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>
67 #include <linux/uaccess.h>
68 #include <asm/unistd.h>
69 #include <asm/pgtable.h>
70 #include <asm/mmu_context.h>
72 static void __unhash_process(struct task_struct
*p
, bool group_dead
)
75 detach_pid(p
, PIDTYPE_PID
);
77 detach_pid(p
, PIDTYPE_TGID
);
78 detach_pid(p
, PIDTYPE_PGID
);
79 detach_pid(p
, PIDTYPE_SID
);
81 list_del_rcu(&p
->tasks
);
82 list_del_init(&p
->sibling
);
83 __this_cpu_dec(process_counts
);
85 list_del_rcu(&p
->thread_group
);
86 list_del_rcu(&p
->thread_node
);
90 * This function expects the tasklist_lock write-locked.
92 static void __exit_signal(struct task_struct
*tsk
)
94 struct signal_struct
*sig
= tsk
->signal
;
95 bool group_dead
= thread_group_leader(tsk
);
96 struct sighand_struct
*sighand
;
97 struct tty_struct
*uninitialized_var(tty
);
100 sighand
= rcu_dereference_check(tsk
->sighand
,
101 lockdep_tasklist_lock_is_held());
102 spin_lock(&sighand
->siglock
);
104 #ifdef CONFIG_POSIX_TIMERS
105 posix_cpu_timers_exit(tsk
);
107 posix_cpu_timers_exit_group(tsk
);
115 * If there is any task waiting for the group exit
118 if (sig
->notify_count
> 0 && !--sig
->notify_count
)
119 wake_up_process(sig
->group_exit_task
);
121 if (tsk
== sig
->curr_target
)
122 sig
->curr_target
= next_thread(tsk
);
125 add_device_randomness((const void*) &tsk
->se
.sum_exec_runtime
,
126 sizeof(unsigned long long));
129 * Accumulate here the counters for all threads as they die. We could
130 * skip the group leader because it is the last user of signal_struct,
131 * but we want to avoid the race with thread_group_cputime() which can
132 * see the empty ->thread_head list.
134 task_cputime(tsk
, &utime
, &stime
);
135 write_seqlock(&sig
->stats_lock
);
138 sig
->gtime
+= task_gtime(tsk
);
139 sig
->min_flt
+= tsk
->min_flt
;
140 sig
->maj_flt
+= tsk
->maj_flt
;
141 sig
->nvcsw
+= tsk
->nvcsw
;
142 sig
->nivcsw
+= tsk
->nivcsw
;
143 sig
->inblock
+= task_io_get_inblock(tsk
);
144 sig
->oublock
+= task_io_get_oublock(tsk
);
145 task_io_accounting_add(&sig
->ioac
, &tsk
->ioac
);
146 sig
->sum_sched_runtime
+= tsk
->se
.sum_exec_runtime
;
148 __unhash_process(tsk
, group_dead
);
149 write_sequnlock(&sig
->stats_lock
);
152 * Do this under ->siglock, we can race with another thread
153 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
155 flush_sigqueue(&tsk
->pending
);
157 spin_unlock(&sighand
->siglock
);
159 __cleanup_sighand(sighand
);
160 clear_tsk_thread_flag(tsk
, TIF_SIGPENDING
);
162 flush_sigqueue(&sig
->shared_pending
);
167 static void delayed_put_task_struct(struct rcu_head
*rhp
)
169 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
171 perf_event_delayed_put(tsk
);
172 trace_sched_process_free(tsk
);
173 put_task_struct(tsk
);
176 void put_task_struct_rcu_user(struct task_struct
*task
)
178 if (refcount_dec_and_test(&task
->rcu_users
))
179 call_rcu(&task
->rcu
, delayed_put_task_struct
);
182 void release_task(struct task_struct
*p
)
184 struct task_struct
*leader
;
187 /* don't need to get the RCU readlock here - the process is dead and
188 * can't be modifying its own credentials. But shut RCU-lockdep up */
190 atomic_dec(&__task_cred(p
)->user
->processes
);
196 write_lock_irq(&tasklist_lock
);
197 ptrace_release_task(p
);
201 * If we are the last non-leader member of the thread
202 * group, and the leader is zombie, then notify the
203 * group leader's parent process. (if it wants notification.)
206 leader
= p
->group_leader
;
207 if (leader
!= p
&& thread_group_empty(leader
)
208 && leader
->exit_state
== EXIT_ZOMBIE
) {
210 * If we were the last child thread and the leader has
211 * exited already, and the leader's parent ignores SIGCHLD,
212 * then we are the one who should release the leader.
214 zap_leader
= do_notify_parent(leader
, leader
->exit_signal
);
216 leader
->exit_state
= EXIT_DEAD
;
219 write_unlock_irq(&tasklist_lock
);
221 put_task_struct_rcu_user(p
);
224 if (unlikely(zap_leader
))
228 void rcuwait_wake_up(struct rcuwait
*w
)
230 struct task_struct
*task
;
235 * Order condition vs @task, such that everything prior to the load
236 * of @task is visible. This is the condition as to why the user called
237 * rcuwait_trywake() in the first place. Pairs with set_current_state()
238 * barrier (A) in rcuwait_wait_event().
241 * [S] tsk = current [S] cond = true
247 task
= rcu_dereference(w
->task
);
249 wake_up_process(task
);
252 EXPORT_SYMBOL_GPL(rcuwait_wake_up
);
255 * Determine if a process group is "orphaned", according to the POSIX
256 * definition in 2.2.2.52. Orphaned process groups are not to be affected
257 * by terminal-generated stop signals. Newly orphaned process groups are
258 * to receive a SIGHUP and a SIGCONT.
