Merge commit 'v3.3-rc1' into stable/for-linus-fixes-3.3
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / kernel / exit.c
blob294b1709170d1e890c80d064c65d38c1b12a88c1
1 /*
2 * linux/kernel/exit.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
7 #include <linux/mm.h>
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/binfmts.h>
24 #include <linux/nsproxy.h>
25 #include <linux/pid_namespace.h>
26 #include <linux/ptrace.h>
27 #include <linux/profile.h>
28 #include <linux/mount.h>
29 #include <linux/proc_fs.h>
30 #include <linux/kthread.h>
31 #include <linux/mempolicy.h>
32 #include <linux/taskstats_kern.h>
33 #include <linux/delayacct.h>
34 #include <linux/freezer.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>
56 #include <asm/uaccess.h>
57 #include <asm/unistd.h>
58 #include <asm/pgtable.h>
59 #include <asm/mmu_context.h>
61 static void exit_mm(struct task_struct * tsk);
63 static void __unhash_process(struct task_struct *p, bool group_dead)
65 nr_threads--;
66 detach_pid(p, PIDTYPE_PID);
67 if (group_dead) {
68 detach_pid(p, PIDTYPE_PGID);
69 detach_pid(p, PIDTYPE_SID);
71 list_del_rcu(&p->tasks);
72 list_del_init(&p->sibling);
73 __this_cpu_dec(process_counts);
75 list_del_rcu(&p->thread_group);
79 * This function expects the tasklist_lock write-locked.
81 static void __exit_signal(struct task_struct *tsk)
83 struct signal_struct *sig = tsk->signal;
84 bool group_dead = thread_group_leader(tsk);
85 struct sighand_struct *sighand;
86 struct tty_struct *uninitialized_var(tty);
88 sighand = rcu_dereference_check(tsk->sighand,
89 lockdep_tasklist_lock_is_held());
90 spin_lock(&sighand->siglock);
92 posix_cpu_timers_exit(tsk);
93 if (group_dead) {
94 posix_cpu_timers_exit_group(tsk);
95 tty = sig->tty;
96 sig->tty = NULL;
97 } else {
99 * This can only happen if the caller is de_thread().
100 * FIXME: this is the temporary hack, we should teach
101 * posix-cpu-timers to handle this case correctly.
103 if (unlikely(has_group_leader_pid(tsk)))
104 posix_cpu_timers_exit_group(tsk);
107 * If there is any task waiting for the group exit
108 * then notify it:
110 if (sig->notify_count > 0 && !--sig->notify_count)
111 wake_up_process(sig->group_exit_task);
113 if (tsk == sig->curr_target)
114 sig->curr_target = next_thread(tsk);
116 * Accumulate here the counters for all threads but the
117 * group leader as they die, so they can be added into
118 * the process-wide totals when those are taken.
119 * The group leader stays around as a zombie as long
120 * as there are other threads. When it gets reaped,
121 * the exit.c code will add its counts into these totals.
122 * We won't ever get here for the group leader, since it
123 * will have been the last reference on the signal_struct.
125 sig->utime += tsk->utime;
126 sig->stime += tsk->stime;
127 sig->gtime += tsk->gtime;
128 sig->min_flt += tsk->min_flt;
129 sig->maj_flt += tsk->maj_flt;
130 sig->nvcsw += tsk->nvcsw;
131 sig->nivcsw += tsk->nivcsw;
132 sig->inblock += task_io_get_inblock(tsk);
133 sig->oublock += task_io_get_oublock(tsk);
134 task_io_accounting_add(&sig->ioac, &tsk->ioac);
135 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
138 sig->nr_threads--;
139 __unhash_process(tsk, group_dead);
142 * Do this under ->siglock, we can race with another thread
143 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
145 flush_sigqueue(&tsk->pending);
146 tsk->sighand = NULL;
147 spin_unlock(&sighand->siglock);
149 __cleanup_sighand(sighand);
150 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
151 if (group_dead) {
152 flush_sigqueue(&sig->shared_pending);
153 tty_kref_put(tty);
157 static void delayed_put_task_struct(struct rcu_head *rhp)
159 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
161 perf_event_delayed_put(tsk);
162 trace_sched_process_free(tsk);
163 put_task_struct(tsk);
167 void release_task(struct task_struct * p)
169 struct task_struct *leader;
170 int zap_leader;
171 repeat:
172 /* don't need to get the RCU readlock here - the process is dead and
173 * can't be modifying its own credentials. But shut RCU-lockdep up */
174 rcu_read_lock();
175 atomic_dec(&__task_cred(p)->user->processes);
176 rcu_read_unlock();
178 proc_flush_task(p);
180 write_lock_irq(&tasklist_lock);
181 ptrace_release_task(p);
182 __exit_signal(p);
185 * If we are the last non-leader member of the thread
186 * group, and the leader is zombie, then notify the
187 * group leader's parent process. (if it wants notification.)
189 zap_leader = 0;
190 leader = p->group_leader;
191 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
193 * If we were the last child thread and the leader has
194 * exited already, and the leader's parent ignores SIGCHLD,
195 * then we are the one who should release the leader.
197 zap_leader = do_notify_parent(leader, leader->exit_signal);
198 if (zap_leader)
199 leader->exit_state = EXIT_DEAD;
202 write_unlock_irq(&tasklist_lock);
203 release_thread(p);
204 call_rcu(&p->rcu, delayed_put_task_struct);
206 p = leader;
207 if (unlikely(zap_leader))
208 goto repeat;
212 * This checks not only the pgrp, but falls back on the pid if no
213 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
214 * without this...
216 * The caller must hold rcu lock or the tasklist lock.
218 struct pid *session_of_pgrp(struct pid *pgrp)
220 struct task_struct *p;
221 struct pid *sid = NULL;
223 p = pid_task(pgrp, PIDTYPE_PGID);
224 if (p == NULL)
225 p = pid_task(pgrp, PIDTYPE_PID);
226 if (p != NULL)
227 sid = task_session(p);
229 return sid;
233 * Determine if a process group is "orphaned", according to the POSIX
234 * definition in 2.2.2.52. Orphaned process groups are not to be affected
235 * by terminal-generated stop signals. Newly orphaned process groups are
236 * to receive a SIGHUP and a SIGCONT.
238 * "I ask you, have you ever known what it is to be an orphan?"
