This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / kernel / exit.c
index d27b470..17987bc 100644 (file)
@@ -4,12 +4,12 @@
  *  Copyright (C) 1991, 1992  Linus Torvalds
  */
 
-#include <linux/config.h>
 #include <linux/mm.h>
 #include <linux/slab.h>
 #include <linux/interrupt.h>
 #include <linux/smp_lock.h>
 #include <linux/module.h>
+#include <linux/capability.h>
 #include <linux/completion.h>
 #include <linux/personality.h>
 #include <linux/tty.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 <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--;
-       /* tasklist_lock is held, is this sufficient? */
-       if (p->vx_info) {
-               atomic_dec(&p->vx_info->cvirt.nr_threads);
-               vx_nproc_dec(p);
-       }
        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);
-               if (p->pid)
-                       __get_cpu_var(process_counts)--;
+
+               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));
 
-       REMOVE_LINKS(p);
+       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);
+       }
+}
+
+static void delayed_put_task_struct(struct rcu_head *rhp)
+{
+       put_task_struct(container_of(rhp, struct task_struct, rcu));
 }
 
 void release_task(struct task_struct * p)
 {
+       struct task_struct *leader;
        int zap_leader;
-       task_t *leader;
-       struct dentry *proc_dentry;
-
-repeat: 
+repeat:
        atomic_dec(&p->user->processes);
-       spin_lock(&p->proc_lock);
-       proc_dentry = proc_pid_unhash(p);
        write_lock_irq(&tasklist_lock);
-       if (unlikely(p->ptrace))
-               __ptrace_unlink(p);
+       ptrace_unlink(p);
        BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
        __exit_signal(p);
-       __exit_sighand(p);
-       __unhash_process(p);
 
        /*
         * If we are the last non-leader member of the thread
@@ -99,33 +171,15 @@ repeat:
 
        sched_exit(p);
        write_unlock_irq(&tasklist_lock);
-       spin_unlock(&p->proc_lock);
-       proc_pid_flush(proc_dentry);
+       proc_flush_task(p);
        release_thread(p);
-       if (p->vx_info)
-               release_vx_info(p->vx_info, p);
-       put_task_struct(p);
+       call_rcu(&p->rcu, delayed_put_task_struct);
 
        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
@@ -160,7 +214,7 @@ out:
  *
  * "I ask you, have you ever known what it is to be an orphan?"
  */
-static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
+static int will_become_orphaned_pgrp(int pgrp, struct task_struct *ignored_task)
 {
        struct task_struct *p;
        int ret = 1;
@@ -190,7 +244,7 @@ int is_orphaned_pgrp(int pgrp)
        return retval;
 }
 
-static inline int has_stopped_jobs(int pgrp)
+static int has_stopped_jobs(int pgrp)
 {
        int retval = 0;
        struct task_struct *p;
@@ -216,7 +270,7 @@ static inline int has_stopped_jobs(int pgrp)
 }
 
 /**
- * reparent_to_init() - Reparent the calling kernel thread to the init task.
+ * reparent_to_init - Reparent the calling kernel thread to the init task.
  *
  * If a kernel thread is launched as a result of a system call, or if
  * it ever exits, it should generally reparent itself to init so that
@@ -227,22 +281,23 @@ static inline int has_stopped_jobs(int pgrp)
  *
  * NOTE that reparent_to_init() gives the caller full capabilities.
  */
-void reparent_to_init(void)
+static void reparent_to_init(void)
 {
        write_lock_irq(&tasklist_lock);
 
        ptrace_unlink(current);
        /* Reparent to init */
-       REMOVE_LINKS(current);
-       /* FIXME handle vchild_reaper/initpid */
+       remove_parent(current);
        current->parent = child_reaper;
        current->real_parent = child_reaper;
-       SET_LINKS(current);
+       add_parent(current);
 
        /* Set the exit signal to SIGCHLD so we signal init on exit */
        current->exit_signal = SIGCHLD;
 
