4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/smp_lock.h>
11 #include <linux/module.h>
12 #include <linux/capability.h>
13 #include <linux/completion.h>
14 #include <linux/personality.h>
15 #include <linux/tty.h>
16 #include <linux/namespace.h>
17 #include <linux/key.h>
18 #include <linux/security.h>
19 #include <linux/cpu.h>
20 #include <linux/acct.h>
21 #include <linux/file.h>
22 #include <linux/binfmts.h>
23 #include <linux/ptrace.h>
24 #include <linux/profile.h>
25 #include <linux/mount.h>
26 #include <linux/proc_fs.h>
27 #include <linux/mempolicy.h>
28 #include <linux/taskstats_kern.h>
29 #include <linux/delayacct.h>
30 #include <linux/cpuset.h>
31 #include <linux/syscalls.h>
32 #include <linux/signal.h>
33 #include <linux/posix-timers.h>
34 #include <linux/cn_proc.h>
35 #include <linux/mutex.h>
36 #include <linux/futex.h>
37 #include <linux/compat.h>
38 #include <linux/pipe_fs_i.h>
39 #include <linux/audit.h> /* for audit_free() */
40 #include <linux/resource.h>
42 #include <asm/uaccess.h>
43 #include <asm/unistd.h>
44 #include <asm/pgtable.h>
45 #include <asm/mmu_context.h>
47 extern void sem_exit (void);
48 extern struct task_struct *child_reaper;
50 static void exit_mm(struct task_struct * tsk);
52 static void __unhash_process(struct task_struct *p)
55 detach_pid(p, PIDTYPE_PID);
56 if (thread_group_leader(p)) {
57 detach_pid(p, PIDTYPE_PGID);
58 detach_pid(p, PIDTYPE_SID);
60 list_del_rcu(&p->tasks);
61 __get_cpu_var(process_counts)--;
63 list_del_rcu(&p->thread_group);
68 * This function expects the tasklist_lock write-locked.
70 static void __exit_signal(struct task_struct *tsk)
72 struct signal_struct *sig = tsk->signal;
73 struct sighand_struct *sighand;
76 BUG_ON(!atomic_read(&sig->count));
79 sighand = rcu_dereference(tsk->sighand);
80 spin_lock(&sighand->siglock);
82 posix_cpu_timers_exit(tsk);
83 if (atomic_dec_and_test(&sig->count))
84 posix_cpu_timers_exit_group(tsk);
87 * If there is any task waiting for the group exit
90 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
91 wake_up_process(sig->group_exit_task);
92 sig->group_exit_task = NULL;
94 if (tsk == sig->curr_target)
95 sig->curr_target = next_thread(tsk);
97 * Accumulate here the counters for all threads but the
98 * group leader as they die, so they can be added into
99 * the process-wide totals when those are taken.
100 * The group leader stays around as a zombie as long
101 * as there are other threads. When it gets reaped,
102 * the exit.c code will add its counts into these totals.
103 * We won't ever get here for the group leader, since it
104 * will have been the last reference on the signal_struct.
106 sig->utime = cputime_add(sig->utime, tsk->utime);
107 sig->stime = cputime_add(sig->stime, tsk->stime);
108 sig->min_flt += tsk->min_flt;
109 sig->maj_flt += tsk->maj_flt;
110 sig->nvcsw += tsk->nvcsw;
111 sig->nivcsw += tsk->nivcsw;
112 sig->sched_time += tsk->sched_time;
113 sig = NULL; /* Marker for below. */
116 __unhash_process(tsk);
120 spin_unlock(&sighand->siglock);
123 __cleanup_sighand(sighand);
124 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
125 flush_sigqueue(&tsk->pending);
127 flush_sigqueue(&sig->shared_pending);
128 __cleanup_signal(sig);
132 static void delayed_put_task_struct(struct rcu_head *rhp)
134 put_task_struct(container_of(rhp, struct task_struct, rcu));
137 void release_task(struct task_struct * p)
139 struct task_struct *leader;
142 atomic_dec(&p->user->processes);
143 write_lock_irq(&tasklist_lock);
145 BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
149 * If we are the last non-leader member of the thread
150 * group, and the leader is zombie, then notify the
151 * group leader's parent process. (if it wants notification.)
154 leader = p->group_leader;
155 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
156 BUG_ON(leader->exit_signal == -1);
157 do_notify_parent(leader, leader->exit_signal);
159 * If we were the last child thread and the leader has
160 * exited already, and the leader's parent ignores SIGCHLD,
161 * then we are the one who should release the leader.
