]>
git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - kernel/exit.c
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/mnt_namespace.h>
16 #include <linux/iocontext.h>
17 #include <linux/key.h>
18 #include <linux/security.h>
19 #include <linux/cpu.h>
20 #include <linux/acct.h>
21 #include <linux/tsacct_kern.h>
22 #include <linux/file.h>
23 #include <linux/fdtable.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/freezer.h>
36 #include <linux/cgroup.h>
37 #include <linux/syscalls.h>
38 #include <linux/signal.h>
39 #include <linux/posix-timers.h>
40 #include <linux/cn_proc.h>
41 #include <linux/mutex.h>
42 #include <linux/futex.h>
43 #include <linux/pipe_fs_i.h>
44 #include <linux/audit.h> /* for audit_free() */
45 #include <linux/resource.h>
46 #include <linux/blkdev.h>
47 #include <linux/task_io_accounting_ops.h>
48 #include <linux/tracehook.h>
49 #include <linux/fs_struct.h>
50 #include <linux/init_task.h>
51 #include <trace/sched.h>
53 #include <asm/uaccess.h>
54 #include <asm/unistd.h>
55 #include <asm/pgtable.h>
56 #include <asm/mmu_context.h>
57 #include "cred-internals.h"
59 DEFINE_TRACE(sched_process_free
);
60 DEFINE_TRACE(sched_process_exit
);
61 DEFINE_TRACE(sched_process_wait
);
63 static void exit_mm(struct task_struct
* tsk
);
65 static void __unhash_process(struct task_struct
*p
)
68 detach_pid(p
, PIDTYPE_PID
);
69 if (thread_group_leader(p
)) {
70 detach_pid(p
, PIDTYPE_PGID
);
71 detach_pid(p
, PIDTYPE_SID
);
73 list_del_rcu(&p
->tasks
);
74 __get_cpu_var(process_counts
)--;
76 list_del_rcu(&p
->thread_group
);
77 list_del_init(&p
->sibling
);
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 struct sighand_struct
*sighand
;
89 BUG_ON(!atomic_read(&sig
->count
));
91 sighand
= rcu_dereference(tsk
->sighand
);
92 spin_lock(&sighand
->siglock
);
94 posix_cpu_timers_exit(tsk
);
95 if (atomic_dec_and_test(&sig
->count
))
96 posix_cpu_timers_exit_group(tsk
);
99 * If there is any task waiting for the group exit
102 if (sig
->group_exit_task
&& atomic_read(&sig
->count
) == sig
->notify_count
)
103 wake_up_process(sig
->group_exit_task
);
105 if (tsk
== sig
->curr_target
)
106 sig
->curr_target
= next_thread(tsk
);
108 * Accumulate here the counters for all threads but the
109 * group leader as they die, so they can be added into
110 * the process-wide totals when those are taken.
111 * The group leader stays around as a zombie as long
112 * as there are other threads. When it gets reaped,
113 * the exit.c code will add its counts into these totals.
114 * We won't ever get here for the group leader, since it
115 * will have been the last reference on the signal_struct.
117 sig
->utime
= cputime_add(sig
->utime
, task_utime(tsk
));
118 sig
->stime
= cputime_add(sig
->stime
, task_stime(tsk
));
119 sig
->gtime
= cputime_add(sig
->gtime
, task_gtime(tsk
));
120 sig
->min_flt
+= tsk
->min_flt
;
121 sig
->maj_flt
+= tsk
->maj_flt
;
122 sig
->nvcsw
+= tsk
->nvcsw
;
123 sig
->nivcsw
+= tsk
->nivcsw
;
124 sig
->inblock
+= task_io_get_inblock(tsk
);
125 sig
->oublock
+= task_io_get_oublock(tsk
);
126 task_io_accounting_add(&sig
->ioac
, &tsk
->ioac
);
127 sig
->sum_sched_runtime
+= tsk
->se
.sum_exec_runtime
;
128 sig
= NULL
; /* Marker for below. */
132 * Flush inherited counters to the parent - before the parent
133 * gets woken up by child-exit notifications.
135 perf_counter_exit_task(tsk
);
137 __unhash_process(tsk
);
140 * Do this under ->siglock, we can race with another thread
141 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
143 flush_sigqueue(&tsk
->pending
);
147 spin_unlock(&sighand
->siglock
);
149 __cleanup_sighand(sighand
);
150 clear_tsk_thread_flag(tsk
,TIF_SIGPENDING
);
152 flush_sigqueue(&sig
->shared_pending
);
153 taskstats_tgid_free(sig
);
155 * Make sure ->signal can't go away under rq->lock,
156 * see account_group_exec_runtime().
158 task_rq_unlock_wait(tsk
);
159 __cleanup_signal(sig
);
163 static void delayed_put_task_struct(struct rcu_head
*rhp
)
165 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
167 #ifdef CONFIG_PERF_COUNTERS
168 WARN_ON_ONCE(!list_empty(&tsk
->perf_counter_ctx
.counter_list
));
170 trace_sched_process_free(tsk
);
171 put_task_struct(tsk
);
175 void release_task(struct task_struct
* p
)
177 struct task_struct
*leader
;
180 tracehook_prepare_release_task(p
);
181 /* don't need to get the RCU readlock here - the process is dead and
182 * can't be modifying its own credentials */
183 atomic_dec(&__task_cred(p
)->user
->processes
);
186 write_lock_irq(&tasklist_lock
);
187 tracehook_finish_release_task(p
);
191 * If we are the last non-leader member of the thread
192 * group, and the leader is zombie, then notify the
193 * group leader's parent process. (if it wants notification.)
196 leader
= p
->group_leader
;
197 if (leader
!= p
&& thread_group_empty(leader
) && leader
->exit_state
== EXIT_ZOMBIE
) {
198 BUG_ON(task_detached(leader
));
199 do_notify_parent(leader
, leader
->exit_signal
);
201 * If we were the last child thread and the leader has
202 * exited already, and the leader's parent ignores SIGCHLD,
203 * then we are the one who should release the leader.
