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);
60 __get_cpu_var(process_counts)--;
66 void release_task(struct task_struct * p)
70 struct dentry *proc_dentry;
73 atomic_dec(&p->user->processes);
74 spin_lock(&p->proc_lock);
75 proc_dentry = proc_pid_unhash(p);
76 write_lock_irq(&tasklist_lock);
77 if (unlikely(p->ptrace))
79 BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
82 * Note that the fastpath in sys_times depends on __exit_signal having
83 * updated the counters before a task is removed from the tasklist of
84 * the process by __unhash_process.
89 * If we are the last non-leader member of the thread
90 * group, and the leader is zombie, then notify the
91 * group leader's parent process. (if it wants notification.)
94 leader = p->group_leader;
95 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
96 BUG_ON(leader->exit_signal == -1);
97 do_notify_parent(leader, leader->exit_signal);
99 * If we were the last child thread and the leader has
100 * exited already, and the leader's parent ignores SIGCHLD,
101 * then we are the one who should release the leader.
103 * do_notify_parent() will have marked it self-reaping in
106 zap_leader = (leader->exit_signal == -1);
110 write_unlock_irq(&tasklist_lock);
111 spin_unlock(&p->proc_lock);
112 proc_pid_flush(proc_dentry);
117 if (unlikely(zap_leader))
121 /* we are using it only for SMP init */
123 void unhash_process(struct task_struct *p)
125 struct dentry *proc_dentry;
127 spin_lock(&p->proc_lock);
128 proc_dentry = proc_pid_unhash(p);
129 write_lock_irq(&tasklist_lock);
131 write_unlock_irq(&tasklist_lock);
132 spin_unlock(&p->proc_lock);
133 proc_pid_flush(proc_dentry);
137 * This checks not only the pgrp, but falls back on the pid if no
138 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
141 int session_of_pgrp(int pgrp)
143 struct task_struct *p;
146 read_lock(&tasklist_lock);
147 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
148 if (p->signal->session > 0) {
149 sid = p->signal->session;
152 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
153 p = find_task_by_pid(pgrp);
155 sid = p->signal->session;
157 read_unlock(&tasklist_lock);
163 * Determine if a process group is "orphaned", according to the POSIX
164 * definition in 2.2.2.52. Orphaned process groups are not to be affected
165 * by terminal-generated stop signals. Newly orphaned process groups are
166 * to receive a SIGHUP and a SIGCONT.
168 * "I ask you, have you ever known what it is to be an orphan?"
170 static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
172 struct task_struct *p;
175 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
176 if (p == ignored_task
178 || p->real_parent->pid == 1)
180 if (process_group(p->real_parent) != pgrp
181 && p->real_parent->signal->session == p->signal->session) {
185 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
186 return ret; /* (sighing) "Often!" */
189 int is_orphaned_pgrp(int pgrp)
193 read_lock(&tasklist_lock);
194 retval = will_become_orphaned_pgrp(pgrp, NULL);
195 read_unlock(&tasklist_lock);
200 static int has_stopped_jobs(int pgrp)
203 struct task_struct *p;
205 do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
206 if (p->state != TASK_STOPPED)
209 /* If p is stopped by a debugger on a signal that won't
210 stop it, then don't count p as stopped. This isn't
211 perfect but it's a good approximation. */
212 if (unlikely (p->ptrace)
213 && p->exit_code != SIGSTOP
214 && p->exit_code != SIGTSTP
215 && p->exit_code != SIGTTOU
216 && p->exit_code != SIGTTIN)
221 } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
226 * reparent_to_init - Reparent the calling kernel thread to the init task.
228 * If a kernel thread is launched as a result of a system call, or if
229 * it ever exits, it should generally reparent itself to init so that
230 * it is correctly cleaned up on exit.
232 * The various task state such as scheduling policy and priority may have
233 * been inherited from a user process, so we reset them to sane values here.
235 * NOTE that reparent_to_init() gives the caller full capabilities.