163 * do_notify_parent() will have marked it self-reaping in
166 zap_leader = (leader->exit_signal == -1);
170 write_unlock_irq(&tasklist_lock);
173 call_rcu(&p->rcu, delayed_put_task_struct);
176 if (unlikely(zap_leader))
181 * This checks not only the pgrp, but falls back on the pid if no
182 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
185 int session_of_pgrp(int pgrp)
187 struct task_struct *p;
190 read_lock(&tasklist_lock);
191 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
192 if (p->signal->session > 0) {
193 sid = p->signal->session;
196 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
197 p = find_task_by_pid(pgrp);
199 sid = p->signal->session;
201 read_unlock(&tasklist_lock);
207 * Determine if a process group is "orphaned", according to the POSIX
208 * definition in 2.2.2.52. Orphaned process groups are not to be affected
209 * by terminal-generated stop signals. Newly orphaned process groups are
210 * to receive a SIGHUP and a SIGCONT.
212 * "I ask you, have you ever known what it is to be an orphan?"
214 static int will_become_orphaned_pgrp(int pgrp, struct task_struct *ignored_task)
216 struct task_struct *p;
219 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
220 if (p == ignored_task
222 || is_init(p->real_parent))
224 if (process_group(p->real_parent) != pgrp
225 && p->real_parent->signal->session == p->signal->session) {
229 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
230 return ret; /* (sighing) "Often!" */
233 int is_orphaned_pgrp(int pgrp)
237 read_lock(&tasklist_lock);
238 retval = will_become_orphaned_pgrp(pgrp, NULL);
239 read_unlock(&tasklist_lock);
244 static int has_stopped_jobs(int pgrp)
247 struct task_struct *p;
249 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
250 if (p->state != TASK_STOPPED)
253 /* If p is stopped by a debugger on a signal that won't
254 stop it, then don't count p as stopped. This isn't
255 perfect but it's a good approximation. */
256 if (unlikely (p->ptrace)
257 && p->exit_code != SIGSTOP
258 && p->exit_code != SIGTSTP
259 && p->exit_code != SIGTTOU
260 && p->exit_code != SIGTTIN)
265 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
270 * reparent_to_init - Reparent the calling kernel thread to the init task.
272 * If a kernel thread is launched as a result of a system call, or if
273 * it ever exits, it should generally reparent itself to init so that
274 * it is correctly cleaned up on exit.
276 * The various task state such as scheduling policy and priority may have
277 * been inherited from a user process, so we reset them to sane values here.
279 * NOTE that reparent_to_init() gives the caller full capabilities.
281 static void reparent_to_init(void)
283 write_lock_irq(&tasklist_lock);
285 ptrace_unlink(current);
286 /* Reparent to init */
287 remove_parent(current);
288 current->parent = child_reaper;
289 current->real_parent = child_reaper;
292 /* Set the exit signal to SIGCHLD so we signal init on exit */
293 current->exit_signal = SIGCHLD;
295 if ((current->policy == SCHED_NORMAL ||
296 current->policy == SCHED_BATCH)
297 && (task_nice(current) < 0))
298 set_user_nice(current, 0);
302 security_task_reparent_to_init(current);
303 memcpy(current->signal->rlim, init_task.signal->rlim,
304 sizeof(current->signal->rlim));
305 atomic_inc(&(INIT_USER->__count));
306 write_unlock_irq(&tasklist_lock);
307 switch_uid(INIT_USER);
310 void __set_special_pids(pid_t session, pid_t pgrp)
312 struct task_struct *curr = current->group_leader;
314 if (curr->signal->session != session) {
315 detach_pid(curr, PIDTYPE_SID);
316 curr->signal->session = session;
317 attach_pid(curr, PIDTYPE_SID, session);
319 if (process_group(curr) != pgrp) {
320 detach_pid(curr, PIDTYPE_PGID);
321 curr->signal->pgrp = pgrp;
322 attach_pid(curr, PIDTYPE_PGID, pgrp);
326 void set_special_pids(pid_t session, pid_t pgrp)
328 write_lock_irq(&tasklist_lock);
329 __set_special_pids(session, pgrp);
330 write_unlock_irq(&tasklist_lock);
334 * Let kernel threads use this to say that they
335 * allow a certain signal (since daemonize() will
336 * have disabled all of them by default).
338 int allow_signal(int sig)
340 if (!valid_signal(sig) || sig < 1)
343 spin_lock_irq(¤t->sighand->siglock);
344 sigdelset(¤t->blocked, sig);
346 /* Kernel threads handle their own signals.
347 Let the signal code know it'll be handled, so
348 that they don't get converted to SIGKILL or
349 just silently dropped */
350 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
353 spin_unlock_irq(¤t->sighand->siglock);
357 EXPORT_SYMBOL(allow_signal);
359 int disallow_signal(int sig)
361 if (!valid_signal(sig) || sig < 1)
364 spin_lock_irq(¤t->sighand->siglock);
365 sigaddset(¤t->blocked, sig);
367 spin_unlock_irq(¤t->sighand->siglock);
371 EXPORT_SYMBOL(disallow_signal);
374 * Put all the gunge required to become a kernel thread without
375 * attached user resources in one place where it belongs.