260 * "I ask you, have you ever known what it is to be an orphan?"
262 static int will_become_orphaned_pgrp(struct pid
*pgrp
,
263 struct task_struct
*ignored_task
)
265 struct task_struct
*p
;
267 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
268 if ((p
== ignored_task
) ||
269 (p
->exit_state
&& thread_group_empty(p
)) ||
270 is_global_init(p
->real_parent
))
273 if (task_pgrp(p
->real_parent
) != pgrp
&&
274 task_session(p
->real_parent
) == task_session(p
))
276 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
281 int is_current_pgrp_orphaned(void)
285 read_lock(&tasklist_lock
);
286 retval
= will_become_orphaned_pgrp(task_pgrp(current
), NULL
);
287 read_unlock(&tasklist_lock
);
292 static bool has_stopped_jobs(struct pid
*pgrp
)
294 struct task_struct
*p
;
296 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
297 if (p
->signal
->flags
& SIGNAL_STOP_STOPPED
)
299 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
305 * Check to see if any process groups have become orphaned as
306 * a result of our exiting, and if they have any stopped jobs,
307 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
310 kill_orphaned_pgrp(struct task_struct
*tsk
, struct task_struct
*parent
)
312 struct pid
*pgrp
= task_pgrp(tsk
);
313 struct task_struct
*ignored_task
= tsk
;
316 /* exit: our father is in a different pgrp than
317 * we are and we were the only connection outside.
319 parent
= tsk
->real_parent
;
321 /* reparent: our child is in a different pgrp than
322 * we are, and it was the only connection outside.
326 if (task_pgrp(parent
) != pgrp
&&
327 task_session(parent
) == task_session(tsk
) &&
328 will_become_orphaned_pgrp(pgrp
, ignored_task
) &&
329 has_stopped_jobs(pgrp
)) {
330 __kill_pgrp_info(SIGHUP
, SEND_SIG_PRIV
, pgrp
);
331 __kill_pgrp_info(SIGCONT
, SEND_SIG_PRIV
, pgrp
);
337 * A task is exiting. If it owned this mm, find a new owner for the mm.
339 void mm_update_next_owner(struct mm_struct
*mm
)
341 struct task_struct
*c
, *g
, *p
= current
;
345 * If the exiting or execing task is not the owner, it's
346 * someone else's problem.
351 * The current owner is exiting/execing and there are no other
352 * candidates. Do not leave the mm pointing to a possibly
353 * freed task structure.
355 if (atomic_read(&mm
->mm_users
) <= 1) {
356 WRITE_ONCE(mm
->owner
, NULL
);
360 read_lock(&tasklist_lock
);
362 * Search in the children
364 list_for_each_entry(c
, &p
->children
, sibling
) {
366 goto assign_new_owner
;
370 * Search in the siblings
372 list_for_each_entry(c
, &p
->real_parent
->children
, sibling
) {
374 goto assign_new_owner
;
378 * Search through everything else, we should not get here often.
380 for_each_process(g
) {
381 if (g
->flags
& PF_KTHREAD
)
383 for_each_thread(g
, c
) {
385 goto assign_new_owner
;
390 read_unlock(&tasklist_lock
);
392 * We found no owner yet mm_users > 1: this implies that we are
393 * most likely racing with swapoff (try_to_unuse()) or /proc or
394 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
396 WRITE_ONCE(mm
->owner
, NULL
);
403 * The task_lock protects c->mm from changing.
404 * We always want mm->owner->mm == mm
408 * Delay read_unlock() till we have the task_lock()
409 * to ensure that c does not slip away underneath us
411 read_unlock(&tasklist_lock
);
417 WRITE_ONCE(mm
->owner
, c
);
421 #endif /* CONFIG_MEMCG */
424 * Turn us into a lazy TLB process if we
427 static void exit_mm(void)
429 struct mm_struct
*mm
= current
->mm
;
430 struct core_state
*core_state
;
432 exit_mm_release(current
, mm
);
437 * Serialize with any possible pending coredump.
438 * We must hold mmap_sem around checking core_state
439 * and clearing tsk->mm. The core-inducing thread
440 * will increment ->nr_threads for each thread in the
441 * group with ->mm != NULL.
443 down_read(&mm
->mmap_sem
);
444 core_state
= mm
->core_state
;
446 struct core_thread self
;
448 up_read(&mm
->mmap_sem
);
451 self
.next
= xchg(&core_state
->dumper
.next
, &self
);
453 * Implies mb(), the result of xchg() must be visible
454 * to core_state->dumper.