-       if ((current->policy == SCHED_NORMAL) && (task_nice(current) < 0))
+       if ((current->policy == SCHED_NORMAL ||
+                       current->policy == SCHED_BATCH)
+                               && (task_nice(current) < 0))
                set_user_nice(current, 0);
        /* cpus_allowed? */
        /* rt_priority? */
@@ -257,7 +312,7 @@ void reparent_to_init(void)
 
 void __set_special_pids(pid_t session, pid_t pgrp)
 {
-       struct task_struct *curr = current;
+       struct task_struct *curr = current->group_leader;
 
        if (curr->signal->session != session) {
                detach_pid(curr, PIDTYPE_SID);
@@ -285,7 +340,7 @@ void set_special_pids(pid_t session, pid_t pgrp)
  */
 int allow_signal(int sig)
 {
-       if (sig < 1 || sig > _NSIG)
+       if (!valid_signal(sig) || sig < 1)
                return -EINVAL;
 
        spin_lock_irq(&current->sighand->siglock);
@@ -306,7 +361,7 @@ EXPORT_SYMBOL(allow_signal);
 
 int disallow_signal(int sig)
 {
-       if (sig < 1 || sig > _NSIG)
+       if (!valid_signal(sig) || sig < 1)
                return -EINVAL;
 
        spin_lock_irq(&current->sighand->siglock);
@@ -341,9 +396,7 @@ void daemonize(const char *name, ...)
        exit_mm(current);
 
        set_special_pids(1, 1);
-       down(&tty_sem);
-       current->signal->tty = NULL;
-       up(&tty_sem);
+       proc_clear_tty(current);
 
        /* Block and flush all signals */
        sigfillset(&blocked);
@@ -356,6 +409,9 @@ void daemonize(const char *name, ...)
        fs = init_task.fs;
        current->fs = fs;
        atomic_inc(&fs->count);
+       exit_namespace(current);
+       current->namespace = init_task.namespace;
+       get_namespace(current->namespace);
        exit_files(current);
        current->files = init_task.files;
        atomic_inc(&current->files->count);
@@ -365,26 +421,35 @@ void daemonize(const char *name, ...)
 
 EXPORT_SYMBOL(daemonize);
 
-static inline void close_files(struct files_struct * files)
+static void close_files(struct files_struct * files)
 {
        int i, j;
+       struct fdtable *fdt;
 
        j = 0;
+
+       /*
+        * It is safe to dereference the fd table without RCU or
+        * ->file_lock because this is the last reference to the
+        * files structure.
+        */
+       fdt = files_fdtable(files);
        for (;;) {
                unsigned long set;
                i = j * __NFDBITS;
-               if (i >= files->max_fdset || i >= files->max_fds)
+               if (i >= fdt->max_fdset || i >= fdt->max_fds)
                        break;
-               set = files->open_fds->fds_bits[j++];
+               set = fdt->open_fds->fds_bits[j++];
                while (set) {
                        if (set & 1) {
-                               struct file * file = xchg(&files->fd[i], NULL);
+                               struct file * file = xchg(&fdt->fd[i], NULL);
                                if (file)
                                        filp_close(file, files);
                                vx_openfd_dec(i);
                        }
                        i++;
                        set >>= 1;
+                       cond_resched();
                }
        }
 }
@@ -404,18 +469,22 @@ struct files_struct *get_files_struct(struct task_struct *task)
 
 void fastcall put_files_struct(struct files_struct *files)
 {
+       struct fdtable *fdt;
+
        if (atomic_dec_and_test(&files->count)) {
                close_files(files);
                /*
                 * Free the fd and fdset arrays if we expanded them.
+                * If the fdtable was embedded, pass files for freeing
+                * at the end of the RCU grace period. Otherwise,
+                * you can free files immediately.
                 */
-               if (files->fd != &files->fd_array[0])
-                       free_fd_array(files->fd, files->max_fds);
-               if (files->max_fdset > __FD_SETSIZE) {
-                       free_fdset(files->open_fds, files->max_fdset);
-                       free_fdset(files->close_on_exec, files->max_fdset);
-               }
-               kmem_cache_free(files_cachep, files);
+               fdt = files_fdtable(files);
+               if (fdt == &files->fdtab)
+                       fdt->free_files = files;
+               else
+                       kmem_cache_free(files_cachep, files);
+               free_fdtable(fdt);
        }
 }
 
@@ -482,7 +551,7 @@ EXPORT_SYMBOL_GPL(exit_fs);
  * Turn us into a lazy TLB process if we
  * aren't already..
  */
-void exit_mm(struct task_struct * tsk)
+static void exit_mm(struct task_struct * tsk)
 {
        struct mm_struct *mm = tsk->mm;
 
@@ -508,7 +577,7 @@ void exit_mm(struct task_struct * tsk)
                down_read(&mm->mmap_sem);
        }
        atomic_inc(&mm->mm_count);
-       if (mm != tsk->active_mm) BUG();
+       BUG_ON(mm != tsk->active_mm);
        /* more a memory barrier than a real lock */
        task_lock(tsk);
        tsk->mm = NULL;
@@ -518,19 +587,24 @@ void exit_mm(struct task_struct * tsk)
        mmput(mm);
 }
 
