This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / kernel / exit.c
index 80530fb..208cbfa 100644 (file)
@@ -4,6 +4,7 @@
  *  Copyright (C) 1991, 1992  Linus Torvalds
  */
 
+#include <linux/config.h>
 #include <linux/mm.h>
 #include <linux/slab.h>
 #include <linux/interrupt.h>
 #include <linux/file.h>
 #include <linux/binfmts.h>
 #include <linux/ptrace.h>
-#include <linux/tracehook.h>
 #include <linux/profile.h>
 #include <linux/mount.h>
 #include <linux/proc_fs.h>
 #include <linux/mempolicy.h>
-#include <linux/taskstats_kern.h>
-#include <linux/delayacct.h>
 #include <linux/cpuset.h>
 #include <linux/syscalls.h>
 #include <linux/signal.h>
-#include <linux/posix-timers.h>
 #include <linux/cn_proc.h>
 #include <linux/mutex.h>
-#include <linux/futex.h>
-#include <linux/compat.h>
-#include <linux/pipe_fs_i.h>
-#include <linux/audit.h> /* for audit_free() */
-#include <linux/resource.h>
 #include <linux/vs_limit.h>
 #include <linux/vs_context.h>
 #include <linux/vs_network.h>
+#include <linux/vs_cvirt.h>
 
 #include <asm/uaccess.h>
 #include <asm/unistd.h>
 extern void sem_exit (void);
 extern struct task_struct *child_reaper;
 
+int getrusage(struct task_struct *, int, struct rusage __user *);
+
 static void exit_mm(struct task_struct * tsk);
 
 static void __unhash_process(struct task_struct *p)
 {
        nr_threads--;
        detach_pid(p, PIDTYPE_PID);
+       detach_pid(p, PIDTYPE_TGID);
        if (thread_group_leader(p)) {
                detach_pid(p, PIDTYPE_PGID);
                detach_pid(p, PIDTYPE_SID);
-
-               list_del_rcu(&p->tasks);
-               __get_cpu_var(process_counts)--;
-       }
-       list_del_rcu(&p->thread_group);
-       remove_parent(p);
-}
-
-/*
- * This function expects the tasklist_lock write-locked.
- */
-static void __exit_signal(struct task_struct *tsk)
-{
-       struct signal_struct *sig = tsk->signal;
-       struct sighand_struct *sighand;
-
-       BUG_ON(!sig);
-       BUG_ON(!atomic_read(&sig->count));
-
-       rcu_read_lock();
-       sighand = rcu_dereference(tsk->sighand);
-       spin_lock(&sighand->siglock);
-
-       posix_cpu_timers_exit(tsk);
-       if (atomic_dec_and_test(&sig->count))
-               posix_cpu_timers_exit_group(tsk);
-       else {
-               /*
-                * If there is any task waiting for the group exit
-                * then notify it:
-                */
-               if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count) {
-                       wake_up_process(sig->group_exit_task);
-                       sig->group_exit_task = NULL;
-               }
-               if (tsk == sig->curr_target)
-                       sig->curr_target = next_thread(tsk);
-               /*
-                * Accumulate here the counters for all threads but the
-                * group leader as they die, so they can be added into
-                * the process-wide totals when those are taken.
-                * The group leader stays around as a zombie as long
-                * as there are other threads.  When it gets reaped,
-                * the exit.c code will add its counts into these totals.
-                * We won't ever get here for the group leader, since it
-                * will have been the last reference on the signal_struct.
-                */
-               sig->utime = cputime_add(sig->utime, tsk->utime);
-               sig->stime = cputime_add(sig->stime, tsk->stime);
-               sig->min_flt += tsk->min_flt;
-               sig->maj_flt += tsk->maj_flt;
-               sig->nvcsw += tsk->nvcsw;
-               sig->nivcsw += tsk->nivcsw;
-               sig->sched_time += tsk->sched_time;
-               sig = NULL; /* Marker for below. */
-       }
-
-       __unhash_process(tsk);
-
-       tsk->signal = NULL;
-       tsk->sighand = NULL;
-       spin_unlock(&sighand->siglock);
-       rcu_read_unlock();
-
-       __cleanup_sighand(sighand);
-       clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
-       flush_sigqueue(&tsk->pending);
-       if (sig) {
-               flush_sigqueue(&sig->shared_pending);
-               __cleanup_signal(sig);
+               if (p->pid)
+                       __get_cpu_var(process_counts)--;
        }
-}
 