378 void daemonize(const char *name, ...)
381 struct fs_struct *fs;
384 va_start(args, name);
385 vsnprintf(current->comm, sizeof(current->comm), name, args);
389 * If we were started as result of loading a module, close all of the
390 * user space pages. We don't need them, and if we didn't close them
391 * they would be locked into memory.
395 set_special_pids(1, 1);
396 mutex_lock(&tty_mutex);
397 current->signal->tty = NULL;
398 mutex_unlock(&tty_mutex);
400 /* Block and flush all signals */
401 sigfillset(&blocked);
402 sigprocmask(SIG_BLOCK, &blocked, NULL);
403 flush_signals(current);
405 /* Become as one with the init task */
407 exit_fs(current); /* current->fs->count--; */
410 atomic_inc(&fs->count);
411 exit_namespace(current);
412 current->namespace = init_task.namespace;
413 get_namespace(current->namespace);
415 current->files = init_task.files;
416 atomic_inc(¤t->files->count);
421 EXPORT_SYMBOL(daemonize);
423 static void close_files(struct files_struct * files)
431 * It is safe to dereference the fd table without RCU or
432 * ->file_lock because this is the last reference to the
435 fdt = files_fdtable(files);
439 if (i >= fdt->max_fdset || i >= fdt->max_fds)
441 set = fdt->open_fds->fds_bits[j++];
444 struct file * file = xchg(&fdt->fd[i], NULL);
446 filp_close(file, files);
454 struct files_struct *get_files_struct(struct task_struct *task)
456 struct files_struct *files;
461 atomic_inc(&files->count);
467 void fastcall put_files_struct(struct files_struct *files)
471 if (atomic_dec_and_test(&files->count)) {
474 * Free the fd and fdset arrays if we expanded them.
475 * If the fdtable was embedded, pass files for freeing
476 * at the end of the RCU grace period. Otherwise,
477 * you can free files immediately.
479 fdt = files_fdtable(files);
480 if (fdt == &files->fdtab)
481 fdt->free_files = files;
483 kmem_cache_free(files_cachep, files);
488 EXPORT_SYMBOL(put_files_struct);
490 void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
492 struct files_struct *old;
498 put_files_struct(old);
500 EXPORT_SYMBOL(reset_files_struct);
502 static inline void __exit_files(struct task_struct *tsk)
504 struct files_struct * files = tsk->files;
510 put_files_struct(files);
514 void exit_files(struct task_struct *tsk)
519 static inline void __put_fs_struct(struct fs_struct *fs)
521 /* No need to hold fs->lock if we are killing it */
522 if (atomic_dec_and_test(&fs->count)) {
529 mntput(fs->altrootmnt);
531 kmem_cache_free(fs_cachep, fs);
535 void put_fs_struct(struct fs_struct *fs)
540 static inline void __exit_fs(struct task_struct *tsk)
542 struct fs_struct * fs = tsk->fs;
552 void exit_fs(struct task_struct *tsk)
557 EXPORT_SYMBOL_GPL(exit_fs);
560 * Turn us into a lazy TLB process if we
563 static void exit_mm(struct task_struct * tsk)
565 struct mm_struct *mm = tsk->mm;
571 * Serialize with any possible pending coredump.
572 * We must hold mmap_sem around checking core_waiters
573 * and clearing tsk->mm. The core-inducing thread
574 * will increment core_waiters for each thread in the
575 * group with ->mm != NULL.
577 down_read(&mm->mmap_sem);
578 if (mm->core_waiters) {
579 up_read(&mm->mmap_sem);
580 down_write(&mm->mmap_sem);
581 if (!--mm->core_waiters)
582 complete(mm->core_startup_done);
583 up_write(&mm->mmap_sem);
585 wait_for_completion(&mm->core_done);
586 down_read(&mm->mmap_sem);
588 atomic_inc(&mm->mm_count);
589 BUG_ON(mm != tsk->active_mm);
590 /* more a memory barrier than a real lock */
593 up_read(&mm->mmap_sem);
594 enter_lazy_tlb(mm, current);
600 choose_new_parent(struct task_struct *p, struct task_struct *reaper)
603 * Make sure we're not reparenting to ourselves and that
604 * the parent is not a zombie.
606 BUG_ON(p == reaper || reaper->exit_state);
607 p->real_parent = reaper;
611 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
613 /* We don't want people slaying init. */
614 if (p->exit_signal != -1)
615 p->exit_signal = SIGCHLD;
617 if (p->pdeath_signal)
618 /* We already hold the tasklist_lock here. */
619 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
621 /* Move the child from its dying parent to the new one. */
622 if (unlikely(traced)) {
623 /* Preserve ptrace links if someone else is tracing this child. */
624 list_del_init(&p->ptrace_list);
625 if (p->parent != p->real_parent)
626 list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
628 /* If this child is being traced, then we're the one tracing it
629 * anyway, so let go of it.