240 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
242 struct task_struct *p;
244 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
245 if ((p == ignored_task) ||
246 (p->exit_state && thread_group_empty(p)) ||
247 is_global_init(p->real_parent))
248 continue;
250 if (task_pgrp(p->real_parent) != pgrp &&
251 task_session(p->real_parent) == task_session(p))
252 return 0;
253 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
255 return 1;
258 int is_current_pgrp_orphaned(void)
260 int retval;
262 read_lock(&tasklist_lock);
263 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
264 read_unlock(&tasklist_lock);
266 return retval;
269 static bool has_stopped_jobs(struct pid *pgrp)
271 struct task_struct *p;
273 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
274 if (p->signal->flags & SIGNAL_STOP_STOPPED)
275 return true;
276 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
278 return false;
282 * Check to see if any process groups have become orphaned as
283 * a result of our exiting, and if they have any stopped jobs,
284 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
286 static void
287 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
289 struct pid *pgrp = task_pgrp(tsk);
290 struct task_struct *ignored_task = tsk;
292 if (!parent)
293 /* exit: our father is in a different pgrp than
294 * we are and we were the only connection outside.
296 parent = tsk->real_parent;
297 else
298 /* reparent: our child is in a different pgrp than
299 * we are, and it was the only connection outside.
301 ignored_task = NULL;
303 if (task_pgrp(parent) != pgrp &&
304 task_session(parent) == task_session(tsk) &&
305 will_become_orphaned_pgrp(pgrp, ignored_task) &&
306 has_stopped_jobs(pgrp)) {
307 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
308 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
313 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
315 * If a kernel thread is launched as a result of a system call, or if
316 * it ever exits, it should generally reparent itself to kthreadd so it
317 * isn't in the way of other processes and is correctly cleaned up on exit.
319 * The various task state such as scheduling policy and priority may have
320 * been inherited from a user process, so we reset them to sane values here.
322 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
324 static void reparent_to_kthreadd(void)
326 write_lock_irq(&tasklist_lock);
328 ptrace_unlink(current);
329 /* Reparent to init */
330 current->real_parent = current->parent = kthreadd_task;
331 list_move_tail(&current->sibling, &current->real_parent->children);
333 /* Set the exit signal to SIGCHLD so we signal init on exit */
334 current->exit_signal = SIGCHLD;
336 if (task_nice(current) < 0)
337 set_user_nice(current, 0);
338 /* cpus_allowed? */
339 /* rt_priority? */
340 /* signals? */
341 memcpy(current->signal->rlim, init_task.signal->rlim,
342 sizeof(current->signal->rlim));
344 atomic_inc(&init_cred.usage);
345 commit_creds(&init_cred);
346 write_unlock_irq(&tasklist_lock);
349 void __set_special_pids(struct pid *pid)
351 struct task_struct *curr = current->group_leader;
353 if (task_session(curr) != pid)
354 change_pid(curr, PIDTYPE_SID, pid);
356 if (task_pgrp(curr) != pid)
357 change_pid(curr, PIDTYPE_PGID, pid);
360 static void set_special_pids(struct pid *pid)
362 write_lock_irq(&tasklist_lock);
363 __set_special_pids(pid);
364 write_unlock_irq(&tasklist_lock);
368 * Let kernel threads use this to say that they allow a certain signal.
369 * Must not be used if kthread was cloned with CLONE_SIGHAND.
371 int allow_signal(int sig)
373 if (!valid_signal(sig) || sig < 1)
374 return -EINVAL;
376 spin_lock_irq(&current->sighand->siglock);
377 /* This is only needed for daemonize()'ed kthreads */
378 sigdelset(&current->blocked, sig);
380 * Kernel threads handle their own signals. Let the signal code
381 * know it'll be handled, so that they don't get converted to
382 * SIGKILL or just silently dropped.
384 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
385 recalc_sigpending();
386 spin_unlock_irq(&current->sighand->siglock);
387 return 0;
390 EXPORT_SYMBOL(allow_signal);
392 int disallow_signal(int sig)
394 if (!valid_signal(sig) || sig < 1)
395 return -EINVAL;
397 spin_lock_irq(&current->sighand->siglock);
398 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
399 recalc_sigpending();
400 spin_unlock_irq(&current->sighand->siglock);
401 return 0;
404 EXPORT_SYMBOL(disallow_signal);
407 * Put all the gunge required to become a kernel thread without
408 * attached user resources in one place where it belongs.
411 void daemonize(const char *name, ...)
413 va_list args;
414 sigset_t blocked;
416 va_start(args, name);
417 vsnprintf(current->comm, sizeof(current->comm), name, args);
418 va_end(args);
421 * If we were started as result of loading a module, close all of the
422 * user space pages. We don't need them, and if we didn't close them
423 * they would be locked into memory.
425 exit_mm(current);
427 * We don't want to have TIF_FREEZE set if the system-wide hibernation
428 * or suspend transition begins right now.
430 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
432 if (current->nsproxy != &init_nsproxy) {
433 get_nsproxy(&init_nsproxy);
434 switch_task_namespaces(current, &init_nsproxy);
436 set_special_pids(&init_struct_pid);
437 proc_clear_tty(current);
439 /* Block and flush all signals */
440 sigfillset(&blocked);
441 sigprocmask(SIG_BLOCK, &blocked, NULL);
442 flush_signals(current);
444 /* Become as one with the init task */
446 daemonize_fs_struct();
447 exit_files(current);
448 current->files = init_task.files;
449 atomic_inc(&current->files->count);
451 reparent_to_kthreadd();
454 EXPORT_SYMBOL(daemonize);
456 static void close_files(struct files_struct * files)
458 int i, j;
459 struct fdtable *fdt;
461 j = 0;
464 * It is safe to dereference the fd table without RCU or
465 * ->file_lock because this is the last reference to the
466 * files structure. But use RCU to shut RCU-lockdep up.
468 rcu_read_lock();
469 fdt = files_fdtable(files);
470 rcu_read_unlock();
471 for (;;) {
472 unsigned long set;
473 i = j * __NFDBITS;
474 if (i >= fdt->max_fds)
475 break;
476 set = fdt->open_fds->fds_bits[j++];
477 while (set) {
478 if (set & 1) {
479 struct file * file = xchg(&fdt->fd[i], NULL);
480 if (file) {
481 filp_close(file, files);
482 cond_resched();
485 i++;
486 set >>= 1;
491 struct files_struct *get_files_struct(struct task_struct *task)
493 struct files_struct *files;
495 task_lock(task);
496 files = task->files;
497 if (files)
498 atomic_inc(&files->count);
499 task_unlock(task);
501 return files;
504 void put_files_struct(struct files_struct *files)
506 struct fdtable *fdt;
508 if (atomic_dec_and_test(&files->count)) {
509 close_files(files);
511 * Free the fd and fdset arrays if we expanded them.