-static void delayed_put_task_struct(struct rcu_head *rhp)
-{
-       put_task_struct(container_of(rhp, struct task_struct, rcu));
+       REMOVE_LINKS(p);
 }
 
 void release_task(struct task_struct * p)
 {
-       struct task_struct *leader;
        int zap_leader;
-repeat:
-       tracehook_release_task(p);
+       task_t *leader;
+       struct dentry *proc_dentry;
+
+repeat: 
        atomic_dec(&p->user->processes);
+       spin_lock(&p->proc_lock);
+       proc_dentry = proc_pid_unhash(p);
        write_lock_irq(&tasklist_lock);
-       BUG_ON(tracehook_check_released(p));
+       if (unlikely(p->ptrace))
+               __ptrace_unlink(p);
+       BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
        __exit_signal(p);
+       /*
+        * Note that the fastpath in sys_times depends on __exit_signal having
+        * updated the counters before a task is removed from the tasklist of
+        * the process by __unhash_process.
+        */
+       __unhash_process(p);
 
        /*
         * If we are the last non-leader member of the thread
@@ -172,15 +108,31 @@ repeat:
 
        sched_exit(p);
        write_unlock_irq(&tasklist_lock);
-       proc_flush_task(p);
+       spin_unlock(&p->proc_lock);
+       proc_pid_flush(proc_dentry);
        release_thread(p);
-       call_rcu(&p->rcu, delayed_put_task_struct);
+       put_task_struct(p);
 
        p = leader;
        if (unlikely(zap_leader))
                goto repeat;
 }
 
+/* we are using it only for SMP init */
+
+void unhash_process(struct task_struct *p)
+{
+       struct dentry *proc_dentry;
+
+       spin_lock(&p->proc_lock);
+       proc_dentry = proc_pid_unhash(p);
+       write_lock_irq(&tasklist_lock);
+       __unhash_process(p);
+       write_unlock_irq(&tasklist_lock);
+       spin_unlock(&p->proc_lock);
+       proc_pid_flush(proc_dentry);
+}
+
 /*
  * This checks not only the pgrp, but falls back on the pid if no
  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
@@ -215,7 +167,7 @@ out:
  *
  * "I ask you, have you ever known what it is to be an orphan?"
  */
-static int will_become_orphaned_pgrp(int pgrp, struct task_struct *ignored_task)
+static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
 {
        struct task_struct *p;
        int ret = 1;
@@ -223,10 +175,10 @@ static int will_become_orphaned_pgrp(int pgrp, struct task_struct *ignored_task)
        do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
                if (p == ignored_task
                                || p->exit_state
-                               || p->parent->pid == 1)
+                               || p->real_parent->pid == 1)
                        continue;
-               if (process_group(p->parent) != pgrp
-                           && p->parent->signal->session == p->signal->session) {
+               if (process_group(p->real_parent) != pgrp
+                           && p->real_parent->signal->session == p->signal->session) {
                        ret = 0;
                        break;
                }
@@ -254,6 +206,16 @@ static int has_stopped_jobs(int pgrp)
                if (p->state != TASK_STOPPED)
                        continue;
 
+               /* If p is stopped by a debugger on a signal that won't
+                  stop it, then don't count p as stopped.  This isn't
+                  perfect but it's a good approximation.  */
+               if (unlikely (p->ptrace)
+                   && p->exit_code != SIGSTOP
+                   && p->exit_code != SIGTSTP
+                   && p->exit_code != SIGTTOU
+                   && p->exit_code != SIGTTIN)
+                       continue;
+
                retval = 1;
                break;
        } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
@@ -276,10 +238,12 @@ static void reparent_to_init(void)
 {
        write_lock_irq(&tasklist_lock);
 
