4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/iocontext.h>
16 #include <linux/key.h>
17 #include <linux/security.h>
18 #include <linux/cpu.h>
19 #include <linux/acct.h>
20 #include <linux/tsacct_kern.h>
21 #include <linux/file.h>
22 #include <linux/fdtable.h>
23 #include <linux/freezer.h>
24 #include <linux/binfmts.h>
25 #include <linux/nsproxy.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/ptrace.h>
28 #include <linux/profile.h>
29 #include <linux/mount.h>
30 #include <linux/proc_fs.h>
31 #include <linux/kthread.h>
32 #include <linux/mempolicy.h>
33 #include <linux/taskstats_kern.h>
34 #include <linux/delayacct.h>
35 #include <linux/cgroup.h>
36 #include <linux/syscalls.h>
37 #include <linux/signal.h>
38 #include <linux/posix-timers.h>
39 #include <linux/cn_proc.h>
40 #include <linux/mutex.h>
41 #include <linux/futex.h>
42 #include <linux/pipe_fs_i.h>
43 #include <linux/audit.h> /* for audit_free() */
44 #include <linux/resource.h>
45 #include <linux/blkdev.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/tracehook.h>
48 #include <linux/fs_struct.h>
49 #include <linux/init_task.h>
50 #include <linux/perf_event.h>
51 #include <trace/events/sched.h>
52 #include <linux/hw_breakpoint.h>
53 #include <linux/oom.h>
54 #include <linux/writeback.h>
55 #include <linux/shm.h>
57 #include <asm/uaccess.h>
58 #include <asm/unistd.h>
59 #include <asm/pgtable.h>
60 #include <asm/mmu_context.h>
62 static void exit_mm(struct task_struct
*tsk
);
64 static void __unhash_process(struct task_struct
*p
, bool group_dead
)
67 detach_pid(p
, PIDTYPE_PID
);
69 detach_pid(p
, PIDTYPE_PGID
);
70 detach_pid(p
, PIDTYPE_SID
);
72 list_del_rcu(&p
->tasks
);
73 list_del_init(&p
->sibling
);
74 __this_cpu_dec(process_counts
);
76 list_del_rcu(&p
->thread_group
);
77 list_del_rcu(&p
->thread_node
);
81 * This function expects the tasklist_lock write-locked.
83 static void __exit_signal(struct task_struct
*tsk
)
85 struct signal_struct
*sig
= tsk
->signal
;
86 bool group_dead
= thread_group_leader(tsk
);
87 struct sighand_struct
*sighand
;
88 struct tty_struct
*uninitialized_var(tty
);
89 cputime_t utime
, stime
;
91 sighand
= rcu_dereference_check(tsk
->sighand
,
92 lockdep_tasklist_lock_is_held());
93 spin_lock(&sighand
->siglock
);
95 posix_cpu_timers_exit(tsk
);
97 posix_cpu_timers_exit_group(tsk
);
102 * This can only happen if the caller is de_thread().
103 * FIXME: this is the temporary hack, we should teach
104 * posix-cpu-timers to handle this case correctly.
106 if (unlikely(has_group_leader_pid(tsk
)))
107 posix_cpu_timers_exit_group(tsk
);
110 * If there is any task waiting for the group exit
113 if (sig
->notify_count
> 0 && !--sig
->notify_count
)
114 wake_up_process(sig
->group_exit_task
);
116 if (tsk
== sig
->curr_target
)
117 sig
->curr_target
= next_thread(tsk
);
121 * Accumulate here the counters for all threads but the group leader
122 * as they die, so they can be added into the process-wide totals
123 * when those are taken. The group leader stays around as a zombie as
124 * long as there are other threads. When it gets reaped, the exit.c
125 * code will add its counts into these totals. We won't ever get here
126 * for the group leader, since it will have been the last reference on
129 task_cputime(tsk
, &utime
, &stime
);
130 write_seqlock(&sig
->stats_lock
);
133 sig
->gtime
+= task_gtime(tsk
);
134 sig
->min_flt
+= tsk
->min_flt
;
135 sig
->maj_flt
+= tsk
->maj_flt
;
136 sig
->nvcsw
+= tsk
->nvcsw
;
137 sig
->nivcsw
+= tsk
->nivcsw
;
138 sig
->inblock
+= task_io_get_inblock(tsk
);
139 sig
->oublock
+= task_io_get_oublock(tsk
);
140 task_io_accounting_add(&sig
->ioac
, &tsk
->ioac
);
141 sig
->sum_sched_runtime
+= tsk
->se
.sum_exec_runtime
;
143 __unhash_process(tsk
, group_dead
);
144 write_sequnlock(&sig
->stats_lock
);
147 * Do this under ->siglock, we can race with another thread
148 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
150 flush_sigqueue(&tsk
->pending
);
152 spin_unlock(&sighand
->siglock
);
154 __cleanup_sighand(sighand
);
155 clear_tsk_thread_flag(tsk
, TIF_SIGPENDING
);
157 flush_sigqueue(&sig
->shared_pending
);
162 static void delayed_put_task_struct(struct rcu_head
*rhp
)
164 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
166 perf_event_delayed_put(tsk
);
167 trace_sched_process_free(tsk
);
168 put_task_struct(tsk
);
172 void release_task(struct task_struct
*p
)
174 struct task_struct
*leader
;
177 /* don't need to get the RCU readlock here - the process is dead and
178 * can't be modifying its own credentials. But shut RCU-lockdep up */
180 atomic_dec(&__task_cred(p
)->user
->processes
);
185 write_lock_irq(&tasklist_lock
);
186 ptrace_release_task(p
);
190 * If we are the last non-leader member of the thread
191 * group, and the leader is zombie, then notify the
192 * group leader's parent process. (if it wants notification.)
