared_pending);
223 static void delayed_put_task_struct(struct rcu_head *rhp)
225 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
227 perf_event_delayed_put(tsk);
228 trace_sched_process_free(tsk);
229 put_task_struct(tsk);
232 void put_task_struct_rcu_user(struct task_struct *task)
234 if (refcount_dec_and_test(&task->rcu_users))
235 call_rcu(&task->rcu, delayed_put_task_struct);
238 void release_task(struct task_struct *p)
240 struct task_struct *leader;
243 /* don't need to get the RCU readlock here - the process is dead and
244 * can't be modifying its own credentials. But shut RCU-lockdep up */
246 atomic_dec(&__task_cred(p)->user->processes);
252 write_lock_irq(&tasklist_lock);
253 ptrace_release_task(p);
257 * If we are the last non-leader member of the thread
258 * group, and the leader is zombie, then notify the
259 * group leader's parent process. (if it wants notification.)
262 leader = p->group_leader;
263 if (leader != p && thread_group_empty(leader)
264 && leader->exit_state == EXIT_ZOMBIE) {
266 * If we were the last child thread and the leader has
267 * exited already, and the leader's parent ignores SIGCHLD,
268 * then we are the one who should release the leader.
270 zap_leader = do_notify_parent(leader, leader->exit_signal);
272 leader->exit_state = EXIT_DEAD;
275 write_unlock_irq(&tasklist_lock);
277 put_task_struct_rcu_user(p);
280 if (unlikely(zap_leader))
284 void rcuwait_wake_up(struct rcuwait *w)
286 struct task_struct *task;
291 * Order condition vs @task, such that everything prior to the load
292 * of @task is visible. This is the condition as to why the user called
293 * rcuwait_trywake() in the first place. Pairs with set_current_state()
294 * barrier (A) in rcuwait_wait_event().
297 * [S] tsk = current [S] cond = true
303 task = rcu_dereference(w->task);
305 wake_up_process(task);
310 * Determine if a process group is "orphaned", according to the POSIX
311 * definition in 2.2.2.52. Orphaned process groups are not to be affected
312 * by terminal-generated stop signals. Newly orphaned process groups are
313 * to receive a SIGHUP and a SIGCONT.
315 * "I ask you, have you ever known what it is to be an orphan?"
317 static int will_become_orphaned_pgrp(struct pid *pgrp,
318 struct task_struct *ignored_task)
320 struct task_struct *p;
322 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
323 if ((p == ignored_task) ||
324 (p->exit_state && thread_group_empty(p)) ||
325 is_global_init(p->real_parent))
328 if (task_pgrp(p->real_parent) != pgrp &&
329 task_session(p->real_parent) == task_session(p))
331 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
336 int is_current_pgrp_orphaned(void)
340 read_lock(&tasklist_lock);
341 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
342 read_unlock(&tasklist_lock);
347 static bool has_stopped_jobs(struct pid *pgrp)
349 struct task_struct *p;
351 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
352 if (p->signal->flags & SIGNAL_STOP_STOPPED)
354 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
360 * Check to see if any process groups have become orphaned as
361 * a result of our exiting, and if they have any stopped jobs,
362 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
365 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
367 struct pid *pgrp = task_pgrp(tsk);
368 struct task_struct *ignored_task = tsk;
371 /* exit: our father is in a different pgrp than
372 * we are and we were the only connection outside.
374 parent = tsk->real_parent;
376 /* reparent: our child is in a different pgrp than
377 * we are, and it was the only connection outside.
381 if (task_pgrp(parent) != pgrp &&
382 task_session(parent) == task_session(tsk) &&
383 will_become_orphaned_pgrp(pgrp, ignored_task) &&
384 has_stopped_jobs(pgrp)) {
385 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
386 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
392 * A task is exiting. If it owned this mm, find a new owner for the mm.
394 void mm_update_next_owner(struct mm_struct *mm)
396 struct task_struct *c, *g, *p = current;
400 * If the exiting or execing task is not the owner, it's
401 * someone else's problem.
406 * The current owner is exiting/execing and there are no other
407 * candidates. Do not leave the mm pointing to a possibly
408 * freed task structure.