237 static void reparent_to_init(void)
239 write_lock_irq(&tasklist_lock);
241 ptrace_unlink(current);
242 /* Reparent to init */
243 remove_parent(current);
244 current->parent = child_reaper;
245 current->real_parent = child_reaper;
248 /* Set the exit signal to SIGCHLD so we signal init on exit */
249 current->exit_signal = SIGCHLD;
251 if ((current->policy == SCHED_NORMAL ||
252 current->policy == SCHED_BATCH)
253 && (task_nice(current) < 0))
254 set_user_nice(current, 0);
258 security_task_reparent_to_init(current);
259 memcpy(current->signal->rlim, init_task.signal->rlim,
260 sizeof(current->signal->rlim));
261 atomic_inc(&(INIT_USER->__count));
262 write_unlock_irq(&tasklist_lock);
263 switch_uid(INIT_USER);
266 void __set_special_pids(pid_t session, pid_t pgrp)
268 struct task_struct *curr = current->group_leader;
270 if (curr->signal->session != session) {
271 detach_pid(curr, PIDTYPE_SID);
272 curr->signal->session = session;
273 attach_pid(curr, PIDTYPE_SID, session);
275 if (process_group(curr) != pgrp) {
276 detach_pid(curr, PIDTYPE_PGID);
277 curr->signal->pgrp = pgrp;
278 attach_pid(curr, PIDTYPE_PGID, pgrp);
282 void set_special_pids(pid_t session, pid_t pgrp)
284 write_lock_irq(&tasklist_lock);
285 __set_special_pids(session, pgrp);
286 write_unlock_irq(&tasklist_lock);
290 * Let kernel threads use this to say that they
291 * allow a certain signal (since daemonize() will
292 * have disabled all of them by default).
294 int allow_signal(int sig)
296 if (!valid_signal(sig) || sig < 1)
299 spin_lock_irq(¤t->sighand->siglock);
300 sigdelset(¤t->blocked, sig);
302 /* Kernel threads handle their own signals.
303 Let the signal code know it'll be handled, so
304 that they don't get converted to SIGKILL or
305 just silently dropped */
306 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
309 spin_unlock_irq(¤t->sighand->siglock);
313 EXPORT_SYMBOL(allow_signal);
315 int disallow_signal(int sig)
317 if (!valid_signal(sig) || sig < 1)
320 spin_lock_irq(¤t->sighand->siglock);
321 sigaddset(¤t->blocked, sig);
323 spin_unlock_irq(¤t->sighand->siglock);
327 EXPORT_SYMBOL(disallow_signal);
330 * Put all the gunge required to become a kernel thread without
331 * attached user resources in one place where it belongs.
334 void daemonize(const char *name, ...)
337 struct fs_struct *fs;
340 va_start(args, name);
341 vsnprintf(current->comm, sizeof(current->comm), name, args);
345 * If we were started as result of loading a module, close all of the
346 * user space pages. We don't need them, and if we didn't close them
347 * they would be locked into memory.
351 set_special_pids(1, 1);
352 mutex_lock(&tty_mutex);
353 current->signal->tty = NULL;
354 mutex_unlock(&tty_mutex);
356 /* Block and flush all signals */
357 sigfillset(&blocked);
358 sigprocmask(SIG_BLOCK, &blocked, NULL);
359 flush_signals(current);
361 /* Become as one with the init task */
363 exit_fs(current); /* current->fs->count--; */
366 atomic_inc(&fs->count);
367 exit_namespace(current);
368 current->namespace = init_task.namespace;
369 get_namespace(current->namespace);
371 current->files = init_task.files;
372 atomic_inc(¤t->files->count);
377 EXPORT_SYMBOL(daemonize);
379 static void close_files(struct files_struct * files)
387 * It is safe to dereference the fd table without RCU or
388 * ->file_lock because this is the last reference to the
391 fdt = files_fdtable(files);
395 if (i >= fdt->max_fdset || i >= fdt->max_fds)
397 set = fdt->open_fds->fds_bits[j++];
400 struct file * file = xchg(&fdt->fd[i], NULL);
402 filp_close(file, files);
410 struct files_struct *get_files_struct(struct task_struct *task)
412 struct files_struct *files;
417 atomic_inc(&files->count);
423 void fastcall put_files_struct(struct files_struct *files)
427 if (atomic_dec_and_test(&files->count)) {
430 * Free the fd and fdset arrays if we expanded them.
431 * If the fdtable was embedded, pass files for freeing
432 * at the end of the RCU grace period. Otherwise,
433 * you can free files immediately.