633 p->parent = p->real_parent;
636 /* If we'd notified the old parent about this child's death,
637 * also notify the new parent.
639 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
640 thread_group_empty(p))
641 do_notify_parent(p, p->exit_signal);
642 else if (p->state == TASK_TRACED) {
644 * If it was at a trace stop, turn it into
645 * a normal stop since it's no longer being
653 * process group orphan check
654 * Case ii: Our child is in a different pgrp
655 * than we are, and it was the only connection
656 * outside, so the child pgrp is now orphaned.
658 if ((process_group(p) != process_group(father)) &&
659 (p->signal->session == father->signal->session)) {
660 int pgrp = process_group(p);
662 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
663 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
664 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
670 * When we die, we re-parent all our children.
671 * Try to give them to another thread in our thread
672 * group, and if no such member exists, give it to
673 * the global child reaper process (ie "init")
676 forget_original_parent(struct task_struct *father, struct list_head *to_release)
678 struct task_struct *p, *reaper = father;
679 struct list_head *_p, *_n;
682 reaper = next_thread(reaper);
683 if (reaper == father) {
684 reaper = child_reaper;
687 } while (reaper->exit_state);
690 * There are only two places where our children can be:
692 * - in our child list
693 * - in our ptraced child list
695 * Search them and reparent children.
697 list_for_each_safe(_p, _n, &father->children) {
699 p = list_entry(_p, struct task_struct, sibling);
703 /* if father isn't the real parent, then ptrace must be enabled */
704 BUG_ON(father != p->real_parent && !ptrace);
706 if (father == p->real_parent) {
707 /* reparent with a reaper, real father it's us */
708 choose_new_parent(p, reaper);
709 reparent_thread(p, father, 0);
711 /* reparent ptraced task to its real parent */
713 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
714 thread_group_empty(p))
715 do_notify_parent(p, p->exit_signal);
719 * if the ptraced child is a zombie with exit_signal == -1
720 * we must collect it before we exit, or it will remain
721 * zombie forever since we prevented it from self-reap itself
722 * while it was being traced by us, to be able to see it in wait4.
724 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
725 list_add(&p->ptrace_list, to_release);
727 list_for_each_safe(_p, _n, &father->ptrace_children) {
728 p = list_entry(_p, struct task_struct, ptrace_list);
729 choose_new_parent(p, reaper);
730 reparent_thread(p, father, 1);
735 * Send signals to all our closest relatives so that they know
736 * to properly mourn us..
738 static void exit_notify(struct task_struct *tsk)
741 struct task_struct *t;
742 struct list_head ptrace_dead, *_p, *_n;
744 if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
745 && !thread_group_empty(tsk)) {
747 * This occurs when there was a race between our exit
748 * syscall and a group signal choosing us as the one to
749 * wake up. It could be that we are the only thread
750 * alerted to check for pending signals, but another thread
751 * should be woken now to take the signal since we will not.
752 * Now we'll wake all the threads in the group just to make
753 * sure someone gets all the pending signals.
755 read_lock(&tasklist_lock);
756 spin_lock_irq(&tsk->sighand->siglock);
757 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
758 if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
759 recalc_sigpending_tsk(t);
760 if (signal_pending(t))
761 signal_wake_up(t, 0);
763 spin_unlock_irq(&tsk->sighand->siglock);
764 read_unlock(&tasklist_lock);
767 write_lock_irq(&tasklist_lock);
770 * This does two things:
772 * A. Make init inherit all the child processes
773 * B. Check to see if any process groups have become orphaned
774 * as a result of our exiting, and if they have any stopped
775 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
778 INIT_LIST_HEAD(&ptrace_dead);
779 forget_original_parent(tsk, &ptrace_dead);
780 BUG_ON(!list_empty(&tsk->children));
781 BUG_ON(!list_empty(&tsk->ptrace_children));
784 * Check to see if any process groups have become orphaned
785 * as a result of our exiting, and if they have any stopped
786 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
788 * Case i: Our father is in a different pgrp than we are
789 * and we were the only connection outside, so our pgrp
790 * is about to become orphaned.
793 t = tsk->real_parent;
795 if ((process_group(t) != process_group(tsk)) &&
796 (t->signal->session == tsk->signal->session) &&
797 will_become_orphaned_pgrp(process_group(tsk), tsk) &&
798 has_stopped_jobs(process_group(tsk))) {
799 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
800 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
803 /* Let father know we died
805 * Thread signals are configurable, but you aren't going to use
806 * that to send signals to arbitary processes.