410 if (atomic_read(&mm->mm_users) <= 1) {
411 WRITE_ONCE(mm->owner, NULL);
415 read_lock(&tasklist_lock);
417 * Search in the children
419 list_for_each_entry(c, &p->children, sibling) {
421 goto assign_new_owner;
425 * Search in the siblings
427 list_for_each_entry(c, &p->real_parent->children, sibling) {
429 goto assign_new_owner;
433 * Search through everything else, we should not get here often.
435 for_each_process(g) {
436 if (g->flags & PF_KTHREAD)
438 for_each_thread(g, c) {
440 goto assign_new_owner;
445 read_unlock(&tasklist_lock);
447 * We found no owner yet mm_users > 1: this implies that we are
448 * most likely racing with swapoff (try_to_unuse()) or /proc or
449 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
451 WRITE_ONCE(mm->owner, NULL);
458 * The task_lock protects c->mm from changing.
459 * We always want mm->owner->mm == mm
463 * Delay read_unlock() till we have the task_lock()
464 * to ensure that c does not slip away underneath us
466 read_unlock(&tasklist_lock);
472 WRITE_ONCE(mm->owner, c);
476 #endif /* CONFIG_MEMCG */
479 * Turn us into a lazy TLB process if we
482 static void exit_mm(void)
484 struct mm_struct *mm = current->mm;
485 struct core_state *core_state;
487 exit_mm_release(current, mm);
492 * Serialize with any possible pending coredump.
493 * We must hold mmap_sem around checking core_state
494 * and clearing tsk->mm. The core-inducing thread
495 * will increment ->nr_threads for each thread in the
496 * group with ->mm != NULL.
498 down_read(&mm->mmap_sem);
499 core_state = mm->core_state;
501 struct core_thread self;
503 up_read(&mm->mmap_sem);
506 if (self.task->flags & PF_SIGNALED)
507 self.next = xchg(&core_state->dumper.next, &self);
511 * Implies mb(), the result of xchg() must be visible
512 * to core_state->dumper.
514 if (atomic_dec_and_test(&core_state->nr_threads))
515 complete(&core_state->startup);
518 set_current_state(TASK_UNINTERRUPTIBLE);
519 if (!self.task) /* see coredump_finish() */
521 freezable_schedule();
523 __set_current_state(TASK_RUNNING);
524 down_read(&mm->mmap_sem);
527 BUG_ON(mm != current->active_mm);
528 /* more a memory barrier than a real lock */
531 up_read(&mm->mmap_sem);
532 enter_lazy_tlb(mm, current);
533 task_unlock(current);
534 mm_update_next_owner(mm);
536 if (test_thread_flag(TIF_MEMDIE))
540 static struct task_struct *find_alive_thread(struct task_struct *p)
542 struct task_struct *t;
544 for_each_thread(p, t) {
545 if (!(t->flags & PF_EXITING))
551 static struct task_struct *find_child_reaper(struct task_struct *father,
552 struct list_head *dead)
553 __releases(&tasklist_lock)
554 __acquires(&tasklist_lock)
556 struct pid_namespace *pid_ns = task_active_pid_ns(father);
557 struct task_struct *reaper = pid_ns->child_reaper;
558 struct task_struct *p, *n;
560 if (likely(reaper != father))
563 reaper = find_alive_thread(father);
565 pid_ns->child_reaper = reaper;
569 write_unlock_irq(&tasklist_lock);
571 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
572 list_del_init(&p->ptrace_entry);
576 zap_pid_ns_processes(pid_ns);
577 write_lock_irq(&tasklist_lock);
583 * When we die, we re-parent all our children, and try to:
584 * 1. give them to another thread in our thread group, if such a member exists
585 * 2. give it to the first ancestor process which prctl'd itself as a
586 * child_subreaper for its children (like a service manager)
587 * 3. give it to the init process (PID 1) in our pid namespace
589 static struct task_struct *find_new_reaper(struct task_struct *father,
590 struct task_struct *child_reaper)
592 struct task_struct *thread, *reaper;
594 thread = find_alive_thread(father);
598 if (father->signal->has_child_subreaper) {
599 unsigned int ns_level = task_pid(father)->level;
601 * Find the first ->is_child_subreaper ancestor in our pid_ns.
602 * We can't check reaper != child_reaper to ensure we do not
603 * cross the namespaces, the exiting parent could be injected
604 * by setns() + fork().