512 * If the fdtable was embedded, pass files for freeing
513 * at the end of the RCU grace period. Otherwise,
514 * you can free files immediately.
516 rcu_read_lock();
517 fdt = files_fdtable(files);
518 if (fdt != &files->fdtab)
519 kmem_cache_free(files_cachep, files);
520 free_fdtable(fdt);
521 rcu_read_unlock();
525 void reset_files_struct(struct files_struct *files)
527 struct task_struct *tsk = current;
528 struct files_struct *old;
530 old = tsk->files;
531 task_lock(tsk);
532 tsk->files = files;
533 task_unlock(tsk);
534 put_files_struct(old);
537 void exit_files(struct task_struct *tsk)
539 struct files_struct * files = tsk->files;
541 if (files) {
542 task_lock(tsk);
543 tsk->files = NULL;
544 task_unlock(tsk);
545 put_files_struct(files);
549 #ifdef CONFIG_MM_OWNER
551 * A task is exiting. If it owned this mm, find a new owner for the mm.
553 void mm_update_next_owner(struct mm_struct *mm)
555 struct task_struct *c, *g, *p = current;
557 retry:
559 * If the exiting or execing task is not the owner, it's
560 * someone else's problem.
562 if (mm->owner != p)
563 return;
565 * The current owner is exiting/execing and there are no other
566 * candidates. Do not leave the mm pointing to a possibly
567 * freed task structure.
569 if (atomic_read(&mm->mm_users) <= 1) {
570 mm->owner = NULL;
571 return;
574 read_lock(&tasklist_lock);
576 * Search in the children
578 list_for_each_entry(c, &p->children, sibling) {
579 if (c->mm == mm)
580 goto assign_new_owner;
584 * Search in the siblings
586 list_for_each_entry(c, &p->real_parent->children, sibling) {
587 if (c->mm == mm)
588 goto assign_new_owner;
592 * Search through everything else. We should not get
593 * here often
595 do_each_thread(g, c) {
596 if (c->mm == mm)
597 goto assign_new_owner;
598 } while_each_thread(g, c);
600 read_unlock(&tasklist_lock);
602 * We found no owner yet mm_users > 1: this implies that we are
603 * most likely racing with swapoff (try_to_unuse()) or /proc or
604 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
606 mm->owner = NULL;
607 return;
609 assign_new_owner:
610 BUG_ON(c == p);
611 get_task_struct(c);
613 * The task_lock protects c->mm from changing.
614 * We always want mm->owner->mm == mm
616 task_lock(c);
618 * Delay read_unlock() till we have the task_lock()
619 * to ensure that c does not slip away underneath us
621 read_unlock(&tasklist_lock);
622 if (c->mm != mm) {
623 task_unlock(c);
624 put_task_struct(c);
625 goto retry;
627 mm->owner = c;
628 task_unlock(c);
629 put_task_struct(c);
631 #endif /* CONFIG_MM_OWNER */
634 * Turn us into a lazy TLB process if we
635 * aren't already..
637 static void exit_mm(struct task_struct * tsk)
639 struct mm_struct *mm = tsk->mm;
640 struct core_state *core_state;
642 mm_release(tsk, mm);
643 if (!mm)
644 return;
646 * Serialize with any possible pending coredump.
647 * We must hold mmap_sem around checking core_state
648 * and clearing tsk->mm. The core-inducing thread
649 * will increment ->nr_threads for each thread in the
650 * group with ->mm != NULL.
652 down_read(&mm->mmap_sem);
653 core_state = mm->core_state;
654 if (core_state) {
655 struct core_thread self;
656 up_read(&mm->mmap_sem);
658 self.task = tsk;
659 self.next = xchg(&core_state->dumper.next, &self);
661 * Implies mb(), the result of xchg() must be visible
662 * to core_state->dumper.
664 if (atomic_dec_and_test(&core_state->nr_threads))
665 complete(&core_state->startup);
667 for (;;) {
668 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
669 if (!self.task) /* see coredump_finish() */
670 break;
671 schedule();
673 __set_task_state(tsk, TASK_RUNNING);
674 down_read(&mm->mmap_sem);
676 atomic_inc(&mm->mm_count);
677 BUG_ON(mm != tsk->active_mm);
678 /* more a memory barrier than a real lock */
679 task_lock(tsk);
680 tsk->mm = NULL;
681 up_read(&mm->mmap_sem);
682 enter_lazy_tlb(mm, current);
683 task_unlock(tsk);
684 mm_update_next_owner(mm);
685 mmput(mm);
689 * When we die, we re-parent all our children.
690 * Try to give them to another thread in our thread
691 * group, and if no such member exists, give it to
692 * the child reaper process (ie "init") in our pid
693 * space.
695 static struct task_struct *find_new_reaper(struct task_struct *father)
696 __releases(&tasklist_lock)
697 __acquires(&tasklist_lock)
699 struct pid_namespace *pid_ns = task_active_pid_ns(father);
700 struct task_struct *thread;
702 thread = father;
703 while_each_thread(father, thread) {
704 if (thread->flags & PF_EXITING)
705 continue;
706 if (unlikely(pid_ns->child_reaper == father))
707 pid_ns->child_reaper = thread;
708 return thread;
711 if (unlikely(pid_ns->child_reaper == father)) {
712 write_unlock_irq(&tasklist_lock);
713 if (unlikely(pid_ns == &init_pid_ns))
714 panic("Attempted to kill init!");
716 zap_pid_ns_processes(pid_ns);
717 write_lock_irq(&tasklist_lock);
719 * We can not clear ->child_reaper or leave it alone.
720 * There may by stealth EXIT_DEAD tasks on ->children,
721 * forget_original_parent() must move them somewhere.
723 pid_ns->child_reaper = init_pid_ns.child_reaper;
726 return pid_ns->child_reaper;
730 * Any that need to be release_task'd are put on the @dead list.
732 static void reparent_leader(struct task_struct *father, struct task_struct *p,
733 struct list_head *dead)
735 list_move_tail(&p->sibling, &p->real_parent->children);
737 if (p->exit_state == EXIT_DEAD)
738 return;
740 * If this is a threaded reparent there is no need to
741 * notify anyone anything has happened.