+       ptrace_unlink(current);
        /* Reparent to init */
-       remove_parent(current);
+       REMOVE_LINKS(current);
        current->parent = child_reaper;
-       add_parent(current);
+       current->real_parent = child_reaper;
+       SET_LINKS(current);
 
        /* Set the exit signal to SIGCHLD so we signal init on exit */
        current->exit_signal = SIGCHLD;
@@ -385,7 +349,9 @@ void daemonize(const char *name, ...)
        exit_mm(current);
 
        set_special_pids(1, 1);
-       proc_clear_tty(current);
+       down(&tty_sem);
+       current->signal->tty = NULL;
+       up(&tty_sem);
 
        /* Block and flush all signals */
        sigfillset(&blocked);
@@ -438,7 +404,6 @@ static void close_files(struct files_struct * files)
                        }
                        i++;
                        set >>= 1;
-                       cond_resched();
                }
        }
 }
@@ -566,7 +531,7 @@ static void exit_mm(struct task_struct * tsk)
                down_read(&mm->mmap_sem);
        }
        atomic_inc(&mm->mm_count);
-       BUG_ON(mm != tsk->active_mm);
+       if (mm != tsk->active_mm) BUG();
        /* more a memory barrier than a real lock */
        task_lock(tsk);
        tsk->mm = NULL;
@@ -576,8 +541,7 @@ static void exit_mm(struct task_struct * tsk)
        mmput(mm);
 }
 
-static inline void
-choose_new_parent(struct task_struct *p, struct task_struct *reaper)
+static inline void choose_new_parent(task_t *p, task_t *reaper)
 {
        /* check for reaper context */
        vxwprintk((p->xid != reaper->xid) && (reaper != child_reaper),
@@ -588,12 +552,11 @@ choose_new_parent(struct task_struct *p, struct task_struct *reaper)
         * Make sure we're not reparenting to ourselves and that
         * the parent is not a zombie.
         */
-       BUG_ON(p == reaper || reaper->exit_state);
-       p->parent = reaper;
+       BUG_ON(p == reaper || reaper->exit_state >= EXIT_ZOMBIE);
+       p->real_parent = reaper;
 }
 
-static void
-reparent_thread(struct task_struct *p, struct task_struct *father)
+static void reparent_thread(task_t *p, task_t *father, int traced)
 {
        /* We don't want people slaying init.  */
        if (p->exit_signal != -1)
@@ -604,14 +567,35 @@ reparent_thread(struct task_struct *p, struct task_struct *father)
                group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 
        /* Move the child from its dying parent to the new one.  */
-       list_move_tail(&p->sibling, &p->parent->children);
+       if (unlikely(traced)) {
+               /* Preserve ptrace links if someone else is tracing this child.  */
+               list_del_init(&p->ptrace_list);
+               if (p->parent != p->real_parent)
+                       list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
+       } else {
+               /* If this child is being traced, then we're the one tracing it
+                * anyway, so let go of it.
+                */
+               p->ptrace = 0;
+               list_del_init(&p->sibling);
+               p->parent = p->real_parent;
+               list_add_tail(&p->sibling, &p->parent->children);
 
-       /* If we'd notified the old parent about this child's death,
-        * also notify the new parent.
-        */
-       if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
-           thread_group_empty(p))
-               do_notify_parent(p, p->exit_signal);
+               /* If we'd notified the old parent about this child's death,
+                * also notify the new parent.
+                */
+               if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
+                   thread_group_empty(p))
+                       do_notify_parent(p, p->exit_signal);
+               else if (p->state == TASK_TRACED) {
+                       /*
+                        * If it was at a trace stop, turn it into
+                        * a normal stop since it's no longer being
+                        * traced.
+                        */
+                       ptrace_untrace(p);
+               }
+       }
 