456 if (atomic_dec_and_test(&core_state
->nr_threads
))
457 complete(&core_state
->startup
);
460 set_current_state(TASK_UNINTERRUPTIBLE
);
461 if (!self
.task
) /* see coredump_finish() */
463 freezable_schedule();
465 __set_current_state(TASK_RUNNING
);
466 down_read(&mm
->mmap_sem
);
469 BUG_ON(mm
!= current
->active_mm
);
470 /* more a memory barrier than a real lock */
473 up_read(&mm
->mmap_sem
);
474 enter_lazy_tlb(mm
, current
);
475 task_unlock(current
);
476 mm_update_next_owner(mm
);
478 if (test_thread_flag(TIF_MEMDIE
))
482 static struct task_struct
*find_alive_thread(struct task_struct
*p
)
484 struct task_struct
*t
;
486 for_each_thread(p
, t
) {
487 if (!(t
->flags
& PF_EXITING
))
493 static struct task_struct
*find_child_reaper(struct task_struct
*father
,
494 struct list_head
*dead
)
495 __releases(&tasklist_lock
)
496 __acquires(&tasklist_lock
)
498 struct pid_namespace
*pid_ns
= task_active_pid_ns(father
);
499 struct task_struct
*reaper
= pid_ns
->child_reaper
;
500 struct task_struct
*p
, *n
;
502 if (likely(reaper
!= father
))
505 reaper
= find_alive_thread(father
);
507 pid_ns
->child_reaper
= reaper
;
511 write_unlock_irq(&tasklist_lock
);
513 list_for_each_entry_safe(p
, n
, dead
, ptrace_entry
) {
514 list_del_init(&p
->ptrace_entry
);
518 zap_pid_ns_processes(pid_ns
);
519 write_lock_irq(&tasklist_lock
);
525 * When we die, we re-parent all our children, and try to:
526 * 1. give them to another thread in our thread group, if such a member exists
527 * 2. give it to the first ancestor process which prctl'd itself as a
528 * child_subreaper for its children (like a service manager)
529 * 3. give it to the init process (PID 1) in our pid namespace
531 static struct task_struct
*find_new_reaper(struct task_struct
*father
,
532 struct task_struct
*child_reaper
)
534 struct task_struct
*thread
, *reaper
;
536 thread
= find_alive_thread(father
);
540 if (father
->signal
->has_child_subreaper
) {
541 unsigned int ns_level
= task_pid(father
)->level
;
543 * Find the first ->is_child_subreaper ancestor in our pid_ns.
544 * We can't check reaper != child_reaper to ensure we do not
545 * cross the namespaces, the exiting parent could be injected
546 * by setns() + fork().
547 * We check pid->level, this is slightly more efficient than
548 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
550 for (reaper
= father
->real_parent
;
551 task_pid(reaper
)->level
== ns_level
;
552 reaper
= reaper
->real_parent
) {
553 if (reaper
== &init_task
)
555 if (!reaper
->signal
->is_child_subreaper
)
557 thread
= find_alive_thread(reaper
);
567 * Any that need to be release_task'd are put on the @dead list.
569 static void reparent_leader(struct task_struct
*father
, struct task_struct
*p
,
570 struct list_head
*dead
)
572 if (unlikely(p
->exit_state
== EXIT_DEAD
))
575 /* We don't want people slaying init. */
576 p
->exit_signal
= SIGCHLD
;
578 /* If it has exited notify the new parent about this child's death. */
580 p
->exit_state
== EXIT_ZOMBIE
&& thread_group_empty(p
)) {
581 if (do_notify_parent(p
, p
->exit_signal
)) {
582 p
->exit_state
= EXIT_DEAD
;
583 list_add(&p
->ptrace_entry
, dead
);
587 kill_orphaned_pgrp(p
, father
);
591 * This does two things:
593 * A. Make init inherit all the child processes
594 * B. Check to see if any process groups have become orphaned
595 * as a result of our exiting, and if they have any stopped
596 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
598 static void forget_original_parent(struct task_struct
*father
,
599 struct list_head
*dead
)
601 struct task_struct
*p
, *t
, *reaper
;
603 if (unlikely(!list_empty(&father
->ptraced
)))
604 exit_ptrace(father
, dead
);
606 /* Can drop and reacquire tasklist_lock */
607 reaper
= find_child_reaper(father
, dead
);
608 if (list_empty(&father
->children
))
611 reaper
= find_new_reaper(father
, reaper
);
612 list_for_each_entry(p
, &father
->children
, sibling
) {
613 for_each_thread(p
, t
) {
614 RCU_INIT_POINTER(t
->real_parent
, reaper
);
615 BUG_ON((!t
->ptrace
) != (rcu_access_pointer(t
->parent
) == father
));
616 if (likely(!t
->ptrace
))
617 t
->parent
= t
->real_parent
;
618 if (t
->pdeath_signal
)
619 group_send_sig_info(t
->pdeath_signal
,
624 * If this is a threaded reparent there is no need to
625 * notify anyone anything has happened.
627 if (!same_thread_group(reaper
, father
))
628 reparent_leader(father
, p
, dead
);
630 list_splice_tail_init(&father
->children
, &reaper
->children
);
634 * Send signals to all our closest relatives so that they know
635 * to properly mourn us..
637 static void exit_notify(struct task_struct
*tsk
, int group_dead
)
640 struct task_struct
*p
, *n
;
643 write_lock_irq(&tasklist_lock
);
644 forget_original_parent(tsk
, &dead
);
647 kill_orphaned_pgrp(tsk
->group_leader
, NULL
);
649 tsk
->exit_state
= EXIT_ZOMBIE
;
650 if (unlikely(tsk
->ptrace
)) {
651 int sig
= thread_group_leader(tsk
) &&
652 thread_group_empty(tsk
) &&
653 !ptrace_reparented(tsk
) ?
