4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/config.h>
9 #include <linux/slab.h>
10 #include <linux/interrupt.h>
11 #include <linux/smp_lock.h>
12 #include <linux/module.h>
13 #include <linux/capability.h>
14 #include <linux/completion.h>
15 #include <linux/personality.h>
16 #include <linux/tty.h>
17 #include <linux/namespace.h>
18 #include <linux/key.h>
19 #include <linux/security.h>
20 #include <linux/cpu.h>
21 #include <linux/acct.h>
22 #include <linux/file.h>
23 #include <linux/binfmts.h>
24 #include <linux/ptrace.h>
25 #include <linux/profile.h>
26 #include <linux/mount.h>
27 #include <linux/proc_fs.h>
28 #include <linux/mempolicy.h>
29 #include <linux/cpuset.h>
30 #include <linux/syscalls.h>
31 #include <linux/signal.h>
32 #include <linux/cn_proc.h>
33 #include <linux/mutex.h>
34 #include <linux/futex.h>
35 #include <linux/compat.h>
37 #include <asm/uaccess.h>
38 #include <asm/unistd.h>
39 #include <asm/pgtable.h>
40 #include <asm/mmu_context.h>
42 extern void sem_exit (void);
43 extern struct task_struct *child_reaper;
45 int getrusage(struct task_struct *, int, struct rusage __user *);
47 static void exit_mm(struct task_struct * tsk);
49 static void __unhash_process(struct task_struct *p)
52 detach_pid(p, PIDTYPE_PID);
53 detach_pid(p, PIDTYPE_TGID);
54 if (thread_group_leader(p)) {
55 detach_pid(p, PIDTYPE_PGID);
56 detach_pid(p, PIDTYPE_SID);
58 list_del_init(&p->tasks);
59 __get_cpu_var(process_counts)--;
65 void release_task(struct task_struct * p)
69 struct dentry *proc_dentry;
72 atomic_dec(&p->user->processes);
73 spin_lock(&p->proc_lock);
74 proc_dentry = proc_pid_unhash(p);
75 write_lock_irq(&tasklist_lock);
77 BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
80 * Note that the fastpath in sys_times depends on __exit_signal having
81 * updated the counters before a task is removed from the tasklist of
82 * the process by __unhash_process.
87 * If we are the last non-leader member of the thread
88 * group, and the leader is zombie, then notify the
89 * group leader's parent process. (if it wants notification.)
92 leader = p->group_leader;
93 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
94 BUG_ON(leader->exit_signal == -1);
95 do_notify_parent(leader, leader->exit_signal);
97 * If we were the last child thread and the leader has
98 * exited already, and the leader's parent ignores SIGCHLD,
99 * then we are the one who should release the leader.
101 * do_notify_parent() will have marked it self-reaping in
104 zap_leader = (leader->exit_signal == -1);
108 write_unlock_irq(&tasklist_lock);
109 spin_unlock(&p->proc_lock);
110 proc_pid_flush(proc_dentry);
115 if (unlikely(zap_leader))
120 * This checks not only the pgrp, but falls back on the pid if no
121 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
124 int session_of_pgrp(int pgrp)
126 struct task_struct *p;
129 read_lock(&tasklist_lock);
130 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
131 if (p->signal->session > 0) {
132 sid = p->signal->session;
135 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
136 p = find_task_by_pid(pgrp);
138 sid = p->signal->session;
140 read_unlock(&tasklist_lock);
146 * Determine if a process group is "orphaned", according to the POSIX
147 * definition in 2.2.2.52. Orphaned process groups are not to be affected
148 * by terminal-generated stop signals. Newly orphaned process groups are
149 * to receive a SIGHUP and a SIGCONT.
151 * "I ask you, have you ever known what it is to be an orphan?"
153 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
155 struct task_struct *p;
158 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
159 if (p == ignored_task
161 || p->real_parent->pid == 1)
163 if (process_group(p->real_parent) != pgrp
164 && p->real_parent->signal->session == p->signal->session) {
168 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
169 return ret; /* (sighing) "Often!" */
172 int is_orphaned_pgrp(int pgrp)
176 read_lock(&tasklist_lock);
177 retval = will_become_orphaned_pgrp(pgrp, NULL);
178 read_unlock(&tasklist_lock);
183 static int has_stopped_jobs(int pgrp)
186 struct task_struct *p;
188 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
189 if (p->state != TASK_STOPPED)
192 /* If p is stopped by a debugger on a signal that won't
193 stop it, then don't count p as stopped. This isn't
194 perfect but it's a good approximation. */
195 if (unlikely (p->ptrace)
196 && p->exit_code != SIGSTOP
197 && p->exit_code != SIGTSTP
198 && p->exit_code != SIGTTOU
199 && p->exit_code != SIGTTIN)
204 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
209 * reparent_to_init - Reparent the calling kernel thread to the init task.