435 fdt = files_fdtable(files);
436 if (fdt == &files->fdtab)
437 fdt->free_files = files;
439 kmem_cache_free(files_cachep, files);
444 EXPORT_SYMBOL(put_files_struct);
446 static inline void __exit_files(struct task_struct *tsk)
448 struct files_struct * files = tsk->files;
454 put_files_struct(files);
458 void exit_files(struct task_struct *tsk)
463 static inline void __put_fs_struct(struct fs_struct *fs)
465 /* No need to hold fs->lock if we are killing it */
466 if (atomic_dec_and_test(&fs->count)) {
473 mntput(fs->altrootmnt);
475 kmem_cache_free(fs_cachep, fs);
479 void put_fs_struct(struct fs_struct *fs)
484 static inline void __exit_fs(struct task_struct *tsk)
486 struct fs_struct * fs = tsk->fs;
496 void exit_fs(struct task_struct *tsk)
501 EXPORT_SYMBOL_GPL(exit_fs);
504 * Turn us into a lazy TLB process if we
507 static void exit_mm(struct task_struct * tsk)
509 struct mm_struct *mm = tsk->mm;
515 * Serialize with any possible pending coredump.
516 * We must hold mmap_sem around checking core_waiters
517 * and clearing tsk->mm. The core-inducing thread
518 * will increment core_waiters for each thread in the
519 * group with ->mm != NULL.
521 down_read(&mm->mmap_sem);
522 if (mm->core_waiters) {
523 up_read(&mm->mmap_sem);
524 down_write(&mm->mmap_sem);
525 if (!--mm->core_waiters)
526 complete(mm->core_startup_done);
527 up_write(&mm->mmap_sem);
529 wait_for_completion(&mm->core_done);
530 down_read(&mm->mmap_sem);
532 atomic_inc(&mm->mm_count);
533 if (mm != tsk->active_mm) BUG();
534 /* more a memory barrier than a real lock */
537 up_read(&mm->mmap_sem);
538 enter_lazy_tlb(mm, current);
543 static inline void choose_new_parent(task_t *p, task_t *reaper)
546 * Make sure we're not reparenting to ourselves and that
547 * the parent is not a zombie.
549 BUG_ON(p == reaper || reaper->exit_state);
550 p->real_parent = reaper;
553 static void reparent_thread(task_t *p, task_t *father, int traced)
555 /* We don't want people slaying init. */
556 if (p->exit_signal != -1)
557 p->exit_signal = SIGCHLD;
559 if (p->pdeath_signal)
560 /* We already hold the tasklist_lock here. */
561 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
563 /* Move the child from its dying parent to the new one. */
564 if (unlikely(traced)) {
565 /* Preserve ptrace links if someone else is tracing this child. */
566 list_del_init(&p->ptrace_list);
567 if (p->parent != p->real_parent)
568 list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
570 /* If this child is being traced, then we're the one tracing it
571 * anyway, so let go of it.
574 list_del_init(&p->sibling);
575 p->parent = p->real_parent;
576 list_add_tail(&p->sibling, &p->parent->children);
578 /* If we'd notified the old parent about this child's death,
579 * also notify the new parent.
581 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
582 thread_group_empty(p))
583 do_notify_parent(p, p->exit_signal);
584 else if (p->state == TASK_TRACED) {
586 * If it was at a trace stop, turn it into
587 * a normal stop since it's no longer being
595 * process group orphan check
596 * Case ii: Our child is in a different pgrp
597 * than we are, and it was the only connection
598 * outside, so the child pgrp is now orphaned.
600 if ((process_group(p) != process_group(father)) &&
601 (p->signal->session == father->signal->session)) {
602 int pgrp = process_group(p);
604 if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
605 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
606 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
612 * When we die, we re-parent all our children.
613 * Try to give them to another thread in our thread
614 * group, and if no such member exists, give it to
615 * the global child reaper process (ie "init")
617 static void forget_original_parent(struct task_struct * father,
618 struct list_head *to_release)
620 struct task_struct *p, *reaper = father;
621 struct list_head *_p, *_n;
624 reaper = next_thread(reaper);
625 if (reaper == father) {
626 reaper = child_reaper;
629 } while (reaper->exit_state);
632 * There are only two places where our children can be:
634 * - in our child list
635 * - in our ptraced child list
637 * Search them and reparent children.