743 if (same_thread_group(p->real_parent, father))
744 return;
746 /* We don't want people slaying init. */
747 p->exit_signal = SIGCHLD;
749 /* If it has exited notify the new parent about this child's death. */
750 if (!p->ptrace &&
751 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
752 if (do_notify_parent(p, p->exit_signal)) {
753 p->exit_state = EXIT_DEAD;
754 list_move_tail(&p->sibling, dead);
758 kill_orphaned_pgrp(p, father);
761 static void forget_original_parent(struct task_struct *father)
763 struct task_struct *p, *n, *reaper;
764 LIST_HEAD(dead_children);
766 write_lock_irq(&tasklist_lock);
768 * Note that exit_ptrace() and find_new_reaper() might
769 * drop tasklist_lock and reacquire it.
771 exit_ptrace(father);
772 reaper = find_new_reaper(father);
774 list_for_each_entry_safe(p, n, &father->children, sibling) {
775 struct task_struct *t = p;
776 do {
777 t->real_parent = reaper;
778 if (t->parent == father) {
779 BUG_ON(t->ptrace);
780 t->parent = t->real_parent;
782 if (t->pdeath_signal)
783 group_send_sig_info(t->pdeath_signal,
784 SEND_SIG_NOINFO, t);
785 } while_each_thread(p, t);
786 reparent_leader(father, p, &dead_children);
788 write_unlock_irq(&tasklist_lock);
790 BUG_ON(!list_empty(&father->children));
792 list_for_each_entry_safe(p, n, &dead_children, sibling) {
793 list_del_init(&p->sibling);
794 release_task(p);
799 * Send signals to all our closest relatives so that they know
800 * to properly mourn us..
802 static void exit_notify(struct task_struct *tsk, int group_dead)
804 bool autoreap;
807 * This does two things:
809 * A. Make init inherit all the child processes
810 * B. Check to see if any process groups have become orphaned
811 * as a result of our exiting, and if they have any stopped
812 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
814 forget_original_parent(tsk);
815 exit_task_namespaces(tsk);
817 write_lock_irq(&tasklist_lock);
818 if (group_dead)
819 kill_orphaned_pgrp(tsk->group_leader, NULL);
821 /* Let father know we died
823 * Thread signals are configurable, but you aren't going to use
824 * that to send signals to arbitrary processes.
825 * That stops right now.
827 * If the parent exec id doesn't match the exec id we saved
828 * when we started then we know the parent has changed security
829 * domain.
831 * If our self_exec id doesn't match our parent_exec_id then
832 * we have changed execution domain as these two values started
833 * the same after a fork.
835 if (thread_group_leader(tsk) && tsk->exit_signal != SIGCHLD &&
836 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
837 tsk->self_exec_id != tsk->parent_exec_id))
838 tsk->exit_signal = SIGCHLD;
840 if (unlikely(tsk->ptrace)) {
841 int sig = thread_group_leader(tsk) &&
842 thread_group_empty(tsk) &&
843 !ptrace_reparented(tsk) ?
844 tsk->exit_signal : SIGCHLD;
845 autoreap = do_notify_parent(tsk, sig);
846 } else if (thread_group_leader(tsk)) {
847 autoreap = thread_group_empty(tsk) &&
848 do_notify_parent(tsk, tsk->exit_signal);
849 } else {
850 autoreap = true;
853 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
855 /* mt-exec, de_thread() is waiting for group leader */
856 if (unlikely(tsk->signal->notify_count < 0))
857 wake_up_process(tsk->signal->group_exit_task);
858 write_unlock_irq(&tasklist_lock);
860 /* If the process is dead, release it - nobody will wait for it */
861 if (autoreap)
862 release_task(tsk);
865 #ifdef CONFIG_DEBUG_STACK_USAGE
866 static void check_stack_usage(void)
868 static DEFINE_SPINLOCK(low_water_lock);
869 static int lowest_to_date = THREAD_SIZE;
870 unsigned long free;
872 free = stack_not_used(current);
874 if (free >= lowest_to_date)
875 return;
877 spin_lock(&low_water_lock);
878 if (free < lowest_to_date) {
879 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
880 "left\n",
881 current->comm, free);
882 lowest_to_date = free;
884 spin_unlock(&low_water_lock);
886 #else
887 static inline void check_stack_usage(void) {}
888 #endif
890 void do_exit(long code)
892 struct task_struct *tsk = current;
893 int group_dead;
895 profile_task_exit(tsk);
897 WARN_ON(blk_needs_flush_plug(tsk));
899 if (unlikely(in_interrupt()))
900 panic("Aiee, killing interrupt handler!");
901 if (unlikely(!tsk->pid))
902 panic("Attempted to kill the idle task!");
905 * If do_exit is called because this processes oopsed, it's possible
906 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
907 * continuing. Amongst other possible reasons, this is to prevent
908 * mm_release()->clear_child_tid() from writing to a user-controlled
909 * kernel address.
911 set_fs(USER_DS);
913 ptrace_event(PTRACE_EVENT_EXIT, code);
915 validate_creds_for_do_exit(tsk);
918 * We're taking recursive faults here in do_exit. Safest is to just
919 * leave this task alone and wait for reboot.
921 if (unlikely(tsk->flags & PF_EXITING)) {
922 printk(KERN_ALERT
923 "Fixing recursive fault but reboot is needed!\n");
925 * We can do this unlocked here. The futex code uses
926 * this flag just to verify whether the pi state
927 * cleanup has been done or not. In the worst case it
928 * loops once more. We pretend that the cleanup was
929 * done as there is no way to return. Either the
930 * OWNER_DIED bit is set by now or we push the blocked
931 * task into the wait for ever nirwana as well.
933 tsk->flags |= PF_EXITPIDONE;
934 set_current_state(TASK_UNINTERRUPTIBLE);
935 schedule();
938 exit_irq_thread();
940 exit_signals(tsk); /* sets PF_EXITING */
942 * tsk->flags are checked in the futex code to protect against
943 * an exiting task cleaning up the robust pi futexes.