205 * do_notify_parent() will have marked it self-reaping in
208 zap_leader
= task_detached(leader
);
211 * This maintains the invariant that release_task()
212 * only runs on a task in EXIT_DEAD, just for sanity.
215 leader
->exit_state
= EXIT_DEAD
;
218 write_unlock_irq(&tasklist_lock
);
220 call_rcu(&p
->rcu
, delayed_put_task_struct
);
223 if (unlikely(zap_leader
))
228 * This checks not only the pgrp, but falls back on the pid if no
229 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
232 * The caller must hold rcu lock or the tasklist lock.
234 struct pid
*session_of_pgrp(struct pid
*pgrp
)
236 struct task_struct
*p
;
237 struct pid
*sid
= NULL
;
239 p
= pid_task(pgrp
, PIDTYPE_PGID
);
241 p
= pid_task(pgrp
, PIDTYPE_PID
);
243 sid
= task_session(p
);
249 * Determine if a process group is "orphaned", according to the POSIX
250 * definition in 2.2.2.52. Orphaned process groups are not to be affected
251 * by terminal-generated stop signals. Newly orphaned process groups are
252 * to receive a SIGHUP and a SIGCONT.
254 * "I ask you, have you ever known what it is to be an orphan?"
256 static int will_become_orphaned_pgrp(struct pid
*pgrp
, struct task_struct
*ignored_task
)
258 struct task_struct
*p
;
260 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
261 if ((p
== ignored_task
) ||
262 (p
->exit_state
&& thread_group_empty(p
)) ||
263 is_global_init(p
->real_parent
))
266 if (task_pgrp(p
->real_parent
) != pgrp
&&
267 task_session(p
->real_parent
) == task_session(p
))
269 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
274 int is_current_pgrp_orphaned(void)
278 read_lock(&tasklist_lock
);
279 retval
= will_become_orphaned_pgrp(task_pgrp(current
), NULL
);
280 read_unlock(&tasklist_lock
);
285 static int has_stopped_jobs(struct pid
*pgrp
)
288 struct task_struct
*p
;
290 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
291 if (!task_is_stopped(p
))
295 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
300 * Check to see if any process groups have become orphaned as
301 * a result of our exiting, and if they have any stopped jobs,
302 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
305 kill_orphaned_pgrp(struct task_struct
*tsk
, struct task_struct
*parent
)
307 struct pid
*pgrp
= task_pgrp(tsk
);
308 struct task_struct
*ignored_task
= tsk
;
311 /* exit: our father is in a different pgrp than
312 * we are and we were the only connection outside.
314 parent
= tsk
->real_parent
;
316 /* reparent: our child is in a different pgrp than
317 * we are, and it was the only connection outside.
321 if (task_pgrp(parent
) != pgrp
&&
322 task_session(parent
) == task_session(tsk
) &&
323 will_become_orphaned_pgrp(pgrp
, ignored_task
) &&
324 has_stopped_jobs(pgrp
)) {
325 __kill_pgrp_info(SIGHUP
, SEND_SIG_PRIV
, pgrp
);
326 __kill_pgrp_info(SIGCONT
, SEND_SIG_PRIV
, pgrp
);
331 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
333 * If a kernel thread is launched as a result of a system call, or if
334 * it ever exits, it should generally reparent itself to kthreadd so it
335 * isn't in the way of other processes and is correctly cleaned up on exit.
337 * The various task state such as scheduling policy and priority may have
338 * been inherited from a user process, so we reset them to sane values here.
340 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
342 static void reparent_to_kthreadd(void)
344 write_lock_irq(&tasklist_lock
);
346 ptrace_unlink(current
);
347 /* Reparent to init */
348 current
->real_parent
= current
->parent
= kthreadd_task
;
349 list_move_tail(¤t
->sibling
, ¤t
->real_parent
->children
);
351 /* Set the exit signal to SIGCHLD so we signal init on exit */
352 current
->exit_signal
= SIGCHLD
;
354 if (task_nice(current
) < 0)
355 set_user_nice(current
, 0);
359 memcpy(current
->signal
->rlim
, init_task
.signal
->rlim
,
360 sizeof(current
->signal
->rlim
));
362 atomic_inc(&init_cred
.usage
);
363 commit_creds(&init_cred
);
364 write_unlock_irq(&tasklist_lock
);
367 void __set_special_pids(struct pid
*pid
)
369 struct task_struct
*curr
= current
->group_leader
;
371 if (task_session(curr
) != pid
)
372 change_pid(curr
, PIDTYPE_SID
, pid
);
374 if (task_pgrp(curr
) != pid
)
375 change_pid(curr
, PIDTYPE_PGID
, pid
);
378 static void set_special_pids(struct pid
*pid
)
380 write_lock_irq(&tasklist_lock
);
381 __set_special_pids(pid
);
382 write_unlock_irq(&tasklist_lock
);
386 * Let kernel threads use this to say that they
387 * allow a certain signal (since daemonize() will
388 * have disabled all of them by default).
390 int allow_signal(int sig
)
392 if (!valid_signal(sig
) || sig
< 1)
395 spin_lock_irq(¤t
->sighand
->siglock
);
396 sigdelset(¤t
->blocked
, sig
);
398 /* Kernel threads handle their own signals.
399 Let the signal code know it'll be handled, so
400 that they don't get converted to SIGKILL or
401 just silently dropped */
402 current
->sighand
->action
[(sig
)-1].sa
.sa_handler
= (void __user
*)2;
405 spin_unlock_irq(¤t
->sighand
->siglock
);
409 EXPORT_SYMBOL(allow_signal
);
411 int disallow_signal(int sig
)
413 if (!valid_signal(sig
) || sig
< 1)
416 spin_lock_irq(¤t
->sighand
->siglock
);
417 current
->sighand
->action
[(sig
)-1].sa
.sa_handler
= SIG_IGN
;
419 spin_unlock_irq(¤t
->sighand
->siglock
);
423 EXPORT_SYMBOL(disallow_signal
);
426 * Put all the gunge required to become a kernel thread without
427 * attached user resources in one place where it belongs.