654 tsk
->exit_signal
: SIGCHLD
;
655 autoreap
= do_notify_parent(tsk
, sig
);
656 } else if (thread_group_leader(tsk
)) {
657 autoreap
= thread_group_empty(tsk
) &&
658 do_notify_parent(tsk
, tsk
->exit_signal
);
664 tsk
->exit_state
= EXIT_DEAD
;
665 list_add(&tsk
->ptrace_entry
, &dead
);
668 /* mt-exec, de_thread() is waiting for group leader */
669 if (unlikely(tsk
->signal
->notify_count
< 0))
670 wake_up_process(tsk
->signal
->group_exit_task
);
671 write_unlock_irq(&tasklist_lock
);
673 list_for_each_entry_safe(p
, n
, &dead
, ptrace_entry
) {
674 list_del_init(&p
->ptrace_entry
);
679 #ifdef CONFIG_DEBUG_STACK_USAGE
680 static void check_stack_usage(void)
682 static DEFINE_SPINLOCK(low_water_lock
);
683 static int lowest_to_date
= THREAD_SIZE
;
686 free
= stack_not_used(current
);
688 if (free
>= lowest_to_date
)
691 spin_lock(&low_water_lock
);
692 if (free
< lowest_to_date
) {
693 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
694 current
->comm
, task_pid_nr(current
), free
);
695 lowest_to_date
= free
;
697 spin_unlock(&low_water_lock
);
700 static inline void check_stack_usage(void) {}
703 void __noreturn
do_exit(long code
)
705 struct task_struct
*tsk
= current
;
708 profile_task_exit(tsk
);
711 WARN_ON(blk_needs_flush_plug(tsk
));
713 if (unlikely(in_interrupt()))
714 panic("Aiee, killing interrupt handler!");
715 if (unlikely(!tsk
->pid
))
716 panic("Attempted to kill the idle task!");
719 * If do_exit is called because this processes oopsed, it's possible
720 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
721 * continuing. Amongst other possible reasons, this is to prevent
722 * mm_release()->clear_child_tid() from writing to a user-controlled
727 ptrace_event(PTRACE_EVENT_EXIT
, code
);
729 validate_creds_for_do_exit(tsk
);
732 * We're taking recursive faults here in do_exit. Safest is to just
733 * leave this task alone and wait for reboot.
735 if (unlikely(tsk
->flags
& PF_EXITING
)) {
736 pr_alert("Fixing recursive fault but reboot is needed!\n");
737 futex_exit_recursive(tsk
);
738 set_current_state(TASK_UNINTERRUPTIBLE
);
742 exit_signals(tsk
); /* sets PF_EXITING */
744 if (unlikely(in_atomic())) {
745 pr_info("note: %s[%d] exited with preempt_count %d\n",
746 current
->comm
, task_pid_nr(current
),
748 preempt_count_set(PREEMPT_ENABLED
);
751 /* sync mm's RSS info before statistics gathering */
753 sync_mm_rss(tsk
->mm
);
754 acct_update_integrals(tsk
);
755 group_dead
= atomic_dec_and_test(&tsk
->signal
->live
);
758 * If the last thread of global init has exited, panic
759 * immediately to get a useable coredump.
761 if (unlikely(is_global_init(tsk
)))
762 panic("Attempted to kill init! exitcode=0x%08x\n",
763 tsk
->signal
->group_exit_code
?: (int)code
);
765 #ifdef CONFIG_POSIX_TIMERS
766 hrtimer_cancel(&tsk
->signal
->real_timer
);
767 exit_itimers(tsk
->signal
);
770 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, tsk
->mm
);
772 acct_collect(code
, group_dead
);
777 tsk
->exit_code
= code
;
778 taskstats_exit(tsk
, group_dead
);
784 trace_sched_process_exit(tsk
);
791 disassociate_ctty(1);
792 exit_task_namespaces(tsk
);
798 * Flush inherited counters to the parent - before the parent
799 * gets woken up by child-exit notifications.
801 * because of cgroup mode, must be called before cgroup_exit()
803 perf_event_exit_task(tsk
);
805 sched_autogroup_exit_task(tsk
);
809 * FIXME: do that only when needed, using sched_exit tracepoint
811 flush_ptrace_hw_breakpoint(tsk
);
813 exit_tasks_rcu_start();
814 exit_notify(tsk
, group_dead
);
815 proc_exit_connector(tsk
);
816 mpol_put_task_policy(tsk
);
818 if (unlikely(current
->pi_state_cache
))
819 kfree(current
->pi_state_cache
);
822 * Make sure we are holding no locks:
824 debug_check_no_locks_held();
827 exit_io_context(tsk
);
829 if (tsk
->splice_pipe
)
830 free_pipe_info(tsk
->splice_pipe
);
832 if (tsk
->task_frag
.page
)
833 put_page(tsk
->task_frag
.page
);
835 validate_creds_for_do_exit(tsk
);
840 __this_cpu_add(dirty_throttle_leaks
, tsk
->nr_dirtied
);
842 exit_tasks_rcu_finish();
844 lockdep_free_task(tsk
);
847 EXPORT_SYMBOL_GPL(do_exit
);
849 void complete_and_exit(struct completion
*comp
, long code
)
856 EXPORT_SYMBOL(complete_and_exit
);
858 SYSCALL_DEFINE1(exit
, int, error_code
)
860 do_exit((error_code
&0xff)<<8);
864 * Take down every thread in the group. This is called by fatal signals
865 * as well as by sys_exit_group (below).
868 do_group_exit(int exit_code
)
870 struct signal_struct
*sig
= current
->signal
;
872 BUG_ON(exit_code
& 0x80); /* core dumps don't get here */
874 if (signal_group_exit(sig
))
875 exit_code
= sig
->group_exit_code
;
876 else if (!thread_group_empty(current
)) {
877 struct sighand_struct
*const sighand
= current
->sighand
;
879 spin_lock_irq(&sighand
->siglock
);
880 if (signal_group_exit(sig
))
881 /* Another thread got here before we took the lock. */
882 exit_code
= sig
->group_exit_code
;
884 sig
->group_exit_code
= exit_code
;
885 sig
->flags
= SIGNAL_GROUP_EXIT
;
886 zap_other_threads(current
);
888 spin_unlock_irq(&sighand
->siglock
);
896 * this kills every thread in the thread group. Note that any externally
897 * wait4()-ing process will get the correct exit code - even if this
898 * thread is not the thread group leader.