211 * If a kernel thread is launched as a result of a system call, or if
212 * it ever exits, it should generally reparent itself to init so that
213 * it is correctly cleaned up on exit.
215 * The various task state such as scheduling policy and priority may have
216 * been inherited from a user process, so we reset them to sane values here.
218 * NOTE that reparent_to_init() gives the caller full capabilities.
220 static void reparent_to_init(void)
222 write_lock_irq(&tasklist_lock);
224 ptrace_unlink(current);
225 /* Reparent to init */
226 remove_parent(current);
227 current->parent = child_reaper;
228 current->real_parent = child_reaper;
231 /* Set the exit signal to SIGCHLD so we signal init on exit */
232 current->exit_signal = SIGCHLD;
234 if ((current->policy == SCHED_NORMAL ||
235 current->policy == SCHED_BATCH)
236 && (task_nice(current) < 0))
237 set_user_nice(current, 0);
241 security_task_reparent_to_init(current);
242 memcpy(current->signal->rlim, init_task.signal->rlim,
243 sizeof(current->signal->rlim));
244 atomic_inc(&(INIT_USER->__count));
245 write_unlock_irq(&tasklist_lock);
246 switch_uid(INIT_USER);
249 void __set_special_pids(pid_t session, pid_t pgrp)
251 struct task_struct *curr = current->group_leader;
253 if (curr->signal->session != session) {
254 detach_pid(curr, PIDTYPE_SID);
255 curr->signal->session = session;
256 attach_pid(curr, PIDTYPE_SID, session);
258 if (process_group(curr) != pgrp) {
259 detach_pid(curr, PIDTYPE_PGID);
260 curr->signal->pgrp = pgrp;
261 attach_pid(curr, PIDTYPE_PGID, pgrp);
265 void set_special_pids(pid_t session, pid_t pgrp)
267 write_lock_irq(&tasklist_lock);
268 __set_special_pids(session, pgrp);
269 write_unlock_irq(&tasklist_lock);
273 * Let kernel threads use this to say that they
274 * allow a certain signal (since daemonize() will
275 * have disabled all of them by default).
277 int allow_signal(int sig)
279 if (!valid_signal(sig) || sig < 1)
282 spin_lock_irq(¤t->sighand->siglock);
283 sigdelset(¤t->blocked, sig);
285 /* Kernel threads handle their own signals.
286 Let the signal code know it'll be handled, so
287 that they don't get converted to SIGKILL or
288 just silently dropped */
289 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
292 spin_unlock_irq(¤t->sighand->siglock);
296 EXPORT_SYMBOL(allow_signal);
298 int disallow_signal(int sig)
300 if (!valid_signal(sig) || sig < 1)
303 spin_lock_irq(¤t->sighand->siglock);
304 sigaddset(¤t->blocked, sig);
306 spin_unlock_irq(¤t->sighand->siglock);
310 EXPORT_SYMBOL(disallow_signal);
313 * Put all the gunge required to become a kernel thread without
314 * attached user resources in one place where it belongs.
317 void daemonize(const char *name, ...)
320 struct fs_struct *fs;
323 va_start(args, name);
324 vsnprintf(current->comm, sizeof(current->comm), name, args);
328 * If we were started as result of loading a module, close all of the
329 * user space pages. We don't need them, and if we didn't close them
330 * they would be locked into memory.
334 set_special_pids(1, 1);
335 mutex_lock(&tty_mutex);
336 current->signal->tty = NULL;
337 mutex_unlock(&tty_mutex);
339 /* Block and flush all signals */
340 sigfillset(&blocked);
341 sigprocmask(SIG_BLOCK, &blocked, NULL);
342 flush_signals(current);
344 /* Become as one with the init task */
346 exit_fs(current); /* current->fs->count--; */
349 atomic_inc(&fs->count);
350 exit_namespace(current);
351 current->namespace = init_task.namespace;
352 get_namespace(current->namespace);
354 current->files = init_task.files;
355 atomic_inc(¤t->files->count);
360 EXPORT_SYMBOL(daemonize);
362 static void close_files(struct files_struct * files)
370 * It is safe to dereference the fd table without RCU or
371 * ->file_lock because this is the last reference to the
374 fdt = files_fdtable(files);
378 if (i >= fdt->max_fdset || i >= fdt->max_fds)
380 set = fdt->open_fds->fds_bits[j++];
383 struct file * file = xchg(&fdt->fd[i], NULL);
385 filp_close(file, files);
393 struct files_struct *get_files_struct(struct task_struct *task)
395 struct files_struct *files;
400 atomic_inc(&files->count);
406 void fastcall put_files_struct(struct files_struct *files)
410 if (atomic_dec_and_test(&files->count)) {
413 * Free the fd and fdset arrays if we expanded them.