195 leader
= p
->group_leader
;
196 if (leader
!= p
&& thread_group_empty(leader
)
197 && leader
->exit_state
== EXIT_ZOMBIE
) {
199 * If we were the last child thread and the leader has
200 * exited already, and the leader's parent ignores SIGCHLD,
201 * then we are the one who should release the leader.
203 zap_leader
= do_notify_parent(leader
, leader
->exit_signal
);
205 leader
->exit_state
= EXIT_DEAD
;
208 write_unlock_irq(&tasklist_lock
);
210 call_rcu(&p
->rcu
, delayed_put_task_struct
);
213 if (unlikely(zap_leader
))
218 * This checks not only the pgrp, but falls back on the pid if no
219 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
222 * The caller must hold rcu lock or the tasklist lock.
224 struct pid
*session_of_pgrp(struct pid
*pgrp
)
226 struct task_struct
*p
;
227 struct pid
*sid
= NULL
;
229 p
= pid_task(pgrp
, PIDTYPE_PGID
);
231 p
= pid_task(pgrp
, PIDTYPE_PID
);
233 sid
= task_session(p
);
239 * Determine if a process group is "orphaned", according to the POSIX
240 * definition in 2.2.2.52. Orphaned process groups are not to be affected
241 * by terminal-generated stop signals. Newly orphaned process groups are
242 * to receive a SIGHUP and a SIGCONT.
244 * "I ask you, have you ever known what it is to be an orphan?"
246 static int will_become_orphaned_pgrp(struct pid
*pgrp
,
247 struct task_struct
*ignored_task
)
249 struct task_struct
*p
;
251 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
252 if ((p
== ignored_task
) ||
253 (p
->exit_state
&& thread_group_empty(p
)) ||
254 is_global_init(p
->real_parent
))
257 if (task_pgrp(p
->real_parent
) != pgrp
&&
258 task_session(p
->real_parent
) == task_session(p
))
260 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
265 int is_current_pgrp_orphaned(void)
269 read_lock(&tasklist_lock
);
270 retval
= will_become_orphaned_pgrp(task_pgrp(current
), NULL
);
271 read_unlock(&tasklist_lock
);
276 static bool has_stopped_jobs(struct pid
*pgrp
)
278 struct task_struct
*p
;
280 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
281 if (p
->signal
->flags
& SIGNAL_STOP_STOPPED
)
283 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
289 * Check to see if any process groups have become orphaned as
290 * a result of our exiting, and if they have any stopped jobs,
291 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
294 kill_orphaned_pgrp(struct task_struct
*tsk
, struct task_struct
*parent
)
296 struct pid
*pgrp
= task_pgrp(tsk
);
297 struct task_struct
*ignored_task
= tsk
;
300 /* exit: our father is in a different pgrp than
301 * we are and we were the only connection outside.
303 parent
= tsk
->real_parent
;
305 /* reparent: our child is in a different pgrp than
306 * we are, and it was the only connection outside.
310 if (task_pgrp(parent
) != pgrp
&&
311 task_session(parent
) == task_session(tsk
) &&
312 will_become_orphaned_pgrp(pgrp
, ignored_task
) &&
313 has_stopped_jobs(pgrp
)) {
314 __kill_pgrp_info(SIGHUP
, SEND_SIG_PRIV
, pgrp
);
315 __kill_pgrp_info(SIGCONT
, SEND_SIG_PRIV
, pgrp
);
321 * A task is exiting. If it owned this mm, find a new owner for the mm.
323 void mm_update_next_owner(struct mm_struct
*mm
)
325 struct task_struct
*c
, *g
, *p
= current
;
329 * If the exiting or execing task is not the owner, it's
330 * someone else's problem.
335 * The current owner is exiting/execing and there are no other
336 * candidates. Do not leave the mm pointing to a possibly
337 * freed task structure.
339 if (atomic_read(&mm
->mm_users
) <= 1) {
344 read_lock(&tasklist_lock
);
346 * Search in the children
348 list_for_each_entry(c
, &p
->children
, sibling
) {
350 goto assign_new_owner
;
354 * Search in the siblings
356 list_for_each_entry(c
, &p
->real_parent
->children
, sibling
) {
358 goto assign_new_owner
;
362 * Search through everything else, we should not get here often.
364 for_each_process(g
) {
365 if (g
->flags
& PF_KTHREAD
)
367 for_each_thread(g
, c
) {
369 goto assign_new_owner
;
374 read_unlock(&tasklist_lock
);
376 * We found no owner yet mm_users > 1: this implies that we are
377 * most likely racing with swapoff (try_to_unuse()) or /proc or
378 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
387 * The task_lock protects c->mm from changing.
388 * We always want mm->owner->mm == mm
392 * Delay read_unlock() till we have the task_lock()
393 * to ensure that c does not slip away underneath us
395 read_unlock(&tasklist_lock
);
405 #endif /* CONFIG_MEMCG */
408 * Turn us into a lazy TLB process if we
411 static void exit_mm(struct task_struct
*tsk
)
413 struct mm_struct
*mm
= tsk
->mm
;
414 struct core_state
*core_state
;
421 * Serialize with any possible pending coredump.
422 * We must hold mmap_sem around checking core_state
423 * and clearing tsk->mm. The core-inducing thread
424 * will increment ->nr_threads for each thread in the
425 * group with ->mm != NULL.
427 down_read(&mm
->mmap_sem
);
428 core_state
= mm
->core_state
;
430 struct core_thread self
;
432 up_read(&mm
->mmap_sem
);
435 self
.next
= xchg(&core_state
->dumper
.next
, &self
);
437 * Implies mb(), the result of xchg() must be visible
438 * to core_state->dumper.