        /*
         * process group orphan check
@@ -636,8 +620,8 @@ reparent_thread(struct task_struct *p, struct task_struct *father)
  * group, and if no such member exists, give it to
  * the global child reaper process (ie "init")
  */
-static void
-forget_original_parent(struct task_struct *father)
+static void forget_original_parent(struct task_struct * father,
+                                         struct list_head *to_release)
 {
        struct task_struct *p, *reaper = father;
        struct list_head *_p, *_n;
@@ -650,10 +634,49 @@ forget_original_parent(struct task_struct *father)
                }
        } while (reaper->exit_state);
 
+       /*
+        * There are only two places where our children can be:
+        *
+        * - in our child list
+        * - in our ptraced child list
+        *
+        * Search them and reparent children.
+        */
        list_for_each_safe(_p, _n, &father->children) {
-               p = list_entry(_p, struct task_struct, sibling);
-               choose_new_parent(p, vx_child_reaper(p));
-               reparent_thread(p, father);
+               int ptrace;
+               p = list_entry(_p,struct task_struct,sibling);
+
+               ptrace = p->ptrace;
+
+               /* if father isn't the real parent, then ptrace must be enabled */
+               BUG_ON(father != p->real_parent && !ptrace);
+
+               if (father == p->real_parent) {
+                       /* reparent with a reaper, real father it's us */
+                       choose_new_parent(p, vx_child_reaper(p));
+                       reparent_thread(p, father, 0);
+               } else {
+                       /* reparent ptraced task to its real parent */
+                       __ptrace_unlink (p);
+                       if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
+                           thread_group_empty(p))
+                               do_notify_parent(p, p->exit_signal);
+               }
+
+               /*
+                * if the ptraced child is a zombie with exit_signal == -1
+                * we must collect it before we exit, or it will remain
+                * zombie forever since we prevented it from self-reap itself
+                * while it was being traced by us, to be able to see it in wait4.
+                */
+               if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
+                       list_add(&p->ptrace_list, to_release);
+       }
+       list_for_each_safe(_p, _n, &father->ptrace_children) {
+               p = list_entry(_p,struct task_struct,ptrace_list);
+
+               choose_new_parent(p, reaper);
+               reparent_thread(p, father, 1);
        }
 }
 
@@ -665,8 +688,7 @@ static void exit_notify(struct task_struct *tsk)
 {
        int state;
        struct task_struct *t;
-       int noreap;
-       void *cookie;
+       struct list_head ptrace_dead, *_p, *_n;
 
        if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
            && !thread_group_empty(tsk)) {
@@ -702,8 +724,10 @@ static void exit_notify(struct task_struct *tsk)
         *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
         */
 
-       forget_original_parent(tsk);
+       INIT_LIST_HEAD(&ptrace_dead);
+       forget_original_parent(tsk, &ptrace_dead);
        BUG_ON(!list_empty(&tsk->children));
+       BUG_ON(!list_empty(&tsk->ptrace_children));
 
        /*
         * Check to see if any process groups have become orphaned
@@ -715,7 +739,7 @@ static void exit_notify(struct task_struct *tsk)
         * is about to become orphaned.
         */
         
-       t = tsk->parent;
+       t = tsk->real_parent;
        
        if ((process_group(t) != process_group(tsk)) &&
            (t->signal->session == tsk->signal->session) &&
@@ -747,18 +771,32 @@ static void exit_notify(struct task_struct *tsk)
            && !capable(CAP_KILL))
                tsk->exit_signal = SIGCHLD;
 
-       if (!tracehook_notify_death(tsk, &noreap, &cookie)
-           && tsk->exit_signal != -1 && thread_group_empty(tsk))
-               do_notify_parent(tsk, tsk->exit_signal);
+
+       /* If something other than our normal parent is ptracing us, then
+        * send it a SIGCHLD instead of honoring exit_signal.  exit_signal
+        * only has special meaning to our real parent.
+        */
+       if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
+               int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
+               do_notify_parent(tsk, signal);
+       } else if (tsk->ptrace) {
+               do_notify_parent(tsk, SIGCHLD);
+       }
 
        state = EXIT_ZOMBIE;
-       if (tsk->exit_signal == -1 && !noreap)
+       if (tsk->exit_signal == -1 &&
+           (likely(tsk->ptrace == 0) ||
+            unlikely(tsk->parent->signal->flags & SIGNAL_GROUP_EXIT)))
                state = EXIT_DEAD;
        tsk->exit_state = state;
 
        write_unlock_irq(&tasklist_lock);
 