430 void daemonize(const char *name
, ...)
435 va_start(args
, name
);
436 vsnprintf(current
->comm
, sizeof(current
->comm
), name
, args
);
440 * If we were started as result of loading a module, close all of the
441 * user space pages. We don't need them, and if we didn't close them
442 * they would be locked into memory.
446 * We don't want to have TIF_FREEZE set if the system-wide hibernation
447 * or suspend transition begins right now.
449 current
->flags
|= (PF_NOFREEZE
| PF_KTHREAD
);
451 if (current
->nsproxy
!= &init_nsproxy
) {
452 get_nsproxy(&init_nsproxy
);
453 switch_task_namespaces(current
, &init_nsproxy
);
455 set_special_pids(&init_struct_pid
);
456 proc_clear_tty(current
);
458 /* Block and flush all signals */
459 sigfillset(&blocked
);
460 sigprocmask(SIG_BLOCK
, &blocked
, NULL
);
461 flush_signals(current
);
463 /* Become as one with the init task */
465 daemonize_fs_struct();
467 current
->files
= init_task
.files
;
468 atomic_inc(¤t
->files
->count
);
470 reparent_to_kthreadd();
473 EXPORT_SYMBOL(daemonize
);
475 static void close_files(struct files_struct
* files
)
483 * It is safe to dereference the fd table without RCU or
484 * ->file_lock because this is the last reference to the
487 fdt
= files_fdtable(files
);
491 if (i
>= fdt
->max_fds
)
493 set
= fdt
->open_fds
->fds_bits
[j
++];
496 struct file
* file
= xchg(&fdt
->fd
[i
], NULL
);
498 filp_close(file
, files
);
508 struct files_struct
*get_files_struct(struct task_struct
*task
)
510 struct files_struct
*files
;
515 atomic_inc(&files
->count
);
521 void put_files_struct(struct files_struct
*files
)
525 if (atomic_dec_and_test(&files
->count
)) {
528 * Free the fd and fdset arrays if we expanded them.
529 * If the fdtable was embedded, pass files for freeing
530 * at the end of the RCU grace period. Otherwise,
531 * you can free files immediately.
533 fdt
= files_fdtable(files
);
534 if (fdt
!= &files
->fdtab
)
535 kmem_cache_free(files_cachep
, files
);
540 void reset_files_struct(struct files_struct
*files
)
542 struct task_struct
*tsk
= current
;
543 struct files_struct
*old
;
549 put_files_struct(old
);
552 void exit_files(struct task_struct
*tsk
)
554 struct files_struct
* files
= tsk
->files
;
560 put_files_struct(files
);
564 #ifdef CONFIG_MM_OWNER
566 * Task p is exiting and it owned mm, lets find a new owner for it
569 mm_need_new_owner(struct mm_struct
*mm
, struct task_struct
*p
)
572 * If there are other users of the mm and the owner (us) is exiting
573 * we need to find a new owner to take on the responsibility.
575 if (atomic_read(&mm
->mm_users
) <= 1)
582 void mm_update_next_owner(struct mm_struct
*mm
)
584 struct task_struct
*c
, *g
, *p
= current
;
587 if (!mm_need_new_owner(mm
, p
))
590 read_lock(&tasklist_lock
);
592 * Search in the children
594 list_for_each_entry(c
, &p
->children
, sibling
) {
596 goto assign_new_owner
;
600 * Search in the siblings
602 list_for_each_entry(c
, &p
->parent
->children
, sibling
) {
604 goto assign_new_owner
;
608 * Search through everything else. We should not get
611 do_each_thread(g
, c
) {
613 goto assign_new_owner
;
614 } while_each_thread(g
, c
);
616 read_unlock(&tasklist_lock
);
618 * We found no owner yet mm_users > 1: this implies that we are
619 * most likely racing with swapoff (try_to_unuse()) or /proc or
620 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
629 * The task_lock protects c->mm from changing.
630 * We always want mm->owner->mm == mm
634 * Delay read_unlock() till we have the task_lock()
635 * to ensure that c does not slip away underneath us
637 read_unlock(&tasklist_lock
);
647 #endif /* CONFIG_MM_OWNER */
650 * Turn us into a lazy TLB process if we
653 static void exit_mm(struct task_struct
* tsk
)
655 struct mm_struct
*mm
= tsk
->mm
;
656 struct core_state
*core_state
;
662 * Serialize with any possible pending coredump.
663 * We must hold mmap_sem around checking core_state
664 * and clearing tsk->mm. The core-inducing thread
665 * will increment ->nr_threads for each thread in the
666 * group with ->mm != NULL.
668 down_read(&mm
->mmap_sem
);
669 core_state
= mm
->core_state
;
671 struct core_thread self
;
672 up_read(&mm
->mmap_sem
);
675 self
.next
= xchg(&core_state
->dumper
.next
, &self
);
677 * Implies mb(), the result of xchg() must be visible
678 * to core_state->dumper.
680 if (atomic_dec_and_test(&core_state
->nr_threads
))
681 complete(&core_state
->startup
);
684 set_task_state(tsk
, TASK_UNINTERRUPTIBLE
);
685 if (!self
.task
) /* see coredump_finish() */
689 __set_task_state(tsk
, TASK_RUNNING
);
690 down_read(&mm
->mmap_sem
);
692 atomic_inc(&mm
->mm_count
);
693 BUG_ON(mm
!= tsk
->active_mm
);
694 /* more a memory barrier than a real lock */
697 up_read(&mm
->mmap_sem
);
698 enter_lazy_tlb(mm
, current
);
699 /* We don't want this task to be frozen prematurely */
700 clear_freeze_flag(tsk
);
702 mm_update_next_owner(mm
);
707 * When we die, we re-parent all our children.
