This commit was manufactured by cvs2svn to create tag
[linux-2.6.git] / kernel / exit.c
index 6ec1f96..2e1f0bb 100644 (file)
 #include <linux/mount.h>
 #include <linux/proc_fs.h>
 #include <linux/mempolicy.h>
+#include <linux/ckrm.h>
+#include <linux/ckrm_tsk.h>
+#include <linux/vs_limit.h>
+#include <linux/ckrm_mem.h>
 
 #include <asm/uaccess.h>
 #include <asm/unistd.h>
@@ -38,6 +42,11 @@ int getrusage(struct task_struct *, int, struct rusage __user *);
 static void __unhash_process(struct task_struct *p)
 {
        nr_threads--;
+       /* tasklist_lock is held, is this sufficient? */
+       if (p->vx_info) {
+               atomic_dec(&p->vx_info->cvirt.nr_threads);
+               vx_nproc_dec(p);
+       }
        detach_pid(p, PIDTYPE_PID);
        detach_pid(p, PIDTYPE_TGID);
        if (thread_group_leader(p)) {
@@ -75,7 +84,7 @@ repeat:
         */
        zap_leader = 0;
        leader = p->group_leader;
-       if (leader != p && thread_group_empty(leader) && leader->state == TASK_ZOMBIE) {
+       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);
                /*
@@ -92,7 +101,7 @@ repeat:
        sched_exit(p);
        write_unlock_irq(&tasklist_lock);
        spin_unlock(&p->proc_lock);
-       proc_pid_flush(proc_dentry);
+       dput(proc_dentry);
        release_thread(p);
        put_task_struct(p);
 
@@ -157,7 +166,7 @@ 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->state >= TASK_ZOMBIE 
+                               || p->exit_state >= EXIT_ZOMBIE
                                || p->real_parent->pid == 1)
                        continue;
                if (process_group(p->real_parent) != pgrp
@@ -224,6 +233,7 @@ void reparent_to_init(void)
        ptrace_unlink(current);
        /* Reparent to init */
        REMOVE_LINKS(current);
+       /* FIXME handle vchild_reaper/initpid */
        current->parent = child_reaper;
        current->real_parent = child_reaper;
        SET_LINKS(current);
@@ -366,8 +376,11 @@ static inline void close_files(struct files_struct * files)
                while (set) {
                        if (set & 1) {
                                struct file * file = xchg(&files->fd[i], NULL);
-                               if (file)
+                               if (file) {
                                        filp_close(file, files);
+                                       cond_resched();
+                               }
+                               // vx_openfd_dec(fd);
                        }
                        i++;
                        set >>= 1;
@@ -499,6 +512,12 @@ static inline void __exit_mm(struct task_struct * tsk)
        task_lock(tsk);
        tsk->mm = NULL;
        up_read(&mm->mmap_sem);
+#ifdef CONFIG_CKRM_RES_MEM
+       spin_lock(&mm->peertask_lock);
+       list_del_init(&tsk->mm_peers);
+       ckrm_mem_evaluate_mm(mm);
+       spin_unlock(&mm->peertask_lock);
+#endif
        enter_lazy_tlb(mm, current);
        task_unlock(tsk);
        mmput(mm);
@@ -509,15 +528,13 @@ void exit_mm(struct task_struct *tsk)
        __exit_mm(tsk);
 }
 
-EXPORT_SYMBOL(exit_mm);
-
 static inline void choose_new_parent(task_t *p, task_t *reaper, task_t *child_reaper)
 {
        /*
         * Make sure we're not reparenting to ourselves and that
         * the parent is not a zombie.
         */
-       BUG_ON(p == reaper || reaper->state >= TASK_ZOMBIE);
+       BUG_ON(p == reaper || reaper->state >= EXIT_ZOMBIE || reaper->exit_state >= EXIT_ZOMBIE);
        p->real_parent = reaper;
        if (p->parent == p->real_parent)
                BUG();
@@ -528,7 +545,6 @@ static inline void reparent_thread(task_t *p, task_t *father, int traced)
        /* We don't want people slaying init.  */
        if (p->exit_signal != -1)
                p->exit_signal = SIGCHLD;
-       p->self_exec_id++;
 
        if (p->pdeath_signal)
                /* We already hold the tasklist_lock here.  */
@@ -552,7 +568,7 @@ static inline void reparent_thread(task_t *p, task_t *father, int traced)
                /* If we'd notified the old parent about this child's death,
                 * also notify the new parent.
                 */
-               if (p->state == TASK_ZOMBIE && p->exit_signal != -1 &&
+               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) {
@@ -600,7 +616,7 @@ static inline void forget_original_parent(struct task_struct * father,
                        reaper = child_reaper;
                        break;
                }
-       } while (reaper->state >= TASK_ZOMBIE);
+       } while (reaper->exit_state >= EXIT_ZOMBIE);
 