900 SYSCALL_DEFINE1(exit_group
, int, error_code
)
902 do_group_exit((error_code
& 0xff) << 8);
915 enum pid_type wo_type
;
919 struct waitid_info
*wo_info
;
921 struct rusage
*wo_rusage
;
923 wait_queue_entry_t child_wait
;
927 static int eligible_pid(struct wait_opts
*wo
, struct task_struct
*p
)
929 return wo
->wo_type
== PIDTYPE_MAX
||
930 task_pid_type(p
, wo
->wo_type
) == wo
->wo_pid
;
934 eligible_child(struct wait_opts
*wo
, bool ptrace
, struct task_struct
*p
)
936 if (!eligible_pid(wo
, p
))
940 * Wait for all children (clone and not) if __WALL is set or
941 * if it is traced by us.
943 if (ptrace
|| (wo
->wo_flags
& __WALL
))
947 * Otherwise, wait for clone children *only* if __WCLONE is set;
948 * otherwise, wait for non-clone children *only*.
950 * Note: a "clone" child here is one that reports to its parent
951 * using a signal other than SIGCHLD, or a non-leader thread which
952 * we can only see if it is traced by us.
954 if ((p
->exit_signal
!= SIGCHLD
) ^ !!(wo
->wo_flags
& __WCLONE
))
961 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
962 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
963 * the lock and this task is uninteresting. If we return nonzero, we have
964 * released the lock and the system call should return.
966 static int wait_task_zombie(struct wait_opts
*wo
, struct task_struct
*p
)
969 pid_t pid
= task_pid_vnr(p
);
970 uid_t uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
971 struct waitid_info
*infop
;
973 if (!likely(wo
->wo_flags
& WEXITED
))
976 if (unlikely(wo
->wo_flags
& WNOWAIT
)) {
977 status
= p
->exit_code
;
979 read_unlock(&tasklist_lock
);
980 sched_annotate_sleep();
982 getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
);
987 * Move the task's state to DEAD/TRACE, only one thread can do this.
989 state
= (ptrace_reparented(p
) && thread_group_leader(p
)) ?
990 EXIT_TRACE
: EXIT_DEAD
;
991 if (cmpxchg(&p
->exit_state
, EXIT_ZOMBIE
, state
) != EXIT_ZOMBIE
)
994 * We own this thread, nobody else can reap it.
996 read_unlock(&tasklist_lock
);
997 sched_annotate_sleep();
1000 * Check thread_group_leader() to exclude the traced sub-threads.
1002 if (state
== EXIT_DEAD
&& thread_group_leader(p
)) {
1003 struct signal_struct
*sig
= p
->signal
;
1004 struct signal_struct
*psig
= current
->signal
;
1005 unsigned long maxrss
;
1006 u64 tgutime
, tgstime
;
1009 * The resource counters for the group leader are in its
1010 * own task_struct. Those for dead threads in the group
1011 * are in its signal_struct, as are those for the child
1012 * processes it has previously reaped. All these
1013 * accumulate in the parent's signal_struct c* fields.
1015 * We don't bother to take a lock here to protect these
1016 * p->signal fields because the whole thread group is dead
1017 * and nobody can change them.
1019 * psig->stats_lock also protects us from our sub-theads
1020 * which can reap other children at the same time. Until
1021 * we change k_getrusage()-like users to rely on this lock
1022 * we have to take ->siglock as well.
1024 * We use thread_group_cputime_adjusted() to get times for
1025 * the thread group, which consolidates times for all threads
1026 * in the group including the group leader.
1028 thread_group_cputime_adjusted(p
, &tgutime
, &tgstime
);
1029 spin_lock_irq(¤t
->sighand
->siglock
);
1030 write_seqlock(&psig
->stats_lock
);
1031 psig
->cutime
+= tgutime
+ sig
->cutime
;
1032 psig
->cstime
+= tgstime
+ sig
->cstime
;
1033 psig
->cgtime
+= task_gtime(p
) + sig
->gtime
+ sig
->cgtime
;
1035 p
->min_flt
+ sig
->min_flt
+ sig
->cmin_flt
;
1037 p
->maj_flt
+ sig
->maj_flt
+ sig
->cmaj_flt
;
1039 p
->nvcsw
+ sig
->nvcsw
+ sig
->cnvcsw
;
1041 p
->nivcsw
+ sig
->nivcsw
+ sig
->cnivcsw
;
1043 task_io_get_inblock(p
) +
1044 sig
->inblock
+ sig
->cinblock
;
1046 task_io_get_oublock(p
) +
1047 sig
->oublock
+ sig
->coublock
;
1048 maxrss
= max(sig
->maxrss
, sig
->cmaxrss
);
1049 if (psig
->cmaxrss
< maxrss
)
1050 psig
->cmaxrss
= maxrss
;
1051 task_io_accounting_add(&psig
->ioac
, &p
->ioac
);
1052 task_io_accounting_add(&psig
->ioac
, &sig
->ioac
);
1053 write_sequnlock(&psig
->stats_lock
);
1054 spin_unlock_irq(¤t
->sighand
->siglock
);
1058 getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
);
1059 status
= (p
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1060 ? p
->signal
->group_exit_code
: p
->exit_code
;
1061 wo
->wo_stat
= status
;
1063 if (state
== EXIT_TRACE
) {
1064 write_lock_irq(&tasklist_lock
);
1065 /* We dropped tasklist, ptracer could die and untrace */
1068 /* If parent wants a zombie, don't release it now */
1069 state
= EXIT_ZOMBIE
;
1070 if (do_notify_parent(p
, p
->exit_signal
))
1072 p
->exit_state
= state
;
1073 write_unlock_irq(&tasklist_lock
);
1075 if (state
== EXIT_DEAD
)
1079 infop
= wo
->wo_info
;
1081 if ((status
& 0x7f) == 0) {
1082 infop
->cause
= CLD_EXITED
;
1083 infop
->status
= status
>> 8;
1085 infop
->cause
= (status
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1086 infop
->status
= status
& 0x7f;
1095 static int *task_stopped_code(struct task_struct
*p
, bool ptrace
)
1098 if (task_is_traced(p
) && !(p
->jobctl
& JOBCTL_LISTENING
))
1099 return &p
->exit_code
;
1101 if (p
->signal
->flags
& SIGNAL_STOP_STOPPED
)
1102 return &p
->signal
->group_exit_code
;
1108 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1110 * @ptrace: is the wait for ptrace
1111 * @p: task to wait for
1113 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1116 * read_lock(&tasklist_lock), which is released if return value is
1117 * non-zero. Also, grabs and releases @p->sighand->siglock.
