fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / kernel / exit.c
index 80530fb..2961702 100644 (file)
 #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/ptrace.h>
+#include <linux/nsproxy.h>
 #include <linux/tracehook.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/pipe_fs_i.h>
 #include <linux/audit.h> /* for audit_free() */
 #include <linux/resource.h>
+#include <linux/blkdev.h>
 #include <linux/vs_limit.h>
 #include <linux/vs_context.h>
 #include <linux/vs_network.h>
+#include <linux/vs_pid.h>
+#include <linux/vserver/global.h>
 
 #include <asm/uaccess.h>
 #include <asm/unistd.h>
@@ -49,7 +55,6 @@
 #include <asm/mmu_context.h>
 
 extern void sem_exit (void);
-extern struct task_struct *child_reaper;
 
 static void exit_mm(struct task_struct * tsk);
 
@@ -129,6 +134,7 @@ 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);
        }
 }
@@ -142,6 +148,7 @@ void release_task(struct task_struct * p)
 {
        struct task_struct *leader;
        int zap_leader;
+       int inhibited_leader;
 repeat:
        tracehook_release_task(p);
        atomic_dec(&p->user->processes);
@@ -155,10 +162,14 @@ repeat:
         * group leader's parent process. (if it wants notification.)
         */
        zap_leader = 0;
+       inhibited_leader = 0;
        leader = p->group_leader;
        if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
                BUG_ON(leader->exit_signal == -1);
-               do_notify_parent(leader, leader->exit_signal);
+               if (tracehook_inhibit_wait_zombie(leader))
+                       inhibited_leader = 1;
+               else
+                       do_notify_parent(leader, leader->exit_signal);
                /*
                 * If we were the last child thread and the leader has
                 * exited already, and the leader's parent ignores SIGCHLD,
@@ -179,6 +190,13 @@ repeat:
        p = leader;
        if (unlikely(zap_leader))
                goto repeat;
+
+       /*
+        * If tracing usurps normal reaping of the leader, tracing needs
+        * to be notified it would normally be reapable now.
+        */
+       if (unlikely(inhibited_leader))
+               tracehook_report_delayed_group_leader(leader);
 }
 
 /*
@@ -189,21 +207,18 @@ repeat:
 int session_of_pgrp(int pgrp)
 {
        struct task_struct *p;
-       int sid = -1;
+       int sid = 0;
 
        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:
+
+       p = find_task_by_pid_type(PIDTYPE_PGID, pgrp);
+       if (p == NULL)
+               p = find_task_by_pid(pgrp);
+       if (p != NULL)
+               sid = process_session(p);
+
        read_unlock(&tasklist_lock);
-       
+
        return sid;
 }
 
@@ -223,10 +238,10 @@ static int will_become_orphaned_pgrp(int pgrp, struct task_struct *ignored_task)
        do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
                if (p == ignored_task
                                || p->exit_state
-                               || p->parent->pid == 1)
+                               || is_init(p->parent))
                        continue;
-               if (process_group(p->parent) != pgrp
-                           && p->parent->signal->session == p->signal->session) {
+               if (process_group(p->parent) != pgrp &&
+                   process_session(p->parent) == process_session(p)) {
                        ret = 0;
                        break;
                }
@@ -253,7 +268,6 @@ static int has_stopped_jobs(int pgrp)
        do_each_task_pid(pgrp, PIDTYPE_PGID, p) {
                if (p->state != TASK_STOPPED)
                        continue;
-
                retval = 1;
                break;
        } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
@@ -261,7 +275,8 @@ static 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
+ * 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
@@ -278,15 +293,13 @@ static void reparent_to_init(void)
 
        /* Reparent to init */
        remove_parent(current);
-       current->parent = child_reaper;
+       current->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? */
@@ -303,9 +316,9 @@ 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) {
@@ -315,7 +328,7 @@ 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);
@@ -398,9 +411,11 @@ 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);
@@ -426,7 +441,7 @@ 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) {
@@ -469,9 +484,7 @@ 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);
        }
@@ -479,6 +492,18 @@ 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;
@@ -508,6 +533,7 @@ static inline void __put_fs_struct(struct fs_struct *fs)
                        dput(fs->altroot);
                        mntput(fs->altrootmnt);
                }
+               atomic_dec(&vs_global_fs);
                kmem_cache_free(fs_cachep, fs);
        }
 }
@@ -580,10 +606,13 @@ 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),
+       vxwprintk((p->xid != reaper->xid) && (reaper != child_reaper(p)),
                "rogue reaper: %p[%d,#%u] <> %p[%d,#%u]",
                p, p->pid, p->xid, reaper, reaper->pid, reaper->xid);
 
