linux 2.6.16.38 w/ vs2.0.3-rc1
[linux-2.6.git] / kernel / exit.c
index 17987bc..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/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)--;
+               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)
 {
-       struct task_struct *leader;
        int zap_leader;
-repeat:
+       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);
-       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
@@ -171,15 +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
@@ -214,7 +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;
@@ -287,10 +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;
@@ -396,7 +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);
@@ -449,7 +404,6 @@ static void close_files(struct files_struct * files)
                        }
                        i++;
                        set >>= 1;
-                       cond_resched();
                }
        }
 }
@@ -577,7 +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;
@@ -587,8 +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),
@@ -599,12 +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);
+       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, int traced)
+static void reparent_thread(task_t *p, task_t *father, int traced)
 {
        /* We don't want people slaying init.  */
        if (p->exit_signal != -1)
@@ -625,9 +577,9 @@ reparent_thread(struct task_struct *p, struct task_struct *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.
@@ -668,8 +620,8 @@ reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
  * 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, 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;
@@ -692,7 +644,7 @@ forget_original_parent(struct task_struct *father, struct list_head *to_release)
         */
        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;
 
@@ -721,7 +673,8 @@ forget_original_parent(struct task_struct *father, struct list_head *to_release)
                        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);
+               p = list_entry(_p,struct task_struct,ptrace_list);
+
                choose_new_parent(p, reaper);
                reparent_thread(p, father, 1);
        }
@@ -841,7 +794,7 @@ static void exit_notify(struct task_struct *tsk)
 
        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);
        }
 
@@ -853,9 +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);
 
@@ -865,8 +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;
@@ -880,8 +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();
        }
@@ -893,8 +844,6 @@ fastcall NORET_TYPE void do_exit(long code)
                                current->comm, current->pid,
                                preempt_count());
 
-       taskstats_exit_alloc(&tidstats, &mycpu);
-
        acct_update_integrals(tsk);
        if (tsk->mm) {
                update_hiwater_rss(tsk->mm);
@@ -904,32 +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);
@@ -955,23 +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);
@@ -1005,6 +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).
@@ -1019,6 +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.  */
@@ -1028,6 +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);
@@ -1044,7 +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)
@@ -1076,7 +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))
@@ -1085,13 +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);
@@ -1116,7 +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)
 {
@@ -1278,8 +1199,8 @@ 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;
@@ -1358,7 +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);
 
@@ -1393,7 +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)
 {
@@ -1479,7 +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)
@@ -1568,7 +1489,8 @@ check_continued:
                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);