708 * Try to give them to another thread in our thread
709 * group, and if no such member exists, give it to
710 * the child reaper process (ie "init") in our pid
713 static struct task_struct
*find_new_reaper(struct task_struct
*father
)
715 struct pid_namespace
*pid_ns
= task_active_pid_ns(father
);
716 struct task_struct
*thread
;
719 while_each_thread(father
, thread
) {
720 if (thread
->flags
& PF_EXITING
)
722 if (unlikely(pid_ns
->child_reaper
== father
))
723 pid_ns
->child_reaper
= thread
;
727 if (unlikely(pid_ns
->child_reaper
== father
)) {
728 write_unlock_irq(&tasklist_lock
);
729 if (unlikely(pid_ns
== &init_pid_ns
))
730 panic("Attempted to kill init!");
732 zap_pid_ns_processes(pid_ns
);
733 write_lock_irq(&tasklist_lock
);
735 * We can not clear ->child_reaper or leave it alone.
736 * There may by stealth EXIT_DEAD tasks on ->children,
737 * forget_original_parent() must move them somewhere.
739 pid_ns
->child_reaper
= init_pid_ns
.child_reaper
;
742 return pid_ns
->child_reaper
;
746 * Any that need to be release_task'd are put on the @dead list.
748 static void reparent_thread(struct task_struct
*father
, struct task_struct
*p
,
749 struct list_head
*dead
)
751 if (p
->pdeath_signal
)
752 group_send_sig_info(p
->pdeath_signal
, SEND_SIG_NOINFO
, p
);
754 list_move_tail(&p
->sibling
, &p
->real_parent
->children
);
756 if (task_detached(p
))
759 * If this is a threaded reparent there is no need to
760 * notify anyone anything has happened.
762 if (same_thread_group(p
->real_parent
, father
))
765 /* We don't want people slaying init. */
766 p
->exit_signal
= SIGCHLD
;
768 /* If it has exited notify the new parent about this child's death. */
770 p
->exit_state
== EXIT_ZOMBIE
&& thread_group_empty(p
)) {
771 do_notify_parent(p
, p
->exit_signal
);
772 if (task_detached(p
)) {
773 p
->exit_state
= EXIT_DEAD
;
774 list_move_tail(&p
->sibling
, dead
);
778 kill_orphaned_pgrp(p
, father
);
781 static void forget_original_parent(struct task_struct
*father
)
783 struct task_struct
*p
, *n
, *reaper
;
784 LIST_HEAD(dead_children
);
788 write_lock_irq(&tasklist_lock
);
789 reaper
= find_new_reaper(father
);
791 list_for_each_entry_safe(p
, n
, &father
->children
, sibling
) {
792 p
->real_parent
= reaper
;
793 if (p
->parent
== father
) {
795 p
->parent
= p
->real_parent
;
797 reparent_thread(father
, p
, &dead_children
);
799 write_unlock_irq(&tasklist_lock
);
801 BUG_ON(!list_empty(&father
->children
));
803 list_for_each_entry_safe(p
, n
, &dead_children
, sibling
) {
804 list_del_init(&p
->sibling
);
810 * Send signals to all our closest relatives so that they know
811 * to properly mourn us..
813 static void exit_notify(struct task_struct
*tsk
, int group_dead
)
819 * This does two things:
821 * A. Make init inherit all the child processes
822 * B. Check to see if any process groups have become orphaned
823 * as a result of our exiting, and if they have any stopped
824 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
826 forget_original_parent(tsk
);
827 exit_task_namespaces(tsk
);
829 write_lock_irq(&tasklist_lock
);
831 kill_orphaned_pgrp(tsk
->group_leader
, NULL
);
833 /* Let father know we died
835 * Thread signals are configurable, but you aren't going to use
836 * that to send signals to arbitary processes.
837 * That stops right now.
839 * If the parent exec id doesn't match the exec id we saved
840 * when we started then we know the parent has changed security
843 * If our self_exec id doesn't match our parent_exec_id then
844 * we have changed execution domain as these two values started
845 * the same after a fork.
847 if (tsk
->exit_signal
!= SIGCHLD
&& !task_detached(tsk
) &&
848 (tsk
->parent_exec_id
!= tsk
->real_parent
->self_exec_id
||
849 tsk
->self_exec_id
!= tsk
->parent_exec_id
))
850 tsk
->exit_signal
= SIGCHLD
;
852 signal
= tracehook_notify_death(tsk
, &cookie
, group_dead
);
854 signal
= do_notify_parent(tsk
, signal
);
856 tsk
->exit_state
= signal
== DEATH_REAP
? EXIT_DEAD
: EXIT_ZOMBIE
;
858 /* mt-exec, de_thread() is waiting for us */
859 if (thread_group_leader(tsk
) &&
860 tsk
->signal
->group_exit_task
&&
861 tsk
->signal
->notify_count
< 0)
862 wake_up_process(tsk
->signal
->group_exit_task
);
864 write_unlock_irq(&tasklist_lock
);
866 tracehook_report_death(tsk
, signal
, cookie
, group_dead
);
868 /* If the process is dead, release it - nobody will wait for it */
869 if (signal
== DEATH_REAP
)
873 #ifdef CONFIG_DEBUG_STACK_USAGE
874 static void check_stack_usage(void)
876 static DEFINE_SPINLOCK(low_water_lock
);
877 static int lowest_to_date
= THREAD_SIZE
;
880 free
= stack_not_used(current
);
882 if (free
>= lowest_to_date
)
885 spin_lock(&low_water_lock
);
886 if (free
< lowest_to_date
) {
887 printk(KERN_WARNING
"%s used greatest stack depth: %lu bytes "
889 current
->comm
, free
);
890 lowest_to_date
= free
;
892 spin_unlock(&low_water_lock
);
895 static inline void check_stack_usage(void) {}
898 NORET_TYPE
void do_exit(long code
)
900 struct task_struct
*tsk
= current
;
903 profile_task_exit(tsk
);
905 WARN_ON(atomic_read(&tsk
->fs_excl
));
907 if (unlikely(in_interrupt()))
908 panic("Aiee, killing interrupt handler!");
909 if (unlikely(!tsk
->pid
))
910 panic("Attempted to kill the idle task!");
912 tracehook_report_exit(&code
);
915 * We're taking recursive faults here in do_exit. Safest is to just
916 * leave this task alone and wait for reboot.