605 * We check pid->level, this is slightly more efficient than
606 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
608 for (reaper = father->real_parent;
609 task_pid(reaper)->level == ns_level;
610 reaper = reaper->real_parent) {
611 if (reaper == &init_task)
613 if (!reaper->signal->is_child_subreaper)
615 thread = find_alive_thread(reaper);
625 * Any that need to be release_task'd are put on the @dead list.
627 static void reparent_leader(struct task_struct *father, struct task_struct *p,
628 struct list_head *dead)
630 if (unlikely(p->exit_state == EXIT_DEAD))
633 /* We don't want people slaying init. */
634 p->exit_signal = SIGCHLD;
636 /* If it has exited notify the new parent about this child's death. */
638 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
639 if (do_notify_parent(p, p->exit_signal)) {
640 p->exit_state = EXIT_DEAD;
641 list_add(&p->ptrace_entry, dead);
645 kill_orphaned_pgrp(p, father);
649 * This does two things:
651 * A. Make init inherit all the child processes
652 * B. Check to see if any process groups have become orphaned
653 * as a result of our exiting, and if they have any stopped
654 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
656 static void forget_original_parent(struct task_struct *father,
657 struct list_head *dead)
659 struct task_struct *p, *t, *reaper;
661 if (unlikely(!list_empty(&father->ptraced)))
662 exit_ptrace(father, dead);
664 /* Can drop and reacquire tasklist_lock */
665 reaper = find_child_reaper(father, dead);
666 if (list_empty(&father->children))
669 reaper = find_new_reaper(father, reaper);
670 list_for_each_entry(p, &father->children, sibling) {
671 for_each_thread(p, t) {
672 t->real_parent = reaper;
673 BUG_ON((!t->ptrace) != (t->parent == father));
674 if (likely(!t->ptrace))
675 t->parent = t->real_parent;
676 if (t->pdeath_signal)
677 group_send_sig_info(t->pdeath_signal,
682 * If this is a threaded reparent there is no need to
683 * notify anyone anything has happened.
685 if (!same_thread_group(reaper, father))
686 reparent_leader(father, p, dead);
688 list_splice_tail_init(&father->children, &reaper->children);
692 * Send signals to all our closest relatives so that they know
693 * to properly mourn us..
695 static void exit_notify(struct task_struct *tsk, int group_dead)
698 struct task_struct *p, *n;
701 write_lock_irq(&tasklist_lock);
702 forget_original_parent(tsk, &dead);
705 kill_orphaned_pgrp(tsk->group_leader, NULL);
707 tsk->exit_state = EXIT_ZOMBIE;
708 if (unlikely(tsk->ptrace)) {
709 int sig = thread_group_leader(tsk) &&
710 thread_group_empty(tsk) &&
711 !ptrace_reparented(tsk) ?
712 tsk->exit_signal : SIGCHLD;
713 autoreap = do_notify_parent(tsk, sig);
714 } else if (thread_group_leader(tsk)) {
715 autoreap = thread_group_empty(tsk) &&
716 do_notify_parent(tsk, tsk->exit_signal);
722 tsk->exit_state = EXIT_DEAD;
723 list_add(&tsk->ptrace_entry, &dead);
726 /* mt-exec, de_thread() is waiting for group leader */
727 if (unlikely(tsk->signal->notify_count < 0))
728 wake_up_process(tsk->signal->group_exit_task);
729 write_unlock_irq(&tasklist_lock);
731 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
732 list_del_init(&p->ptrace_entry);
737 #ifdef CONFIG_DEBUG_STACK_USAGE
738 static void check_stack_usage(void)
740 static DEFINE_SPINLOCK(low_water_lock);
741 static int lowest_to_date = THREAD_SIZE;
744 free = stack_not_used(current);
746 if (free >= lowest_to_date)
749 spin_lock(&low_water_lock);
750 if (free < lowest_to_date) {
751 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
752 current->comm, task_pid_nr(current), free);
753 lowest_to_date = free;
755 spin_unlock(&low_water_lock);
758 static inline void check_stack_usage(void) {}
761 void __noreturn do_exit(long code)
763 struct task_struct *tsk = current;
767 * We can get here from a kernel oops, sometimes with preemption off.
768 * Start by checking for critical errors.
769 * Then fix up important state like USER_DS and preemption.
770 * Then do everything else.
