[PATCH] Add irq flag to disable balancing for an interrupt
[safe/jmp/linux-2.6] / kernel / exit.c
index ab06b9f..f132349 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/completion.h>
 #include <linux/personality.h>
 #include <linux/tty.h>
-#include <linux/namespace.h>
+#include <linux/mnt_namespace.h>
 #include <linux/key.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/pid_namespace.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>
@@ -37,6 +41,7 @@
 #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>
@@ -44,7 +49,6 @@
 #include <asm/mmu_context.h>
 
 extern void sem_exit (void);
-extern struct task_struct *child_reaper;
 
 static void exit_mm(struct task_struct * tsk);
 
@@ -124,6 +128,7 @@ static void __exit_signal(struct task_struct *tsk)
        flush_sigqueue(&tsk->pending);
        if (sig) {
                flush_sigqueue(&sig->shared_pending);
+               taskstats_tgid_free(sig);
                __cleanup_signal(sig);
        }
 }
@@ -135,8 +140,8 @@ static void delayed_put_task_struct(struct rcu_head *rhp)
 
 void release_task(struct task_struct * p)
 {
+       struct task_struct *leader;
        int zap_leader;
-       task_t *leader;
 repeat:
        atomic_dec(&p->user->processes);
        write_lock_irq(&tasklist_lock);
@@ -180,25 +185,20 @@ repeat:
  * 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
  * without this...
+ *
+ * The caller must hold rcu lock or the tasklist lock.
  */
-int session_of_pgrp(int pgrp)
+struct pid *session_of_pgrp(struct pid *pgrp)
 {
        struct task_struct *p;
-       int sid = -1;
+       struct pid *sid = NULL;
+
+       p = pid_task(pgrp, PIDTYPE_PGID);
+       if (p == NULL)
+               p = pid_task(pgrp, PIDTYPE_PID);
+       if (p != NULL)
+               sid = task_session(p);
 
-       read_lock(&tasklist_lock);
-       do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
-               if (p->signal->session > 0) {
-                       sid = p->signal->session;
-                       goto out;
-               }
-       } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
-       p = find_task_by_pid(pgrp);
-       if (p)
-               sid = p->signal->session;
-out:
-       read_unlock(&tasklist_lock);
-       
        return sid;
 }
 
@@ -210,63 +210,52 @@ 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(struct pid *pgrp, struct task_struct *ignored_task)
 {
        struct task_struct *p;
        int ret = 1;
 
-       do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
+       do_each_pid_task(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) {
+               if (task_pgrp(p->real_parent) != pgrp &&
+                   task_session(p->real_parent) == task_session(p)) {
                        ret = 0;
                        break;
                }
-       } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
+       } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
        return ret;     /* (sighing) "Often!" */
 }
 
-int is_orphaned_pgrp(int pgrp)
+int is_current_pgrp_orphaned(void)
 {
        int retval;
 
        read_lock(&tasklist_lock);
-       retval = will_become_orphaned_pgrp(pgrp, NULL);
+       retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
        read_unlock(&tasklist_lock);
 
        return retval;
 }
 
-static int has_stopped_jobs(int pgrp)
+static int has_stopped_jobs(struct pid *pgrp)
 {
        int retval = 0;
        struct task_struct *p;
 
-       do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
+       do_each_pid_task(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);
+       } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
        return retval;
 }
 
 /**
- * reparent_to_init - Reparent the calling kernel thread to the init task.
+ * reparent_to_init - Reparent the calling kernel thread to the init task of the pid space that the thread belongs to.
  *
  * 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
@@ -284,16 +273,14 @@ static void reparent_to_init(void)
        ptrace_unlink(current);
        /* Reparent to init */
        remove_parent(current);
-       current->parent = child_reaper;
-       current->real_parent = child_reaper;
+       current->parent = child_reaper(current);
+       current->real_parent = child_reaper(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 ||
-                       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? */
@@ -310,9 +297,9 @@ void __set_special_pids(pid_t session, pid_t pgrp)
 {
        struct task_struct *curr = current->group_leader;
 