1120 * 0 if wait condition didn't exist and search for other wait conditions
1121 * should continue. Non-zero return, -errno on failure and @p's pid on
1122 * success, implies that tasklist_lock is released and wait condition
1123 * search should terminate.
1125 static int wait_task_stopped(struct wait_opts
*wo
,
1126 int ptrace
, struct task_struct
*p
)
1128 struct waitid_info
*infop
;
1129 int exit_code
, *p_code
, why
;
1130 uid_t uid
= 0; /* unneeded, required by compiler */
1134 * Traditionally we see ptrace'd stopped tasks regardless of options.
1136 if (!ptrace
&& !(wo
->wo_flags
& WUNTRACED
))
1139 if (!task_stopped_code(p
, ptrace
))
1143 spin_lock_irq(&p
->sighand
->siglock
);
1145 p_code
= task_stopped_code(p
, ptrace
);
1146 if (unlikely(!p_code
))
1149 exit_code
= *p_code
;
1153 if (!unlikely(wo
->wo_flags
& WNOWAIT
))
1156 uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1158 spin_unlock_irq(&p
->sighand
->siglock
);
1163 * Now we are pretty sure this task is interesting.
1164 * Make sure it doesn't get reaped out from under us while we
1165 * give up the lock and then examine it below. We don't want to
1166 * keep holding onto the tasklist_lock while we call getrusage and
1167 * possibly take page faults for user memory.
1170 pid
= task_pid_vnr(p
);
1171 why
= ptrace
? CLD_TRAPPED
: CLD_STOPPED
;
1172 read_unlock(&tasklist_lock
);
1173 sched_annotate_sleep();
1175 getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
);
1178 if (likely(!(wo
->wo_flags
& WNOWAIT
)))
1179 wo
->wo_stat
= (exit_code
<< 8) | 0x7f;
1181 infop
= wo
->wo_info
;
1184 infop
->status
= exit_code
;
1192 * Handle do_wait work for one task in a live, non-stopped state.
1193 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1194 * the lock and this task is uninteresting. If we return nonzero, we have
1195 * released the lock and the system call should return.
1197 static int wait_task_continued(struct wait_opts
*wo
, struct task_struct
*p
)
1199 struct waitid_info
*infop
;
1203 if (!unlikely(wo
->wo_flags
& WCONTINUED
))
1206 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
))
1209 spin_lock_irq(&p
->sighand
->siglock
);
1210 /* Re-check with the lock held. */
1211 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
)) {
1212 spin_unlock_irq(&p
->sighand
->siglock
);
1215 if (!unlikely(wo
->wo_flags
& WNOWAIT
))
1216 p
->signal
->flags
&= ~SIGNAL_STOP_CONTINUED
;
1217 uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1218 spin_unlock_irq(&p
->sighand
->siglock
);
1220 pid
= task_pid_vnr(p
);
1222 read_unlock(&tasklist_lock
);
1223 sched_annotate_sleep();
1225 getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
);
1228 infop
= wo
->wo_info
;
1230 wo
->wo_stat
= 0xffff;
1232 infop
->cause
= CLD_CONTINUED
;
1235 infop
->status
= SIGCONT
;
1241 * Consider @p for a wait by @parent.
1243 * -ECHILD should be in ->notask_error before the first call.
1244 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1245 * Returns zero if the search for a child should continue;
1246 * then ->notask_error is 0 if @p is an eligible child,
1249 static int wait_consider_task(struct wait_opts
*wo
, int ptrace
,
1250 struct task_struct
*p
)
1253 * We can race with wait_task_zombie() from another thread.
1254 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1255 * can't confuse the checks below.
1257 int exit_state
= READ_ONCE(p
->exit_state
);
1260 if (unlikely(exit_state
== EXIT_DEAD
))
1263 ret
= eligible_child(wo
, ptrace
, p
);
1267 if (unlikely(exit_state
== EXIT_TRACE
)) {
1269 * ptrace == 0 means we are the natural parent. In this case
1270 * we should clear notask_error, debugger will notify us.
1272 if (likely(!ptrace
))
1273 wo
->notask_error
= 0;
1277 if (likely(!ptrace
) && unlikely(p
->ptrace
)) {
1279 * If it is traced by its real parent's group, just pretend
1280 * the caller is ptrace_do_wait() and reap this child if it
1283 * This also hides group stop state from real parent; otherwise
1284 * a single stop can be reported twice as group and ptrace stop.