773 WARN_ON(blk_needs_flush_plug(tsk));
775 if (unlikely(in_interrupt()))
776 panic("Aiee, killing interrupt handler!");
777 if (unlikely(!tsk->pid))
778 panic("Attempted to kill the idle task!");
781 * If do_exit is called because this processes oopsed, it's possible
782 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
783 * continuing. Amongst other possible reasons, this is to prevent
784 * mm_release()->clear_child_tid() from writing to a user-controlled
789 if (unlikely(in_atomic())) {
790 pr_info("note: %s[%d] exited with preempt_count %d\n",
791 current->comm, task_pid_nr(current),
793 preempt_count_set(PREEMPT_ENABLED);
796 profile_task_exit(tsk);
799 ptrace_event(PTRACE_EVENT_EXIT, code);
801 validate_creds_for_do_exit(tsk);
804 * We're taking recursive faults here in do_exit. Safest is to just
805 * leave this task alone and wait for reboot.
807 if (unlikely(tsk->flags & PF_EXITING)) {
808 pr_alert("Fixing recursive fault but reboot is needed!\n");
809 futex_exit_recursive(tsk);
810 set_current_state(TASK_UNINTERRUPTIBLE);
814 exit_signals(tsk); /* sets PF_EXITING */
816 /* sync mm's RSS info before statistics gathering */
818 sync_mm_rss(tsk->mm);
819 acct_update_integrals(tsk);
820 group_dead = atomic_dec_and_test(&tsk->signal->live);
823 * If the last thread of global init has exited, panic
824 * immediately to get a useable coredump.
826 if (unlikely(is_global_init(tsk)))
827 panic("Attempted to kill init! exitcode=0x%08x\n",
828 tsk->signal->group_exit_code ?: (int)code);
830 #ifdef CONFIG_POSIX_TIMERS
831 hrtimer_cancel(&tsk->signal->real_timer);
832 exit_itimers(tsk->signal);
835 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
837 acct_collect(code, group_dead);
842 tsk->exit_code = code;
843 taskstats_exit(tsk, group_dead);
849 trace_sched_process_exit(tsk);
856 disassociate_ctty(1);
857 exit_task_namespaces(tsk);
863 * Flush inherited counters to the parent - before the parent
864 * gets woken up by child-exit notifications.
866 * because of cgroup mode, must be called before cgroup_exit()
868 perf_event_exit_task(tsk);
870 sched_autogroup_exit_task(tsk);
874 * FIXME: do that only when needed, using sched_exit tracepoint
876 flush_ptrace_hw_breakpoint(tsk);
878 exit_tasks_rcu_start();
879 exit_notify(tsk, group_dead);
880 proc_exit_connector(tsk);
881 mpol_put_task_policy(tsk);
883 if (unlikely(current->pi_state_cache))
884 kfree(current->pi_state_cache);
887 * Make sure we are holding no locks:
889 debug_check_no_locks_held();
892 exit_io_context(tsk);
894 if (tsk->splice_pipe)
895 free_pipe_info(tsk->splice_pipe);
897 if (tsk->task_frag.page)
898 put_page(tsk->task_frag.page);
900 validate_creds_for_do_exit(tsk);
905 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
907 exit_tasks_rcu_finish();
909 lockdep_free_task(tsk);
912 EXPORT_SYMBOL_GPL(do_exit);
914 void __noreturn make_task_dead(int signr)
917 * Take the task off the cpu after something catastrophic has
923 * Every time the system oopses, if the oops happens while a reference
924 * to an object was held, the reference leaks.
925 * If the oops doesn't also leak memory, repeated oopsing can cause
926 * reference counters to wrap around (if they're not using refcount_t).
927 * This means that repeated oopsing can make unexploitable-looking bugs
928 * exploitable through repeated oopsing.
929 * To make sure this can't happen, place an upper bound on how often the
930 * kernel may oops without panic().
932 limit = READ_ONCE(oops_limit);
933 if (atomic_inc_return(&oops_count) >= limit && limit)
934 panic("Oopsed too often (kernel.oops_limit is %d)", limit);
939 void complete_and_exit(struct completion *comp, long code)
946 EXPORT_SYMBOL(complete_and_exit);
948 SYSCALL_DEFINE1(exit, int, error_code)
950 do_exit((error_code&0xff)<<8);
954 * Take down every thread in the group. This is called by fatal signals
955 * as well as by sys_exit_group (below).