918 if (unlikely(tsk
->flags
& PF_EXITING
)) {
920 "Fixing recursive fault but reboot is needed!\n");
922 * We can do this unlocked here. The futex code uses
923 * this flag just to verify whether the pi state
924 * cleanup has been done or not. In the worst case it
925 * loops once more. We pretend that the cleanup was
926 * done as there is no way to return. Either the
927 * OWNER_DIED bit is set by now or we push the blocked
928 * task into the wait for ever nirwana as well.
930 tsk
->flags
|= PF_EXITPIDONE
;
931 set_current_state(TASK_UNINTERRUPTIBLE
);
937 exit_signals(tsk
); /* sets PF_EXITING */
939 * tsk->flags are checked in the futex code to protect against
940 * an exiting task cleaning up the robust pi futexes.
943 spin_unlock_wait(&tsk
->pi_lock
);
945 if (unlikely(in_atomic()))
946 printk(KERN_INFO
"note: %s[%d] exited with preempt_count %d\n",
947 current
->comm
, task_pid_nr(current
),
950 acct_update_integrals(tsk
);
952 group_dead
= atomic_dec_and_test(&tsk
->signal
->live
);
954 hrtimer_cancel(&tsk
->signal
->real_timer
);
955 exit_itimers(tsk
->signal
);
957 acct_collect(code
, group_dead
);
960 if (unlikely(tsk
->audit_context
))
963 tsk
->exit_code
= code
;
964 taskstats_exit(tsk
, group_dead
);
970 trace_sched_process_exit(tsk
);
979 if (group_dead
&& tsk
->signal
->leader
)
980 disassociate_ctty(1);
982 module_put(task_thread_info(tsk
)->exec_domain
->module
);
984 module_put(tsk
->binfmt
->module
);
986 proc_exit_connector(tsk
);
987 exit_notify(tsk
, group_dead
);
989 mpol_put(tsk
->mempolicy
);
990 tsk
->mempolicy
= NULL
;
993 if (unlikely(!list_empty(&tsk
->pi_state_list
)))
994 exit_pi_state_list(tsk
);
995 if (unlikely(current
->pi_state_cache
))
996 kfree(current
->pi_state_cache
);
999 * Make sure we are holding no locks:
1001 debug_check_no_locks_held(tsk
);
1003 * We can do this unlocked here. The futex code uses this flag
1004 * just to verify whether the pi state cleanup has been done
1005 * or not. In the worst case it loops once more.
1007 tsk
->flags
|= PF_EXITPIDONE
;
1009 if (tsk
->io_context
)
1012 if (tsk
->splice_pipe
)
1013 __free_pipe_info(tsk
->splice_pipe
);
1016 /* causes final put_task_struct in finish_task_switch(). */
1017 tsk
->state
= TASK_DEAD
;
1020 /* Avoid "noreturn function does return". */
1022 cpu_relax(); /* For when BUG is null */
1025 EXPORT_SYMBOL_GPL(do_exit
);
1027 NORET_TYPE
void complete_and_exit(struct completion
*comp
, long code
)
1035 EXPORT_SYMBOL(complete_and_exit
);
1037 SYSCALL_DEFINE1(exit
, int, error_code
)
1039 do_exit((error_code
&0xff)<<8);
1043 * Take down every thread in the group. This is called by fatal signals
1044 * as well as by sys_exit_group (below).
1047 do_group_exit(int exit_code
)
1049 struct signal_struct
*sig
= current
->signal
;
1051 BUG_ON(exit_code
& 0x80); /* core dumps don't get here */
1053 if (signal_group_exit(sig
))
1054 exit_code
= sig
->group_exit_code
;
1055 else if (!thread_group_empty(current
)) {
1056 struct sighand_struct
*const sighand
= current
->sighand
;
1057 spin_lock_irq(&sighand
->siglock
);
1058 if (signal_group_exit(sig
))
1059 /* Another thread got here before we took the lock. */
1060 exit_code
= sig
->group_exit_code
;
1062 sig
->group_exit_code
= exit_code
;
1063 sig
->flags
= SIGNAL_GROUP_EXIT
;
1064 zap_other_threads(current
);
1066 spin_unlock_irq(&sighand
->siglock
);
1074 * this kills every thread in the thread group. Note that any externally
1075 * wait4()-ing process will get the correct exit code - even if this
1076 * thread is not the thread group leader.
1078 SYSCALL_DEFINE1(exit_group
, int, error_code
)
1080 do_group_exit((error_code
& 0xff) << 8);
1085 static struct pid
*task_pid_type(struct task_struct
*task
, enum pid_type type
)
1087 struct pid
*pid
= NULL
;
1088 if (type
== PIDTYPE_PID
)
1089 pid
= task
->pids
[type
].pid
;
1090 else if (type
< PIDTYPE_MAX
)
1091 pid
= task
->group_leader
->pids
[type
].pid
;
1095 static int eligible_child(enum pid_type type
, struct pid
*pid
, int options
,
1096 struct task_struct
*p
)
1100 if (type
< PIDTYPE_MAX
) {
1101 if (task_pid_type(p
, type
) != pid
)
1105 /* Wait for all children (clone and not) if __WALL is set;
1106 * otherwise, wait for clone children *only* if __WCLONE is
1107 * set; otherwise, wait for non-clone children *only*. (Note:
1108 * A "clone" child here is one that reports to its parent
1109 * using a signal other than SIGCHLD.) */
1110 if (((p
->exit_signal
!= SIGCHLD
) ^ ((options
& __WCLONE
) != 0))
1111 && !(options
& __WALL
))
1114 err
= security_task_wait(p
);
1121 static int wait_noreap_copyout(struct task_struct
*p
, pid_t pid
, uid_t uid
,
1122 int why
, int status
,
1123 struct siginfo __user
*infop
,
1124 struct rusage __user
*rusagep
)
1126 int retval
= rusagep
? getrusage(p
, RUSAGE_BOTH
, rusagep
) : 0;
1130 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1132 retval
= put_user(0, &infop
->si_errno
);
1134 retval
= put_user((short)why
, &infop
->si_code
);
1136 retval
= put_user(pid
, &infop
->si_pid
);
1138 retval
= put_user(uid
, &infop
->si_uid
);
1140 retval
= put_user(status
, &infop
->si_status
);
1147 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1148 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1149 * the lock and this task is uninteresting. If we return nonzero, we have
1150 * released the lock and the system call should return.