-static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
+static inline void
+choose_new_parent(struct task_struct *p, struct task_struct *reaper)
 {
+       /* check for reaper context */
+       vxwprintk((p->xid != reaper->xid) && (reaper != child_reaper),
+               "rogue reaper: %p[%d,#%u] <> %p[%d,#%u]",
+               p, p->pid, p->xid, reaper, reaper->pid, reaper->xid);
+
        /*
         * Make sure we're not reparenting to ourselves and that
         * the parent is not a zombie.
         */
-       BUG_ON(p == reaper || reaper->exit_state >= EXIT_ZOMBIE);
+       BUG_ON(p == reaper || reaper->exit_state);
        p->real_parent = reaper;
-       if (p->parent == p->real_parent)
-               BUG();
 }
 
-static inline void reparent_thread(task_t *p, task_t *father, int traced)
+static void
+reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
 {
        /* We don't want people slaying init.  */
        if (p->exit_signal != -1)
@@ -538,7 +612,7 @@ static inline void reparent_thread(task_t *p, task_t *father, int traced)
 
        if (p->pdeath_signal)
                /* We already hold the tasklist_lock here.  */
-               group_send_sig_info(p->pdeath_signal, (void *) 0, p);
+               group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
 
        /* Move the child from its dying parent to the new one.  */
        if (unlikely(traced)) {
@@ -551,9 +625,9 @@ static inline void reparent_thread(task_t *p, task_t *father, int traced)
                 * anyway, so let go of it.
                 */
                p->ptrace = 0;
-               list_del_init(&p->sibling);
+               remove_parent(p);
                p->parent = p->real_parent;
-               list_add_tail(&p->sibling, &p->parent->children);
+               add_parent(p);
 
                /* If we'd notified the old parent about this child's death,
                 * also notify the new parent.
@@ -582,8 +656,8 @@ static inline void reparent_thread(task_t *p, task_t *father, int traced)
                int pgrp = process_group(p);
 
                if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
-                       __kill_pg_info(SIGHUP, (void *)1, pgrp);
-                       __kill_pg_info(SIGCONT, (void *)1, pgrp);
+                       __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
+                       __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
                }
        }
 }
@@ -594,17 +668,16 @@ static inline void reparent_thread(task_t *p, task_t *father, int traced)
  * group, and if no such member exists, give it to
  * the global child reaper process (ie "init")
  */
-static inline void forget_original_parent(struct task_struct * father,
-                                         struct list_head *to_release)
+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;
 
-       /* FIXME handle vchild_reaper/initpid */
        do {
                reaper = next_thread(reaper);
                if (reaper == father) {
-                       reaper = child_reaper;
+                       reaper = vx_child_reaper(father);
                        break;
                }
        } while (reaper->exit_state);
@@ -619,7 +692,7 @@ static inline void forget_original_parent(struct task_struct * father,
         */
        list_for_each_safe(_p, _n, &father->children) {
                int ptrace;
-               p = list_entry(_p,struct task_struct,sibling);
+               p = list_entry(_p, struct task_struct, sibling);
 
                ptrace = p->ptrace;
 
@@ -628,7 +701,7 @@ static inline void forget_original_parent(struct task_struct * father,
 
                if (father == p->real_parent) {
                        /* reparent with a reaper, real father it's us */
-                       choose_new_parent(p, reaper, child_reaper);
+                       choose_new_parent(p, vx_child_reaper(p));
                        reparent_thread(p, father, 0);
                } else {
                        /* reparent ptraced task to its real parent */
@@ -648,8 +721,8 @@ static inline void forget_original_parent(struct task_struct * father,
                        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, child_reaper);
+               p = list_entry(_p, struct task_struct, ptrace_list);
+               choose_new_parent(p, reaper);
                reparent_thread(p, father, 1);
        }
 }
@@ -719,8 +792,8 @@ static void exit_notify(struct task_struct *tsk)
            (t->signal->session == tsk->signal->session) &&
            will_become_orphaned_pgrp(process_group(tsk), tsk) &&
            has_stopped_jobs(process_group(tsk))) {
-               __kill_pg_info(SIGHUP, (void *)1, process_group(tsk));
-               __kill_pg_info(SIGCONT, (void *)1, process_group(tsk));
+               __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
+               __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
        }
 
        /* Let father know we died 
@@ -764,94 +837,159 @@ static void exit_notify(struct task_struct *tsk)
                state = EXIT_DEAD;
        tsk->exit_state = state;
 