440 if (atomic_dec_and_test(&core_state
->nr_threads
))
441 complete(&core_state
->startup
);
444 set_task_state(tsk
, TASK_UNINTERRUPTIBLE
);
445 if (!self
.task
) /* see coredump_finish() */
447 freezable_schedule();
449 __set_task_state(tsk
, TASK_RUNNING
);
450 down_read(&mm
->mmap_sem
);
452 atomic_inc(&mm
->mm_count
);
453 BUG_ON(mm
!= tsk
->active_mm
);
454 /* more a memory barrier than a real lock */
457 up_read(&mm
->mmap_sem
);
458 enter_lazy_tlb(mm
, current
);
460 mm_update_next_owner(mm
);
462 clear_thread_flag(TIF_MEMDIE
);
466 * When we die, we re-parent all our children, and try to:
467 * 1. give them to another thread in our thread group, if such a member exists
468 * 2. give it to the first ancestor process which prctl'd itself as a
469 * child_subreaper for its children (like a service manager)
470 * 3. give it to the init process (PID 1) in our pid namespace
472 static struct task_struct
*find_new_reaper(struct task_struct
*father
)
473 __releases(&tasklist_lock
)
474 __acquires(&tasklist_lock
)
476 struct pid_namespace
*pid_ns
= task_active_pid_ns(father
);
477 struct task_struct
*thread
;
480 while_each_thread(father
, thread
) {
481 if (thread
->flags
& PF_EXITING
)
483 if (unlikely(pid_ns
->child_reaper
== father
))
484 pid_ns
->child_reaper
= thread
;
488 if (unlikely(pid_ns
->child_reaper
== father
)) {
489 write_unlock_irq(&tasklist_lock
);
490 if (unlikely(pid_ns
== &init_pid_ns
)) {
491 panic("Attempted to kill init! exitcode=0x%08x\n",
492 father
->signal
->group_exit_code
?:
496 zap_pid_ns_processes(pid_ns
);
497 write_lock_irq(&tasklist_lock
);
498 } else if (father
->signal
->has_child_subreaper
) {
499 struct task_struct
*reaper
;
502 * Find the first ancestor marked as child_subreaper.
503 * Note that the code below checks same_thread_group(reaper,
504 * pid_ns->child_reaper). This is what we need to DTRT in a
505 * PID namespace. However we still need the check above, see
506 * http://marc.info/?l=linux-kernel&m=131385460420380
508 for (reaper
= father
->real_parent
;
509 reaper
!= &init_task
;
510 reaper
= reaper
->real_parent
) {
511 if (same_thread_group(reaper
, pid_ns
->child_reaper
))
513 if (!reaper
->signal
->is_child_subreaper
)
517 if (!(thread
->flags
& PF_EXITING
))
519 } while_each_thread(reaper
, thread
);
523 return pid_ns
->child_reaper
;
527 * Any that need to be release_task'd are put on the @dead list.
529 static void reparent_leader(struct task_struct
*father
, struct task_struct
*p
,
530 struct list_head
*dead
)
532 if (unlikely(p
->exit_state
== EXIT_DEAD
))
535 /* We don't want people slaying init. */
536 p
->exit_signal
= SIGCHLD
;
538 /* If it has exited notify the new parent about this child's death. */
540 p
->exit_state
== EXIT_ZOMBIE
&& thread_group_empty(p
)) {
541 if (do_notify_parent(p
, p
->exit_signal
)) {
542 p
->exit_state
= EXIT_DEAD
;
543 list_add(&p
->ptrace_entry
, dead
);
547 kill_orphaned_pgrp(p
, father
);
550 static void forget_original_parent(struct task_struct
*father
)
552 struct task_struct
*p
, *t
, *n
, *reaper
;
553 LIST_HEAD(dead_children
);
555 write_lock_irq(&tasklist_lock
);
557 * Note that exit_ptrace() and find_new_reaper() might
558 * drop tasklist_lock and reacquire it.
561 reaper
= find_new_reaper(father
);
563 list_for_each_entry(p
, &father
->children
, sibling
) {
564 for_each_thread(p
, t
) {
565 t
->real_parent
= reaper
;
566 BUG_ON((!t
->ptrace
) != (t
->parent
== father
));
567 if (likely(!t
->ptrace
))
568 t
->parent
= t
->real_parent
;
569 if (t
->pdeath_signal
)
570 group_send_sig_info(t
->pdeath_signal
,
574 * If this is a threaded reparent there is no need to
575 * notify anyone anything has happened.
577 if (!same_thread_group(reaper
, father
))
578 reparent_leader(father
, p
, &dead_children
);
580 list_splice_tail_init(&father
->children
, &reaper
->children
);
581 write_unlock_irq(&tasklist_lock
);
583 list_for_each_entry_safe(p
, n
, &dead_children
, ptrace_entry
) {
584 list_del_init(&p
->ptrace_entry
);
590 * Send signals to all our closest relatives so that they know
591 * to properly mourn us..
593 static void exit_notify(struct task_struct
*tsk
, int group_dead
)
598 * This does two things:
600 * A. Make init inherit all the child processes
601 * B. Check to see if any process groups have become orphaned
602 * as a result of our exiting, and if they have any stopped
603 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
605 forget_original_parent(tsk
);
607 write_lock_irq(&tasklist_lock
);
609 kill_orphaned_pgrp(tsk
->group_leader
, NULL
);
611 if (unlikely(tsk
->ptrace
)) {
612 int sig
= thread_group_leader(tsk
) &&
613 thread_group_empty(tsk
) &&
614 !ptrace_reparented(tsk
) ?