-       tracehook_report_death(tsk, state, cookie);
+       list_for_each_safe(_p, _n, &ptrace_dead) {
+               list_del_init(_p);
+               t = list_entry(_p,struct task_struct,ptrace_list);
+               release_task(t);
+       }
 
        /* If the process is dead, release it - nobody will wait for it */
        if (state == EXIT_DEAD)
@@ -768,9 +806,7 @@ static void exit_notify(struct task_struct *tsk)
 fastcall NORET_TYPE void do_exit(long code)
 {
        struct task_struct *tsk = current;
-       struct taskstats *tidstats;
        int group_dead;
-       unsigned int mycpu;
 
        profile_task_exit(tsk);
 
@@ -780,10 +816,15 @@ fastcall NORET_TYPE void do_exit(long code)
                panic("Aiee, killing interrupt handler!");
        if (unlikely(!tsk->pid))
                panic("Attempted to kill the idle task!");
-       if (unlikely(tsk == child_reaper))
+       if (unlikely(tsk->pid == 1))
                panic("Attempted to kill init!");
+       if (tsk->io_context)
+               exit_io_context();
 
-       tracehook_report_exit(&code);
+       if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
+               current->ptrace_message = code;
+               ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
+       }
 
        /*
         * We're taking recursive faults here in do_exit. Safest is to just
@@ -792,23 +833,17 @@ fastcall NORET_TYPE void do_exit(long code)
        if (unlikely(tsk->flags & PF_EXITING)) {
                printk(KERN_ALERT
                        "Fixing recursive fault but reboot is needed!\n");
-               if (tsk->io_context)
-                       exit_io_context();
                set_current_state(TASK_UNINTERRUPTIBLE);
                schedule();
        }
 
        tsk->flags |= PF_EXITING;
 
-       ptrace_exit(tsk);
-
        if (unlikely(in_atomic()))
                printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
                                current->comm, current->pid,
                                preempt_count());
 
-       taskstats_exit_alloc(&tidstats, &mycpu);
-
        acct_update_integrals(tsk);
        if (tsk->mm) {
                update_hiwater_rss(tsk->mm);
@@ -818,32 +853,10 @@ fastcall NORET_TYPE void do_exit(long code)
        if (group_dead) {
                hrtimer_cancel(&tsk->signal->real_timer);
                exit_itimers(tsk->signal);
+               acct_process(code);
        }
-
-       if (current->tux_info) {
-#ifdef CONFIG_TUX_DEBUG
-               printk("Possibly unexpected TUX-thread exit(%ld) at %p?\n",
-                       code, __builtin_return_address(0));
-#endif
-               current->tux_exit();
-       }
-
-       acct_collect(code, group_dead);
-       if (unlikely(tsk->robust_list))
-               exit_robust_list(tsk);
-#if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
-       if (unlikely(tsk->compat_robust_list))
-               compat_exit_robust_list(tsk);
-#endif
-       if (unlikely(tsk->audit_context))
-               audit_free(tsk);
-       taskstats_exit_send(tsk, tidstats, group_dead, mycpu);
-       taskstats_exit_free(tidstats);
-
        exit_mm(tsk);
 
-       if (group_dead)
-               acct_process();
        exit_sem(tsk);
        __exit_files(tsk);
        __exit_fs(tsk);
@@ -869,23 +882,9 @@ fastcall NORET_TYPE void do_exit(long code)
        tsk->mempolicy = NULL;
 #endif
        /*
-        * This must happen late, after the PID is not
-        * hashed anymore:
-        */
-       if (unlikely(!list_empty(&tsk->pi_state_list)))
-               exit_pi_state_list(tsk);
-       if (unlikely(current->pi_state_cache))
-               kfree(current->pi_state_cache);
-       /*
-        * Make sure we are holding no locks:
+        * If DEBUG_MUTEXES is on, make sure we are holding no locks:
         */
-       debug_check_no_locks_held(tsk);
-
-       if (tsk->io_context)
-               exit_io_context();
-
-       if (tsk->splice_pipe)
-               __free_pipe_info(tsk->splice_pipe);
+       mutex_debug_check_no_locks_held(tsk);
 