+       if (p == reaper)
+               reaper = vx_child_reaper(p);
+
        /*
         * Make sure we're not reparenting to ourselves and that
         * the parent is not a zombie.
@@ -595,10 +624,6 @@ choose_new_parent(struct task_struct *p, struct task_struct *reaper)
 static void
 reparent_thread(struct task_struct *p, struct task_struct *father)
 {
-       /* 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);
@@ -606,11 +631,22 @@ reparent_thread(struct task_struct *p, struct task_struct *father)
        /* Move the child from its dying parent to the new one.  */
        list_move_tail(&p->sibling, &p->parent->children);
 
+       /* If this is a threaded reparent there is no need to
+        * notify anyone anything has happened.
+        */
+       if (p->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 (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
-           thread_group_empty(p))
+       if (!tracehook_inhibit_wait_zombie(p) &&
+           p->exit_state == EXIT_ZOMBIE &&
+           p->exit_signal != -1 && thread_group_empty(p))
                do_notify_parent(p, p->exit_signal);
 
        /*
@@ -620,10 +656,11 @@ reparent_thread(struct task_struct *p, struct task_struct *father)
         * outside, so the child pgrp is now orphaned.
         */
        if ((process_group(p) != process_group(father)) &&
-           (p->signal->session == father->signal->session)) {
+           (process_session(p) == process_session(father))) {
                int pgrp = process_group(p);
 
-               if (will_become_orphaned_pgrp(pgrp, NULL) && has_stopped_jobs(pgrp)) {
+               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);
                }
@@ -634,7 +671,8 @@ reparent_thread(struct task_struct *p, struct task_struct *father)
  * 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)
@@ -645,6 +683,7 @@ forget_original_parent(struct task_struct *father)
        do {
                reaper = next_thread(reaper);
                if (reaper == father) {
+                       // reaper = child_reaper(father);
                        reaper = vx_child_reaper(father);
                        break;
                }
@@ -652,7 +691,7 @@ forget_original_parent(struct task_struct *father)
 
        list_for_each_safe(_p, _n, &father->children) {
                p = list_entry(_p, struct task_struct, sibling);
-               choose_new_parent(p, vx_child_reaper(p));
+               choose_new_parent(p, reaper);
                reparent_thread(p, father);
        }
 }
@@ -718,7 +757,7 @@ static void exit_notify(struct task_struct *tsk)
        t = tsk->parent;
        
        if ((process_group(t) != process_group(tsk)) &&
-           (t->signal->session == tsk->signal->session) &&
+           (process_session(t) == process_session(tsk)) &&
            will_become_orphaned_pgrp(process_group(tsk), tsk) &&
            has_stopped_jobs(process_group(tsk))) {
                __kill_pg_info(SIGHUP, SEND_SIG_PRIV, process_group(tsk));
@@ -768,9 +807,7 @@ 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);
 
@@ -780,8 +817,13 @@ 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!");
+       }
+
 
        tracehook_report_exit(&code);
 
@@ -807,8 +849,6 @@ 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);
@@ -819,15 +859,6 @@ fastcall NORET_TYPE void do_exit(long code)
                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);
@@ -837,8 +868,8 @@ fastcall NORET_TYPE void do_exit(long code)
 #endif
        if (unlikely(tsk->audit_context))
                audit_free(tsk);
-       taskstats_exit_send(tsk, tidstats, group_dead, mycpu);
-       taskstats_exit_free(tidstats);
+
+       taskstats_exit(tsk, group_dead);
 
        exit_mm(tsk);
 
@@ -847,7 +878,6 @@ 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);
@@ -861,9 +891,12 @@ fastcall NORET_TYPE void do_exit(long code)
 
        tsk->exit_code = code;
        proc_exit_connector(tsk);
+       exit_task_namespaces(tsk);
+       // ns = exit_task_namespaces_early(tsk);
        /* needs to stay before exit_notify() */
        exit_vx_info_early(tsk, code);
        exit_notify(tsk);
+       // exit_task_namespaces(tsk, ns);
 #ifdef CONFIG_NUMA
        mpol_free(tsk->mempolicy);
        tsk->mempolicy = NULL;
@@ -891,15 +924,15 @@ fastcall NORET_TYPE void do_exit(long code)
        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;
+       /* 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);
@@ -908,7 +941,7 @@ NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 {
        if (comp)
                complete(comp);
-       
+
        do_exit(code);
 }