linux 2.6.16.38 w/ vs2.0.3-rc1
[linux-2.6.git] / kernel / exit.c
index 08eee6d..208cbfa 100644 (file)
 #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/vs_limit.h>
 #include <linux/vs_context.h>
 #include <linux/vs_network.h>
@@ -57,85 +52,15 @@ 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)--;
+               if (p->pid)
+                       __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);
-       }
-}
-
-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)
@@ -144,14 +69,21 @@ void release_task(struct task_struct * p)
        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);
-       ptrace_unlink(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
@@ -179,13 +111,28 @@ repeat:
        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
@@ -293,10 +240,10 @@ static void reparent_to_init(void)
 
        ptrace_unlink(current);
        /* Reparent to init */
-       remove_parent(current);
+       REMOVE_LINKS(current);
        current->parent = child_reaper;
        current->real_parent = child_reaper;
-       add_parent(current);
+       SET_LINKS(current);
 
        /* Set the exit signal to SIGCHLD so we signal init on exit */
        current->exit_signal = SIGCHLD;
@@ -402,9 +349,9 @@ void daemonize(const char *name, ...)
        exit_mm(current);
 
        set_special_pids(1, 1);
-       mutex_lock(&tty_mutex);
+       down(&tty_sem);
        current->signal->tty = NULL;
-       mutex_unlock(&tty_mutex);
+       up(&tty_sem);
 
        /* Block and flush all signals */
        sigfillset(&blocked);
@@ -457,7 +404,6 @@ static void close_files(struct files_struct * files)
                        }
                        i++;
                        set >>= 1;
-                       cond_resched();
                }
        }
 }
@@ -606,7 +552,7 @@ static inline void choose_new_parent(task_t *p, task_t *reaper)
         * Make sure we're not reparenting to ourselves and that
         * the parent is not a zombie.
         */
-       BUG_ON(p == reaper || reaper->exit_state);
+       BUG_ON(p == reaper || reaper->exit_state >= EXIT_ZOMBIE);
        p->real_parent = reaper;
 }
 
@@ -631,9 +577,9 @@ static void reparent_thread(task_t *p, task_t *father, int traced)
                 * anyway, so let go of it.
                 */
                p->ptrace = 0;
-               remove_parent(p);
+               list_del_init(&p->sibling);
                p->parent = p->real_parent;
-               add_parent(p);
+               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.
@@ -728,6 +674,7 @@ static void forget_original_parent(struct task_struct * father,
        }
        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);
        }
@@ -869,8 +816,10 @@ 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();
 
        if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
                current->ptrace_message = code;
@@ -884,8 +833,6 @@ 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();
        }
@@ -907,30 +854,13 @@ fastcall NORET_TYPE void do_exit(long code)
                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();
-               }
        }
-       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);
        exit_mm(tsk);
 
        exit_sem(tsk);
        __exit_files(tsk);
        __exit_fs(tsk);
        exit_namespace(tsk);
-       exit_vx_info(tsk, code);
-       exit_nx_info(tsk);
        exit_thread();
        cpuset_exit(tsk);
        exit_keys(tsk);
@@ -944,6 +874,8 @@ fastcall NORET_TYPE void do_exit(long code)
 
        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);
@@ -954,11 +886,9 @@ fastcall NORET_TYPE void do_exit(long code)
         */
        mutex_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();
@@ -988,6 +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).
@@ -1002,6 +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.  */
@@ -1011,6 +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);
@@ -1340,7 +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);