        /* needs to stay after exit_notify() */
        exit_vx_info(tsk, code);
@@ -919,6 +918,13 @@ asmlinkage long sys_exit(int error_code)
        do_exit((error_code&0xff)<<8);
 }
 
+task_t fastcall *next_thread(const task_t *p)
+{
+       return pid_task(p->pids[PIDTYPE_TGID].pid_list.next, PIDTYPE_TGID);
+}
+
+EXPORT_SYMBOL(next_thread);
+
 /*
  * Take down every thread in the group.  This is called by fatal signals
  * as well as by sys_exit_group (below).
@@ -933,6 +939,7 @@ do_group_exit(int exit_code)
        else if (!thread_group_empty(current)) {
                struct signal_struct *const sig = current->signal;
                struct sighand_struct *const sighand = current->sighand;
+               read_lock(&tasklist_lock);
                spin_lock_irq(&sighand->siglock);
                if (sig->flags & SIGNAL_GROUP_EXIT)
                        /* Another thread got here before we took the lock.  */
@@ -942,6 +949,7 @@ do_group_exit(int exit_code)
                        zap_other_threads(current);
                }
                spin_unlock_irq(&sighand->siglock);
+               read_unlock(&tasklist_lock);
        }
 
        do_exit(exit_code);
@@ -958,7 +966,7 @@ asmlinkage void sys_exit_group(int error_code)
        do_group_exit((error_code & 0xff) << 8);
 }
 
-static int eligible_child(pid_t pid, int options, struct task_struct *p)
+static int eligible_child(pid_t pid, int options, task_t *p)
 {
        if (pid > 0) {
                if (p->pid != pid)
@@ -972,9 +980,10 @@ static int eligible_child(pid_t pid, int options, struct task_struct *p)
        }
 
        /*
-        * Do not consider detached threads.
+        * Do not consider detached threads that are
+        * not ptraced:
         */
-       if (p->exit_signal == -1)
+       if (p->exit_signal == -1 && !p->ptrace)
                return 0;
 
        /* Wait for all children (clone and not) if __WALL is set;
@@ -989,7 +998,7 @@ static int eligible_child(pid_t pid, int options, struct task_struct *p)
         * Do not consider thread group leaders that are
         * in a non-empty thread group:
         */
-       if (delay_group_leader(p))
+       if (current->tgid != p->tgid && delay_group_leader(p))
                return 2;
 
        if (security_task_wait(p))
@@ -998,13 +1007,12 @@ static int eligible_child(pid_t pid, int options, struct task_struct *p)
        return 1;
 }
 
-static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
+static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
                               int why, int status,
                               struct siginfo __user *infop,
                               struct rusage __user *rusagep)
 {
        int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
-
        put_task_struct(p);
        if (!retval)
                retval = put_user(SIGCHLD, &infop->si_signo);
@@ -1029,7 +1037,7 @@ static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
  * the lock and this task is uninteresting.  If we return nonzero, we have
  * released the lock and the system call should return.
  */
-static int wait_task_zombie(struct task_struct *p, int noreap,
+static int wait_task_zombie(task_t *p, int noreap,
                            struct siginfo __user *infop,
                            int __user *stat_addr, struct rusage __user *ru)
 {
@@ -1045,7 +1053,7 @@ static int wait_task_zombie(struct task_struct *p, int noreap,
 
                if (unlikely(p->exit_state != EXIT_ZOMBIE))
                        return 0;
-               if (unlikely(p->exit_signal == -1))
+               if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
                        return 0;
                get_task_struct(p);
                read_unlock(&tasklist_lock);
@@ -1069,9 +1077,15 @@ static int wait_task_zombie(struct task_struct *p, int noreap,
                BUG_ON(state != EXIT_DEAD);
                return 0;
        }
-       BUG_ON(p->exit_signal == -1);
+       if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
+               /*
+                * This can only happen in a race with a ptraced thread
+                * dying on another processor.
+                */
+               return 0;
+       }
 