639 list_for_each_safe(_p, _n, &father->children) {
641 p = list_entry(_p,struct task_struct,sibling);
645 /* if father isn't the real parent, then ptrace must be enabled */
646 BUG_ON(father != p->real_parent && !ptrace);
648 if (father == p->real_parent) {
649 /* reparent with a reaper, real father it's us */
650 choose_new_parent(p, reaper);
651 reparent_thread(p, father, 0);
653 /* reparent ptraced task to its real parent */
655 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
656 thread_group_empty(p))
657 do_notify_parent(p, p->exit_signal);
661 * if the ptraced child is a zombie with exit_signal == -1
662 * we must collect it before we exit, or it will remain
663 * zombie forever since we prevented it from self-reap itself
664 * while it was being traced by us, to be able to see it in wait4.
666 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
667 list_add(&p->ptrace_list, to_release);
669 list_for_each_safe(_p, _n, &father->ptrace_children) {
670 p = list_entry(_p,struct task_struct,ptrace_list);
671 choose_new_parent(p, reaper);
672 reparent_thread(p, father, 1);
677 * Send signals to all our closest relatives so that they know
678 * to properly mourn us..
680 static void exit_notify(struct task_struct *tsk)
683 struct task_struct *t;
684 struct list_head ptrace_dead, *_p, *_n;
686 if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
687 && !thread_group_empty(tsk)) {
689 * This occurs when there was a race between our exit
690 * syscall and a group signal choosing us as the one to
691 * wake up. It could be that we are the only thread
692 * alerted to check for pending signals, but another thread
693 * should be woken now to take the signal since we will not.
694 * Now we'll wake all the threads in the group just to make
695 * sure someone gets all the pending signals.
697 read_lock(&tasklist_lock);
698 spin_lock_irq(&tsk->sighand->siglock);
699 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
700 if (!signal_pending(t) && !(t->flags & PF_EXITING)) {
701 recalc_sigpending_tsk(t);
702 if (signal_pending(t))
703 signal_wake_up(t, 0);
705 spin_unlock_irq(&tsk->sighand->siglock);
706 read_unlock(&tasklist_lock);
709 write_lock_irq(&tasklist_lock);
712 * This does two things:
714 * A. Make init inherit all the child processes
715 * B. Check to see if any process groups have become orphaned
716 * as a result of our exiting, and if they have any stopped
717 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
720 INIT_LIST_HEAD(&ptrace_dead);
721 forget_original_parent(tsk, &ptrace_dead);
722 BUG_ON(!list_empty(&tsk->children));
723 BUG_ON(!list_empty(&tsk->ptrace_children));
726 * Check to see if any process groups have become orphaned
727 * as a result of our exiting, and if they have any stopped
728 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
730 * Case i: Our father is in a different pgrp than we are
731 * and we were the only connection outside, so our pgrp
732 * is about to become orphaned.
735 t = tsk->real_parent;
737 if ((process_group(t) != process_group(tsk)) &&
738 (t->signal->session == tsk->signal->session) &&
739 will_become_orphaned_pgrp(process_group(tsk), tsk) &&
740 has_stopped_jobs(process_group(tsk))) {
741 __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
742 __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
745 /* Let father know we died
747 * Thread signals are configurable, but you aren't going to use
748 * that to send signals to arbitary processes.
749 * That stops right now.
751 * If the parent exec id doesn't match the exec id we saved
752 * when we started then we know the parent has changed security
755 * If our self_exec id doesn't match our parent_exec_id then
756 * we have changed execution domain as these two values started
757 * the same after a fork.
761 if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
762 ( tsk->parent_exec_id != t->self_exec_id ||
763 tsk->self_exec_id != tsk->parent_exec_id)
764 && !capable(CAP_KILL))
765 tsk->exit_signal = SIGCHLD;
768 /* If something other than our normal parent is ptracing us, then
769 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
770 * only has special meaning to our real parent.
772 if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
773 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
774 do_notify_parent(tsk, signal);
775 } else if (tsk->ptrace) {
776 do_notify_parent(tsk, SIGCHLD);
780 if (tsk->exit_signal == -1 &&
781 (likely(tsk->ptrace == 0) ||
782 unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
784 tsk->exit_state = state;
786 write_unlock_irq(&tasklist_lock);
788 list_for_each_safe(_p, _n, &ptrace_dead) {
790 t = list_entry(_p,struct task_struct,ptrace_list);
794 /* If the process is dead, release it - nobody will wait for it */
795 if (state == EXIT_DEAD)
799 fastcall NORET_TYPE void do_exit(long code)
801 struct task_struct *tsk = current;
804 profile_task_exit(tsk);
806 WARN_ON(atomic_read(&tsk->fs_excl));
808 if (unlikely(in_interrupt()))
809 panic("Aiee, killing interrupt handler!");
810 if (unlikely(!tsk->pid))
811 panic("Attempted to kill the idle task!");
812 if (unlikely(tsk == child_reaper))
813 panic("Attempted to kill init!");
815 if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
816 current->ptrace_message = code;
817 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
821 * We're taking recursive faults here in do_exit. Safest is to just
822 * leave this task alone and wait for reboot.