-       /*
-        * Clear these here so that update_process_times() won't try to deliver
-        * itimer, profile or rlimit signals to this task while it is in late exit.
-        */
-       tsk->it_virt_value = cputime_zero;
-       tsk->it_prof_value = cputime_zero;
-
        write_unlock_irq(&tasklist_lock);
 
        list_for_each_safe(_p, _n, &ptrace_dead) {
                list_del_init(_p);
-               t = list_entry(_p,struct task_struct,ptrace_list);
+               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)
                release_task(tsk);
-
-       /* PF_DEAD causes final put_task_struct after we schedule. */
-       preempt_disable();
-       tsk->flags |= PF_DEAD;
 }
 
 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);
 
+       WARN_ON(atomic_read(&tsk->fs_excl));
+
        if (unlikely(in_interrupt()))
                panic("Aiee, killing interrupt handler!");
        if (unlikely(!tsk->pid))
                panic("Attempted to kill the idle task!");
-       if (unlikely(tsk->pid == 1))
+       if (unlikely(tsk == child_reaper))
                panic("Attempted to kill init!");
-       if (tsk->io_context)
-               exit_io_context();
 
        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
+        * leave this task alone and wait for reboot.
+        */
+       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;
-       del_timer_sync(&tsk->real_timer);
 
        if (unlikely(in_atomic()))
                printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
                                current->comm, current->pid,
                                preempt_count());
 
-       acct_update_integrals();
-       update_mem_hiwater();
+       taskstats_exit_alloc(&tidstats, &mycpu);
+
+       acct_update_integrals(tsk);
+       if (tsk->mm) {
+               update_hiwater_rss(tsk->mm);
+               update_hiwater_vm(tsk->mm);
+       }
        group_dead = atomic_dec_and_test(&tsk->signal->live);
-       if (group_dead)
-               acct_process(code);
+       if (group_dead) {
+               hrtimer_cancel(&tsk->signal->real_timer);
+               exit_itimers(tsk->signal);
+       }
+
+       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);
        exit_namespace(tsk);
        exit_thread();
+       cpuset_exit(tsk);
        exit_keys(tsk);
 
        if (group_dead && tsk->signal->leader)
                disassociate_ctty(1);
 
-       module_put(tsk->thread_info->exec_domain->module);
+       module_put(task_thread_info(tsk)->exec_domain->module);
        if (tsk->binfmt)
                module_put(tsk->binfmt->module);
 
        tsk->exit_code = code;
+       proc_exit_connector(tsk);
+       /* needs to stay before exit_notify() */
+       exit_vx_info_early(tsk, code);
        exit_notify(tsk);
 #ifdef CONFIG_NUMA
        mpol_free(tsk->mempolicy);
        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:
+        */
+       debug_check_no_locks_held(tsk);
+
+       if (tsk->io_context)
+               exit_io_context();
+
+       if (tsk->splice_pipe)
+               __free_pipe_info(tsk->splice_pipe);
+
+       /* needs to stay after exit_notify() */
+       exit_vx_info(tsk, code);
+       exit_nx_info(tsk);
+
+       /* PF_DEAD causes final put_task_struct after we schedule. */
+       preempt_disable();
+       BUG_ON(tsk->flags & PF_DEAD);
+       tsk->flags |= PF_DEAD;
 
-       BUG_ON(!(current->flags & PF_DEAD));
        schedule();
        BUG();
        /* Avoid "noreturn function does return".  */
        for (;;) ;
 }
 
+EXPORT_SYMBOL_GPL(do_exit);
+
 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 {
        if (comp)
@@ -867,13 +1005,6 @@ 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).
@@ -888,18 +1019,15 @@ 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.  */
                        exit_code = sig->group_exit_code;
                else {
-                       sig->flags = SIGNAL_GROUP_EXIT;
                        sig->group_exit_code = exit_code;
                        zap_other_threads(current);
                }
                spin_unlock_irq(&sighand->siglock);
-               read_unlock(&tasklist_lock);
        }
 
        do_exit(exit_code);
@@ -916,7 +1044,7 @@ asmlinkage void sys_exit_group(int error_code)
        do_group_exit((error_code & 0xff) << 8);
 }
 
-static int eligible_child(pid_t pid, int options, task_t *p)
+static int eligible_child(pid_t pid, int options, struct task_struct *p)
 {
        if (pid > 0) {
                if (p->pid != pid)
@@ -948,7 +1076,7 @@ static int eligible_child(pid_t pid, int options, task_t *p)
         * Do not consider thread group leaders that are
         * in a non-empty thread group:
         */
-       if (current->tgid != p->tgid && delay_group_leader(p))
+       if (delay_group_leader(p))
                return 2;
 
        if (security_task_wait(p))
@@ -957,12 +1085,13 @@ static int eligible_child(pid_t pid, int options, task_t *p)
        return 1;
 }
 