414 * If the fdtable was embedded, pass files for freeing
415 * at the end of the RCU grace period. Otherwise,
416 * you can free files immediately.
418 fdt = files_fdtable(files);
419 if (fdt == &files->fdtab)
420 fdt->free_files = files;
422 kmem_cache_free(files_cachep, files);
427 EXPORT_SYMBOL(put_files_struct);
429 static inline void __exit_files(struct task_struct *tsk)
431 struct files_struct * files = tsk->files;
437 put_files_struct(files);
441 void exit_files(struct task_struct *tsk)
446 static inline void __put_fs_struct(struct fs_struct *fs)
448 /* No need to hold fs->lock if we are killing it */
449 if (atomic_dec_and_test(&fs->count)) {
456 mntput(fs->altrootmnt);
458 kmem_cache_free(fs_cachep, fs);
462 void put_fs_struct(struct fs_struct *fs)
467 static inline void __exit_fs(struct task_struct *tsk)
469 struct fs_struct * fs = tsk->fs;
479 void exit_fs(struct task_struct *tsk)
484 EXPORT_SYMBOL_GPL(exit_fs);
487 * Turn us into a lazy TLB process if we
490 static void exit_mm(struct task_struct * tsk)
492 struct mm_struct *mm = tsk->mm;
498 * Serialize with any possible pending coredump.
499 * We must hold mmap_sem around checking core_waiters
500 * and clearing tsk->mm. The core-inducing thread
501 * will increment core_waiters for each thread in the
502 * group with ->mm != NULL.
504 down_read(&mm->mmap_sem);
505 if (mm->core_waiters) {
506 up_read(&mm->mmap_sem);
507 down_write(&mm->mmap_sem);
508 if (!--mm->core_waiters)
509 complete(mm->core_startup_done);
510 up_write(&mm->mmap_sem);
512 wait_for_completion(&mm->core_done);
513 down_read(&mm->mmap_sem);
515 atomic_inc(&mm->mm_count);
516 if (mm != tsk->active_mm) BUG();
517 /* more a memory barrier than a real lock */
520 up_read(&mm->mmap_sem);
521 enter_lazy_tlb(mm, current);
526 static inline void choose_new_parent(task_t *p, task_t *reaper)
529 * Make sure we're not reparenting to ourselves and that
530 * the parent is not a zombie.
532 BUG_ON(p == reaper || reaper->exit_state);
533 p->real_parent = reaper;
536 static void reparent_thread(task_t *p, task_t *father, int traced)
538 /* We don't want people slaying init. */
539 if (p->exit_signal != -1)
540 p->exit_signal = SIGCHLD;
542 if (p->pdeath_signal)
543 /* We already hold the tasklist_lock here. */
544 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
546 /* Move the child from its dying parent to the new one. */
547 if (unlikely(traced)) {
548 /* Preserve ptrace links if someone else is tracing this child. */
549 list_del_init(&p->ptrace_list);
550 if (p->parent != p->real_parent)
551 list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
553 /* If this child is being traced, then we're the one tracing it
554 * anyway, so let go of it.
558 p->parent = p->real_parent;
561 /* If we'd notified the old parent about this child's death,
562 * also notify the new parent.
564 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
565 thread_group_empty(p))
566 do_notify_parent(p, p->exit_signal);
567 else if (p->state == TASK_TRACED) {
569 * If it was at a trace stop, turn it into
570 * a normal stop since it's no longer being
578 * process group orphan check
579 * Case ii: Our child is in a different pgrp
580 * than we are, and it was the only connection
581 * outside, so the child pgrp is now orphaned.
583 if ((process_group(p) != process_group(father)) &&
584 (p->signal->session == father->signal->session)) {
585 int pgrp = process_group(p);
587 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
588 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
589 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
595 * When we die, we re-parent all our children.