824 if (unlikely(tsk->flags & PF_EXITING)) {
826 "Fixing recursive fault but reboot is needed!\n");
829 set_current_state(TASK_UNINTERRUPTIBLE);
833 tsk->flags |= PF_EXITING;
836 * Make sure we don't try to process any timer firings
837 * while we are already exiting.
839 tsk->it_virt_expires = cputime_zero;
840 tsk->it_prof_expires = cputime_zero;
841 tsk->it_sched_expires = 0;
843 if (unlikely(in_atomic()))
844 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
845 current->comm, current->pid,
848 acct_update_integrals(tsk);
850 update_hiwater_rss(tsk->mm);
851 update_hiwater_vm(tsk->mm);
853 group_dead = atomic_dec_and_test(&tsk->signal->live);
855 hrtimer_cancel(&tsk->signal->real_timer);
856 exit_itimers(tsk->signal);
859 if (unlikely(tsk->robust_list))
860 exit_robust_list(tsk);
862 if (unlikely(tsk->compat_robust_list))
863 compat_exit_robust_list(tsk);
875 if (group_dead && tsk->signal->leader)
876 disassociate_ctty(1);
878 module_put(task_thread_info(tsk)->exec_domain->module);
880 module_put(tsk->binfmt->module);
882 tsk->exit_code = code;
883 proc_exit_connector(tsk);
886 mpol_free(tsk->mempolicy);
887 tsk->mempolicy = NULL;
890 * If DEBUG_MUTEXES is on, make sure we are holding no locks:
892 mutex_debug_check_no_locks_held(tsk);
897 /* PF_DEAD causes final put_task_struct after we schedule. */
899 BUG_ON(tsk->flags & PF_DEAD);
900 tsk->flags |= PF_DEAD;
904 /* Avoid "noreturn function does return". */
908 EXPORT_SYMBOL_GPL(do_exit);
910 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
918 EXPORT_SYMBOL(complete_and_exit);
920 asmlinkage long sys_exit(int error_code)
922 do_exit((error_code&0xff)<<8);
925 task_t fastcall *next_thread(const task_t *p)
927 return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
930 EXPORT_SYMBOL(next_thread);
933 * Take down every thread in the group. This is called by fatal signals
934 * as well as by sys_exit_group (below).
937 do_group_exit(int exit_code)
939 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
941 if (current->signal->flags & SIGNAL_GROUP_EXIT)
942 exit_code = current->signal->group_exit_code;
943 else if (!thread_group_empty(current)) {
944 struct signal_struct *const sig = current->signal;
945 struct sighand_struct *const sighand = current->sighand;
946 read_lock(&tasklist_lock);
947 spin_lock_irq(&sighand->siglock);
948 if (sig->flags & SIGNAL_GROUP_EXIT)
949 /* Another thread got here before we took the lock. */
950 exit_code = sig->group_exit_code;
952 sig->group_exit_code = exit_code;
953 zap_other_threads(current);
955 spin_unlock_irq(&sighand->siglock);
956 read_unlock(&tasklist_lock);
964 * this kills every thread in the thread group. Note that any externally
965 * wait4()-ing process will get the correct exit code - even if this
966 * thread is not the thread group leader.
968 asmlinkage void sys_exit_group(int error_code)
970 do_group_exit((error_code & 0xff) << 8);
973 static int eligible_child(pid_t pid, int options, task_t *p)
979 if (process_group(p) != process_group(current))
981 } else if (pid != -1) {
982 if (process_group(p) != -pid)
987 * Do not consider detached threads that are
990 if (p->exit_signal == -1 && !p->ptrace)
993 /* Wait for all children (clone and not) if __WALL is set;
994 * otherwise, wait for clone children *only* if __WCLONE is
995 * set; otherwise, wait for non-clone children *only*. (Note:
996 * A "clone" child here is one that reports to its parent
997 * using a signal other than SIGCHLD.) */
998 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
999 && !(options & __WALL))
1002 * Do not consider thread group leaders that are
1003 * in a non-empty thread group:
1005 if (current->tgid != p->tgid && delay_group_leader(p))
1008 if (security_task_wait(p))
1014 static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
1015 int why, int status,
1016 struct siginfo __user *infop,
1017 struct rusage __user *rusagep)
1019 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1022 retval = put_user(SIGCHLD, &infop->si_signo);
1024 retval = put_user(0, &infop->si_errno);
1026 retval = put_user((short)why, &infop->si_code);
1028 retval = put_user(pid, &infop->si_pid);
1030 retval = put_user(uid, &infop->si_uid);
1032 retval = put_user(status, &infop->si_status);
1039 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1040 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1041 * the lock and this task is uninteresting. If we return nonzero, we have
1042 * released the lock and the system call should return.