615 tsk
->exit_signal
: SIGCHLD
;
616 autoreap
= do_notify_parent(tsk
, sig
);
617 } else if (thread_group_leader(tsk
)) {
618 autoreap
= thread_group_empty(tsk
) &&
619 do_notify_parent(tsk
, tsk
->exit_signal
);
624 tsk
->exit_state
= autoreap
? EXIT_DEAD
: EXIT_ZOMBIE
;
626 /* mt-exec, de_thread() is waiting for group leader */
627 if (unlikely(tsk
->signal
->notify_count
< 0))
628 wake_up_process(tsk
->signal
->group_exit_task
);
629 write_unlock_irq(&tasklist_lock
);
631 /* If the process is dead, release it - nobody will wait for it */
636 #ifdef CONFIG_DEBUG_STACK_USAGE
637 static void check_stack_usage(void)
639 static DEFINE_SPINLOCK(low_water_lock
);
640 static int lowest_to_date
= THREAD_SIZE
;
643 free
= stack_not_used(current
);
645 if (free
>= lowest_to_date
)
648 spin_lock(&low_water_lock
);
649 if (free
< lowest_to_date
) {
650 pr_warn("%s (%d) used greatest stack depth: %lu bytes left\n",
651 current
->comm
, task_pid_nr(current
), free
);
652 lowest_to_date
= free
;
654 spin_unlock(&low_water_lock
);
657 static inline void check_stack_usage(void) {}
660 void do_exit(long code
)
662 struct task_struct
*tsk
= current
;
664 TASKS_RCU(int tasks_rcu_i
);
666 profile_task_exit(tsk
);
668 WARN_ON(blk_needs_flush_plug(tsk
));
670 if (unlikely(in_interrupt()))
671 panic("Aiee, killing interrupt handler!");
672 if (unlikely(!tsk
->pid
))
673 panic("Attempted to kill the idle task!");
676 * If do_exit is called because this processes oopsed, it's possible
677 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
678 * continuing. Amongst other possible reasons, this is to prevent
679 * mm_release()->clear_child_tid() from writing to a user-controlled
684 ptrace_event(PTRACE_EVENT_EXIT
, code
);
686 validate_creds_for_do_exit(tsk
);
689 * We're taking recursive faults here in do_exit. Safest is to just
690 * leave this task alone and wait for reboot.
692 if (unlikely(tsk
->flags
& PF_EXITING
)) {
693 pr_alert("Fixing recursive fault but reboot is needed!\n");
695 * We can do this unlocked here. The futex code uses
696 * this flag just to verify whether the pi state
697 * cleanup has been done or not. In the worst case it
698 * loops once more. We pretend that the cleanup was
699 * done as there is no way to return. Either the
700 * OWNER_DIED bit is set by now or we push the blocked
701 * task into the wait for ever nirwana as well.
703 tsk
->flags
|= PF_EXITPIDONE
;
704 set_current_state(TASK_UNINTERRUPTIBLE
);
708 exit_signals(tsk
); /* sets PF_EXITING */
710 * tsk->flags are checked in the futex code to protect against
711 * an exiting task cleaning up the robust pi futexes.
714 raw_spin_unlock_wait(&tsk
->pi_lock
);
716 if (unlikely(in_atomic()))
717 pr_info("note: %s[%d] exited with preempt_count %d\n",
718 current
->comm
, task_pid_nr(current
),
721 acct_update_integrals(tsk
);
722 /* sync mm's RSS info before statistics gathering */
724 sync_mm_rss(tsk
->mm
);
725 group_dead
= atomic_dec_and_test(&tsk
->signal
->live
);
727 hrtimer_cancel(&tsk
->signal
->real_timer
);
728 exit_itimers(tsk
->signal
);
730 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, tsk
->mm
);
732 acct_collect(code
, group_dead
);
737 tsk
->exit_code
= code
;
738 taskstats_exit(tsk
, group_dead
);
744 trace_sched_process_exit(tsk
);
751 disassociate_ctty(1);
752 exit_task_namespaces(tsk
);
757 * Flush inherited counters to the parent - before the parent
758 * gets woken up by child-exit notifications.
760 * because of cgroup mode, must be called before cgroup_exit()
762 perf_event_exit_task(tsk
);
766 module_put(task_thread_info(tsk
)->exec_domain
->module
);
769 * FIXME: do that only when needed, using sched_exit tracepoint
771 flush_ptrace_hw_breakpoint(tsk
);
773 TASKS_RCU(tasks_rcu_i
= __srcu_read_lock(&tasks_rcu_exit_srcu
));
774 exit_notify(tsk
, group_dead
);
775 proc_exit_connector(tsk
);
778 mpol_put(tsk
->mempolicy
);
779 tsk
->mempolicy
= NULL
;
783 if (unlikely(current
->pi_state_cache
))
784 kfree(current
->pi_state_cache
);
787 * Make sure we are holding no locks:
789 debug_check_no_locks_held();
791 * We can do this unlocked here. The futex code uses this flag
792 * just to verify whether the pi state cleanup has been done
793 * or not. In the worst case it loops once more.
795 tsk
->flags
|= PF_EXITPIDONE
;
798 exit_io_context(tsk
);
800 if (tsk
->splice_pipe
)
801 free_pipe_info(tsk
->splice_pipe
);
803 if (tsk
->task_frag
.page
)
804 put_page(tsk
->task_frag
.page
);
806 validate_creds_for_do_exit(tsk
);
811 __this_cpu_add(dirty_throttle_leaks
, tsk
->nr_dirtied
);
813 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu
, tasks_rcu_i
));
816 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
817 * when the following two conditions become true.
818 * - There is race condition of mmap_sem (It is acquired by
820 * - SMI occurs before setting TASK_RUNINNG.