-       if (curr->signal->session != session) {
+       if (process_session(curr) != session) {
                detach_pid(curr, PIDTYPE_SID);
-               curr->signal->session = session;
+               set_signal_session(curr->signal, session);
                attach_pid(curr, PIDTYPE_SID, session);
        }
        if (process_group(curr) != pgrp) {
@@ -322,7 +309,7 @@ void __set_special_pids(pid_t session, pid_t pgrp)
        }
 }
 
-void set_special_pids(pid_t session, pid_t pgrp)
+static void set_special_pids(pid_t session, pid_t pgrp)
 {
        write_lock_irq(&tasklist_lock);
        __set_special_pids(session, pgrp);
@@ -392,9 +379,7 @@ void daemonize(const char *name, ...)
        exit_mm(current);
 
        set_special_pids(1, 1);
-       mutex_lock(&tty_mutex);
-       current->signal->tty = NULL;
-       mutex_unlock(&tty_mutex);
+       proc_clear_tty(current);
 
        /* Block and flush all signals */
        sigfillset(&blocked);
@@ -407,9 +392,11 @@ 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);
@@ -435,14 +422,16 @@ static void close_files(struct files_struct * files)
        for (;;) {
                unsigned long set;
                i = j * __NFDBITS;
-               if (i >= fdt->max_fdset || i >= fdt->max_fds)
+               if (i >= fdt->max_fds)
                        break;
                set = fdt->open_fds->fds_bits[j++];
                while (set) {
                        if (set & 1) {
                                struct file * file = xchg(&fdt->fd[i], NULL);
-                               if (file)
+                               if (file) {
                                        filp_close(file, files);
+                                       cond_resched();
+                               }
                        }
                        i++;
                        set >>= 1;
@@ -476,9 +465,7 @@ void fastcall put_files_struct(struct files_struct *files)
                 * you can free files immediately.
                 */
                fdt = files_fdtable(files);
-               if (fdt == &files->fdtab)
-                       fdt->free_files = files;
-               else
+               if (fdt != &files->fdtab)
                        kmem_cache_free(files_cachep, files);
                free_fdtable(fdt);
        }
@@ -486,6 +473,18 @@ 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;
@@ -583,7 +582,8 @@ 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
@@ -593,12 +593,9 @@ 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)
-               p->exit_signal = SIGCHLD;
-
        if (p->pdeath_signal)
                /* We already hold the tasklist_lock here.  */
                group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
@@ -618,13 +615,7 @@ static void reparent_thread(task_t *p, task_t *father, int traced)
                p->parent = p->real_parent;
                add_parent(p);
 
-               /* 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 (p->state == TASK_TRACED) {
                        /*
                         * If it was at a trace stop, turn it into
                         * a normal stop since it's no longer being
@@ -634,19 +625,37 @@ static void reparent_thread(task_t *p, task_t *father, int traced)
                }
        }
 
+       /* If this is a threaded reparent there is no need to
+        * notify anyone anything has happened.
+        */
+       if (p->real_parent->group_leader == father->group_leader)
+               return;
+
+       /* We don't want people slaying init.  */
+       if (p->exit_signal != -1)
+               p->exit_signal = SIGCHLD;
+
+       /* If we'd notified the old parent about this child's death,
+        * also notify the new parent.
+        */
+       if (!traced && p->exit_state == EXIT_ZOMBIE &&
+           p->exit_signal != -1 && thread_group_empty(p))
+               do_notify_parent(p, p->exit_signal);
+
        /*
         * process group orphan check
         * Case ii: Our child is in a different pgrp
         * than we are, and it was the only connection
         * outside, so the child pgrp is now orphaned.
         */
-       if ((process_group(p) != process_group(father)) &&
-           (p->signal->session == father->signal->session)) {
-               int pgrp = process_group(p);
-
-               if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
-                       __kill_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
-                       __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
+       if ((task_pgrp(p) != task_pgrp(father)) &&
+           (task_session(p) == task_session(father))) {
+               struct pid *pgrp = task_pgrp(p);
+
+               if (will_become_orphaned_pgrp(pgrp, NULL) &&
+                   has_stopped_jobs(pgrp)) {
+                       __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
+                       __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
                }
        }
 }
@@ -655,10 +664,11 @@ static void reparent_thread(task_t *p, task_t *father, int traced)
  * When we die, we re-parent all our children.
  * Try to give them to another thread in our thread
  * group, and if no such member exists, give it to
- * the global child reaper process (ie "init")
+ * the child reaper process (ie "init") in our pid
+ * space.
  */
-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;
@@ -666,7 +676,7 @@ static void forget_original_parent(struct task_struct * father,
        do {
                reaper = next_thread(reaper);
                if (reaper == father) {
-                       reaper = child_reaper;
+                       reaper = child_reaper(father);
                        break;
                }
        } while (reaper->exit_state);
@@ -681,7 +691,7 @@ 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;
 