807 * That stops right now.
809 * If the parent exec id doesn't match the exec id we saved
810 * when we started then we know the parent has changed security
813 * If our self_exec id doesn't match our parent_exec_id then
814 * we have changed execution domain as these two values started
815 * the same after a fork.
819 if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
820 ( tsk->parent_exec_id != t->self_exec_id ||
821 tsk->self_exec_id != tsk->parent_exec_id)
822 && !capable(CAP_KILL))
823 tsk->exit_signal = SIGCHLD;
826 /* If something other than our normal parent is ptracing us, then
827 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
828 * only has special meaning to our real parent.
830 if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
831 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
832 do_notify_parent(tsk, signal);
833 } else if (tsk->ptrace) {
834 do_notify_parent(tsk, SIGCHLD);
838 if (tsk->exit_signal == -1 &&
839 (likely(tsk->ptrace == 0) ||
840 unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
842 tsk->exit_state = state;
844 write_unlock_irq(&tasklist_lock);
846 list_for_each_safe(_p, _n, &ptrace_dead) {
848 t = list_entry(_p, struct task_struct, ptrace_list);
852 /* If the process is dead, release it - nobody will wait for it */
853 if (state == EXIT_DEAD)
857 fastcall NORET_TYPE void do_exit(long code)
859 struct task_struct *tsk = current;
860 struct taskstats *tidstats;
864 profile_task_exit(tsk);
866 WARN_ON(atomic_read(&tsk->fs_excl));
868 if (unlikely(in_interrupt()))
869 panic("Aiee, killing interrupt handler!");
870 if (unlikely(!tsk->pid))
871 panic("Attempted to kill the idle task!");
872 if (unlikely(tsk == child_reaper))
873 panic("Attempted to kill init!");
875 if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
876 current->ptrace_message = code;
877 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
881 * We're taking recursive faults here in do_exit. Safest is to just
882 * leave this task alone and wait for reboot.
884 if (unlikely(tsk->flags & PF_EXITING)) {
886 "Fixing recursive fault but reboot is needed!\n");
889 set_current_state(TASK_UNINTERRUPTIBLE);
893 tsk->flags |= PF_EXITING;
895 if (unlikely(in_atomic()))
896 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
897 current->comm, current->pid,
900 taskstats_exit_alloc(&tidstats, &mycpu);
902 acct_update_integrals(tsk);
904 update_hiwater_rss(tsk->mm);
905 update_hiwater_vm(tsk->mm);
907 group_dead = atomic_dec_and_test(&tsk->signal->live);
909 hrtimer_cancel(&tsk->signal->real_timer);
910 exit_itimers(tsk->signal);
912 acct_collect(code, group_dead);
913 if (unlikely(tsk->robust_list))
914 exit_robust_list(tsk);
915 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
916 if (unlikely(tsk->compat_robust_list))
917 compat_exit_robust_list(tsk);
919 if (unlikely(tsk->audit_context))
921 taskstats_exit_send(tsk, tidstats, group_dead, mycpu);
922 taskstats_exit_free(tidstats);
936 if (group_dead && tsk->signal->leader)
937 disassociate_ctty(1);
939 module_put(task_thread_info(tsk)->exec_domain->module);
941 module_put(tsk->binfmt->module);
943 tsk->exit_code = code;
944 proc_exit_connector(tsk);
947 mpol_free(tsk->mempolicy);
948 tsk->mempolicy = NULL;
951 * This must happen late, after the PID is not
954 if (unlikely(!list_empty(&tsk->pi_state_list)))
955 exit_pi_state_list(tsk);
956 if (unlikely(current->pi_state_cache))
957 kfree(current->pi_state_cache);
959 * Make sure we are holding no locks:
961 debug_check_no_locks_held(tsk);
966 if (tsk->splice_pipe)
967 __free_pipe_info(tsk->splice_pipe);
969 /* PF_DEAD causes final put_task_struct after we schedule. */
971 BUG_ON(tsk->flags & PF_DEAD);
972 tsk->flags |= PF_DEAD;
976 /* Avoid "noreturn function does return". */
980 EXPORT_SYMBOL_GPL(do_exit);
982 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
990 EXPORT_SYMBOL(complete_and_exit);
992 asmlinkage long sys_exit(int error_code)
994 do_exit((error_code&0xff)<<8);
998 * Take down every thread in the group. This is called by fatal signals
999 * as well as by sys_exit_group (below).