821 * (or hypervisor of virtual machine switches to other guest)
822 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
824 * To avoid it, we have to wait for releasing tsk->pi_lock which
825 * is held by try_to_wake_up()
828 raw_spin_unlock_wait(&tsk
->pi_lock
);
830 /* causes final put_task_struct in finish_task_switch(). */
831 tsk
->state
= TASK_DEAD
;
832 tsk
->flags
|= PF_NOFREEZE
; /* tell freezer to ignore us */
835 /* Avoid "noreturn function does return". */
837 cpu_relax(); /* For when BUG is null */
839 EXPORT_SYMBOL_GPL(do_exit
);
841 void complete_and_exit(struct completion
*comp
, long code
)
848 EXPORT_SYMBOL(complete_and_exit
);
850 SYSCALL_DEFINE1(exit
, int, error_code
)
852 do_exit((error_code
&0xff)<<8);
856 * Take down every thread in the group. This is called by fatal signals
857 * as well as by sys_exit_group (below).
860 do_group_exit(int exit_code
)
862 struct signal_struct
*sig
= current
->signal
;
864 BUG_ON(exit_code
& 0x80); /* core dumps don't get here */
866 if (signal_group_exit(sig
))
867 exit_code
= sig
->group_exit_code
;
868 else if (!thread_group_empty(current
)) {
869 struct sighand_struct
*const sighand
= current
->sighand
;
871 spin_lock_irq(&sighand
->siglock
);
872 if (signal_group_exit(sig
))
873 /* Another thread got here before we took the lock. */
874 exit_code
= sig
->group_exit_code
;
876 sig
->group_exit_code
= exit_code
;
877 sig
->flags
= SIGNAL_GROUP_EXIT
;
878 zap_other_threads(current
);
880 spin_unlock_irq(&sighand
->siglock
);
888 * this kills every thread in the thread group. Note that any externally
889 * wait4()-ing process will get the correct exit code - even if this
890 * thread is not the thread group leader.
892 SYSCALL_DEFINE1(exit_group
, int, error_code
)
894 do_group_exit((error_code
& 0xff) << 8);
900 enum pid_type wo_type
;
904 struct siginfo __user
*wo_info
;
906 struct rusage __user
*wo_rusage
;
908 wait_queue_t child_wait
;
913 struct pid
*task_pid_type(struct task_struct
*task
, enum pid_type type
)
915 if (type
!= PIDTYPE_PID
)
916 task
= task
->group_leader
;
917 return task
->pids
[type
].pid
;
920 static int eligible_pid(struct wait_opts
*wo
, struct task_struct
*p
)
922 return wo
->wo_type
== PIDTYPE_MAX
||
923 task_pid_type(p
, wo
->wo_type
) == wo
->wo_pid
;
926 static int eligible_child(struct wait_opts
*wo
, struct task_struct
*p
)
928 if (!eligible_pid(wo
, p
))
930 /* Wait for all children (clone and not) if __WALL is set;
931 * otherwise, wait for clone children *only* if __WCLONE is
932 * set; otherwise, wait for non-clone children *only*. (Note:
933 * A "clone" child here is one that reports to its parent
934 * using a signal other than SIGCHLD.) */
935 if (((p
->exit_signal
!= SIGCHLD
) ^ !!(wo
->wo_flags
& __WCLONE
))
936 && !(wo
->wo_flags
& __WALL
))
942 static int wait_noreap_copyout(struct wait_opts
*wo
, struct task_struct
*p
,
943 pid_t pid
, uid_t uid
, int why
, int status
)
945 struct siginfo __user
*infop
;
946 int retval
= wo
->wo_rusage
947 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
953 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
955 retval
= put_user(0, &infop
->si_errno
);
957 retval
= put_user((short)why
, &infop
->si_code
);
959 retval
= put_user(pid
, &infop
->si_pid
);
961 retval
= put_user(uid
, &infop
->si_uid
);
963 retval
= put_user(status
, &infop
->si_status
);
971 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
972 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
973 * the lock and this task is uninteresting. If we return nonzero, we have
974 * released the lock and the system call should return.
976 static int wait_task_zombie(struct wait_opts
*wo
, struct task_struct
*p
)
979 int retval
, status
, traced
;
980 pid_t pid
= task_pid_vnr(p
);
981 uid_t uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
982 struct siginfo __user
*infop
;
984 if (!likely(wo
->wo_flags
& WEXITED
))
987 if (unlikely(wo
->wo_flags
& WNOWAIT
)) {
988 int exit_code
= p
->exit_code
;
992 read_unlock(&tasklist_lock
);
993 sched_annotate_sleep();
995 if ((exit_code
& 0x7f) == 0) {
997 status
= exit_code
>> 8;
999 why
= (exit_code
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1000 status
= exit_code
& 0x7f;
1002 return wait_noreap_copyout(wo
, p
, pid
, uid
, why
, status
);
1005 traced
= ptrace_reparented(p
);
1007 * Move the task's state to DEAD/TRACE, only one thread can do this.
1009 state
= traced
&& thread_group_leader(p
) ? EXIT_TRACE
: EXIT_DEAD
;
1010 if (cmpxchg(&p
->exit_state
, EXIT_ZOMBIE
, state
) != EXIT_ZOMBIE
)
1013 * It can be ptraced but not reparented, check
1014 * thread_group_leader() to filter out sub-threads.
1016 if (likely(!traced
) && thread_group_leader(p
)) {
1017 struct signal_struct
*psig
;
1018 struct signal_struct
*sig
;
1019 unsigned long maxrss
;
1020 cputime_t tgutime
, tgstime
;
1023 * The resource counters for the group leader are in its
1024 * own task_struct. Those for dead threads in the group
1025 * are in its signal_struct, as are those for the child
1026 * processes it has previously reaped. All these
1027 * accumulate in the parent's signal_struct c* fields.
1029 * We don't bother to take a lock here to protect these
1030 * p->signal fields, because they are only touched by
1031 * __exit_signal, which runs with tasklist_lock
1032 * write-locked anyway, and so is excluded here. We do
1033 * need to protect the access to parent->signal fields,
1034 * as other threads in the parent group can be right
1035 * here reaping other children at the same time.