596 * Try to give them to another thread in our thread
597 * group, and if no such member exists, give it to
598 * the global child reaper process (ie "init")
600 static void forget_original_parent(struct task_struct * father,
601 struct list_head *to_release)
603 struct task_struct *p, *reaper = father;
604 struct list_head *_p, *_n;
607 reaper = next_thread(reaper);
608 if (reaper == father) {
609 reaper = child_reaper;
612 } while (reaper->exit_state);
615 * There are only two places where our children can be:
617 * - in our child list
618 * - in our ptraced child list
620 * Search them and reparent children.
622 list_for_each_safe(_p, _n, &father->children) {
624 p = list_entry(_p,struct task_struct,sibling);
628 /* if father isn't the real parent, then ptrace must be enabled */
629 BUG_ON(father != p->real_parent && !ptrace);
631 if (father == p->real_parent) {
632 /* reparent with a reaper, real father it's us */
633 choose_new_parent(p, reaper);
634 reparent_thread(p, father, 0);
636 /* reparent ptraced task to its real parent */
638 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
639 thread_group_empty(p))
640 do_notify_parent(p, p->exit_signal);
644 * if the ptraced child is a zombie with exit_signal == -1
645 * we must collect it before we exit, or it will remain
646 * zombie forever since we prevented it from self-reap itself
647 * while it was being traced by us, to be able to see it in wait4.
649 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
650 list_add(&p->ptrace_list, to_release);
652 list_for_each_safe(_p, _n, &father->ptrace_children) {
653 p = list_entry(_p,struct task_struct,ptrace_list);
654 choose_new_parent(p, reaper);
655 reparent_thread(p, father, 1);
660 * Send signals to all our closest relatives so that they know
661 * to properly mourn us..
663 static void exit_notify(struct task_struct *tsk)
666 struct task_struct *t;
667 struct list_head ptrace_dead, *_p, *_n;
669 if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
670 && !thread_group_empty(tsk)) {
672 * This occurs when there was a race between our exit
673 * syscall and a group signal choosing us as the one to
674 * wake up. It could be that we are the only thread
675 * alerted to check for pending signals, but another thread
676 * should be woken now to take the signal since we will not.
677 * Now we'll wake all the threads in the group just to make
678 * sure someone gets all the pending signals.
680 read_lock(&tasklist_lock);
681 spin_lock_irq(&tsk->sighand->siglock);
682 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
683 if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
684 recalc_sigpending_tsk(t);
685 if (signal_pending(t))
686 signal_wake_up(t, 0);
688 spin_unlock_irq(&tsk->sighand->siglock);
689 read_unlock(&tasklist_lock);
692 write_lock_irq(&tasklist_lock);
695 * This does two things:
697 * A. Make init inherit all the child processes
698 * B. Check to see if any process groups have become orphaned
699 * as a result of our exiting, and if they have any stopped
700 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
703 INIT_LIST_HEAD(&ptrace_dead);
704 forget_original_parent(tsk, &ptrace_dead);
705 BUG_ON(!list_empty(&tsk->children));
706 BUG_ON(!list_empty(&tsk->ptrace_children));
709 * Check to see if any process groups have become orphaned
710 * as a result of our exiting, and if they have any stopped
711 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
713 * Case i: Our father is in a different pgrp than we are
714 * and we were the only connection outside, so our pgrp
715 * is about to become orphaned.
718 t = tsk->real_parent;
720 if ((process_group(t) != process_group(tsk)) &&
721 (t->signal->session == tsk->signal->session) &&
722 will_become_orphaned_pgrp(process_group(tsk), tsk) &&
723 has_stopped_jobs(process_group(tsk))) {
724 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
725 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
728 /* Let father know we died
730 * Thread signals are configurable, but you aren't going to use
731 * that to send signals to arbitary processes.
732 * That stops right now.
734 * If the parent exec id doesn't match the exec id we saved
735 * when we started then we know the parent has changed security
738 * If our self_exec id doesn't match our parent_exec_id then
739 * we have changed execution domain as these two values started
740 * the same after a fork.
744 if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
745 ( tsk->parent_exec_id != t->self_exec_id ||
746 tsk->self_exec_id != tsk->parent_exec_id)
747 && !capable(CAP_KILL))
748 tsk->exit_signal = SIGCHLD;
751 /* If something other than our normal parent is ptracing us, then
752 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
753 * only has special meaning to our real parent.