1002 do_group_exit(int exit_code)
1004 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1006 if (current->signal->flags & SIGNAL_GROUP_EXIT)
1007 exit_code = current->signal->group_exit_code;
1008 else if (!thread_group_empty(current)) {
1009 struct signal_struct *const sig = current->signal;
1010 struct sighand_struct *const sighand = current->sighand;
1011 spin_lock_irq(&sighand->siglock);
1012 if (sig->flags & SIGNAL_GROUP_EXIT)
1013 /* Another thread got here before we took the lock. */
1014 exit_code = sig->group_exit_code;
1016 sig->group_exit_code = exit_code;
1017 zap_other_threads(current);
1019 spin_unlock_irq(&sighand->siglock);
1027 * this kills every thread in the thread group. Note that any externally
1028 * wait4()-ing process will get the correct exit code - even if this
1029 * thread is not the thread group leader.
1031 asmlinkage void sys_exit_group(int error_code)
1033 do_group_exit((error_code & 0xff) << 8);
1036 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1042 if (process_group(p) != process_group(current))
1044 } else if (pid != -1) {
1045 if (process_group(p) != -pid)
1050 * Do not consider detached threads that are
1053 if (p->exit_signal == -1 && !p->ptrace)
1056 /* Wait for all children (clone and not) if __WALL is set;
1057 * otherwise, wait for clone children *only* if __WCLONE is
1058 * set; otherwise, wait for non-clone children *only*. (Note:
1059 * A "clone" child here is one that reports to its parent
1060 * using a signal other than SIGCHLD.) */
1061 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1062 && !(options & __WALL))
1065 * Do not consider thread group leaders that are
1066 * in a non-empty thread group:
1068 if (delay_group_leader(p))
1071 if (security_task_wait(p))
1077 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1078 int why, int status,
1079 struct siginfo __user *infop,
1080 struct rusage __user *rusagep)
1082 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1086 retval = put_user(SIGCHLD, &infop->si_signo);
1088 retval = put_user(0, &infop->si_errno);
1090 retval = put_user((short)why, &infop->si_code);
1092 retval = put_user(pid, &infop->si_pid);
1094 retval = put_user(uid, &infop->si_uid);
1096 retval = put_user(status, &infop->si_status);
1103 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1104 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1105 * the lock and this task is uninteresting. If we return nonzero, we have
1106 * released the lock and the system call should return.
1108 static int wait_task_zombie(struct task_struct *p, int noreap,
1109 struct siginfo __user *infop,
1110 int __user *stat_addr, struct rusage __user *ru)
1112 unsigned long state;
1116 if (unlikely(noreap)) {
1119 int exit_code = p->exit_code;
1122 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1124 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1127 read_unlock(&tasklist_lock);
1128 if ((exit_code & 0x7f) == 0) {
1130 status = exit_code >> 8;
1132 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1133 status = exit_code & 0x7f;
1135 return wait_noreap_copyout(p, pid, uid, why,
1140 * Try to move the task's state to DEAD
1141 * only one thread is allowed to do this:
1143 state = xchg(&p->exit_state, EXIT_DEAD);
1144 if (state != EXIT_ZOMBIE) {
1145 BUG_ON(state != EXIT_DEAD);
1148 if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1150 * This can only happen in a race with a ptraced thread
1151 * dying on another processor.
1156 if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1157 struct signal_struct *psig;
1158 struct signal_struct *sig;
1161 * The resource counters for the group leader are in its
1162 * own task_struct. Those for dead threads in the group
1163 * are in its signal_struct, as are those for the child
1164 * processes it has previously reaped. All these
1165 * accumulate in the parent's signal_struct c* fields.
1167 * We don't bother to take a lock here to protect these
1168 * p->signal fields, because they are only touched by
1169 * __exit_signal, which runs with tasklist_lock
1170 * write-locked anyway, and so is excluded here. We do
1171 * need to protect the access to p->parent->signal fields,
1172 * as other threads in the parent group can be right
1173 * here reaping other children at the same time.
1175 spin_lock_irq(&p->parent->sighand->siglock);
1176 psig = p->parent->signal;
1179 cputime_add(psig->cutime,
1180 cputime_add(p->utime,
1181 cputime_add(sig->utime,
1184 cputime_add(psig->cstime,
1185 cputime_add(p->stime,
1186 cputime_add(sig->stime,
1189 p->min_flt + sig->min_flt + sig->cmin_flt;
1191 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1193 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1195 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1196 spin_unlock_irq(&p->parent->sighand->siglock);
1200 * Now we are sure this task is interesting, and no other
1201 * thread can reap it because we set its state to EXIT_DEAD.
1203 read_unlock(&tasklist_lock);
1205 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1206 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1207 ? p->signal->group_exit_code : p->exit_code;
1208 if (!retval && stat_addr)
1209 retval = put_user(status, stat_addr);
1210 if (!retval && infop)
1211 retval = put_user(SIGCHLD, &infop->si_signo);
1212 if (!retval && infop)
1213 retval = put_user(0, &infop->si_errno);
1214 if (!retval && infop) {
1217 if ((status & 0x7f) == 0) {
1221 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1224 retval = put_user((short)why, &infop->si_code);
1226 retval = put_user(status, &infop->si_status);
1228 if (!retval && infop)
1229 retval = put_user(p->pid, &infop->si_pid);
1230 if (!retval && infop)
1231 retval = put_user(p->uid, &infop->si_uid);
1233 // TODO: is this safe?