945 smp_mb();
946 raw_spin_unlock_wait(&tsk->pi_lock);
948 if (unlikely(in_atomic()))
949 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
950 current->comm, task_pid_nr(current),
951 preempt_count());
953 acct_update_integrals(tsk);
954 /* sync mm's RSS info before statistics gathering */
955 if (tsk->mm)
956 sync_mm_rss(tsk, tsk->mm);
957 group_dead = atomic_dec_and_test(&tsk->signal->live);
958 if (group_dead) {
959 hrtimer_cancel(&tsk->signal->real_timer);
960 exit_itimers(tsk->signal);
961 if (tsk->mm)
962 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
964 acct_collect(code, group_dead);
965 if (group_dead)
966 tty_audit_exit();
967 audit_free(tsk);
969 tsk->exit_code = code;
970 taskstats_exit(tsk, group_dead);
972 exit_mm(tsk);
974 if (group_dead)
975 acct_process();
976 trace_sched_process_exit(tsk);
978 exit_sem(tsk);
979 exit_shm(tsk);
980 exit_files(tsk);
981 exit_fs(tsk);
982 check_stack_usage();
983 exit_thread();
986 * Flush inherited counters to the parent - before the parent
987 * gets woken up by child-exit notifications.
989 * because of cgroup mode, must be called before cgroup_exit()
991 perf_event_exit_task(tsk);
993 cgroup_exit(tsk, 1);
995 if (group_dead)
996 disassociate_ctty(1);
998 module_put(task_thread_info(tsk)->exec_domain->module);
1000 proc_exit_connector(tsk);
1003 * FIXME: do that only when needed, using sched_exit tracepoint
1005 ptrace_put_breakpoints(tsk);
1007 exit_notify(tsk, group_dead);
1008 #ifdef CONFIG_NUMA
1009 task_lock(tsk);
1010 mpol_put(tsk->mempolicy);
1011 tsk->mempolicy = NULL;
1012 task_unlock(tsk);
1013 #endif
1014 #ifdef CONFIG_FUTEX
1015 if (unlikely(current->pi_state_cache))
1016 kfree(current->pi_state_cache);
1017 #endif
1019 * Make sure we are holding no locks:
1021 debug_check_no_locks_held(tsk);
1023 * We can do this unlocked here. The futex code uses this flag
1024 * just to verify whether the pi state cleanup has been done
1025 * or not. In the worst case it loops once more.
1027 tsk->flags |= PF_EXITPIDONE;
1029 if (tsk->io_context)
1030 exit_io_context(tsk);
1032 if (tsk->splice_pipe)
1033 __free_pipe_info(tsk->splice_pipe);
1035 validate_creds_for_do_exit(tsk);
1037 preempt_disable();
1038 if (tsk->nr_dirtied)
1039 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
1040 exit_rcu();
1041 /* causes final put_task_struct in finish_task_switch(). */
1042 tsk->state = TASK_DEAD;
1043 tsk->flags |= PF_NOFREEZE; /* tell freezer to ignore us */
1044 schedule();
1045 BUG();
1046 /* Avoid "noreturn function does return". */
1047 for (;;)
1048 cpu_relax(); /* For when BUG is null */
1051 EXPORT_SYMBOL_GPL(do_exit);
1053 void complete_and_exit(struct completion *comp, long code)
1055 if (comp)
1056 complete(comp);
1058 do_exit(code);
1061 EXPORT_SYMBOL(complete_and_exit);
1063 SYSCALL_DEFINE1(exit, int, error_code)
1065 do_exit((error_code&0xff)<<8);
1069 * Take down every thread in the group. This is called by fatal signals
1070 * as well as by sys_exit_group (below).
1072 void
1073 do_group_exit(int exit_code)
1075 struct signal_struct *sig = current->signal;
1077 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1079 if (signal_group_exit(sig))
1080 exit_code = sig->group_exit_code;
1081 else if (!thread_group_empty(current)) {
1082 struct sighand_struct *const sighand = current->sighand;
1083 spin_lock_irq(&sighand->siglock);
1084 if (signal_group_exit(sig))
1085 /* Another thread got here before we took the lock. */
1086 exit_code = sig->group_exit_code;
1087 else {
1088 sig->group_exit_code = exit_code;
1089 sig->flags = SIGNAL_GROUP_EXIT;
1090 zap_other_threads(current);
1092 spin_unlock_irq(&sighand->siglock);
1095 do_exit(exit_code);
1096 /* NOTREACHED */
1100 * this kills every thread in the thread group. Note that any externally
1101 * wait4()-ing process will get the correct exit code - even if this
1102 * thread is not the thread group leader.
1104 SYSCALL_DEFINE1(exit_group, int, error_code)
1106 do_group_exit((error_code & 0xff) << 8);
1107 /* NOTREACHED */
1108 return 0;
1111 struct wait_opts {
1112 enum pid_type wo_type;
1113 int wo_flags;
1114 struct pid *wo_pid;
1116 struct siginfo __user *wo_info;
1117 int __user *wo_stat;
1118 struct rusage __user *wo_rusage;
1120 wait_queue_t child_wait;
1121 int notask_error;
1124 static inline
1125 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1127 if (type != PIDTYPE_PID)
1128 task = task->group_leader;
1129 return task->pids[type].pid;
1132 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1134 return wo->wo_type == PIDTYPE_MAX ||
1135 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1138 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1140 if (!eligible_pid(wo, p))
1141 return 0;
1142 /* Wait for all children (clone and not) if __WALL is set;
1143 * otherwise, wait for clone children *only* if __WCLONE is
1144 * set; otherwise, wait for non-clone children *only*. (Note:
1145 * A "clone" child here is one that reports to its parent
1146 * using a signal other than SIGCHLD.) */
1147 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1148 && !(wo->wo_flags & __WALL))
1149 return 0;
1151 return 1;
1154 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1155 pid_t pid, uid_t uid, int why, int status)
1157 struct siginfo __user *infop;
1158 int retval = wo->wo_rusage
1159 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1161 put_task_struct(p);
1162 infop = wo->wo_info;
1163 if (infop) {
1164 if (!retval)
1165 retval = put_user(SIGCHLD, &infop->si_signo);
1166 if (!retval)
1167 retval = put_user(0, &infop->si_errno);
1168 if (!retval)
1169 retval = put_user((short)why, &infop->si_code);
1170 if (!retval)
1171 retval = put_user(pid, &infop->si_pid);
1172 if (!retval)
1173 retval = put_user(uid, &infop->si_uid);
1174 if (!retval)
1175 retval = put_user(status, &infop->si_status);
1177 if (!retval)
1178 retval = pid;
1179 return retval;
1183 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1184 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1185 * the lock and this task is uninteresting. If we return nonzero, we have
1186 * released the lock and the system call should return.