@@ -710,7 +720,7 @@ 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);
        }
@@ -725,6 +735,7 @@ static void exit_notify(struct task_struct *tsk)
        int state;
        struct task_struct *t;
        struct list_head ptrace_dead, *_p, *_n;
+       struct pid *pgrp;
 
        if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
            && !thread_group_empty(tsk)) {
@@ -777,12 +788,13 @@ static void exit_notify(struct task_struct *tsk)
         
        t = tsk->real_parent;
        
-       if ((process_group(t) != process_group(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, SEND_SIG_PRIV, process_group(tsk));
-               __kill_pg_info(SIGCONT, SEND_SIG_PRIV, process_group(tsk));
+       pgrp = task_pgrp(tsk);
+       if ((task_pgrp(t) != pgrp) &&
+           (task_session(t) != task_session(tsk)) &&
+           will_become_orphaned_pgrp(pgrp, tsk) &&
+           has_stopped_jobs(pgrp)) {
+               __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
+               __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
        }
 
        /* Let father know we died 
@@ -830,7 +842,7 @@ 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);
        }
 
@@ -852,8 +864,13 @@ 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))
-               panic("Attempted to kill init!");
+       if (unlikely(tsk == child_reaper(tsk))) {
+               if (tsk->nsproxy->pid_ns != &init_pid_ns)
+                       tsk->nsproxy->pid_ns->child_reaper = init_pid_ns.child_reaper;
+               else
+                       panic("Attempted to kill init!");
+       }
+
 
        if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
                current->ptrace_message = code;
@@ -899,6 +916,9 @@ fastcall NORET_TYPE void do_exit(long code)
 #endif
        if (unlikely(tsk->audit_context))
                audit_free(tsk);
+
+       taskstats_exit(tsk, group_dead);
+
        exit_mm(tsk);
 
        if (group_dead)
@@ -906,7 +926,6 @@ fastcall NORET_TYPE void do_exit(long code)
        exit_sem(tsk);
        __exit_files(tsk);
        __exit_fs(tsk);
-       exit_namespace(tsk);
        exit_thread();
        cpuset_exit(tsk);
        exit_keys(tsk);
@@ -920,6 +939,7 @@ fastcall NORET_TYPE void do_exit(long code)
 
        tsk->exit_code = code;
        proc_exit_connector(tsk);
+       exit_task_namespaces(tsk);
        exit_notify(tsk);
 #ifdef CONFIG_NUMA
        mpol_free(tsk->mempolicy);
@@ -934,10 +954,9 @@ fastcall NORET_TYPE void do_exit(long code)
        if (unlikely(current->pi_state_cache))
                kfree(current->pi_state_cache);
        /*
-        * If DEBUG_MUTEXES is on, make sure we are holding no locks:
+        * Make sure we are holding no locks:
         */
-       mutex_debug_check_no_locks_held(tsk);
-       rt_mutex_debug_check_no_locks_held(tsk);
+       debug_check_no_locks_held(tsk);
 
        if (tsk->io_context)
                exit_io_context();
@@ -945,15 +964,15 @@ fastcall NORET_TYPE void do_exit(long code)
        if (tsk->splice_pipe)
                __free_pipe_info(tsk->splice_pipe);
 
-       /* PF_DEAD causes final put_task_struct after we schedule. */
        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);
@@ -962,7 +981,7 @@ NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 {
        if (comp)
                complete(comp);
-       
+
        do_exit(code);
 }
 
@@ -1012,7 +1031,7 @@ 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)
@@ -1044,7 +1063,7 @@ 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))
@@ -1053,12 +1072,13 @@ 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);
@@ -1083,7 +1103,7 @@ 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)
 {
@@ -1245,8 +1265,8 @@ 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;
@@ -1360,7 +1380,7 @@ 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)
 {
@@ -1446,7 +1466,7 @@ 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)