958 do_group_exit(int exit_code)
960 struct signal_struct *sig = current->signal;
962 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
964 if (signal_group_exit(sig))
965 exit_code = sig->group_exit_code;
966 else if (!thread_group_empty(current)) {
967 struct sighand_struct *const sighand = current->sighand;
969 spin_lock_irq(&sighand->siglock);
970 if (signal_group_exit(sig))
971 /* Another thread got here before we took the lock. */
972 exit_code = sig->group_exit_code;
974 sig->group_exit_code = exit_code;
975 sig->flags = SIGNAL_GROUP_EXIT;
976 zap_other_threads(current);
978 spin_unlock_irq(&sighand->siglock);
986 * this kills every thread in the thread group. Note that any externally
987 * wait4()-ing process will get the correct exit code - even if this
988 * thread is not the thread group leader.
990 SYSCALL_DEFINE1(exit_group, int, error_code)
992 do_group_exit((error_code & 0xff) << 8);
1005 enum pid_type wo_type;
1009 struct waitid_info *wo_info;
1011 struct rusage *wo_rusage;
1013 wait_queue_entry_t child_wait;
1017 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1019 return wo->wo_type == PIDTYPE_MAX ||
1020 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1024 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1026 if (!eligible_pid(wo, p))
1030 * Wait for all children (clone and not) if __WALL is set or
1031 * if it is traced by us.
1033 if (ptrace || (wo->wo_flags & __WALL))
1037 * Otherwise, wait for clone children *only* if __WCLONE is set;
1038 * otherwise, wait for non-clone children *only*.
1040 * Note: a "clone" child here is one that reports to its parent
1041 * using a signal other than SIGCHLD, or a non-leader thread which
1042 * we can only see if it is traced by us.
1044 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1051 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1052 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1053 * the lock and this task is uninteresting. If we return nonzero, we have
1054 * released the lock and the system call should return.
1056 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1059 pid_t pid = task_pid_vnr(p);
1060 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1061 struct waitid_info *infop;
1063 if (!likely(wo->wo_flags & WEXITED))
1066 if (unlikely(wo->wo_flags & WNOWAIT)) {
1067 status = p->exit_code;
1069 read_unlock(&tasklist_lock);
1070 sched_annotate_sleep();
1072 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1077 * Move the task's state to DEAD/TRACE, only one thread can do this.
1079 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1080 EXIT_TRACE : EXIT_DEAD;
1081 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1084 * We own this thread, nobody else can reap it.
1086 read_unlock(&tasklist_lock);
1087 sched_annotate_sleep();
1090 * Check thread_group_leader() to exclude the traced sub-threads.
1092 if (state == EXIT_DEAD && thread_group_leader(p)) {
1093 struct signal_struct *sig = p->signal;
1094 struct signal_struct *psig = current->signal;
1095 unsigned long maxrss;
1096 u64 tgutime, tgstime;
1099 * The resource counters for the group leader are in its
1100 * own task_struct. Those for dead threads in the group
1101 * are in its signal_struct, as are those for the child
1102 * processes it has previously reaped. All these
1103 * accumulate in the parent's signal_struct c* fields.
1105 * We don't bother to take a lock here to protect these
1106 * p->signal fields because the whole thread group is dead
1107 * and nobody can change them.
1109 * psig->stats_lock also protects us from our sub-theads
1110 * which can reap other children at the same time. Until
1111 * we change k_getrusage()-like users to rely on this lock
1112 * we have to take ->siglock as well.
1114 * We use thread_group_cputime_adjusted() to get times for
1115 * the thread group, which consolidates times for all threads
1116 * in the group including the group leader.