1152 static int wait_task_zombie(struct task_struct
*p
, int options
,
1153 struct siginfo __user
*infop
,
1154 int __user
*stat_addr
, struct rusage __user
*ru
)
1156 unsigned long state
;
1157 int retval
, status
, traced
;
1158 pid_t pid
= task_pid_vnr(p
);
1159 uid_t uid
= __task_cred(p
)->uid
;
1161 if (!likely(options
& WEXITED
))
1164 if (unlikely(options
& WNOWAIT
)) {
1165 int exit_code
= p
->exit_code
;
1169 read_unlock(&tasklist_lock
);
1170 if ((exit_code
& 0x7f) == 0) {
1172 status
= exit_code
>> 8;
1174 why
= (exit_code
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1175 status
= exit_code
& 0x7f;
1177 return wait_noreap_copyout(p
, pid
, uid
, why
,
1182 * Try to move the task's state to DEAD
1183 * only one thread is allowed to do this:
1185 state
= xchg(&p
->exit_state
, EXIT_DEAD
);
1186 if (state
!= EXIT_ZOMBIE
) {
1187 BUG_ON(state
!= EXIT_DEAD
);
1191 traced
= ptrace_reparented(p
);
1193 if (likely(!traced
)) {
1194 struct signal_struct
*psig
;
1195 struct signal_struct
*sig
;
1196 struct task_cputime cputime
;
1199 * The resource counters for the group leader are in its
1200 * own task_struct. Those for dead threads in the group
1201 * are in its signal_struct, as are those for the child
1202 * processes it has previously reaped. All these
1203 * accumulate in the parent's signal_struct c* fields.
1205 * We don't bother to take a lock here to protect these
1206 * p->signal fields, because they are only touched by
1207 * __exit_signal, which runs with tasklist_lock
1208 * write-locked anyway, and so is excluded here. We do
1209 * need to protect the access to p->parent->signal fields,
1210 * as other threads in the parent group can be right
1211 * here reaping other children at the same time.
1213 * We use thread_group_cputime() to get times for the thread
1214 * group, which consolidates times for all threads in the
1215 * group including the group leader.
1217 thread_group_cputime(p
, &cputime
);
1218 spin_lock_irq(&p
->parent
->sighand
->siglock
);
1219 psig
= p
->parent
->signal
;
1222 cputime_add(psig
->cutime
,
1223 cputime_add(cputime
.utime
,
1226 cputime_add(psig
->cstime
,
1227 cputime_add(cputime
.stime
,
1230 cputime_add(psig
->cgtime
,
1231 cputime_add(p
->gtime
,
1232 cputime_add(sig
->gtime
,
1235 p
->min_flt
+ sig
->min_flt
+ sig
->cmin_flt
;
1237 p
->maj_flt
+ sig
->maj_flt
+ sig
->cmaj_flt
;
1239 p
->nvcsw
+ sig
->nvcsw
+ sig
->cnvcsw
;
1241 p
->nivcsw
+ sig
->nivcsw
+ sig
->cnivcsw
;
1243 task_io_get_inblock(p
) +
1244 sig
->inblock
+ sig
->cinblock
;
1246 task_io_get_oublock(p
) +
1247 sig
->oublock
+ sig
->coublock
;
1248 task_io_accounting_add(&psig
->ioac
, &p
->ioac
);
1249 task_io_accounting_add(&psig
->ioac
, &sig
->ioac
);
1250 spin_unlock_irq(&p
->parent
->sighand
->siglock
);
1254 * Now we are sure this task is interesting, and no other
1255 * thread can reap it because we set its state to EXIT_DEAD.
1257 read_unlock(&tasklist_lock
);
1260 * Flush inherited counters to the parent - before the parent
1261 * gets woken up by child-exit notifications.
1263 perf_counter_exit_task(p
);
1265 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1266 status
= (p
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1267 ? p
->signal
->group_exit_code
: p
->exit_code
;
1268 if (!retval
&& stat_addr
)
1269 retval
= put_user(status
, stat_addr
);
1270 if (!retval
&& infop
)
1271 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1272 if (!retval
&& infop
)
1273 retval
= put_user(0, &infop
->si_errno
);
1274 if (!retval
&& infop
) {
1277 if ((status
& 0x7f) == 0) {
1281 why
= (status
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1284 retval
= put_user((short)why
, &infop
->si_code
);
1286 retval
= put_user(status
, &infop
->si_status
);
1288 if (!retval
&& infop
)
1289 retval
= put_user(pid
, &infop
->si_pid
);
1290 if (!retval
&& infop
)
1291 retval
= put_user(uid
, &infop
->si_uid
);
1296 write_lock_irq(&tasklist_lock
);
1297 /* We dropped tasklist, ptracer could die and untrace */
1300 * If this is not a detached task, notify the parent.
