[PATCH] Make number of IDE interfaces configurable
[safe/jmp/linux-2.6] / kernel / exit.c
index f436a6b..f250a5e 100644 (file)
@@ -4,7 +4,6 @@
  *  Copyright (C) 1991, 1992  Linus Torvalds
  */
 
-#include <linux/config.h>
 #include <linux/mm.h>
 #include <linux/slab.h>
 #include <linux/interrupt.h>
 #include <linux/security.h>
 #include <linux/cpu.h>
 #include <linux/acct.h>
+#include <linux/tsacct_kern.h>
 #include <linux/file.h>
 #include <linux/binfmts.h>
+#include <linux/nsproxy.h>
 #include <linux/ptrace.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/blkdev.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_init(&p->tasks);
-               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));
+
+       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);
-       /*
-        * 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
@@ -108,31 +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);
-       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
@@ -167,7 +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;
@@ -175,7 +222,7 @@ static int will_become_orphaned_pgrp(int pgrp, task_t *ignored_task)
        do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
                if (p == ignored_task
                                || p->exit_state
-                               || p->real_parent->pid == 1)
+                               || is_init(p->real_parent))
                        continue;
                if (process_group(p->real_parent) != pgrp
                            && p->real_parent->signal->session == p->signal->session) {
@@ -205,17 +252,6 @@ static int has_stopped_jobs(int pgrp)
        do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
                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);
@@ -248,9 +284,7 @@ static void reparent_to_init(void)
        /* Set the exit signal to SIGCHLD so we signal init on exit */
        current->exit_signal = SIGCHLD;
 
-       if ((current->policy == SCHED_NORMAL ||
-                       current->policy == SCHED_BATCH)
-                               && (task_nice(current) < 0))
+       if (!has_rt_policy(current) && (task_nice(current) < 0))
                set_user_nice(current, 0);
        /* cpus_allowed? */
        /* rt_priority? */
@@ -364,9 +398,11 @@ 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_task_namespaces(current);
+       current->nsproxy = init_task.nsproxy;
+       get_task_namespaces(current);
+
        exit_files(current);
        current->files = init_task.files;
        atomic_inc(&current->files->count);
@@ -443,6 +479,18 @@ void fastcall put_files_struct(struct files_struct *files)
 
 EXPORT_SYMBOL(put_files_struct);
 
+void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
+{
+       struct files_struct *old;
+
+       old = tsk->files;
+       task_lock(tsk);
+       tsk->files = files;
+       task_unlock(tsk);
+       put_files_struct(old);
+}
+EXPORT_SYMBOL(reset_files_struct);
+
 static inline void __exit_files(struct task_struct *tsk)
 {
        struct files_struct * files = tsk->files;
@@ -530,7 +578,7 @@ static 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;
@@ -540,7 +588,8 @@ static void exit_mm(struct task_struct * tsk)
        mmput(mm);
 }
 
-static inline void choose_new_parent(task_t *p, task_t *reaper)
+static inline void
+choose_new_parent(struct task_struct *p, struct task_struct *reaper)
 {
        /*
         * Make sure we're not reparenting to ourselves and that
@@ -550,7 +599,8 @@ static inline void choose_new_parent(task_t *p, task_t *reaper)
        p->real_parent = reaper;
 }
 
-static 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)
@@ -571,9 +621,9 @@ static 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.
@@ -614,8 +664,8 @@ static 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 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;
@@ -638,7 +688,7 @@ static 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;
 
@@ -667,7 +717,7 @@ static 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);
+               p = list_entry(_p, struct task_struct, ptrace_list);
                choose_new_parent(p, reaper);
                reparent_thread(p, father, 1);
        }
@@ -787,7 +837,7 @@ 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);
        }
 
@@ -799,7 +849,9 @@ 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);
 
@@ -832,19 +884,13 @@ fastcall NORET_TYPE void do_exit(long code)
 
        tsk->flags |= PF_EXITING;
 
-       /*
-        * Make sure we don't try to process any timer firings
-        * while we are already exiting.
-        */
-       tsk->it_virt_expires = cputime_zero;
-       tsk->it_prof_expires = cputime_zero;
-       tsk->it_sched_expires = 0;
-
        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);
@@ -854,20 +900,26 @@ fastcall NORET_TYPE void do_exit(long code)
        if (group_dead) {
                hrtimer_cancel(&tsk->signal->real_timer);
                exit_itimers(tsk->signal);
-               acct_process(code);
        }
+       acct_collect(code, group_dead);
        if (unlikely(tsk->robust_list))
                exit_robust_list(tsk);
-#ifdef CONFIG_COMPAT
+#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);
@@ -882,27 +934,39 @@ fastcall NORET_TYPE void do_exit(long code)
        tsk->exit_code = code;
        proc_exit_connector(tsk);
        exit_notify(tsk);
+       exit_task_namespaces(tsk);
 #ifdef CONFIG_NUMA
        mpol_free(tsk->mempolicy);
        tsk->mempolicy = NULL;
 #endif
        /*
-        * If DEBUG_MUTEXES is on, make sure we are holding no locks:
+        * 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:
         */
-       mutex_debug_check_no_locks_held(tsk);
+       debug_check_no_locks_held(tsk);
 
        if (tsk->io_context)
                exit_io_context();
 
-       /* PF_DEAD causes final put_task_struct after we schedule. */
+       if (tsk->splice_pipe)
+               __free_pipe_info(tsk->splice_pipe);
+
        preempt_disable();
-       BUG_ON(tsk->flags & PF_DEAD);
-       tsk->flags |= PF_DEAD;
+       /* causes final put_task_struct in finish_task_switch(). */
+       tsk->state = TASK_DEAD;
 
        schedule();
        BUG();
        /* Avoid "noreturn function does return".  */
-       for (;;) ;
+       for (;;)
+               cpu_relax();    /* For when BUG is null */
 }
 
 EXPORT_SYMBOL_GPL(do_exit);
@@ -911,7 +975,7 @@ NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 {
        if (comp)
                complete(comp);
-       
+
        do_exit(code);
 }
 
@@ -922,13 +986,6 @@ 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).
@@ -943,7 +1000,6 @@ 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.  */
@@ -953,7 +1009,6 @@ do_group_exit(int exit_code)
                        zap_other_threads(current);
                }
                spin_unlock_irq(&sighand->siglock);
-               read_unlock(&tasklist_lock);
        }
 
        do_exit(exit_code);
@@ -970,7 +1025,7 @@ 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)
@@ -1002,7 +1057,7 @@ 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))
@@ -1011,12 +1066,13 @@ 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);
@@ -1041,7 +1097,7 @@ 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)
 {
@@ -1203,8 +1259,8 @@ 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;
@@ -1318,7 +1374,7 @@ 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)
 {
@@ -1404,7 +1460,7 @@ 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)
@@ -1493,8 +1549,7 @@ 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);