1234 p->exit_state = EXIT_ZOMBIE;
1238 if (p->real_parent != p->parent) {
1239 write_lock_irq(&tasklist_lock);
1240 /* Double-check with lock held. */
1241 if (p->real_parent != p->parent) {
1243 // TODO: is this safe?
1244 p->exit_state = EXIT_ZOMBIE;
1246 * If this is not a detached task, notify the parent.
1247 * If it's still not detached after that, don't release
1250 if (p->exit_signal != -1) {
1251 do_notify_parent(p, p->exit_signal);
1252 if (p->exit_signal != -1)
1256 write_unlock_irq(&tasklist_lock);
1265 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1266 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1267 * the lock and this task is uninteresting. If we return nonzero, we have
1268 * released the lock and the system call should return.
1270 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1271 int noreap, struct siginfo __user *infop,
1272 int __user *stat_addr, struct rusage __user *ru)
1274 int retval, exit_code;
1278 if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1279 p->signal && p->signal->group_stop_count > 0)
1281 * A group stop is in progress and this is the group leader.
1282 * We won't report until all threads have stopped.
1287 * Now we are pretty sure this task is interesting.
1288 * Make sure it doesn't get reaped out from under us while we
1289 * give up the lock and then examine it below. We don't want to
1290 * keep holding onto the tasklist_lock while we call getrusage and
1291 * possibly take page faults for user memory.
1294 read_unlock(&tasklist_lock);
1296 if (unlikely(noreap)) {
1299 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1301 exit_code = p->exit_code;
1302 if (unlikely(!exit_code) ||
1303 unlikely(p->state & TASK_TRACED))
1305 return wait_noreap_copyout(p, pid, uid,
1306 why, (exit_code << 8) | 0x7f,
1310 write_lock_irq(&tasklist_lock);
1313 * This uses xchg to be atomic with the thread resuming and setting
1314 * it. It must also be done with the write lock held to prevent a
1315 * race with the EXIT_ZOMBIE case.
1317 exit_code = xchg(&p->exit_code, 0);
1318 if (unlikely(p->exit_state)) {
1320 * The task resumed and then died. Let the next iteration
1321 * catch it in EXIT_ZOMBIE. Note that exit_code might
1322 * already be zero here if it resumed and did _exit(0).
1323 * The task itself is dead and won't touch exit_code again;
1324 * other processors in this function are locked out.
1326 p->exit_code = exit_code;
1329 if (unlikely(exit_code == 0)) {
1331 * Another thread in this function got to it first, or it
1332 * resumed, or it resumed and then died.
1334 write_unlock_irq(&tasklist_lock);
1338 * We are returning to the wait loop without having successfully
1339 * removed the process and having released the lock. We cannot
1340 * continue, since the "p" task pointer is potentially stale.
1342 * Return -EAGAIN, and do_wait() will restart the loop from the
1343 * beginning. Do _not_ re-acquire the lock.
1348 /* move to end of parent's list to avoid starvation */
1352 write_unlock_irq(&tasklist_lock);
1354 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1355 if (!retval && stat_addr)
1356 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1357 if (!retval && infop)
1358 retval = put_user(SIGCHLD, &infop->si_signo);
1359 if (!retval && infop)
1360 retval = put_user(0, &infop->si_errno);
1361 if (!retval && infop)
1362 retval = put_user((short)((p->ptrace & PT_PTRACED)
1363 ? CLD_TRAPPED : CLD_STOPPED),
1365 if (!retval && infop)
1366 retval = put_user(exit_code, &infop->si_status);
1367 if (!retval && infop)
1368 retval = put_user(p->pid, &infop->si_pid);
1369 if (!retval && infop)
1370 retval = put_user(p->uid, &infop->si_uid);
1380 * Handle do_wait work for one task in a live, non-stopped state.
1381 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1382 * the lock and this task is uninteresting. If we return nonzero, we have
1383 * released the lock and the system call should return.