-static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
+static int wait_noreap_copyout(struct task_struct *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);
@@ -987,7 +1116,7 @@ static int wait_noreap_copyout(task_t *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(task_t *p, int noreap,
+static int wait_task_zombie(struct task_struct *p, int noreap,
                            struct siginfo __user *infop,
                            int __user *stat_addr, struct rusage __user *ru)
 {
@@ -1036,6 +1165,9 @@ static int wait_task_zombie(task_t *p, int noreap,
        }
 
        if (likely(p->real_parent == p->parent) && likely(p->signal)) {
+               struct signal_struct *psig;
+               struct signal_struct *sig;
+
                /*
                 * The resource counters for the group leader are in its
                 * own task_struct.  Those for dead threads in the group
@@ -1052,24 +1184,26 @@ static int wait_task_zombie(task_t *p, int noreap,
                 * here reaping other children at the same time.
                 */
                spin_lock_irq(&p->parent->sighand->siglock);
-               p->parent->signal->cutime =
-                       cputime_add(p->parent->signal->cutime,
+               psig = p->parent->signal;
+               sig = p->signal;
+               psig->cutime =
+                       cputime_add(psig->cutime,
                        cputime_add(p->utime,
-                       cputime_add(p->signal->utime,
-                                   p->signal->cutime)));
-               p->parent->signal->cstime =
-                       cputime_add(p->parent->signal->cstime,
+                       cputime_add(sig->utime,
+                                   sig->cutime)));
+               psig->cstime =
+                       cputime_add(psig->cstime,
                        cputime_add(p->stime,
-                       cputime_add(p->signal->stime,
-                                   p->signal->cstime)));
-               p->parent->signal->cmin_flt +=
-                       p->min_flt + p->signal->min_flt + p->signal->cmin_flt;
-               p->parent->signal->cmaj_flt +=
-                       p->maj_flt + p->signal->maj_flt + p->signal->cmaj_flt;
-               p->parent->signal->cnvcsw +=
-                       p->nvcsw + p->signal->nvcsw + p->signal->cnvcsw;
-               p->parent->signal->cnivcsw +=
-                       p->nivcsw + p->signal->nivcsw + p->signal->cnivcsw;
+                       cputime_add(sig->stime,
+                                   sig->cstime)));
+               psig->cmin_flt +=
+                       p->min_flt + sig->min_flt + sig->cmin_flt;
+               psig->cmaj_flt +=
+                       p->maj_flt + sig->maj_flt + sig->cmaj_flt;
+               psig->cnvcsw +=
+                       p->nvcsw + sig->nvcsw + sig->cnvcsw;
+               psig->cnivcsw +=
+                       p->nivcsw + sig->nivcsw + sig->cnivcsw;
                spin_unlock_irq(&p->parent->sighand->siglock);
        }
 
@@ -1144,8 +1278,8 @@ static int wait_task_zombie(task_t *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(task_t *p, int delayed_group_leader, int noreap,
-                            struct siginfo __user *infop,
+static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
+                            int noreap, struct siginfo __user *infop,
                             int __user *stat_addr, struct rusage __user *ru)
 {
        int retval, exit_code;
@@ -1177,7 +1311,7 @@ static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
 
                exit_code = p->exit_code;
                if (unlikely(!exit_code) ||
-                   unlikely(p->state > TASK_STOPPED))
+                   unlikely(p->state & TASK_TRACED))
                        goto bail_ref;
                return wait_noreap_copyout(p, pid, uid,
                                           why, (exit_code << 8) | 0x7f,
@@ -1224,7 +1358,7 @@ bail_ref:
 
        /* move to end of parent's list to avoid starvation */
        remove_parent(p);
-       add_parent(p, p->parent);
+       add_parent(p);
 
        write_unlock_irq(&tasklist_lock);
 
@@ -1259,7 +1393,7 @@ 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(task_t *p, int noreap,
+static int wait_task_continued(struct task_struct *p, int noreap,
                               struct siginfo __user *infop,
                               int __user *stat_addr, struct rusage __user *ru)
 {
@@ -1345,7 +1479,7 @@ 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)
@@ -1353,6 +1487,15 @@ 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;
                                if (!my_ptrace_child(p))
                                        continue;
                                /*FALLTHROUGH*/
@@ -1425,8 +1568,7 @@ check_continued:
                if (options & __WNOTHREAD)
                        break;
                tsk = next_thread(tsk);
-               if (tsk->signal != current->signal)
-                       BUG();
+               BUG_ON(tsk->signal != current->signal);
        } while (tsk != current);
 
        read_unlock(&tasklist_lock);