        /*
         * There are only two places where our children can be:
@@ -626,7 +642,7 @@ static inline void forget_original_parent(struct task_struct * father,
                } else {
                        /* reparent ptraced task to its real parent */
                        __ptrace_unlink (p);
-                       if (p->state == TASK_ZOMBIE && p->exit_signal != -1 &&
+                       if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
                            thread_group_empty(p))
                                do_notify_parent(p, p->exit_signal);
                }
@@ -637,7 +653,7 @@ static inline void forget_original_parent(struct task_struct * father,
                 * zombie forever since we prevented it from self-reap itself
                 * while it was being traced by us, to be able to see it in wait4.
                 */
-               if (unlikely(ptrace && p->state == TASK_ZOMBIE && p->exit_signal == -1))
+               if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
                        list_add(&p->ptrace_list, to_release);
        }
        list_for_each_safe(_p, _n, &father->ptrace_children) {
@@ -657,6 +673,8 @@ static void exit_notify(struct task_struct *tsk)
        struct task_struct *t;
        struct list_head ptrace_dead, *_p, *_n;
 
+       ckrm_cb_exit(tsk);
+
        if (signal_pending(tsk) && !tsk->signal->group_exit
            && !thread_group_empty(tsk)) {
                /*
@@ -750,10 +768,10 @@ static void exit_notify(struct task_struct *tsk)
                do_notify_parent(tsk, SIGCHLD);
        }
 
-       state = TASK_ZOMBIE;
+       state = EXIT_ZOMBIE;
        if (tsk->exit_signal == -1 && tsk->ptrace == 0)
-               state = TASK_DEAD;
-       tsk->state = state;
+               state = EXIT_DEAD;
+       tsk->exit_state = state;
 
        /*
         * Clear these here so that update_process_times() won't try to deliver
@@ -772,7 +790,7 @@ static void exit_notify(struct task_struct *tsk)
        }
 
        /* If the process is dead, release it - nobody will wait for it */
-       if (state == TASK_DEAD)
+       if (state == EXIT_DEAD)
                release_task(tsk);
 
        /* PF_DEAD causes final put_task_struct after we schedule. */
@@ -808,6 +826,13 @@ asmlinkage NORET_TYPE void do_exit(long code)
        }
 
        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();
+       }
        __exit_mm(tsk);
 
        exit_sem(tsk);
@@ -829,6 +854,8 @@ asmlinkage NORET_TYPE void do_exit(long code)
        mpol_free(tsk->mempolicy);
        tsk->mempolicy = NULL;
 #endif
+
+       BUG_ON(!(current->flags & PF_DEAD));
        schedule();
        BUG();
        /* Avoid "noreturn function does return".  */
@@ -972,7 +999,7 @@ static int wait_noreap_copyout(task_t *p, pid_t pid, uid_t uid,
 }
 
 /*
- * Handle sys_wait4 work for one task in state TASK_ZOMBIE.  We hold
+ * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
  * the lock and this task is uninteresting.  If we return nonzero, we have
  * released the lock and the system call should return.
@@ -991,7 +1018,7 @@ static int wait_task_zombie(task_t *p, int noreap,
                int exit_code = p->exit_code;
                int why, status;
 
-               if (unlikely(p->state != TASK_ZOMBIE))
+               if (unlikely(p->exit_state != EXIT_ZOMBIE))
                        return 0;
                if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
                        return 0;
@@ -1012,9 +1039,9 @@ static int wait_task_zombie(task_t *p, int noreap,
         * Try to move the task's state to DEAD
         * only one thread is allowed to do this:
         */
-       state = xchg(&p->state, TASK_DEAD);
-       if (state != TASK_ZOMBIE) {
-               BUG_ON(state != TASK_DEAD);
+       state = xchg(&p->exit_state, EXIT_DEAD);
+       if (state != EXIT_ZOMBIE) {
+               BUG_ON(state != EXIT_DEAD);
                return 0;
        }
        if (unlikely(p->exit_signal == -1 && p->ptrace == 0)) {
@@ -1059,7 +1086,7 @@ static int wait_task_zombie(task_t *p, int noreap,
 