-       if (likely(p->signal)) {
+       if (likely(p->real_parent == p->parent) && likely(p->signal)) {
                struct signal_struct *psig;
                struct signal_struct *sig;
 
@@ -1153,8 +1167,28 @@ static int wait_task_zombie(struct task_struct *p, int noreap,
                return retval;
        }
        retval = p->pid;
-       release_task(p);
-
+       if (p->real_parent != p->parent) {
+               write_lock_irq(&tasklist_lock);
+               /* Double-check with lock held.  */
+               if (p->real_parent != p->parent) {
+                       __ptrace_unlink(p);
+                       // TODO: is this safe?
+                       p->exit_state = EXIT_ZOMBIE;
+                       /*
+                        * If this is not a detached task, notify the parent.
+                        * If it's still not detached after that, don't release
+                        * it now.
+                        */
+                       if (p->exit_signal != -1) {
+                               do_notify_parent(p, p->exit_signal);
+                               if (p->exit_signal != -1)
+                                       p = NULL;
+                       }
+               }
+               write_unlock_irq(&tasklist_lock);
+       }
+       if (p != NULL)
+               release_task(p);
        BUG_ON(!retval);
        return retval;
 }
@@ -1165,15 +1199,15 @@ static int wait_task_zombie(struct task_struct *p, int noreap,
  * the lock and this task is uninteresting.  If we return nonzero, we have
  * released the lock and the system call should return.
  */
-static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
-                            int noreap, struct siginfo __user *infop,
+static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
+                            struct siginfo __user *infop,
                             int __user *stat_addr, struct rusage __user *ru)
 {
        int retval, exit_code;
 
        if (!p->exit_code)
                return 0;
-       if (delayed_group_leader &&
+       if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
            p->signal && p->signal->group_stop_count > 0)
                /*
                 * A group stop is in progress and this is the group leader.
@@ -1194,13 +1228,14 @@ static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
        if (unlikely(noreap)) {
                pid_t pid = p->pid;
                uid_t uid = p->uid;
+               int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
 
                exit_code = p->exit_code;
                if (unlikely(!exit_code) ||
                    unlikely(p->state & TASK_TRACED))
                        goto bail_ref;
-               return wait_noreap_copyout(p, pid, uid, CLD_STOPPED,
-                                          (exit_code << 8) | 0x7f,
+               return wait_noreap_copyout(p, pid, uid,
+                                          why, (exit_code << 8) | 0x7f,
                                           infop, ru);
        }
 
@@ -1244,7 +1279,7 @@ bail_ref:
 
        /* move to end of parent's list to avoid starvation */
        remove_parent(p);
-       add_parent(p);
+       add_parent(p, p->parent);
 
        write_unlock_irq(&tasklist_lock);
 
@@ -1256,7 +1291,9 @@ bail_ref:
        if (!retval && infop)
                retval = put_user(0, &infop->si_errno);
        if (!retval && infop)
-               retval = put_user((short)CLD_STOPPED, &infop->si_code);
+               retval = put_user((short)((p->ptrace & PT_PTRACED)
+                                         ? CLD_TRAPPED : CLD_STOPPED),
+                                 &infop->si_code);
        if (!retval && infop)
                retval = put_user(exit_code, &infop->si_status);
        if (!retval && infop)
@@ -1277,7 +1314,7 @@ bail_ref:
  * the lock and this task is uninteresting.  If we return nonzero, we have
  * released the lock and the system call should return.
  */
-static int wait_task_continued(struct task_struct *p, int noreap,
+static int wait_task_continued(task_t *p, int noreap,
                               struct siginfo __user *infop,
                               int __user *stat_addr, struct rusage __user *ru)
 {
@@ -1324,6 +1361,22 @@ static int wait_task_continued(struct task_struct *p, int noreap,
 }
 