755 if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
756 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
757 do_notify_parent(tsk, signal);
758 } else if (tsk->ptrace) {
759 do_notify_parent(tsk, SIGCHLD);
763 if (tsk->exit_signal == -1 &&
764 (likely(tsk->ptrace == 0) ||
765 unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
767 tsk->exit_state = state;
769 write_unlock_irq(&tasklist_lock);
771 list_for_each_safe(_p, _n, &ptrace_dead) {
773 t = list_entry(_p,struct task_struct,ptrace_list);
777 /* If the process is dead, release it - nobody will wait for it */
778 if (state == EXIT_DEAD)
782 fastcall NORET_TYPE void do_exit(long code)
784 struct task_struct *tsk = current;
787 profile_task_exit(tsk);
789 WARN_ON(atomic_read(&tsk->fs_excl));
791 if (unlikely(in_interrupt()))
792 panic("Aiee, killing interrupt handler!");
793 if (unlikely(!tsk->pid))
794 panic("Attempted to kill the idle task!");
795 if (unlikely(tsk == child_reaper))
796 panic("Attempted to kill init!");
798 if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
799 current->ptrace_message = code;
800 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
804 * We're taking recursive faults here in do_exit. Safest is to just
805 * leave this task alone and wait for reboot.
807 if (unlikely(tsk->flags & PF_EXITING)) {
809 "Fixing recursive fault but reboot is needed!\n");
812 set_current_state(TASK_UNINTERRUPTIBLE);
816 tsk->flags |= PF_EXITING;
819 * Make sure we don't try to process any timer firings
820 * while we are already exiting.
822 tsk->it_virt_expires = cputime_zero;
823 tsk->it_prof_expires = cputime_zero;
824 tsk->it_sched_expires = 0;
826 if (unlikely(in_atomic()))
827 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
828 current->comm, current->pid,
831 acct_update_integrals(tsk);
833 update_hiwater_rss(tsk->mm);
834 update_hiwater_vm(tsk->mm);
836 group_dead = atomic_dec_and_test(&tsk->signal->live);
838 hrtimer_cancel(&tsk->signal->real_timer);
839 exit_itimers(tsk->signal);
842 if (unlikely(tsk->robust_list))
843 exit_robust_list(tsk);
845 if (unlikely(tsk->compat_robust_list))
846 compat_exit_robust_list(tsk);
858 if (group_dead && tsk->signal->leader)
859 disassociate_ctty(1);
861 module_put(task_thread_info(tsk)->exec_domain->module);
863 module_put(tsk->binfmt->module);
865 tsk->exit_code = code;
866 proc_exit_connector(tsk);
869 mpol_free(tsk->mempolicy);
870 tsk->mempolicy = NULL;
873 * If DEBUG_MUTEXES is on, make sure we are holding no locks:
875 mutex_debug_check_no_locks_held(tsk);
880 /* PF_DEAD causes final put_task_struct after we schedule. */
882 BUG_ON(tsk->flags & PF_DEAD);
883 tsk->flags |= PF_DEAD;
887 /* Avoid "noreturn function does return". */
891 EXPORT_SYMBOL_GPL(do_exit);
893 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
901 EXPORT_SYMBOL(complete_and_exit);
903 asmlinkage long sys_exit(int error_code)
905 do_exit((error_code&0xff)<<8);
908 task_t fastcall *next_thread(const task_t *p)
910 return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
913 EXPORT_SYMBOL(next_thread);
916 * Take down every thread in the group. This is called by fatal signals
917 * as well as by sys_exit_group (below).
920 do_group_exit(int exit_code)
922 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
924 if (current->signal->flags & SIGNAL_GROUP_EXIT)
925 exit_code = current->signal->group_exit_code;
926 else if (!thread_group_empty(current)) {
927 struct signal_struct *const sig = current->signal;
928 struct sighand_struct *const sighand = current->sighand;
929 read_lock(&tasklist_lock);
930 spin_lock_irq(&sighand->siglock);
931 if (sig->flags & SIGNAL_GROUP_EXIT)
932 /* Another thread got here before we took the lock. */
933 exit_code = sig->group_exit_code;
935 sig->group_exit_code = exit_code;
936 zap_other_threads(current);
938 spin_unlock_irq(&sighand->siglock);
939 read_unlock(&tasklist_lock);
947 * this kills every thread in the thread group. Note that any externally
948 * wait4()-ing process will get the correct exit code - even if this
949 * thread is not the thread group leader.