1188 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1190 unsigned long state;
1191 int retval, status, traced;
1192 pid_t pid = task_pid_vnr(p);
1193 uid_t uid = __task_cred(p)->uid;
1194 struct siginfo __user *infop;
1196 if (!likely(wo->wo_flags & WEXITED))
1197 return 0;
1199 if (unlikely(wo->wo_flags & WNOWAIT)) {
1200 int exit_code = p->exit_code;
1201 int why;
1203 get_task_struct(p);
1204 read_unlock(&tasklist_lock);
1205 if ((exit_code & 0x7f) == 0) {
1206 why = CLD_EXITED;
1207 status = exit_code >> 8;
1208 } else {
1209 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1210 status = exit_code & 0x7f;
1212 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1216 * Try to move the task's state to DEAD
1217 * only one thread is allowed to do this:
1219 state = xchg(&p->exit_state, EXIT_DEAD);
1220 if (state != EXIT_ZOMBIE) {
1221 BUG_ON(state != EXIT_DEAD);
1222 return 0;
1225 traced = ptrace_reparented(p);
1227 * It can be ptraced but not reparented, check
1228 * thread_group_leader() to filter out sub-threads.
1230 if (likely(!traced) && thread_group_leader(p)) {
1231 struct signal_struct *psig;
1232 struct signal_struct *sig;
1233 unsigned long maxrss;
1234 cputime_t tgutime, tgstime;
1237 * The resource counters for the group leader are in its
1238 * own task_struct. Those for dead threads in the group
1239 * are in its signal_struct, as are those for the child
1240 * processes it has previously reaped. All these
1241 * accumulate in the parent's signal_struct c* fields.
1243 * We don't bother to take a lock here to protect these
1244 * p->signal fields, because they are only touched by
1245 * __exit_signal, which runs with tasklist_lock
1246 * write-locked anyway, and so is excluded here. We do
1247 * need to protect the access to parent->signal fields,
1248 * as other threads in the parent group can be right
1249 * here reaping other children at the same time.
1251 * We use thread_group_times() to get times for the thread
1252 * group, which consolidates times for all threads in the
1253 * group including the group leader.
1255 thread_group_times(p, &tgutime, &tgstime);
1256 spin_lock_irq(&p->real_parent->sighand->siglock);
1257 psig = p->real_parent->signal;
1258 sig = p->signal;
1259 psig->cutime += tgutime + sig->cutime;
1260 psig->cstime += tgstime + sig->cstime;
1261 psig->cgtime += p->gtime + sig->gtime + sig->cgtime;
1262 psig->cmin_flt +=
1263 p->min_flt + sig->min_flt + sig->cmin_flt;
1264 psig->cmaj_flt +=
1265 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1266 psig->cnvcsw +=
1267 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1268 psig->cnivcsw +=
1269 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1270 psig->cinblock +=
1271 task_io_get_inblock(p) +
1272 sig->inblock + sig->cinblock;
1273 psig->coublock +=
1274 task_io_get_oublock(p) +
1275 sig->oublock + sig->coublock;
1276 maxrss = max(sig->maxrss, sig->cmaxrss);
1277 if (psig->cmaxrss < maxrss)
1278 psig->cmaxrss = maxrss;
1279 task_io_accounting_add(&psig->ioac, &p->ioac);
1280 task_io_accounting_add(&psig->ioac, &sig->ioac);
1281 spin_unlock_irq(&p->real_parent->sighand->siglock);
1285 * Now we are sure this task is interesting, and no other
1286 * thread can reap it because we set its state to EXIT_DEAD.
1288 read_unlock(&tasklist_lock);
1290 retval = wo->wo_rusage
1291 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1292 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1293 ? p->signal->group_exit_code : p->exit_code;
1294 if (!retval && wo->wo_stat)
1295 retval = put_user(status, wo->wo_stat);
1297 infop = wo->wo_info;
1298 if (!retval && infop)
1299 retval = put_user(SIGCHLD, &infop->si_signo);
1300 if (!retval && infop)
1301 retval = put_user(0, &infop->si_errno);
1302 if (!retval && infop) {
1303 int why;
1305 if ((status & 0x7f) == 0) {
1306 why = CLD_EXITED;
1307 status >>= 8;
1308 } else {
1309 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1310 status &= 0x7f;
1312 retval = put_user((short)why, &infop->si_code);
1313 if (!retval)
1314 retval = put_user(status, &infop->si_status);
1316 if (!retval && infop)
1317 retval = put_user(pid, &infop->si_pid);
1318 if (!retval && infop)
1319 retval = put_user(uid, &infop->si_uid);
1320 if (!retval)
1321 retval = pid;
1323 if (traced) {
1324 write_lock_irq(&tasklist_lock);
1325 /* We dropped tasklist, ptracer could die and untrace */
1326 ptrace_unlink(p);
1328 * If this is not a sub-thread, notify the parent.
1329 * If parent wants a zombie, don't release it now.
1331 if (thread_group_leader(p) &&
1332 !do_notify_parent(p, p->exit_signal)) {
1333 p->exit_state = EXIT_ZOMBIE;
1334 p = NULL;
1336 write_unlock_irq(&tasklist_lock);
1338 if (p != NULL)
1339 release_task(p);
1341 return retval;
1344 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1346 if (ptrace) {
1347 if (task_is_stopped_or_traced(p) &&
1348 !(p->jobctl & JOBCTL_LISTENING))
1349 return &p->exit_code;
1350 } else {
1351 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1352 return &p->signal->group_exit_code;
1354 return NULL;
1358 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1359 * @wo: wait options
1360 * @ptrace: is the wait for ptrace
1361 * @p: task to wait for
1363 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1365 * CONTEXT:
1366 * read_lock(&tasklist_lock), which is released if return value is
1367 * non-zero. Also, grabs and releases @p->sighand->siglock.
1369 * RETURNS:
1370 * 0 if wait condition didn't exist and search for other wait conditions
1371 * should continue. Non-zero return, -errno on failure and @p's pid on
1372 * success, implies that tasklist_lock is released and wait condition
1373 * search should terminate.
1375 static int wait_task_stopped(struct wait_opts *wo,
1376 int ptrace, struct task_struct *p)
1378 struct siginfo __user *infop;
1379 int retval, exit_code, *p_code, why;
1380 uid_t uid = 0; /* unneeded, required by compiler */
1381 pid_t pid;
1384 * Traditionally we see ptrace'd stopped tasks regardless of options.