1301 * If it's still not detached after that, don't release
1304 if (!task_detached(p
)) {
1305 do_notify_parent(p
, p
->exit_signal
);
1306 if (!task_detached(p
)) {
1307 p
->exit_state
= EXIT_ZOMBIE
;
1311 write_unlock_irq(&tasklist_lock
);
1319 static int *task_stopped_code(struct task_struct
*p
, bool ptrace
)
1322 if (task_is_stopped_or_traced(p
))
1323 return &p
->exit_code
;
1325 if (p
->signal
->flags
& SIGNAL_STOP_STOPPED
)
1326 return &p
->signal
->group_exit_code
;
1332 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1333 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1334 * the lock and this task is uninteresting. If we return nonzero, we have
1335 * released the lock and the system call should return.
1337 static int wait_task_stopped(int ptrace
, struct task_struct
*p
,
1338 int options
, struct siginfo __user
*infop
,
1339 int __user
*stat_addr
, struct rusage __user
*ru
)
1341 int retval
, exit_code
, *p_code
, why
;
1342 uid_t uid
= 0; /* unneeded, required by compiler */
1345 if (!(options
& WUNTRACED
))
1349 spin_lock_irq(&p
->sighand
->siglock
);
1351 p_code
= task_stopped_code(p
, ptrace
);
1352 if (unlikely(!p_code
))
1355 exit_code
= *p_code
;
1359 if (!unlikely(options
& WNOWAIT
))
1362 /* don't need the RCU readlock here as we're holding a spinlock */
1363 uid
= __task_cred(p
)->uid
;
1365 spin_unlock_irq(&p
->sighand
->siglock
);
1370 * Now we are pretty sure this task is interesting.
1371 * Make sure it doesn't get reaped out from under us while we
1372 * give up the lock and then examine it below. We don't want to
1373 * keep holding onto the tasklist_lock while we call getrusage and
1374 * possibly take page faults for user memory.
1377 pid
= task_pid_vnr(p
);
1378 why
= ptrace
? CLD_TRAPPED
: CLD_STOPPED
;
1379 read_unlock(&tasklist_lock
);
1381 if (unlikely(options
& WNOWAIT
))
1382 return wait_noreap_copyout(p
, pid
, uid
,
1386 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1387 if (!retval
&& stat_addr
)
1388 retval
= put_user((exit_code
<< 8) | 0x7f, stat_addr
);
1389 if (!retval
&& infop
)
1390 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1391 if (!retval
&& infop
)
1392 retval
= put_user(0, &infop
->si_errno
);
1393 if (!retval
&& infop
)
1394 retval
= put_user((short)why
, &infop
->si_code
);
1395 if (!retval
&& infop
)
1396 retval
= put_user(exit_code
, &infop
->si_status
);
1397 if (!retval
&& infop
)
1398 retval
= put_user(pid
, &infop
->si_pid
);
1399 if (!retval
&& infop
)
1400 retval
= put_user(uid
, &infop
->si_uid
);
1410 * Handle do_wait work for one task in a live, non-stopped state.
1411 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1412 * the lock and this task is uninteresting. If we return nonzero, we have
1413 * released the lock and the system call should return.
1415 static int wait_task_continued(struct task_struct
*p
, int options
,
1416 struct siginfo __user
*infop
,
1417 int __user
*stat_addr
, struct rusage __user
*ru
)
1423 if (!unlikely(options
& WCONTINUED
))
1426 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
))
1429 spin_lock_irq(&p
->sighand
->siglock
);
1430 /* Re-check with the lock held. */
1431 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
)) {
1432 spin_unlock_irq(&p
->sighand
->siglock
);
1435 if (!unlikely(options
& WNOWAIT
))
1436 p
->signal
->flags
&= ~SIGNAL_STOP_CONTINUED
;
1437 uid
= __task_cred(p
)->uid
;
1438 spin_unlock_irq(&p
->sighand
->siglock
);
1440 pid
= task_pid_vnr(p
);
1442 read_unlock(&tasklist_lock
);
1445 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1447 if (!retval
&& stat_addr
)
1448 retval
= put_user(0xffff, stat_addr
);
1452 retval
= wait_noreap_copyout(p
, pid
, uid
,
1453 CLD_CONTINUED
, SIGCONT
,
1455 BUG_ON(retval
== 0);
1462 * Consider @p for a wait by @parent.
1464 * -ECHILD should be in *@notask_error before the first call.
1465 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1466 * Returns zero if the search for a child should continue;
1467 * then *@notask_error is 0 if @p is an eligible child,
1468 * or another error from security_task_wait(), or still -ECHILD.
1470 static int wait_consider_task(struct task_struct
*parent
, int ptrace
,
1471 struct task_struct
*p
, int *notask_error
,
1472 enum pid_type type
, struct pid
*pid
, int options
,
1473 struct siginfo __user
*infop
,
1474 int __user
*stat_addr
, struct rusage __user
*ru
)
1476 int ret
= eligible_child(type
, pid
, options
, p
);
1480 if (unlikely(ret
< 0)) {
1482 * If we have not yet seen any eligible child,
1483 * then let this error code replace -ECHILD.
1484 * A permission error will give the user a clue
1485 * to look for security policy problems, rather
1486 * than for mysterious wait bugs.
1489 *notask_error
= ret
;
1492 if (likely(!ptrace
) && unlikely(p
->ptrace
)) {
1494 * This child is hidden by ptrace.
1495 * We aren't allowed to see it now, but eventually we will.
1501 if (p
->exit_state
== EXIT_DEAD
)
1505 * We don't reap group leaders with subthreads.
1507 if (p
->exit_state
== EXIT_ZOMBIE
&& !delay_group_leader(p
))
1508 return wait_task_zombie(p
, options
, infop
, stat_addr
, ru
);
1511 * It's stopped or running now, so it might
1512 * later continue, exit, or stop again.
1516 if (task_stopped_code(p
, ptrace
))
1517 return wait_task_stopped(ptrace
, p
, options
,
1518 infop
, stat_addr
, ru
);
1520 return wait_task_continued(p
, options
, infop
, stat_addr
, ru
);
1524 * Do the work of do_wait() for one thread in the group, @tsk.
1526 * -ECHILD should be in *@notask_error before the first call.