1118 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1119 spin_lock_irq(¤t->sighand->siglock);
1120 write_seqlock(&psig->stats_lock);
1121 psig->cutime += tgutime + sig->cutime;
1122 psig->cstime += tgstime + sig->cstime;
1123 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1125 p->min_flt + sig->min_flt + sig->cmin_flt;
1127 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1129 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1131 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1133 task_io_get_inblock(p) +
1134 sig->inblock + sig->cinblock;
1136 task_io_get_oublock(p) +
1137 sig->oublock + sig->coublock;
1138 maxrss = max(sig->maxrss, sig->cmaxrss);
1139 if (psig->cmaxrss < maxrss)
1140 psig->cmaxrss = maxrss;
1141 task_io_accounting_add(&psig->ioac, &p->ioac);
1142 task_io_accounting_add(&psig->ioac, &sig->ioac);
1143 write_sequnlock(&psig->stats_lock);
1144 spin_unlock_irq(¤t->sighand->siglock);
1148 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1149 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1150 ? p->signal->group_exit_code : p->exit_code;
1151 wo->wo_stat = status;
1153 if (state == EXIT_TRACE) {
1154 write_lock_irq(&tasklist_lock);
1155 /* We dropped tasklist, ptracer could die and untrace */
1158 /* If parent wants a zombie, don't release it now */
1159 state = EXIT_ZOMBIE;
1160 if (do_notify_parent(p, p->exit_signal))
1162 p->exit_state = state;
1163 write_unlock_irq(&tasklist_lock);
1165 if (state == EXIT_DEAD)
1169 infop = wo->wo_info;
1171 if ((status & 0x7f) == 0) {
1172 infop->cause = CLD_EXITED;
1173 infop->status = status >> 8;
1175 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1176 infop->status = status & 0x7f;
1185 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1188 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1189 return &p->exit_code;
1191 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1192 return &p->signal->group_exit_code;
1198 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1200 * @ptrace: is the wait for ptrace
1201 * @p: task to wait for
1203 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1206 * read_lock(&tasklist_lock), which is released if return value is
1207 * non-zero. Also, grabs and releases @p->sighand->siglock.
1210 * 0 if wait condition didn't exist and search for other wait conditions
1211 * should continue. Non-zero return, -errno on failure and @p's pid on
1212 * success, implies that tasklist_lock is released and wait condition
1213 * search should terminate.
1215 static int wait_task_stopped(struct wait_opts *wo,
1216 int ptrace, struct task_struct *p)
1218 struct waitid_info *infop;
1219 int exit_code, *p_code, why;
1220 uid_t uid = 0; /* unneeded, required by compiler */
1224 * Traditionally we see ptrace'd stopped tasks regardless of options.
1226 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1229 if (!task_stopped_code(p, ptrace))
1233 spin_lock_irq(&p->sighand->siglock);
1235 p_code = task_stopped_code(p, ptrace);
1236 if (unlikely(!p_code))
1239 exit_code = *p_code;
1243 if (!unlikely(wo->wo_flags & WNOWAIT))
1246 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1248 spin_unlock_irq(&p->sighand->siglock);
1253 * Now we are pretty sure this task is interesting.
1254 * Make sure it doesn't get reaped out from under us while we
1255 * give up the lock and then examine it below. We don't want to
1256 * keep holding onto the tasklist_lock while we call getrusage and
1257 * possibly take page faults for user memory.
1260 pid = task_pid_vnr(p);
1261 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1262 read_unlock(&tasklist_lock);
1263 sched_annotate_sleep();
1265 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1268 if (likely(!(wo->wo_flags & WNOWAIT)))
1269 wo->wo_stat = (exit_code << 8) | 0x7f;
1271 infop = wo->wo_info;
1274 infop->status = exit_code;
1282 * Handle do_wait work for one task in a live, non-stopped state.
1283 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1284 * the lock and this task is uninteresting. If we return nonzero, we have
1285 * released the lock and the system call should return.
1287 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1289 struct waitid_info *infop;
1293 if (!unlikely(wo->wo_flags & WCONTINUED))
1296 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1299 spin_lock_irq(&p->sighand->siglock);
1300 /* Re-check with the lock held. */
1301 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1302 spin_unlock_irq(&p->sighand->siglock);
1305 if (!unlikely(wo->wo_flags & WNOWAIT))
1306 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1307 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1308 spin_unlock_irq(&p->sighand->siglock);
1310 pid = task_pid_vnr(p);
1312 read_unlock(&tasklist_lock);
1313 sched_annotate_sleep();
1315 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1318 infop = wo->wo_info;
1320 wo->wo_stat = 0xffff;
1322 infop->cause = CLD_CONTINUED;
1325 infop->status = SIGCONT;
1331 * Consider @p for a wait by @parent.
1333 * -ECHILD should be in ->notask_error before the first call.
