and changed files
[powerpc.git] / kernel / exit.c
index 6267a6c..b0c6f0c 100644 (file)
@@ -7,13 +7,12 @@
 #include <linux/mm.h>
 #include <linux/slab.h>
 #include <linux/interrupt.h>
-#include <linux/smp_lock.h>
 #include <linux/module.h>
 #include <linux/capability.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/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/kthread.h>
 #include <linux/mempolicy.h>
 #include <linux/taskstats_kern.h>
 #include <linux/delayacct.h>
@@ -48,7 +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);
 
@@ -185,25 +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 (process_session(p) > 0) {
-                       sid = process_session(p);
-                       goto out;
-               }
-       } while_each_task_pid(pgrp, PIDTYPE_PGID, p);
-       p = find_task_by_pid(pgrp);
-       if (p)
-               sid = process_session(p);
-out:
-       read_unlock(&tasklist_lock);
-       
        return sid;
 }
 
@@ -215,71 +210,70 @@ 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(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
                                || is_init(p->real_parent))
                        continue;
-               if (process_group(p->real_parent) != pgrp &&
-                   process_session(p->real_parent) == process_session(p)) {
+               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;
                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_kthreadd - Reparent the calling kernel thread to kthreadd
  *
  * 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
- * it is correctly cleaned up on exit.
+ * it ever exits, it should generally reparent itself to kthreadd so it
+ * isn't in the way of other processes and is correctly cleaned up on exit.
  *
  * The various task state such as scheduling policy and priority may have
  * been inherited from a user process, so we reset them to sane values here.
  *
- * NOTE that reparent_to_init() gives the caller full capabilities.
+ * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
  */
-static void reparent_to_init(void)
+static void reparent_to_kthreadd(void)
 {
        write_lock_irq(&tasklist_lock);
 
        ptrace_unlink(current);
        /* Reparent to init */
        remove_parent(current);
-       current->parent = child_reaper;
-       current->real_parent = child_reaper;
+       current->real_parent = current->parent = kthreadd_task;
        add_parent(current);
 
        /* Set the exit signal to SIGCHLD so we signal init on exit */
@@ -353,7 +347,7 @@ int disallow_signal(int sig)
                return -EINVAL;
 
        spin_lock_irq(&current->sighand->siglock);
-       sigaddset(&current->blocked, sig);
+       current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
        recalc_sigpending();
        spin_unlock_irq(&current->sighand->siglock);
        return 0;
@@ -406,7 +400,7 @@ void daemonize(const char *name, ...)
        current->files = init_task.files;
        atomic_inc(&current->files->count);
 
-       reparent_to_init();
+       reparent_to_kthreadd();
 }
 
 EXPORT_SYMBOL(daemonize);
@@ -427,14 +421,16 @@ 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;
@@ -468,9 +464,7 @@ 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);
        }
@@ -601,10 +595,6 @@ choose_new_parent(struct task_struct *p, struct task_struct *reaper)
 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);
@@ -624,13 +614,7 @@ reparent_thread(struct task_struct *p, struct task_struct *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
@@ -640,20 +624,37 @@ reparent_thread(struct task_struct *p, struct task_struct *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)) &&
-           (process_session(p) == process_session(father))) {
-               int pgrp = process_group(p);
+       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_pg_info(SIGHUP, SEND_SIG_PRIV, pgrp);
-                       __kill_pg_info(SIGCONT, SEND_SIG_PRIV, pgrp);
+                       __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
+                       __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
                }
        }
 }
@@ -662,7 +663,8 @@ reparent_thread(struct task_struct *p, struct task_struct *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)
@@ -673,7 +675,7 @@ forget_original_parent(struct task_struct *father, struct list_head *to_release)
        do {
                reaper = next_thread(reaper);
                if (reaper == father) {
-                       reaper = child_reaper;
+                       reaper = child_reaper(father);
                        break;
                }
        } while (reaper->exit_state);
@@ -732,6 +734,7 @@ 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)) {
@@ -784,12 +787,13 @@ static void exit_notify(struct task_struct *tsk)
         
        t = tsk->real_parent;
        
-       if ((process_group(t) != process_group(tsk)) &&
-           (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));
-               __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 
@@ -859,8 +863,13 @@ 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;
@@ -929,8 +938,8 @@ fastcall NORET_TYPE void do_exit(long code)
 
        tsk->exit_code = code;
        proc_exit_connector(tsk);
-       exit_notify(tsk);
        exit_task_namespaces(tsk);
+       exit_notify(tsk);
 #ifdef CONFIG_NUMA
        mpol_free(tsk->mempolicy);
        tsk->mempolicy = NULL;
@@ -1023,6 +1032,8 @@ asmlinkage void sys_exit_group(int error_code)
 
 static int eligible_child(pid_t pid, int options, struct task_struct *p)
 {
+       int err;
+
        if (pid > 0) {
                if (p->pid != pid)
                        return 0;
@@ -1056,8 +1067,9 @@ static int eligible_child(pid_t pid, int options, struct task_struct *p)
        if (delay_group_leader(p))
                return 2;
 
-       if (security_task_wait(p))
-               return 0;
+       err = security_task_wait(p);
+       if (err)
+               return err;
 
        return 1;
 }
@@ -1439,6 +1451,7 @@ static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
        DECLARE_WAITQUEUE(wait, current);
        struct task_struct *tsk;
        int flag, retval;
+       int allowed, denied;
 
        add_wait_queue(&current->signal->wait_chldexit,&wait);
 repeat:
@@ -1447,6 +1460,7 @@ repeat:
         * match our criteria, even if we are not able to reap it yet.
         */
        flag = 0;
+       allowed = denied = 0;
        current->state = TASK_INTERRUPTIBLE;
        read_lock(&tasklist_lock);
        tsk = current;
@@ -1462,6 +1476,12 @@ repeat:
                        if (!ret)
                                continue;
 
+                       if (unlikely(ret < 0)) {
+                               denied = ret;
+                               continue;
+                       }
+                       allowed = 1;
+
                        switch (p->state) {
                        case TASK_TRACED:
                                /*
@@ -1560,6 +1580,8 @@ check_continued:
                goto repeat;
        }
        retval = -ECHILD;
+       if (unlikely(denied) && !allowed)
+               retval = denied;
 end:
        current->state = TASK_RUNNING;
        remove_wait_queue(&current->signal->wait_chldexit,&wait);