951 asmlinkage void sys_exit_group(int error_code)
953 do_group_exit((error_code & 0xff) << 8);
956 static int eligible_child(pid_t pid, int options, task_t *p)
962 if (process_group(p) != process_group(current))
964 } else if (pid != -1) {
965 if (process_group(p) != -pid)
970 * Do not consider detached threads that are
973 if (p->exit_signal == -1 && !p->ptrace)
976 /* Wait for all children (clone and not) if __WALL is set;
977 * otherwise, wait for clone children *only* if __WCLONE is
978 * set; otherwise, wait for non-clone children *only*. (Note:
979 * A "clone" child here is one that reports to its parent
980 * using a signal other than SIGCHLD.) */
981 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
982 && !(options & __WALL))
985 * Do not consider thread group leaders that are
986 * in a non-empty thread group:
988 if (current->tgid != p->tgid && delay_group_leader(p))
991 if (security_task_wait(p))
997 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
999 struct siginfo __user *infop,
1000 struct rusage __user *rusagep)
1002 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1005 retval = put_user(SIGCHLD, &infop->si_signo);
1007 retval = put_user(0, &infop->si_errno);
1009 retval = put_user((short)why, &infop->si_code);
1011 retval = put_user(pid, &infop->si_pid);
1013 retval = put_user(uid, &infop->si_uid);
1015 retval = put_user(status, &infop->si_status);
1022 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1023 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1024 * the lock and this task is uninteresting. If we return nonzero, we have
1025 * released the lock and the system call should return.
1027 static int wait_task_zombie(task_t *p, int noreap,
1028 struct siginfo __user *infop,
1029 int __user *stat_addr, struct rusage __user *ru)
1031 unsigned long state;
1035 if (unlikely(noreap)) {
1038 int exit_code = p->exit_code;
1041 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1043 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1046 read_unlock(&tasklist_lock);
1047 if ((exit_code & 0x7f) == 0) {
1049 status = exit_code >> 8;
1051 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1052 status = exit_code & 0x7f;
1054 return wait_noreap_copyout(p, pid, uid, why,
1059 * Try to move the task's state to DEAD
1060 * only one thread is allowed to do this:
1062 state = xchg(&p->exit_state, EXIT_DEAD);
1063 if (state != EXIT_ZOMBIE) {
1064 BUG_ON(state != EXIT_DEAD);
1067 if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1069 * This can only happen in a race with a ptraced thread
1070 * dying on another processor.
1075 if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1076 struct signal_struct *psig;
1077 struct signal_struct *sig;
1080 * The resource counters for the group leader are in its
1081 * own task_struct. Those for dead threads in the group
1082 * are in its signal_struct, as are those for the child
1083 * processes it has previously reaped. All these
1084 * accumulate in the parent's signal_struct c* fields.
1086 * We don't bother to take a lock here to protect these
1087 * p->signal fields, because they are only touched by
1088 * __exit_signal, which runs with tasklist_lock
1089 * write-locked anyway, and so is excluded here. We do
1090 * need to protect the access to p->parent->signal fields,
1091 * as other threads in the parent group can be right
1092 * here reaping other children at the same time.
1094 spin_lock_irq(&p->parent->sighand->siglock);
1095 psig = p->parent->signal;
1098 cputime_add(psig->cutime,
1099 cputime_add(p->utime,
1100 cputime_add(sig->utime,
1103 cputime_add(psig->cstime,
1104 cputime_add(p->stime,
1105 cputime_add(sig->stime,
1108 p->min_flt + sig->min_flt + sig->cmin_flt;
1110 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1112 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1114 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1115 spin_unlock_irq(&p->parent->sighand->siglock);
1119 * Now we are sure this task is interesting, and no other
1120 * thread can reap it because we set its state to EXIT_DEAD.
1122 read_unlock(&tasklist_lock);
1124 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1125 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1126 ? p->signal->group_exit_code : p->exit_code;
1127 if (!retval && stat_addr)
1128 retval = put_user(status, stat_addr);
1129 if (!retval && infop)
1130 retval = put_user(SIGCHLD, &infop->si_signo);
1131 if (!retval && infop)
1132 retval = put_user(0, &infop->si_errno);
1133 if (!retval && infop) {
1136 if ((status & 0x7f) == 0) {
1140 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1143 retval = put_user((short)why, &infop->si_code);
1145 retval = put_user(status, &infop->si_status);
1147 if (!retval && infop)
1148 retval = put_user(p->pid, &infop->si_pid);
1149 if (!retval && infop)
1150 retval = put_user(p->uid, &infop->si_uid);
1152 // TODO: is this safe?
1153 p->exit_state = EXIT_ZOMBIE;
1157 if (p->real_parent != p->parent) {
1158 write_lock_irq(&tasklist_lock);
1159 /* Double-check with lock held. */
1160 if (p->real_parent != p->parent) {
1162 // TODO: is this safe?