1334 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1335 * Returns zero if the search for a child should continue;
1336 * then ->notask_error is 0 if @p is an eligible child,
1339 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1340 struct task_struct *p)
1343 * We can race with wait_task_zombie() from another thread.
1344 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1345 * can't confuse the checks below.
1347 int exit_state = READ_ONCE(p->exit_state);
1350 if (unlikely(exit_state == EXIT_DEAD))
1353 ret = eligible_child(wo, ptrace, p);
1357 if (unlikely(exit_state == EXIT_TRACE)) {
1359 * ptrace == 0 means we are the natural parent. In this case
1360 * we should clear notask_error, debugger will notify us.
1362 if (likely(!ptrace))
1363 wo->notask_error = 0;
1367 if (likely(!ptrace) && unlikely(p->ptrace)) {
1369 * If it is traced by its real parent's group, just pretend
1370 * the caller is ptrace_do_wait() and reap this child if it
1373 * This also hides group stop state from real parent; otherwise
1374 * a single stop can be reported twice as group and ptrace stop.
1375 * If a ptracer wants to distinguish these two events for its
1376 * own children it should create a separate process which takes
1377 * the role of real parent.
1379 if (!ptrace_reparented(p))
1384 if (exit_state == EXIT_ZOMBIE) {
1385 /* we don't reap group leaders with subthreads */
1386 if (!delay_group_leader(p)) {
1388 * A zombie ptracee is only visible to its ptracer.
1389 * Notification and reaping will be cascaded to the
1390 * real parent when the ptracer detaches.
1392 if (unlikely(ptrace) || likely(!p->ptrace))
1393 return wait_task_zombie(wo, p);
1397 * Allow access to stopped/continued state via zombie by
1398 * falling through. Clearing of notask_error is complex.
1402 * If WEXITED is set, notask_error should naturally be
1403 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1404 * so, if there are live subthreads, there are events to
1405 * wait for. If all subthreads are dead, it's still safe
1406 * to clear - this function will be called again in finite
1407 * amount time once all the subthreads are released and
1408 * will then return without clearing.
1412 * Stopped state is per-task and thus can't change once the
1413 * target task dies. Only continued and exited can happen.
1414 * Clear notask_error if WCONTINUED | WEXITED.
1416 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1417 wo->notask_error = 0;
1420 * @p is alive and it's gonna stop, continue or exit, so
1421 * there always is something to wait for.
1423 wo->notask_error = 0;
1427 * Wait for stopped. Depending on @ptrace, different stopped state
1428 * is used and the two don't interact with each other.
1430 ret = wait_task_stopped(wo, ptrace, p);
1435 * Wait for continued. There's only one continued state and the
1436 * ptracer can consume it which can confuse the real parent. Don't
1437 * use WCONTINUED from ptracer. You don't need or want it.
1439 return wait_task_continued(wo, p);
1443 * Do the work of do_wait() for one thread in the group, @tsk.
1445 * -ECHILD should be in ->notask_error before the first call.
1446 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1447 * Returns zero if the search for a child should continue; then
1448 * ->notask_error is 0 if there were any eligible children,
1451 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1453 struct task_struct *p;
1455 list_for_each_entry(p, &tsk->children, sibling) {
1456 int ret = wait_consider_task(wo, 0, p);
1465 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1467 struct task_struct *p;
1469 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1470 int ret = wait_consider_task(wo, 1, p);
1479 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1480 int sync, void *key)
1482 struct wait_opts *wo = container_of(wait, struct wait_opts,
1484 struct task_struct *p = key;
1486 if (!eligible_pid(wo, p))
1489 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1492 return default_wake_function(wait, mode, sync, key);
1495 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1497 __wake_up_sync_key(&parent->signal->wait_chldexit,
1498 TASK_INTERRUPTIBLE, 1, p);
1501 static long do_wait(struct wait_opts *wo)
1503 struct task_struct *tsk;
1506 trace_sched_process_wait(wo->wo_pid);
1508 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1509 wo->child_wait.private = current;
1510 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1513 * If there is nothing that can match our criteria, just get out.