1285 * If a ptracer wants to distinguish these two events for its
1286 * own children it should create a separate process which takes
1287 * the role of real parent.
1289 if (!ptrace_reparented(p
))
1294 if (exit_state
== EXIT_ZOMBIE
) {
1295 /* we don't reap group leaders with subthreads */
1296 if (!delay_group_leader(p
)) {
1298 * A zombie ptracee is only visible to its ptracer.
1299 * Notification and reaping will be cascaded to the
1300 * real parent when the ptracer detaches.
1302 if (unlikely(ptrace
) || likely(!p
->ptrace
))
1303 return wait_task_zombie(wo
, p
);
1307 * Allow access to stopped/continued state via zombie by
1308 * falling through. Clearing of notask_error is complex.
1312 * If WEXITED is set, notask_error should naturally be
1313 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1314 * so, if there are live subthreads, there are events to
1315 * wait for. If all subthreads are dead, it's still safe
1316 * to clear - this function will be called again in finite
1317 * amount time once all the subthreads are released and
1318 * will then return without clearing.
1322 * Stopped state is per-task and thus can't change once the
1323 * target task dies. Only continued and exited can happen.
1324 * Clear notask_error if WCONTINUED | WEXITED.
1326 if (likely(!ptrace
) || (wo
->wo_flags
& (WCONTINUED
| WEXITED
)))
1327 wo
->notask_error
= 0;
1330 * @p is alive and it's gonna stop, continue or exit, so
1331 * there always is something to wait for.
1333 wo
->notask_error
= 0;
1337 * Wait for stopped. Depending on @ptrace, different stopped state
1338 * is used and the two don't interact with each other.
1340 ret
= wait_task_stopped(wo
, ptrace
, p
);
1345 * Wait for continued. There's only one continued state and the
1346 * ptracer can consume it which can confuse the real parent. Don't
1347 * use WCONTINUED from ptracer. You don't need or want it.
1349 return wait_task_continued(wo
, p
);
1353 * Do the work of do_wait() for one thread in the group, @tsk.
1355 * -ECHILD should be in ->notask_error before the first call.
1356 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1357 * Returns zero if the search for a child should continue; then
1358 * ->notask_error is 0 if there were any eligible children,
1361 static int do_wait_thread(struct wait_opts
*wo
, struct task_struct
*tsk
)
1363 struct task_struct
*p
;
1365 list_for_each_entry(p
, &tsk
->children
, sibling
) {
1366 int ret
= wait_consider_task(wo
, 0, p
);
1375 static int ptrace_do_wait(struct wait_opts
*wo
, struct task_struct
*tsk
)
1377 struct task_struct
*p
;
1379 list_for_each_entry(p
, &tsk
->ptraced
, ptrace_entry
) {
1380 int ret
= wait_consider_task(wo
, 1, p
);
1389 static int child_wait_callback(wait_queue_entry_t
*wait
, unsigned mode
,
1390 int sync
, void *key
)
1392 struct wait_opts
*wo
= container_of(wait
, struct wait_opts
,
1394 struct task_struct
*p
= key
;
1396 if (!eligible_pid(wo
, p
))
1399 if ((wo
->wo_flags
& __WNOTHREAD
) && wait
->private != p
->parent
)
1402 return default_wake_function(wait
, mode
, sync
, key
);
1405 void __wake_up_parent(struct task_struct
*p
, struct task_struct
*parent
)
1407 __wake_up_sync_key(&parent
->signal
->wait_chldexit
,
1408 TASK_INTERRUPTIBLE
, p
);
1411 static long do_wait(struct wait_opts
*wo
)
1413 struct task_struct
*tsk
;
1416 trace_sched_process_wait(wo
->wo_pid
);
1418 init_waitqueue_func_entry(&wo
->child_wait
, child_wait_callback
);
1419 wo
->child_wait
.private = current
;
1420 add_wait_queue(¤t
->signal
->wait_chldexit
, &wo
->child_wait
);
1423 * If there is nothing that can match our criteria, just get out.