        /*
         * Now we are sure this task is interesting, and no other
-        * thread can reap it because we set its state to TASK_DEAD.
+        * thread can reap it because we set its state to EXIT_DEAD.
         */
        read_unlock(&tasklist_lock);
 
@@ -1091,7 +1118,8 @@ static int wait_task_zombie(task_t *p, int noreap,
        if (!retval && infop)
                retval = put_user(p->uid, &infop->si_uid);
        if (retval) {
-               p->state = TASK_ZOMBIE;
+               // TODO: is this safe?
+               p->exit_state = EXIT_ZOMBIE;
                return retval;
        }
        retval = p->pid;
@@ -1100,7 +1128,8 @@ static int wait_task_zombie(task_t *p, int noreap,
                /* Double-check with lock held.  */
                if (p->real_parent != p->parent) {
                        __ptrace_unlink(p);
-                       p->state = TASK_ZOMBIE;
+                       // TODO: is this safe?
+                       p->exit_state = EXIT_ZOMBIE;
                        /*
                         * If this is not a detached task, notify the parent.
                         * If it's still not detached after that, don't release
@@ -1171,13 +1200,13 @@ static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
        /*
         * This uses xchg to be atomic with the thread resuming and setting
         * it.  It must also be done with the write lock held to prevent a
-        * race with the TASK_ZOMBIE case.
+        * race with the EXIT_ZOMBIE case.
         */
        exit_code = xchg(&p->exit_code, 0);
-       if (unlikely(p->state >= TASK_ZOMBIE)) {
+       if (unlikely(p->exit_state >= EXIT_ZOMBIE)) {
                /*
                 * The task resumed and then died.  Let the next iteration
-                * catch it in TASK_ZOMBIE.  Note that exit_code might
+                * catch it in EXIT_ZOMBIE.  Note that exit_code might
                 * already be zero here if it resumed and did _exit(0).
                 * The task itself is dead and won't touch exit_code again;
                 * other processors in this function are locked out.
@@ -1193,8 +1222,15 @@ static int wait_task_stopped(task_t *p, int delayed_group_leader, int noreap,
                write_unlock_irq(&tasklist_lock);
 bail_ref:
                put_task_struct(p);
-               read_lock(&tasklist_lock);
-               return 0;
+               /*
+                * We are returning to the wait loop without having successfully
+                * removed the process and having released the lock. We cannot
+                * continue, since the "p" task pointer is potentially stale.
+                *
+                * Return -EAGAIN, and do_wait() will restart the loop from the
+                * beginning. Do _not_ re-acquire the lock.
+                */
+               return -EAGAIN;
        }
 
        /* move to end of parent's list to avoid starvation */
@@ -1228,6 +1264,72 @@ bail_ref:
        return retval;
 }
 
+/*
+ * Handle do_wait work for one task in a live, non-stopped state.
+ * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
+ * 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,
+                              struct siginfo __user *infop,
+                              int __user *stat_addr, struct rusage __user *ru)
+{
+       int retval;
+       pid_t pid;
+       uid_t uid;
+
+       if (unlikely(!p->signal))
+               return 0;
+
+       if (p->signal->stop_state >= 0)
+               return 0;
+
+       spin_lock_irq(&p->sighand->siglock);
+       if (p->signal->stop_state >= 0) { /* Re-check with the lock held.  */
+               spin_unlock_irq(&p->sighand->siglock);
+               return 0;
+       }
+       if (!noreap)
+               p->signal->stop_state = 0;
+       spin_unlock_irq(&p->sighand->siglock);
+
+       pid = p->pid;
+       uid = p->uid;
+       get_task_struct(p);
+       read_unlock(&tasklist_lock);
+
+       if (!infop) {
+               retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
+               put_task_struct(p);
+               if (!retval && stat_addr)
+                       retval = put_user(0xffff, stat_addr);
+               if (!retval)
+                       retval = p->pid;
+       } else {
+               retval = wait_noreap_copyout(p, pid, uid,
+                                            CLD_CONTINUED, SIGCONT,
+                                            infop, ru);
+               BUG_ON(retval == 0);
+       }
+
+       return retval;
+}
+
+
+static inline int my_ptrace_child(struct task_struct *p)
+{
+       if (!(p->ptrace & PT_PTRACED))
+               return 0;
+       /*
+        * This child was PTRACE_ATTACH'd.  We should be seeing it only if
+        * we are the attacher.  If we are the real parent, this is a race
+        * inside ptrace_attach.  It is waiting for the tasklist_lock,
+        * which we have to switch the parent links, but has already set
+        * the flags in p->ptrace.
+        */
+       return (p->parent != p->real_parent);
+}
+
 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
                    int __user *stat_addr, struct rusage __user *ru)
 {
@@ -1237,6 +1339,10 @@ static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
 