1385 static int wait_task_continued(struct task_struct *p, int noreap,
1386 struct siginfo __user *infop,
1387 int __user *stat_addr, struct rusage __user *ru)
1393 if (unlikely(!p->signal))
1396 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1399 spin_lock_irq(&p->sighand->siglock);
1400 /* Re-check with the lock held. */
1401 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1402 spin_unlock_irq(&p->sighand->siglock);
1406 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1407 spin_unlock_irq(&p->sighand->siglock);
1412 read_unlock(&tasklist_lock);
1415 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1417 if (!retval && stat_addr)
1418 retval = put_user(0xffff, stat_addr);
1422 retval = wait_noreap_copyout(p, pid, uid,
1423 CLD_CONTINUED, SIGCONT,
1425 BUG_ON(retval == 0);
1432 static inline int my_ptrace_child(struct task_struct *p)
1434 if (!(p->ptrace & PT_PTRACED))
1436 if (!(p->ptrace & PT_ATTACHED))
1439 * This child was PTRACE_ATTACH'd. We should be seeing it only if
1440 * we are the attacher. If we are the real parent, this is a race
1441 * inside ptrace_attach. It is waiting for the tasklist_lock,
1442 * which we have to switch the parent links, but has already set
1443 * the flags in p->ptrace.
1445 return (p->parent != p->real_parent);
1448 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1449 int __user *stat_addr, struct rusage __user *ru)
1451 DECLARE_WAITQUEUE(wait, current);
1452 struct task_struct *tsk;
1455 add_wait_queue(¤t->signal->wait_chldexit,&wait);
1458 * We will set this flag if we see any child that might later
1459 * match our criteria, even if we are not able to reap it yet.
1462 current->state = TASK_INTERRUPTIBLE;
1463 read_lock(&tasklist_lock);
1466 struct task_struct *p;
1467 struct list_head *_p;
1470 list_for_each(_p,&tsk->children) {
1471 p = list_entry(_p, struct task_struct, sibling);
1473 ret = eligible_child(pid, options, p);
1480 * When we hit the race with PTRACE_ATTACH,
1481 * we will not report this child. But the
1482 * race means it has not yet been moved to
1483 * our ptrace_children list, so we need to
1484 * set the flag here to avoid a spurious ECHILD
1485 * when the race happens with the only child.
1488 if (!my_ptrace_child(p))
1493 * It's stopped now, so it might later
1494 * continue, exit, or stop again.
1497 if (!(options & WUNTRACED) &&
1498 !my_ptrace_child(p))
1500 retval = wait_task_stopped(p, ret == 2,
1501 (options & WNOWAIT),
1504 if (retval == -EAGAIN)
1506 if (retval != 0) /* He released the lock. */
1511 if (p->exit_state == EXIT_DEAD)
1513 // case EXIT_ZOMBIE:
1514 if (p->exit_state == EXIT_ZOMBIE) {
1516 * Eligible but we cannot release
1520 goto check_continued;
1521 if (!likely(options & WEXITED))
1523 retval = wait_task_zombie(
1524 p, (options & WNOWAIT),
1525 infop, stat_addr, ru);
1526 /* He released the lock. */
1533 * It's running now, so it might later
1534 * exit, stop, or stop and then continue.
1537 if (!unlikely(options & WCONTINUED))
1539 retval = wait_task_continued(
1540 p, (options & WNOWAIT),
1541 infop, stat_addr, ru);
1542 if (retval != 0) /* He released the lock. */
1548 list_for_each(_p, &tsk->ptrace_children) {
1549 p = list_entry(_p, struct task_struct,
1551 if (!eligible_child(pid, options, p))
1557 if (options & __WNOTHREAD)
1559 tsk = next_thread(tsk);
1560 BUG_ON(tsk->signal != current->signal);
1561 } while (tsk != current);
1563 read_unlock(&tasklist_lock);
1566 if (options & WNOHANG)
1568 retval = -ERESTARTSYS;
1569 if (signal_pending(current))
1576 current->state = TASK_RUNNING;
1577 remove_wait_queue(¤t->signal->wait_chldexit,&wait);
1583 * For a WNOHANG return, clear out all the fields
1584 * we would set so the user can easily tell the
1588 retval = put_user(0, &infop->si_signo);
1590 retval = put_user(0, &infop->si_errno);
1592 retval = put_user(0, &infop->si_code);
1594 retval = put_user(0, &infop->si_pid);
1596 retval = put_user(0, &infop->si_uid);
1598 retval = put_user(0, &infop->si_status);
1604 asmlinkage long sys_waitid(int which, pid_t pid,
1605 struct siginfo __user *infop, int options,
1606 struct rusage __user *ru)
1610 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1612 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1632 ret = do_wait(pid, options, infop, NULL, ru);
1634 /* avoid REGPARM breakage on x86: */
1635 prevent_tail_call(ret);
1639 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1640 int options, struct rusage __user *ru)
1644 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1645 __WNOTHREAD|__WCLONE|__WALL))
1647 ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1649 /* avoid REGPARM breakage on x86: */
1650 prevent_tail_call(ret);
1654 #ifdef __ARCH_WANT_SYS_WAITPID
1657 * sys_waitpid() remains for compatibility. waitpid() should be
1658 * implemented by calling sys_wait4() from libc.a.
1660 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1662 return sys_wait4(pid, stat_addr, options, NULL);