1386 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1387 return 0;
1389 if (!task_stopped_code(p, ptrace))
1390 return 0;
1392 exit_code = 0;
1393 spin_lock_irq(&p->sighand->siglock);
1395 p_code = task_stopped_code(p, ptrace);
1396 if (unlikely(!p_code))
1397 goto unlock_sig;
1399 exit_code = *p_code;
1400 if (!exit_code)
1401 goto unlock_sig;
1403 if (!unlikely(wo->wo_flags & WNOWAIT))
1404 *p_code = 0;
1406 uid = task_uid(p);
1407 unlock_sig:
1408 spin_unlock_irq(&p->sighand->siglock);
1409 if (!exit_code)
1410 return 0;
1413 * Now we are pretty sure this task is interesting.
1414 * Make sure it doesn't get reaped out from under us while we
1415 * give up the lock and then examine it below. We don't want to
1416 * keep holding onto the tasklist_lock while we call getrusage and
1417 * possibly take page faults for user memory.
1419 get_task_struct(p);
1420 pid = task_pid_vnr(p);
1421 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1422 read_unlock(&tasklist_lock);
1424 if (unlikely(wo->wo_flags & WNOWAIT))
1425 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1427 retval = wo->wo_rusage
1428 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1429 if (!retval && wo->wo_stat)
1430 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1432 infop = wo->wo_info;
1433 if (!retval && infop)
1434 retval = put_user(SIGCHLD, &infop->si_signo);
1435 if (!retval && infop)
1436 retval = put_user(0, &infop->si_errno);
1437 if (!retval && infop)
1438 retval = put_user((short)why, &infop->si_code);
1439 if (!retval && infop)
1440 retval = put_user(exit_code, &infop->si_status);
1441 if (!retval && infop)
1442 retval = put_user(pid, &infop->si_pid);
1443 if (!retval && infop)
1444 retval = put_user(uid, &infop->si_uid);
1445 if (!retval)
1446 retval = pid;
1447 put_task_struct(p);
1449 BUG_ON(!retval);
1450 return retval;
1454 * Handle do_wait work for one task in a live, non-stopped state.
1455 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1456 * the lock and this task is uninteresting. If we return nonzero, we have
1457 * released the lock and the system call should return.
1459 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1461 int retval;
1462 pid_t pid;
1463 uid_t uid;
1465 if (!unlikely(wo->wo_flags & WCONTINUED))
1466 return 0;
1468 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1469 return 0;
1471 spin_lock_irq(&p->sighand->siglock);
1472 /* Re-check with the lock held. */
1473 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1474 spin_unlock_irq(&p->sighand->siglock);
1475 return 0;
1477 if (!unlikely(wo->wo_flags & WNOWAIT))
1478 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1479 uid = task_uid(p);
1480 spin_unlock_irq(&p->sighand->siglock);
1482 pid = task_pid_vnr(p);
1483 get_task_struct(p);
1484 read_unlock(&tasklist_lock);
1486 if (!wo->wo_info) {
1487 retval = wo->wo_rusage
1488 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1489 put_task_struct(p);
1490 if (!retval && wo->wo_stat)
1491 retval = put_user(0xffff, wo->wo_stat);
1492 if (!retval)
1493 retval = pid;
1494 } else {
1495 retval = wait_noreap_copyout(wo, p, pid, uid,
1496 CLD_CONTINUED, SIGCONT);
1497 BUG_ON(retval == 0);
1500 return retval;
1504 * Consider @p for a wait by @parent.
1506 * -ECHILD should be in ->notask_error before the first call.
1507 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1508 * Returns zero if the search for a child should continue;
1509 * then ->notask_error is 0 if @p is an eligible child,
1510 * or another error from security_task_wait(), or still -ECHILD.
1512 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1513 struct task_struct *p)
1515 int ret = eligible_child(wo, p);
1516 if (!ret)
1517 return ret;
1519 ret = security_task_wait(p);
1520 if (unlikely(ret < 0)) {
1522 * If we have not yet seen any eligible child,
1523 * then let this error code replace -ECHILD.
1524 * A permission error will give the user a clue
1525 * to look for security policy problems, rather
1526 * than for mysterious wait bugs.
1528 if (wo->notask_error)
1529 wo->notask_error = ret;
1530 return 0;
1533 /* dead body doesn't have much to contribute */
1534 if (unlikely(p->exit_state == EXIT_DEAD)) {
1536 * But do not ignore this task until the tracer does
1537 * wait_task_zombie()->do_notify_parent().
1539 if (likely(!ptrace) && unlikely(ptrace_reparented(p)))
1540 wo->notask_error = 0;
1541 return 0;
1544 /* slay zombie? */
1545 if (p->exit_state == EXIT_ZOMBIE) {
1547 * A zombie ptracee is only visible to its ptracer.
1548 * Notification and reaping will be cascaded to the real
1549 * parent when the ptracer detaches.
1551 if (likely(!ptrace) && unlikely(p->ptrace)) {
1552 /* it will become visible, clear notask_error */
1553 wo->notask_error = 0;
1554 return 0;
1557 /* we don't reap group leaders with subthreads */
1558 if (!delay_group_leader(p))
1559 return wait_task_zombie(wo, p);
1562 * Allow access to stopped/continued state via zombie by
1563 * falling through. Clearing of notask_error is complex.
1565 * When !@ptrace:
1567 * If WEXITED is set, notask_error should naturally be
1568 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1569 * so, if there are live subthreads, there are events to
1570 * wait for. If all subthreads are dead, it's still safe
1571 * to clear - this function will be called again in finite
1572 * amount time once all the subthreads are released and
1573 * will then return without clearing.
1575 * When @ptrace:
1577 * Stopped state is per-task and thus can't change once the
1578 * target task dies. Only continued and exited can happen.
1579 * Clear notask_error if WCONTINUED | WEXITED.
1581 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1582 wo->notask_error = 0;
1583 } else {
1585 * If @p is ptraced by a task in its real parent's group,
1586 * hide group stop/continued state when looking at @p as
1587 * the real parent; otherwise, a single stop can be
1588 * reported twice as group and ptrace stops.
1590 * If a ptracer wants to distinguish the two events for its
1591 * own children, it should create a separate process which
1592 * takes the role of real parent.
1594 if (likely(!ptrace) && p->ptrace && !ptrace_reparented(p))
1595 return 0;
1598 * @p is alive and it's gonna stop, continue or exit, so
1599 * there always is something to wait for.
1601 wo->notask_error = 0;
1605 * Wait for stopped. Depending on @ptrace, different stopped state
1606 * is used and the two don't interact with each other.