1514 * We will clear ->notask_error to zero if we see any child that
1515 * might later match our criteria, even if we are not able to reap
1518 wo->notask_error = -ECHILD;
1519 if ((wo->wo_type < PIDTYPE_MAX) &&
1520 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1523 set_current_state(TASK_INTERRUPTIBLE);
1524 read_lock(&tasklist_lock);
1527 retval = do_wait_thread(wo, tsk);
1531 retval = ptrace_do_wait(wo, tsk);
1535 if (wo->wo_flags & __WNOTHREAD)
1537 } while_each_thread(current, tsk);
1538 read_unlock(&tasklist_lock);
1541 retval = wo->notask_error;
1542 if (!retval && !(wo->wo_flags & WNOHANG)) {
1543 retval = -ERESTARTSYS;
1544 if (!signal_pending(current)) {
1550 __set_current_state(TASK_RUNNING);
1551 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1555 static struct pid *pidfd_get_pid(unsigned int fd)
1562 return ERR_PTR(-EBADF);
1564 pid = pidfd_pid(f.file);
1572 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1573 int options, struct rusage *ru)
1575 struct wait_opts wo;
1576 struct pid *pid = NULL;
1580 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1581 __WNOTHREAD|__WCLONE|__WALL))
1583 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1595 pid = find_get_pid(upid);
1598 type = PIDTYPE_PGID;
1603 pid = find_get_pid(upid);
1605 pid = get_task_pid(current, PIDTYPE_PGID);
1612 pid = pidfd_get_pid(upid);
1614 return PTR_ERR(pid);
1622 wo.wo_flags = options;
1631 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1632 infop, int, options, struct rusage __user *, ru)
1635 struct waitid_info info = {.status = 0};
1636 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1642 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1648 if (!user_access_begin(infop, sizeof(*infop)))
1651 unsafe_put_user(signo, &infop->si_signo, Efault);
1652 unsafe_put_user(0, &infop->si_errno, Efault);
1653 unsafe_put_user(info.cause, &infop->si_code, Efault);
1654 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1655 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1656 unsafe_put_user(info.status, &infop->si_status, Efault);
1664 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1667 struct wait_opts wo;
1668 struct pid *pid = NULL;
1672 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1673 __WNOTHREAD|__WCLONE|__WALL))
1676 /* -INT_MIN is not defined */
1677 if (upid == INT_MIN)
1682 else if (upid < 0) {
1683 type = PIDTYPE_PGID;
1684 pid = find_get_pid(-upid);
1685 } else if (upid == 0) {
1686 type = PIDTYPE_PGID;
1687 pid = get_task_pid(current, PIDTYPE_PGID);
1688 } else /* upid > 0 */ {
1690 pid = find_get_pid(upid);
1695 wo.wo_flags = options | WEXITED;
1701 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1707 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1708 int, options, struct rusage __user *, ru)
1711 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1714 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1720 #ifdef __ARCH_WANT_SYS_WAITPID
1723 * sys_waitpid() remains for compatibility. waitpid() should be
1724 * implemented by calling sys_wait4() from libc.a.
1726 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1728 return kernel_wait4(pid, stat_addr, options, NULL);
1733 #ifdef CONFIG_COMPAT
1734 COMPAT_SYSCALL_DEFINE4(wait4,
1736 compat_uint_t __user *, stat_addr,
1738 struct compat_rusage __user *, ru)
1741 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1743 if (ru && put_compat_rusage(&r, ru))
1749 COMPAT_SYSCALL_DEFINE5(waitid,
1750 int, which, compat_pid_t, pid,
1751 struct compat_siginfo __user *, infop, int, options,
1752 struct compat_rusage __user *, uru)
1755 struct waitid_info info = {.status = 0};
1756 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1762 /* kernel_waitid() overwrites everything in ru */
1763 if (COMPAT_USE_64BIT_TIME)
1764 err = copy_to_user(uru, &ru, sizeof(ru));
1766 err = put_compat_rusage(&ru, uru);
1775 if (!user_access_begin(infop, sizeof(*infop)))
1778 unsafe_put_user(signo, &infop->si_signo, Efault);
1779 unsafe_put_user(0, &infop->si_errno, Efault);
1780 unsafe_put_user(info.cause, &infop->si_code, Efault);
1781 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1782 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1783 unsafe_put_user(info.status, &infop->si_status, Efault);
1792 __weak void abort(void)
1796 /* if that doesn't kill us, halt */
1797 panic("Oops failed to kill thread");
1799 EXPORT_SYMBOL(abort);