1037 * We use thread_group_cputime_adjusted() to get times for
1038 * the thread group, which consolidates times for all threads
1039 * in the group including the group leader.
1041 thread_group_cputime_adjusted(p
, &tgutime
, &tgstime
);
1042 spin_lock_irq(&p
->real_parent
->sighand
->siglock
);
1043 psig
= p
->real_parent
->signal
;
1045 write_seqlock(&psig
->stats_lock
);
1046 psig
->cutime
+= tgutime
+ sig
->cutime
;
1047 psig
->cstime
+= tgstime
+ sig
->cstime
;
1048 psig
->cgtime
+= task_gtime(p
) + sig
->gtime
+ sig
->cgtime
;
1050 p
->min_flt
+ sig
->min_flt
+ sig
->cmin_flt
;
1052 p
->maj_flt
+ sig
->maj_flt
+ sig
->cmaj_flt
;
1054 p
->nvcsw
+ sig
->nvcsw
+ sig
->cnvcsw
;
1056 p
->nivcsw
+ sig
->nivcsw
+ sig
->cnivcsw
;
1058 task_io_get_inblock(p
) +
1059 sig
->inblock
+ sig
->cinblock
;
1061 task_io_get_oublock(p
) +
1062 sig
->oublock
+ sig
->coublock
;
1063 maxrss
= max(sig
->maxrss
, sig
->cmaxrss
);
1064 if (psig
->cmaxrss
< maxrss
)
1065 psig
->cmaxrss
= maxrss
;
1066 task_io_accounting_add(&psig
->ioac
, &p
->ioac
);
1067 task_io_accounting_add(&psig
->ioac
, &sig
->ioac
);
1068 write_sequnlock(&psig
->stats_lock
);
1069 spin_unlock_irq(&p
->real_parent
->sighand
->siglock
);
1073 * Now we are sure this task is interesting, and no other
1074 * thread can reap it because we its state == DEAD/TRACE.
1076 read_unlock(&tasklist_lock
);
1077 sched_annotate_sleep();
1079 retval
= wo
->wo_rusage
1080 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
1081 status
= (p
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1082 ? p
->signal
->group_exit_code
: p
->exit_code
;
1083 if (!retval
&& wo
->wo_stat
)
1084 retval
= put_user(status
, wo
->wo_stat
);
1086 infop
= wo
->wo_info
;
1087 if (!retval
&& infop
)
1088 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1089 if (!retval
&& infop
)
1090 retval
= put_user(0, &infop
->si_errno
);
1091 if (!retval
&& infop
) {
1094 if ((status
& 0x7f) == 0) {
1098 why
= (status
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1101 retval
= put_user((short)why
, &infop
->si_code
);
1103 retval
= put_user(status
, &infop
->si_status
);
1105 if (!retval
&& infop
)
1106 retval
= put_user(pid
, &infop
->si_pid
);
1107 if (!retval
&& infop
)
1108 retval
= put_user(uid
, &infop
->si_uid
);
1112 if (state
== EXIT_TRACE
) {
1113 write_lock_irq(&tasklist_lock
);
1114 /* We dropped tasklist, ptracer could die and untrace */
1117 /* If parent wants a zombie, don't release it now */
1118 state
= EXIT_ZOMBIE
;
1119 if (do_notify_parent(p
, p
->exit_signal
))
1121 p
->exit_state
= state
;
1122 write_unlock_irq(&tasklist_lock
);
1124 if (state
== EXIT_DEAD
)
1130 static int *task_stopped_code(struct task_struct
*p
, bool ptrace
)
1133 if (task_is_stopped_or_traced(p
) &&
1134 !(p
->jobctl
& JOBCTL_LISTENING
))
1135 return &p
->exit_code
;
1137 if (p
->signal
->flags
& SIGNAL_STOP_STOPPED
)
1138 return &p
->signal
->group_exit_code
;
1144 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1146 * @ptrace: is the wait for ptrace
1147 * @p: task to wait for
1149 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1152 * read_lock(&tasklist_lock), which is released if return value is
1153 * non-zero. Also, grabs and releases @p->sighand->siglock.
1156 * 0 if wait condition didn't exist and search for other wait conditions
1157 * should continue. Non-zero return, -errno on failure and @p's pid on
1158 * success, implies that tasklist_lock is released and wait condition
1159 * search should terminate.
1161 static int wait_task_stopped(struct wait_opts
*wo
,
1162 int ptrace
, struct task_struct
*p
)
1164 struct siginfo __user
*infop
;
1165 int retval
, exit_code
, *p_code
, why
;
1166 uid_t uid
= 0; /* unneeded, required by compiler */
1170 * Traditionally we see ptrace'd stopped tasks regardless of options.
1172 if (!ptrace
&& !(wo
->wo_flags
& WUNTRACED
))
1175 if (!task_stopped_code(p
, ptrace
))
1179 spin_lock_irq(&p
->sighand
->siglock
);
1181 p_code
= task_stopped_code(p
, ptrace
);
1182 if (unlikely(!p_code
))
1185 exit_code
= *p_code
;
1189 if (!unlikely(wo
->wo_flags
& WNOWAIT
))
1192 uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1194 spin_unlock_irq(&p
->sighand
->siglock
);
1199 * Now we are pretty sure this task is interesting.
1200 * Make sure it doesn't get reaped out from under us while we
1201 * give up the lock and then examine it below. We don't want to
1202 * keep holding onto the tasklist_lock while we call getrusage and
1203 * possibly take page faults for user memory.