1044 static int wait_task_zombie(task_t *p, int noreap,
1045 struct siginfo __user *infop,
1046 int __user *stat_addr, struct rusage __user *ru)
1048 unsigned long state;
1052 if (unlikely(noreap)) {
1055 int exit_code = p->exit_code;
1058 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1060 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1063 read_unlock(&tasklist_lock);
1064 if ((exit_code & 0x7f) == 0) {
1066 status = exit_code >> 8;
1068 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1069 status = exit_code & 0x7f;
1071 return wait_noreap_copyout(p, pid, uid, why,
1076 * Try to move the task's state to DEAD
1077 * only one thread is allowed to do this:
1079 state = xchg(&p->exit_state, EXIT_DEAD);
1080 if (state != EXIT_ZOMBIE) {
1081 BUG_ON(state != EXIT_DEAD);
1084 if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
1086 * This can only happen in a race with a ptraced thread
1087 * dying on another processor.
1092 if (likely(p->real_parent == p->parent) && likely(p->signal)) {
1093 struct signal_struct *psig;
1094 struct signal_struct *sig;
1097 * The resource counters for the group leader are in its
1098 * own task_struct. Those for dead threads in the group
1099 * are in its signal_struct, as are those for the child
1100 * processes it has previously reaped. All these
1101 * accumulate in the parent's signal_struct c* fields.
1103 * We don't bother to take a lock here to protect these
1104 * p->signal fields, because they are only touched by
1105 * __exit_signal, which runs with tasklist_lock
1106 * write-locked anyway, and so is excluded here. We do
1107 * need to protect the access to p->parent->signal fields,
1108 * as other threads in the parent group can be right
1109 * here reaping other children at the same time.
1111 spin_lock_irq(&p->parent->sighand->siglock);
1112 psig = p->parent->signal;
1115 cputime_add(psig->cutime,
1116 cputime_add(p->utime,
1117 cputime_add(sig->utime,
1120 cputime_add(psig->cstime,
1121 cputime_add(p->stime,
1122 cputime_add(sig->stime,
1125 p->min_flt + sig->min_flt + sig->cmin_flt;
1127 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1129 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1131 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1132 spin_unlock_irq(&p->parent->sighand->siglock);
1136 * Now we are sure this task is interesting, and no other
1137 * thread can reap it because we set its state to EXIT_DEAD.
1139 read_unlock(&tasklist_lock);
1141 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1142 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1143 ? p->signal->group_exit_code : p->exit_code;
1144 if (!retval && stat_addr)
1145 retval = put_user(status, stat_addr);
1146 if (!retval && infop)
1147 retval = put_user(SIGCHLD, &infop->si_signo);
1148 if (!retval && infop)
1149 retval = put_user(0, &infop->si_errno);
1150 if (!retval && infop) {
1153 if ((status & 0x7f) == 0) {
1157 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1160 retval = put_user((short)why, &infop->si_code);
1162 retval = put_user(status, &infop->si_status);
1164 if (!retval && infop)
1165 retval = put_user(p->pid, &infop->si_pid);
1166 if (!retval && infop)
1167 retval = put_user(p->uid, &infop->si_uid);
1169 // TODO: is this safe?
1170 p->exit_state = EXIT_ZOMBIE;
1174 if (p->real_parent != p->parent) {
1175 write_lock_irq(&tasklist_lock);
1176 /* Double-check with lock held. */
1177 if (p->real_parent != p->parent) {
1179 // TODO: is this safe?
1180 p->exit_state = EXIT_ZOMBIE;
1182 * If this is not a detached task, notify the parent.