 
+static inline int my_ptrace_child(struct task_struct *p)
+{
+       if (!(p->ptrace & PT_PTRACED))
+               return 0;
+       if (!(p->ptrace & PT_ATTACHED))
+               return 1;
+       /*
+        * This child was PTRACE_ATTACH'd.  We should be seeing it only if
+        * we are the attacher.  If we are the real parent, this is a race
+        * inside ptrace_attach.  It is waiting for the tasklist_lock,
+        * which we have to switch the parent links, but has already set
+        * the flags in p->ptrace.
+        */
+       return (p->parent != p->real_parent);
+}
+
 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
                    int __user *stat_addr, struct rusage __user *ru)
 {
@@ -1347,7 +1400,7 @@ repeat:
                int ret;
 
                list_for_each(_p,&tsk->children) {
-                       p = list_entry(_p, struct task_struct, sibling);
+                       p = list_entry(_p,struct task_struct,sibling);
 
                        ret = eligible_child(pid, options, p);
                        if (!ret)
@@ -1355,17 +1408,26 @@ repeat:
 
                        switch (p->state) {
                        case TASK_TRACED:
+                               /*
+                                * When we hit the race with PTRACE_ATTACH,
+                                * we will not report this child.  But the
+                                * race means it has not yet been moved to
+                                * our ptrace_children list, so we need to
+                                * set the flag here to avoid a spurious ECHILD
+                                * when the race happens with the only child.
+                                */
                                flag = 1;
-                               continue;
+                               if (!my_ptrace_child(p))
+                                       continue;
+                               /*FALLTHROUGH*/
                        case TASK_STOPPED:
                                /*
                                 * It's stopped now, so it might later
                                 * continue, exit, or stop again.
                                 */
                                flag = 1;
-                               if (!(options & WUNTRACED))
-                                       continue;
-                               if (tracehook_inhibit_wait_stopped(p))
+                               if (!(options & WUNTRACED) &&
+                                   !my_ptrace_child(p))
                                        continue;
                                retval = wait_task_stopped(p, ret == 2,
                                                           (options & WNOWAIT),
@@ -1390,10 +1452,6 @@ repeat:
                                                goto check_continued;
                                        if (!likely(options & WEXITED))
                                                continue;
-                                       if (tracehook_inhibit_wait_zombie(p)) {
-                                               flag = 1;
-                                               continue;
-                                       }
                                        retval = wait_task_zombie(
                                                p, (options & WNOWAIT),
                                                infop, stat_addr, ru);
@@ -1410,8 +1468,6 @@ check_continued:
                                flag = 1;
                                if (!unlikely(options & WCONTINUED))
                                        continue;
-                               if (tracehook_inhibit_wait_continued(p))
-                                       continue;
                                retval = wait_task_continued(
                                        p, (options & WNOWAIT),
                                        infop, stat_addr, ru);
@@ -1420,19 +1476,21 @@ check_continued:
                                break;
                        }
                }
-
-               retval = ptrace_do_wait(tsk, pid, options,
-                                       infop, stat_addr, ru);
-               if (retval != -ECHILD) {
-                       flag = 1;
-                       if (retval != 0) /* He released the lock.  */
-                               goto end;
+               if (!flag) {
+                       list_for_each(_p, &tsk->ptrace_children) {
+                               p = list_entry(_p, struct task_struct,
+                                               ptrace_list);
+                               if (!eligible_child(pid, options, p))
+                                       continue;
+                               flag = 1;
+                               break;
+                       }
                }
-
                if (options & __WNOTHREAD)
                        break;
                tsk = next_thread(tsk);
-               BUG_ON(tsk->signal != current->signal);
+               if (tsk->signal != current->signal)
+                       BUG();
        } while (tsk != current);
 
        read_unlock(&tasklist_lock);
@@ -1452,7 +1510,7 @@ end:
        remove_wait_queue(&current->signal->wait_chldexit,&wait);
        if (infop) {
                if (retval > 0)
-                       retval = 0;
+               retval = 0;
                else {
                        /*
                         * For a WNOHANG return, clear out all the fields