1206 pid
= task_pid_vnr(p
);
1207 why
= ptrace
? CLD_TRAPPED
: CLD_STOPPED
;
1208 read_unlock(&tasklist_lock
);
1209 sched_annotate_sleep();
1211 if (unlikely(wo
->wo_flags
& WNOWAIT
))
1212 return wait_noreap_copyout(wo
, p
, pid
, uid
, why
, exit_code
);
1214 retval
= wo
->wo_rusage
1215 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
1216 if (!retval
&& wo
->wo_stat
)
1217 retval
= put_user((exit_code
<< 8) | 0x7f, wo
->wo_stat
);
1219 infop
= wo
->wo_info
;
1220 if (!retval
&& infop
)
1221 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1222 if (!retval
&& infop
)
1223 retval
= put_user(0, &infop
->si_errno
);
1224 if (!retval
&& infop
)
1225 retval
= put_user((short)why
, &infop
->si_code
);
1226 if (!retval
&& infop
)
1227 retval
= put_user(exit_code
, &infop
->si_status
);
1228 if (!retval
&& infop
)
1229 retval
= put_user(pid
, &infop
->si_pid
);
1230 if (!retval
&& infop
)
1231 retval
= put_user(uid
, &infop
->si_uid
);
1241 * Handle do_wait work for one task in a live, non-stopped state.
1242 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1243 * the lock and this task is uninteresting. If we return nonzero, we have
1244 * released the lock and the system call should return.
1246 static int wait_task_continued(struct wait_opts
*wo
, struct task_struct
*p
)
1252 if (!unlikely(wo
->wo_flags
& WCONTINUED
))
1255 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
))
1258 spin_lock_irq(&p
->sighand
->siglock
);
1259 /* Re-check with the lock held. */
1260 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
)) {
1261 spin_unlock_irq(&p
->sighand
->siglock
);
1264 if (!unlikely(wo
->wo_flags
& WNOWAIT
))
1265 p
->signal
->flags
&= ~SIGNAL_STOP_CONTINUED
;
1266 uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1267 spin_unlock_irq(&p
->sighand
->siglock
);
1269 pid
= task_pid_vnr(p
);
1271 read_unlock(&tasklist_lock
);
1272 sched_annotate_sleep();
1275 retval
= wo
->wo_rusage
1276 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
1278 if (!retval
&& wo
->wo_stat
)
1279 retval
= put_user(0xffff, wo
->wo_stat
);
1283 retval
= wait_noreap_copyout(wo
, p
, pid
, uid
,
1284 CLD_CONTINUED
, SIGCONT
);
1285 BUG_ON(retval
== 0);
1292 * Consider @p for a wait by @parent.
1294 * -ECHILD should be in ->notask_error before the first call.
1295 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1296 * Returns zero if the search for a child should continue;
1297 * then ->notask_error is 0 if @p is an eligible child,
1298 * or another error from security_task_wait(), or still -ECHILD.
1300 static int wait_consider_task(struct wait_opts
*wo
, int ptrace
,
1301 struct task_struct
*p
)
1305 if (unlikely(p
->exit_state
== EXIT_DEAD
))
1308 ret
= eligible_child(wo
, p
);
1312 ret
= security_task_wait(p
);
1313 if (unlikely(ret
< 0)) {
1315 * If we have not yet seen any eligible child,
1316 * then let this error code replace -ECHILD.
1317 * A permission error will give the user a clue
1318 * to look for security policy problems, rather
1319 * than for mysterious wait bugs.
1321 if (wo
->notask_error
)
1322 wo
->notask_error
= ret
;
1326 if (unlikely(p
->exit_state
== EXIT_TRACE
)) {
1328 * ptrace == 0 means we are the natural parent. In this case
1329 * we should clear notask_error, debugger will notify us.
1331 if (likely(!ptrace
))
1332 wo
->notask_error
= 0;
1336 if (likely(!ptrace
) && unlikely(p
->ptrace
)) {
1338 * If it is traced by its real parent's group, just pretend
1339 * the caller is ptrace_do_wait() and reap this child if it
1342 * This also hides group stop state from real parent; otherwise
1343 * a single stop can be reported twice as group and ptrace stop.
1344 * If a ptracer wants to distinguish these two events for its
1345 * own children it should create a separate process which takes
1346 * the role of real parent.
1348 if (!ptrace_reparented(p
))
1353 if (p
->exit_state
== EXIT_ZOMBIE
) {
1354 /* we don't reap group leaders with subthreads */
1355 if (!delay_group_leader(p
)) {
1357 * A zombie ptracee is only visible to its ptracer.
1358 * Notification and reaping will be cascaded to the
1359 * real parent when the ptracer detaches.
1361 if (unlikely(ptrace
) || likely(!p
->ptrace
))
1362 return wait_task_zombie(wo
, p
);
1366 * Allow access to stopped/continued state via zombie by
1367 * falling through. Clearing of notask_error is complex.
1371 * If WEXITED is set, notask_error should naturally be
1372 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1373 * so, if there are live subthreads, there are events to
1374 * wait for. If all subthreads are dead, it's still safe
1375 * to clear - this function will be called again in finite
1376 * amount time once all the subthreads are released and
1377 * will then return without clearing.
1381 * Stopped state is per-task and thus can't change once the
1382 * target task dies. Only continued and exited can happen.
1383 * Clear notask_error if WCONTINUED | WEXITED.
1385 if (likely(!ptrace
) || (wo
->wo_flags
& (WCONTINUED
| WEXITED
)))
1386 wo
->notask_error
= 0;
1389 * @p is alive and it's gonna stop, continue or exit, so
1390 * there always is something to wait for.
1392 wo
->notask_error
= 0;
1396 * Wait for stopped. Depending on @ptrace, different stopped state
1397 * is used and the two don't interact with each other.
1399 ret
= wait_task_stopped(wo
, ptrace
, p
);
1404 * Wait for continued. There's only one continued state and the
1405 * ptracer can consume it which can confuse the real parent. Don't
1406 * use WCONTINUED from ptracer. You don't need or want it.
1408 return wait_task_continued(wo
, p
);
1412 * Do the work of do_wait() for one thread in the group, @tsk.