1527 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1528 * Returns zero if the search for a child should continue; then
1529 * *@notask_error is 0 if there were any eligible children,
1530 * or another error from security_task_wait(), or still -ECHILD.
1532 static int do_wait_thread(struct task_struct
*tsk
, int *notask_error
,
1533 enum pid_type type
, struct pid
*pid
, int options
,
1534 struct siginfo __user
*infop
, int __user
*stat_addr
,
1535 struct rusage __user
*ru
)
1537 struct task_struct
*p
;
1539 list_for_each_entry(p
, &tsk
->children
, sibling
) {
1541 * Do not consider detached threads.
1543 if (!task_detached(p
)) {
1544 int ret
= wait_consider_task(tsk
, 0, p
, notask_error
,
1546 infop
, stat_addr
, ru
);
1555 static int ptrace_do_wait(struct task_struct
*tsk
, int *notask_error
,
1556 enum pid_type type
, struct pid
*pid
, int options
,
1557 struct siginfo __user
*infop
, int __user
*stat_addr
,
1558 struct rusage __user
*ru
)
1560 struct task_struct
*p
;
1563 * Traditionally we see ptrace'd stopped tasks regardless of options.
1565 options
|= WUNTRACED
;
1567 list_for_each_entry(p
, &tsk
->ptraced
, ptrace_entry
) {
1568 int ret
= wait_consider_task(tsk
, 1, p
, notask_error
,
1570 infop
, stat_addr
, ru
);
1578 static long do_wait(enum pid_type type
, struct pid
*pid
, int options
,
1579 struct siginfo __user
*infop
, int __user
*stat_addr
,
1580 struct rusage __user
*ru
)
1582 DECLARE_WAITQUEUE(wait
, current
);
1583 struct task_struct
*tsk
;
1586 trace_sched_process_wait(pid
);
1588 add_wait_queue(¤t
->signal
->wait_chldexit
,&wait
);
1591 * If there is nothing that can match our critiera just get out.
1592 * We will clear @retval to zero if we see any child that might later
1593 * match our criteria, even if we are not able to reap it yet.
1596 if ((type
< PIDTYPE_MAX
) && (!pid
|| hlist_empty(&pid
->tasks
[type
])))
1599 current
->state
= TASK_INTERRUPTIBLE
;
1600 read_lock(&tasklist_lock
);
1603 int tsk_result
= do_wait_thread(tsk
, &retval
,
1605 infop
, stat_addr
, ru
);
1607 tsk_result
= ptrace_do_wait(tsk
, &retval
,
1609 infop
, stat_addr
, ru
);
1612 * tasklist_lock is unlocked and we have a final result.
1614 retval
= tsk_result
;
1618 if (options
& __WNOTHREAD
)
1620 tsk
= next_thread(tsk
);
1621 BUG_ON(tsk
->signal
!= current
->signal
);
1622 } while (tsk
!= current
);
1623 read_unlock(&tasklist_lock
);
1625 if (!retval
&& !(options
& WNOHANG
)) {
1626 retval
= -ERESTARTSYS
;
1627 if (!signal_pending(current
)) {
1634 current
->state
= TASK_RUNNING
;
1635 remove_wait_queue(¤t
->signal
->wait_chldexit
,&wait
);
1641 * For a WNOHANG return, clear out all the fields
1642 * we would set so the user can easily tell the
1646 retval
= put_user(0, &infop
->si_signo
);
1648 retval
= put_user(0, &infop
->si_errno
);
1650 retval
= put_user(0, &infop
->si_code
);
1652 retval
= put_user(0, &infop
->si_pid
);
1654 retval
= put_user(0, &infop
->si_uid
);
1656 retval
= put_user(0, &infop
->si_status
);
1662 SYSCALL_DEFINE5(waitid
, int, which
, pid_t
, upid
, struct siginfo __user
*,
1663 infop
, int, options
, struct rusage __user
*, ru
)
1665 struct pid
*pid
= NULL
;
1669 if (options
& ~(WNOHANG
|WNOWAIT
|WEXITED
|WSTOPPED
|WCONTINUED
))
1671 if (!(options
& (WEXITED
|WSTOPPED
|WCONTINUED
)))
1684 type
= PIDTYPE_PGID
;
1692 if (type
< PIDTYPE_MAX
)
1693 pid
= find_get_pid(upid
);
1694 ret
= do_wait(type
, pid
, options
, infop
, NULL
, ru
);
1697 /* avoid REGPARM breakage on x86: */
1698 asmlinkage_protect(5, ret
, which
, upid
, infop
, options
, ru
);
1702 SYSCALL_DEFINE4(wait4
, pid_t
, upid
, int __user
*, stat_addr
,
1703 int, options
, struct rusage __user
*, ru
)
1705 struct pid
*pid
= NULL
;
1709 if (options
& ~(WNOHANG
|WUNTRACED
|WCONTINUED
|
1710 __WNOTHREAD
|__WCLONE
|__WALL
))
1715 else if (upid
< 0) {
1716 type
= PIDTYPE_PGID
;
1717 pid
= find_get_pid(-upid
);
1718 } else if (upid
== 0) {
1719 type
= PIDTYPE_PGID
;
1720 pid
= get_task_pid(current
, PIDTYPE_PGID
);
1721 } else /* upid > 0 */ {
1723 pid
= find_get_pid(upid
);
1726 ret
= do_wait(type
, pid
, options
| WEXITED
, NULL
, stat_addr
, ru
);
1729 /* avoid REGPARM breakage on x86: */
1730 asmlinkage_protect(4, ret
, upid
, stat_addr
, options
, ru
);
1734 #ifdef __ARCH_WANT_SYS_WAITPID
1737 * sys_waitpid() remains for compatibility. waitpid() should be
1738 * implemented by calling sys_wait4() from libc.a.
1740 SYSCALL_DEFINE3(waitpid
, pid_t
, pid
, int __user
*, stat_addr
, int, options
)
1742 return sys_wait4(pid
, stat_addr
, options
, NULL
);