1183 * If it's still not detached after that, don't release
1186 if (p->exit_signal != -1) {
1187 do_notify_parent(p, p->exit_signal);
1188 if (p->exit_signal != -1)
1192 write_unlock_irq(&tasklist_lock);
1201 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1202 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1203 * the lock and this task is uninteresting. If we return nonzero, we have
1204 * released the lock and the system call should return.
1206 static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
1207 struct siginfo __user *infop,
1208 int __user *stat_addr, struct rusage __user *ru)
1210 int retval, exit_code;
1214 if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1215 p->signal && p->signal->group_stop_count > 0)
1217 * A group stop is in progress and this is the group leader.
1218 * We won't report until all threads have stopped.
1223 * Now we are pretty sure this task is interesting.
1224 * Make sure it doesn't get reaped out from under us while we
1225 * give up the lock and then examine it below. We don't want to
1226 * keep holding onto the tasklist_lock while we call getrusage and
1227 * possibly take page faults for user memory.
1230 read_unlock(&tasklist_lock);
1232 if (unlikely(noreap)) {
1235 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1237 exit_code = p->exit_code;
1238 if (unlikely(!exit_code) ||
1239 unlikely(p->state & TASK_TRACED))
1241 return wait_noreap_copyout(p, pid, uid,
1242 why, (exit_code << 8) | 0x7f,
1246 write_lock_irq(&tasklist_lock);
1249 * This uses xchg to be atomic with the thread resuming and setting
1250 * it. It must also be done with the write lock held to prevent a
1251 * race with the EXIT_ZOMBIE case.
1253 exit_code = xchg(&p->exit_code, 0);
1254 if (unlikely(p->exit_state)) {
1256 * The task resumed and then died. Let the next iteration
1257 * catch it in EXIT_ZOMBIE. Note that exit_code might
1258 * already be zero here if it resumed and did _exit(0).
1259 * The task itself is dead and won't touch exit_code again;
1260 * other processors in this function are locked out.
1262 p->exit_code = exit_code;
1265 if (unlikely(exit_code == 0)) {
1267 * Another thread in this function got to it first, or it
1268 * resumed, or it resumed and then died.
1270 write_unlock_irq(&tasklist_lock);
1274 * We are returning to the wait loop without having successfully
1275 * removed the process and having released the lock. We cannot
1276 * continue, since the "p" task pointer is potentially stale.
1278 * Return -EAGAIN, and do_wait() will restart the loop from the
1279 * beginning. Do _not_ re-acquire the lock.
1284 /* move to end of parent's list to avoid starvation */
1288 write_unlock_irq(&tasklist_lock);
1290 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1291 if (!retval && stat_addr)
1292 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1293 if (!retval && infop)
1294 retval = put_user(SIGCHLD, &infop->si_signo);
1295 if (!retval && infop)
1296 retval = put_user(0, &infop->si_errno);
1297 if (!retval && infop)
1298 retval = put_user((short)((p->ptrace & PT_PTRACED)
1299 ? CLD_TRAPPED : CLD_STOPPED),
1301 if (!retval && infop)
1302 retval = put_user(exit_code, &infop->si_status);
1303 if (!retval && infop)
1304 retval = put_user(p->pid, &infop->si_pid);
1305 if (!retval && infop)
1306 retval = put_user(p->uid, &infop->si_uid);
1316 * Handle do_wait work for one task in a live, non-stopped state.
1317 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1318 * the lock and this task is uninteresting. If we return nonzero, we have
1319 * released the lock and the system call should return.
1321 static int wait_task_continued(task_t *p, int noreap,
1322 struct siginfo __user *infop,
1323 int __user *stat_addr, struct rusage __user *ru)
1329 if (unlikely(!p->signal))
1332 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1335 spin_lock_irq(&p->sighand->siglock);
1336 /* Re-check with the lock held. */
1337 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1338 spin_unlock_irq(&p->sighand->siglock);
1342 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1343 spin_unlock_irq(&p->sighand->siglock);
1348 read_unlock(&tasklist_lock);
1351 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1353 if (!retval && stat_addr)
1354 retval = put_user(0xffff, stat_addr);
1358 retval = wait_noreap_copyout(p, pid, uid,
1359 CLD_CONTINUED, SIGCONT,
1361 BUG_ON(retval == 0);
1368 static inline int my_ptrace_child(struct task_struct *p)
1370 if (!(p->ptrace & PT_PTRACED))
1372 if (!(p->ptrace & PT_ATTACHED))
1375 * This child was PTRACE_ATTACH'd. We should be seeing it only if
1376 * we are the attacher. If we are the real parent, this is a race
1377 * inside ptrace_attach. It is waiting for the tasklist_lock,
1378 * which we have to switch the parent links, but has already set
1379 * the flags in p->ptrace.