1414 * -ECHILD should be in ->notask_error before the first call.
1415 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1416 * Returns zero if the search for a child should continue; then
1417 * ->notask_error is 0 if there were any eligible children,
1418 * or another error from security_task_wait(), or still -ECHILD.
1420 static int do_wait_thread(struct wait_opts
*wo
, struct task_struct
*tsk
)
1422 struct task_struct
*p
;
1424 list_for_each_entry(p
, &tsk
->children
, sibling
) {
1425 int ret
= wait_consider_task(wo
, 0, p
);
1434 static int ptrace_do_wait(struct wait_opts
*wo
, struct task_struct
*tsk
)
1436 struct task_struct
*p
;
1438 list_for_each_entry(p
, &tsk
->ptraced
, ptrace_entry
) {
1439 int ret
= wait_consider_task(wo
, 1, p
);
1448 static int child_wait_callback(wait_queue_t
*wait
, unsigned mode
,
1449 int sync
, void *key
)
1451 struct wait_opts
*wo
= container_of(wait
, struct wait_opts
,
1453 struct task_struct
*p
= key
;
1455 if (!eligible_pid(wo
, p
))
1458 if ((wo
->wo_flags
& __WNOTHREAD
) && wait
->private != p
->parent
)
1461 return default_wake_function(wait
, mode
, sync
, key
);
1464 void __wake_up_parent(struct task_struct
*p
, struct task_struct
*parent
)
1466 __wake_up_sync_key(&parent
->signal
->wait_chldexit
,
1467 TASK_INTERRUPTIBLE
, 1, p
);
1470 static long do_wait(struct wait_opts
*wo
)
1472 struct task_struct
*tsk
;
1475 trace_sched_process_wait(wo
->wo_pid
);
1477 init_waitqueue_func_entry(&wo
->child_wait
, child_wait_callback
);
1478 wo
->child_wait
.private = current
;
1479 add_wait_queue(¤t
->signal
->wait_chldexit
, &wo
->child_wait
);
1482 * If there is nothing that can match our critiera just get out.
1483 * We will clear ->notask_error to zero if we see any child that
1484 * might later match our criteria, even if we are not able to reap
1487 wo
->notask_error
= -ECHILD
;
1488 if ((wo
->wo_type
< PIDTYPE_MAX
) &&
1489 (!wo
->wo_pid
|| hlist_empty(&wo
->wo_pid
->tasks
[wo
->wo_type
])))
1492 set_current_state(TASK_INTERRUPTIBLE
);
1493 read_lock(&tasklist_lock
);
1496 retval
= do_wait_thread(wo
, tsk
);
1500 retval
= ptrace_do_wait(wo
, tsk
);
1504 if (wo
->wo_flags
& __WNOTHREAD
)
1506 } while_each_thread(current
, tsk
);
1507 read_unlock(&tasklist_lock
);
1510 retval
= wo
->notask_error
;
1511 if (!retval
&& !(wo
->wo_flags
& WNOHANG
)) {
1512 retval
= -ERESTARTSYS
;
1513 if (!signal_pending(current
)) {
1519 __set_current_state(TASK_RUNNING
);
1520 remove_wait_queue(¤t
->signal
->wait_chldexit
, &wo
->child_wait
);
1524 SYSCALL_DEFINE5(waitid
, int, which
, pid_t
, upid
, struct siginfo __user
*,
1525 infop
, int, options
, struct rusage __user
*, ru
)
1527 struct wait_opts wo
;
1528 struct pid
*pid
= NULL
;
1532 if (options
& ~(WNOHANG
|WNOWAIT
|WEXITED
|WSTOPPED
|WCONTINUED
))
1534 if (!(options
& (WEXITED
|WSTOPPED
|WCONTINUED
)))
1547 type
= PIDTYPE_PGID
;
1555 if (type
< PIDTYPE_MAX
)
1556 pid
= find_get_pid(upid
);
1560 wo
.wo_flags
= options
;
1570 * For a WNOHANG return, clear out all the fields
1571 * we would set so the user can easily tell the
1575 ret
= put_user(0, &infop
->si_signo
);
1577 ret
= put_user(0, &infop
->si_errno
);
1579 ret
= put_user(0, &infop
->si_code
);
1581 ret
= put_user(0, &infop
->si_pid
);
1583 ret
= put_user(0, &infop
->si_uid
);
1585 ret
= put_user(0, &infop
->si_status
);
1592 SYSCALL_DEFINE4(wait4
, pid_t
, upid
, int __user
*, stat_addr
,
1593 int, options
, struct rusage __user
*, ru
)
1595 struct wait_opts wo
;
1596 struct pid
*pid
= NULL
;
1600 if (options
& ~(WNOHANG
|WUNTRACED
|WCONTINUED
|
1601 __WNOTHREAD
|__WCLONE
|__WALL
))
1606 else if (upid
< 0) {
1607 type
= PIDTYPE_PGID
;
1608 pid
= find_get_pid(-upid
);
1609 } else if (upid
== 0) {
1610 type
= PIDTYPE_PGID
;
1611 pid
= get_task_pid(current
, PIDTYPE_PGID
);
1612 } else /* upid > 0 */ {
1614 pid
= find_get_pid(upid
);
1619 wo
.wo_flags
= options
| WEXITED
;
1621 wo
.wo_stat
= stat_addr
;
1629 #ifdef __ARCH_WANT_SYS_WAITPID
1632 * sys_waitpid() remains for compatibility. waitpid() should be
1633 * implemented by calling sys_wait4() from libc.a.
1635 SYSCALL_DEFINE3(waitpid
, pid_t
, pid
, int __user
*, stat_addr
, int, options
)
1637 return sys_wait4(pid
, stat_addr
, options
, NULL
);
This page took 0.066771 seconds and 6 git commands to generate.