        add_wait_queue(&current->wait_chldexit,&wait);
 repeat:
+       /*
+        * We will set this flag if we see any child that might later
+        * match our criteria, even if we are not able to reap it yet.
+        */
        flag = 0;
        current->state = TASK_INTERRUPTIBLE;
        read_lock(&tasklist_lock);
@@ -1252,66 +1358,67 @@ repeat:
                        ret = eligible_child(pid, options, p);
                        if (!ret)
                                continue;
-                       flag = 1;
 
                        switch (p->state) {
                        case TASK_TRACED:
-                               if (!(p->ptrace & PT_PTRACED))
+                               if (!my_ptrace_child(p))
                                        continue;
                                /*FALLTHROUGH*/
                        case TASK_STOPPED:
+                               /*
+                                * It's stopped now, so it might later
+                                * continue, exit, or stop again.
+                                */
+                               flag = 1;
                                if (!(options & WUNTRACED) &&
-                                   !(p->ptrace & PT_PTRACED))
+                                   !my_ptrace_child(p))
                                        continue;
                                retval = wait_task_stopped(p, ret == 2,
                                                           (options & WNOWAIT),
                                                           infop,
                                                           stat_addr, ru);
+                               if (retval == -EAGAIN)
+                                       goto repeat;
                                if (retval != 0) /* He released the lock.  */
                                        goto end;
                                break;
-                       case TASK_ZOMBIE:
+                       default:
+                       // case EXIT_DEAD:
+                               if (p->exit_state == EXIT_DEAD)
+                                       continue;
+                       // case EXIT_ZOMBIE:
+                               if (p->exit_state == EXIT_ZOMBIE) {
+                                       /*
+                                        * Eligible but we cannot release
+                                        * it yet:
+                                        */
+                                       if (ret == 2)
+                                               goto check_continued;
+                                       if (!likely(options & WEXITED))
+                                               continue;
+                                       retval = wait_task_zombie(
+                                               p, (options & WNOWAIT),
+                                               infop, stat_addr, ru);
+                                       /* He released the lock.  */
+                                       if (retval != 0)
+                                               goto end;
+                                       break;
+                               }
+check_continued:
                                /*
-                                * Eligible but we cannot release it yet:
+                                * It's running now, so it might later
+                                * exit, stop, or stop and then continue.
                                 */
-                               if (ret == 2)
-                                       goto check_continued;
-                               if (!likely(options & WEXITED))
+                               flag = 1;
+                               if (!unlikely(options & WCONTINUED))
                                        continue;
-                               retval = wait_task_zombie(
+
+                               retval = wait_task_continued(
                                        p, (options & WNOWAIT),
                                        infop, stat_addr, ru);
                                if (retval != 0) /* He released the lock.  */
                                        goto end;
                                break;
-                       case TASK_DEAD:
-                               continue;
-                       default:
-check_continued:
-                               if (!unlikely(options & WCONTINUED))
-                                       continue;
-                               if (unlikely(!p->signal))
-                                       continue;
-                               spin_lock_irq(&p->sighand->siglock);
-                               if (p->signal->stop_state < 0) {
-                                       pid_t pid;
-                                       uid_t uid;
-
-                                       if (!(options & WNOWAIT))
-                                               p->signal->stop_state = 0;
-                                       spin_unlock_irq(&p->sighand->siglock);
-                                       pid = p->pid;
-                                       uid = p->uid;
-                                       get_task_struct(p);
-                                       read_unlock(&tasklist_lock);
-                                       retval = wait_noreap_copyout(p, pid,
-                                                       uid, CLD_CONTINUED,
-                                                       SIGCONT, infop, ru);
-                                       BUG_ON(retval == 0);
-                                       goto end;
-                               }
-                               spin_unlock_irq(&p->sighand->siglock);
-                               break;
                        }
                }
                if (!flag) {
@@ -1412,7 +1519,8 @@ asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
 {
        long ret;
 
-       if (options & ~(WNOHANG|WUNTRACED|__WNOTHREAD|__WCLONE|__WALL))
+       if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
+                       __WNOTHREAD|__WCLONE|__WALL))
                return -EINVAL;
        ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);