1424 * We will clear ->notask_error to zero if we see any child that
1425 * might later match our criteria, even if we are not able to reap
1428 wo
->notask_error
= -ECHILD
;
1429 if ((wo
->wo_type
< PIDTYPE_MAX
) &&
1430 (!wo
->wo_pid
|| !pid_has_task(wo
->wo_pid
, wo
->wo_type
)))
1433 set_current_state(TASK_INTERRUPTIBLE
);
1434 read_lock(&tasklist_lock
);
1437 retval
= do_wait_thread(wo
, tsk
);
1441 retval
= ptrace_do_wait(wo
, tsk
);
1445 if (wo
->wo_flags
& __WNOTHREAD
)
1447 } while_each_thread(current
, tsk
);
1448 read_unlock(&tasklist_lock
);
1451 retval
= wo
->notask_error
;
1452 if (!retval
&& !(wo
->wo_flags
& WNOHANG
)) {
1453 retval
= -ERESTARTSYS
;
1454 if (!signal_pending(current
)) {
1460 __set_current_state(TASK_RUNNING
);
1461 remove_wait_queue(¤t
->signal
->wait_chldexit
, &wo
->child_wait
);
1465 static struct pid
*pidfd_get_pid(unsigned int fd
)
1472 return ERR_PTR(-EBADF
);
1474 pid
= pidfd_pid(f
.file
);
1482 static long kernel_waitid(int which
, pid_t upid
, struct waitid_info
*infop
,
1483 int options
, struct rusage
*ru
)
1485 struct wait_opts wo
;
1486 struct pid
*pid
= NULL
;
1490 if (options
& ~(WNOHANG
|WNOWAIT
|WEXITED
|WSTOPPED
|WCONTINUED
|
1491 __WNOTHREAD
|__WCLONE
|__WALL
))
1493 if (!(options
& (WEXITED
|WSTOPPED
|WCONTINUED
)))
1505 pid
= find_get_pid(upid
);
1508 type
= PIDTYPE_PGID
;
1513 pid
= find_get_pid(upid
);
1515 pid
= get_task_pid(current
, PIDTYPE_PGID
);
1522 pid
= pidfd_get_pid(upid
);
1524 return PTR_ERR(pid
);
1532 wo
.wo_flags
= options
;
1541 SYSCALL_DEFINE5(waitid
, int, which
, pid_t
, upid
, struct siginfo __user
*,
1542 infop
, int, options
, struct rusage __user
*, ru
)
1545 struct waitid_info info
= {.status
= 0};
1546 long err
= kernel_waitid(which
, upid
, &info
, options
, ru
? &r
: NULL
);
1552 if (ru
&& copy_to_user(ru
, &r
, sizeof(struct rusage
)))
1558 if (!user_access_begin(infop
, sizeof(*infop
)))
1561 unsafe_put_user(signo
, &infop
->si_signo
, Efault
);
1562 unsafe_put_user(0, &infop
->si_errno
, Efault
);
1563 unsafe_put_user(info
.cause
, &infop
->si_code
, Efault
);
1564 unsafe_put_user(info
.pid
, &infop
->si_pid
, Efault
);
1565 unsafe_put_user(info
.uid
, &infop
->si_uid
, Efault
);
1566 unsafe_put_user(info
.status
, &infop
->si_status
, Efault
);
1574 long kernel_wait4(pid_t upid
, int __user
*stat_addr
, int options
,
1577 struct wait_opts wo
;
1578 struct pid
*pid
= NULL
;
1582 if (options
& ~(WNOHANG
|WUNTRACED
|WCONTINUED
|
1583 __WNOTHREAD
|__WCLONE
|__WALL
))
1586 /* -INT_MIN is not defined */
1587 if (upid
== INT_MIN
)
1592 else if (upid
< 0) {
1593 type
= PIDTYPE_PGID
;
1594 pid
= find_get_pid(-upid
);
1595 } else if (upid
== 0) {
1596 type
= PIDTYPE_PGID
;
1597 pid
= get_task_pid(current
, PIDTYPE_PGID
);
1598 } else /* upid > 0 */ {
1600 pid
= find_get_pid(upid
);
1605 wo
.wo_flags
= options
| WEXITED
;
1611 if (ret
> 0 && stat_addr
&& put_user(wo
.wo_stat
, stat_addr
))
1617 SYSCALL_DEFINE4(wait4
, pid_t
, upid
, int __user
*, stat_addr
,
1618 int, options
, struct rusage __user
*, ru
)
1621 long err
= kernel_wait4(upid
, stat_addr
, options
, ru
? &r
: NULL
);
1624 if (ru
&& copy_to_user(ru
, &r
, sizeof(struct rusage
)))
1630 #ifdef __ARCH_WANT_SYS_WAITPID
1633 * sys_waitpid() remains for compatibility. waitpid() should be
1634 * implemented by calling sys_wait4() from libc.a.
1636 SYSCALL_DEFINE3(waitpid
, pid_t
, pid
, int __user
*, stat_addr
, int, options
)
1638 return kernel_wait4(pid
, stat_addr
, options
, NULL
);
1643 #ifdef CONFIG_COMPAT
1644 COMPAT_SYSCALL_DEFINE4(wait4
,
1646 compat_uint_t __user
*, stat_addr
,
1648 struct compat_rusage __user
*, ru
)
1651 long err
= kernel_wait4(pid
, stat_addr
, options
, ru
? &r
: NULL
);
1653 if (ru
&& put_compat_rusage(&r
, ru
))
1659 COMPAT_SYSCALL_DEFINE5(waitid
,
1660 int, which
, compat_pid_t
, pid
,
1661 struct compat_siginfo __user
*, infop
, int, options
,
1662 struct compat_rusage __user
*, uru
)
1665 struct waitid_info info
= {.status
= 0};
1666 long err
= kernel_waitid(which
, pid
, &info
, options
, uru
? &ru
: NULL
);
1672 /* kernel_waitid() overwrites everything in ru */
1673 if (COMPAT_USE_64BIT_TIME
)
1674 err
= copy_to_user(uru
, &ru
, sizeof(ru
));
1676 err
= put_compat_rusage(&ru
, uru
);
1685 if (!user_access_begin(infop
, sizeof(*infop
)))
1688 unsafe_put_user(signo
, &infop
->si_signo
, Efault
);
1689 unsafe_put_user(0, &infop
->si_errno
, Efault
);
1690 unsafe_put_user(info
.cause
, &infop
->si_code
, Efault
);
1691 unsafe_put_user(info
.pid
, &infop
->si_pid
, Efault
);
1692 unsafe_put_user(info
.uid
, &infop
->si_uid
, Efault
);
1693 unsafe_put_user(info
.status
, &infop
->si_status
, Efault
);
1702 __weak
void abort(void)
1706 /* if that doesn't kill us, halt */
1707 panic("Oops failed to kill thread");
1709 EXPORT_SYMBOL(abort
);