1163 p->exit_state = EXIT_ZOMBIE;
1165 * If this is not a detached task, notify the parent.
1166 * If it's still not detached after that, don't release
1169 if (p->exit_signal != -1) {
1170 do_notify_parent(p, p->exit_signal);
1171 if (p->exit_signal != -1)
1175 write_unlock_irq(&tasklist_lock);
1184 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1185 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1186 * the lock and this task is uninteresting. If we return nonzero, we have
1187 * released the lock and the system call should return.
1189 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1190 struct siginfo __user *infop,
1191 int __user *stat_addr, struct rusage __user *ru)
1193 int retval, exit_code;
1197 if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1198 p->signal && p->signal->group_stop_count > 0)
1200 * A group stop is in progress and this is the group leader.
1201 * We won't report until all threads have stopped.
1206 * Now we are pretty sure this task is interesting.
1207 * Make sure it doesn't get reaped out from under us while we
1208 * give up the lock and then examine it below. We don't want to
1209 * keep holding onto the tasklist_lock while we call getrusage and
1210 * possibly take page faults for user memory.
1213 read_unlock(&tasklist_lock);
1215 if (unlikely(noreap)) {
1218 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1220 exit_code = p->exit_code;
1221 if (unlikely(!exit_code) ||
1222 unlikely(p->state & TASK_TRACED))
1224 return wait_noreap_copyout(p, pid, uid,
1225 why, (exit_code << 8) | 0x7f,
1229 write_lock_irq(&tasklist_lock);
1232 * This uses xchg to be atomic with the thread resuming and setting
1233 * it. It must also be done with the write lock held to prevent a
1234 * race with the EXIT_ZOMBIE case.
1236 exit_code = xchg(&p->exit_code, 0);
1237 if (unlikely(p->exit_state)) {
1239 * The task resumed and then died. Let the next iteration
1240 * catch it in EXIT_ZOMBIE. Note that exit_code might
1241 * already be zero here if it resumed and did _exit(0).
1242 * The task itself is dead and won't touch exit_code again;
1243 * other processors in this function are locked out.
1245 p->exit_code = exit_code;
1248 if (unlikely(exit_code == 0)) {
1250 * Another thread in this function got to it first, or it
1251 * resumed, or it resumed and then died.
1253 write_unlock_irq(&tasklist_lock);
1257 * We are returning to the wait loop without having successfully
1258 * removed the process and having released the lock. We cannot
1259 * continue, since the "p" task pointer is potentially stale.
1261 * Return -EAGAIN, and do_wait() will restart the loop from the
1262 * beginning. Do _not_ re-acquire the lock.
1267 /* move to end of parent's list to avoid starvation */
1271 write_unlock_irq(&tasklist_lock);
1273 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1274 if (!retval && stat_addr)
1275 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1276 if (!retval && infop)
1277 retval = put_user(SIGCHLD, &infop->si_signo);
1278 if (!retval && infop)
1279 retval = put_user(0, &infop->si_errno);
1280 if (!retval && infop)
1281 retval = put_user((short)((p->ptrace & PT_PTRACED)
1282 ? CLD_TRAPPED : CLD_STOPPED),
1284 if (!retval && infop)
1285 retval = put_user(exit_code, &infop->si_status);
1286 if (!retval && infop)
1287 retval = put_user(p->pid, &infop->si_pid);
1288 if (!retval && infop)
1289 retval = put_user(p->uid, &infop->si_uid);
1299 * Handle do_wait work for one task in a live, non-stopped state.
1300 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1301 * the lock and this task is uninteresting. If we return nonzero, we have
1302 * released the lock and the system call should return.
1304 static int wait_task_continued(task_t *p, int noreap,
1305 struct siginfo __user *infop,
1306 int __user *stat_addr, struct rusage __user *ru)
1312 if (unlikely(!p->signal))
1315 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1318 spin_lock_irq(&p->sighand->siglock);
1319 /* Re-check with the lock held. */
1320 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1321 spin_unlock_irq(&p->sighand->siglock);
1325 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1326 spin_unlock_irq(&p->sighand->siglock);
1331 read_unlock(&tasklist_lock);
1334 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1336 if (!retval && stat_addr)
1337 retval = put_user(0xffff, stat_addr);
1341 retval = wait_noreap_copyout(p, pid, uid,
1342 CLD_CONTINUED, SIGCONT,
1344 BUG_ON(retval == 0);
1351 static inline int my_ptrace_child(struct task_struct *p)
1353 if (!(p->ptrace & PT_PTRACED))
1355 if (!(p->ptrace & PT_ATTACHED))
1358 * This child was PTRACE_ATTACH'd. We should be seeing it only if
1359 * we are the attacher. If we are the real parent, this is a race
1360 * inside ptrace_attach. It is waiting for the tasklist_lock,
1361 * which we have to switch the parent links, but has already set
1362 * the flags in p->ptrace.