1608 ret = wait_task_stopped(wo, ptrace, p);
1609 if (ret)
1610 return ret;
1613 * Wait for continued. There's only one continued state and the
1614 * ptracer can consume it which can confuse the real parent. Don't
1615 * use WCONTINUED from ptracer. You don't need or want it.
1617 return wait_task_continued(wo, p);
1621 * Do the work of do_wait() for one thread in the group, @tsk.
1623 * -ECHILD should be in ->notask_error before the first call.
1624 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1625 * Returns zero if the search for a child should continue; then
1626 * ->notask_error is 0 if there were any eligible children,
1627 * or another error from security_task_wait(), or still -ECHILD.
1629 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1631 struct task_struct *p;
1633 list_for_each_entry(p, &tsk->children, sibling) {
1634 int ret = wait_consider_task(wo, 0, p);
1635 if (ret)
1636 return ret;
1639 return 0;
1642 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1644 struct task_struct *p;
1646 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1647 int ret = wait_consider_task(wo, 1, p);
1648 if (ret)
1649 return ret;
1652 return 0;
1655 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1656 int sync, void *key)
1658 struct wait_opts *wo = container_of(wait, struct wait_opts,
1659 child_wait);
1660 struct task_struct *p = key;
1662 if (!eligible_pid(wo, p))
1663 return 0;
1665 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1666 return 0;
1668 return default_wake_function(wait, mode, sync, key);
1671 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1673 __wake_up_sync_key(&parent->signal->wait_chldexit,
1674 TASK_INTERRUPTIBLE, 1, p);
1677 static long do_wait(struct wait_opts *wo)
1679 struct task_struct *tsk;
1680 int retval;
1682 trace_sched_process_wait(wo->wo_pid);
1684 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1685 wo->child_wait.private = current;
1686 add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1687 repeat:
1689 * If there is nothing that can match our critiera just get out.
1690 * We will clear ->notask_error to zero if we see any child that
1691 * might later match our criteria, even if we are not able to reap
1692 * it yet.
1694 wo->notask_error = -ECHILD;
1695 if ((wo->wo_type < PIDTYPE_MAX) &&
1696 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1697 goto notask;
1699 set_current_state(TASK_INTERRUPTIBLE);
1700 read_lock(&tasklist_lock);
1701 tsk = current;
1702 do {
1703 retval = do_wait_thread(wo, tsk);
1704 if (retval)
1705 goto end;
1707 retval = ptrace_do_wait(wo, tsk);
1708 if (retval)
1709 goto end;
1711 if (wo->wo_flags & __WNOTHREAD)
1712 break;
1713 } while_each_thread(current, tsk);
1714 read_unlock(&tasklist_lock);
1716 notask:
1717 retval = wo->notask_error;
1718 if (!retval && !(wo->wo_flags & WNOHANG)) {
1719 retval = -ERESTARTSYS;
1720 if (!signal_pending(current)) {
1721 schedule();
1722 goto repeat;
1725 end:
1726 __set_current_state(TASK_RUNNING);
1727 remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1728 return retval;
1731 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1732 infop, int, options, struct rusage __user *, ru)
1734 struct wait_opts wo;
1735 struct pid *pid = NULL;
1736 enum pid_type type;
1737 long ret;
1739 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1740 return -EINVAL;
1741 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1742 return -EINVAL;
1744 switch (which) {
1745 case P_ALL:
1746 type = PIDTYPE_MAX;
1747 break;
1748 case P_PID:
1749 type = PIDTYPE_PID;
1750 if (upid <= 0)
1751 return -EINVAL;
1752 break;
1753 case P_PGID:
1754 type = PIDTYPE_PGID;
1755 if (upid <= 0)
1756 return -EINVAL;
1757 break;
1758 default:
1759 return -EINVAL;
1762 if (type < PIDTYPE_MAX)
1763 pid = find_get_pid(upid);
1765 wo.wo_type = type;
1766 wo.wo_pid = pid;
1767 wo.wo_flags = options;
1768 wo.wo_info = infop;
1769 wo.wo_stat = NULL;
1770 wo.wo_rusage = ru;
1771 ret = do_wait(&wo);
1773 if (ret > 0) {
1774 ret = 0;
1775 } else if (infop) {
1777 * For a WNOHANG return, clear out all the fields
1778 * we would set so the user can easily tell the
1779 * difference.
1781 if (!ret)
1782 ret = put_user(0, &infop->si_signo);
1783 if (!ret)
1784 ret = put_user(0, &infop->si_errno);
1785 if (!ret)
1786 ret = put_user(0, &infop->si_code);
1787 if (!ret)
1788 ret = put_user(0, &infop->si_pid);
1789 if (!ret)
1790 ret = put_user(0, &infop->si_uid);
1791 if (!ret)
1792 ret = put_user(0, &infop->si_status);
1795 put_pid(pid);
1797 /* avoid REGPARM breakage on x86: */
1798 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1799 return ret;
1802 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1803 int, options, struct rusage __user *, ru)
1805 struct wait_opts wo;
1806 struct pid *pid = NULL;
1807 enum pid_type type;
1808 long ret;
1810 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1811 __WNOTHREAD|__WCLONE|__WALL))
1812 return -EINVAL;
1814 if (upid == -1)
1815 type = PIDTYPE_MAX;
1816 else if (upid < 0) {
1817 type = PIDTYPE_PGID;
1818 pid = find_get_pid(-upid);
1819 } else if (upid == 0) {
1820 type = PIDTYPE_PGID;
1821 pid = get_task_pid(current, PIDTYPE_PGID);
1822 } else /* upid > 0 */ {
1823 type = PIDTYPE_PID;
1824 pid = find_get_pid(upid);
1827 wo.wo_type = type;
1828 wo.wo_pid = pid;
1829 wo.wo_flags = options | WEXITED;
1830 wo.wo_info = NULL;
1831 wo.wo_stat = stat_addr;
1832 wo.wo_rusage = ru;
1833 ret = do_wait(&wo);
1834 put_pid(pid);
1836 /* avoid REGPARM breakage on x86: */
1837 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1838 return ret;
1841 #ifdef __ARCH_WANT_SYS_WAITPID
1844 * sys_waitpid() remains for compatibility. waitpid() should be
1845 * implemented by calling sys_wait4() from libc.a.
1847 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1849 return sys_wait4(pid, stat_addr, options, NULL);
1852 #endif