1381 return (p->parent != p->real_parent);
1384 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1385 int __user *stat_addr, struct rusage __user *ru)
1387 DECLARE_WAITQUEUE(wait, current);
1388 struct task_struct *tsk;
1391 add_wait_queue(¤t->signal->wait_chldexit,&wait);
1394 * We will set this flag if we see any child that might later
1395 * match our criteria, even if we are not able to reap it yet.
1398 current->state = TASK_INTERRUPTIBLE;
1399 read_lock(&tasklist_lock);
1402 struct task_struct *p;
1403 struct list_head *_p;
1406 list_for_each(_p,&tsk->children) {
1407 p = list_entry(_p,struct task_struct,sibling);
1409 ret = eligible_child(pid, options, p);
1416 * When we hit the race with PTRACE_ATTACH,
1417 * we will not report this child. But the
1418 * race means it has not yet been moved to
1419 * our ptrace_children list, so we need to
1420 * set the flag here to avoid a spurious ECHILD
1421 * when the race happens with the only child.
1424 if (!my_ptrace_child(p))
1429 * It's stopped now, so it might later
1430 * continue, exit, or stop again.
1433 if (!(options & WUNTRACED) &&
1434 !my_ptrace_child(p))
1436 retval = wait_task_stopped(p, ret == 2,
1437 (options & WNOWAIT),
1440 if (retval == -EAGAIN)
1442 if (retval != 0) /* He released the lock. */
1447 if (p->exit_state == EXIT_DEAD)
1449 // case EXIT_ZOMBIE:
1450 if (p->exit_state == EXIT_ZOMBIE) {
1452 * Eligible but we cannot release
1456 goto check_continued;
1457 if (!likely(options & WEXITED))
1459 retval = wait_task_zombie(
1460 p, (options & WNOWAIT),
1461 infop, stat_addr, ru);
1462 /* He released the lock. */
1469 * It's running now, so it might later
1470 * exit, stop, or stop and then continue.
1473 if (!unlikely(options & WCONTINUED))
1475 retval = wait_task_continued(
1476 p, (options & WNOWAIT),
1477 infop, stat_addr, ru);
1478 if (retval != 0) /* He released the lock. */
1484 list_for_each(_p, &tsk->ptrace_children) {
1485 p = list_entry(_p, struct task_struct,
1487 if (!eligible_child(pid, options, p))
1493 if (options & __WNOTHREAD)
1495 tsk = next_thread(tsk);
1496 if (tsk->signal != current->signal)
1498 } while (tsk != current);
1500 read_unlock(&tasklist_lock);
1503 if (options & WNOHANG)
1505 retval = -ERESTARTSYS;
1506 if (signal_pending(current))
1513 current->state = TASK_RUNNING;
1514 remove_wait_queue(¤t->signal->wait_chldexit,&wait);
1520 * For a WNOHANG return, clear out all the fields
1521 * we would set so the user can easily tell the
1525 retval = put_user(0, &infop->si_signo);
1527 retval = put_user(0, &infop->si_errno);
1529 retval = put_user(0, &infop->si_code);
1531 retval = put_user(0, &infop->si_pid);
1533 retval = put_user(0, &infop->si_uid);
1535 retval = put_user(0, &infop->si_status);
1541 asmlinkage long sys_waitid(int which, pid_t pid,
1542 struct siginfo __user *infop, int options,
1543 struct rusage __user *ru)
1547 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1549 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1569 ret = do_wait(pid, options, infop, NULL, ru);
1571 /* avoid REGPARM breakage on x86: */
1572 prevent_tail_call(ret);
1576 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1577 int options, struct rusage __user *ru)
1581 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1582 __WNOTHREAD|__WCLONE|__WALL))
1584 ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1586 /* avoid REGPARM breakage on x86: */
1587 prevent_tail_call(ret);
1591 #ifdef __ARCH_WANT_SYS_WAITPID
1594 * sys_waitpid() remains for compatibility. waitpid() should be
1595 * implemented by calling sys_wait4() from libc.a.
1597 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1599 return sys_wait4(pid, stat_addr, options, NULL);