1364 return (p->parent != p->real_parent);
1367 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1368 int __user *stat_addr, struct rusage __user *ru)
1370 DECLARE_WAITQUEUE(wait, current);
1371 struct task_struct *tsk;
1374 add_wait_queue(¤t->signal->wait_chldexit,&wait);
1377 * We will set this flag if we see any child that might later
1378 * match our criteria, even if we are not able to reap it yet.
1381 current->state = TASK_INTERRUPTIBLE;
1382 read_lock(&tasklist_lock);
1385 struct task_struct *p;
1386 struct list_head *_p;
1389 list_for_each(_p,&tsk->children) {
1390 p = list_entry(_p,struct task_struct,sibling);
1392 ret = eligible_child(pid, options, p);
1399 * When we hit the race with PTRACE_ATTACH,
1400 * we will not report this child. But the
1401 * race means it has not yet been moved to
1402 * our ptrace_children list, so we need to
1403 * set the flag here to avoid a spurious ECHILD
1404 * when the race happens with the only child.
1407 if (!my_ptrace_child(p))
1412 * It's stopped now, so it might later
1413 * continue, exit, or stop again.
1416 if (!(options & WUNTRACED) &&
1417 !my_ptrace_child(p))
1419 retval = wait_task_stopped(p, ret == 2,
1420 (options & WNOWAIT),
1423 if (retval == -EAGAIN)
1425 if (retval != 0) /* He released the lock. */
1430 if (p->exit_state == EXIT_DEAD)
1432 // case EXIT_ZOMBIE:
1433 if (p->exit_state == EXIT_ZOMBIE) {
1435 * Eligible but we cannot release
1439 goto check_continued;
1440 if (!likely(options & WEXITED))
1442 retval = wait_task_zombie(
1443 p, (options & WNOWAIT),
1444 infop, stat_addr, ru);
1445 /* He released the lock. */
1452 * It's running now, so it might later
1453 * exit, stop, or stop and then continue.
1456 if (!unlikely(options & WCONTINUED))
1458 retval = wait_task_continued(
1459 p, (options & WNOWAIT),
1460 infop, stat_addr, ru);
1461 if (retval != 0) /* He released the lock. */
1467 list_for_each(_p, &tsk->ptrace_children) {
1468 p = list_entry(_p, struct task_struct,
1470 if (!eligible_child(pid, options, p))
1476 if (options & __WNOTHREAD)
1478 tsk = next_thread(tsk);
1479 if (tsk->signal != current->signal)
1481 } while (tsk != current);
1483 read_unlock(&tasklist_lock);
1486 if (options & WNOHANG)
1488 retval = -ERESTARTSYS;
1489 if (signal_pending(current))
1496 current->state = TASK_RUNNING;
1497 remove_wait_queue(¤t->signal->wait_chldexit,&wait);
1503 * For a WNOHANG return, clear out all the fields
1504 * we would set so the user can easily tell the
1508 retval = put_user(0, &infop->si_signo);
1510 retval = put_user(0, &infop->si_errno);
1512 retval = put_user(0, &infop->si_code);
1514 retval = put_user(0, &infop->si_pid);
1516 retval = put_user(0, &infop->si_uid);
1518 retval = put_user(0, &infop->si_status);
1524 asmlinkage long sys_waitid(int which, pid_t pid,
1525 struct siginfo __user *infop, int options,
1526 struct rusage __user *ru)
1530 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1532 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1552 ret = do_wait(pid, options, infop, NULL, ru);
1554 /* avoid REGPARM breakage on x86: */
1555 prevent_tail_call(ret);
1559 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1560 int options, struct rusage __user *ru)
1564 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1565 __WNOTHREAD|__WCLONE|__WALL))
1567 ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1569 /* avoid REGPARM breakage on x86: */
1570 prevent_tail_call(ret);
1574 #ifdef __ARCH_WANT_SYS_WAITPID
1577 * sys_waitpid() remains for compatibility. waitpid() should be
1578 * implemented by calling sys_wait4() from libc.a.
1580 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1582 return sys_wait4(pid, stat_addr, options, NULL);