[PATCH] mm: rmap optimisation
[linux-2.6.git] / fs / exec.c
index 9bb55c8..e75a954 100644 (file)
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -48,6 +48,7 @@
 #include <linux/syscalls.h>
 #include <linux/rmap.h>
 #include <linux/acct.h>
+#include <linux/cn_proc.h>
 
 #include <asm/uaccess.h>
 #include <asm/mmu_context.h>
@@ -134,7 +135,7 @@ asmlinkage long sys_uselib(const char __user * library)
        if (!S_ISREG(nd.dentry->d_inode->i_mode))
                goto exit;
 
-       error = permission(nd.dentry->d_inode, MAY_READ | MAY_EXEC, &nd);
+       error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
        if (error)
                goto exit;
 
@@ -305,44 +306,30 @@ void install_arg_page(struct vm_area_struct *vma,
                        struct page *page, unsigned long address)
 {
        struct mm_struct *mm = vma->vm_mm;
-       pgd_t * pgd;
-       pud_t * pud;
-       pmd_t * pmd;
        pte_t * pte;
+       spinlock_t *ptl;
 
        if (unlikely(anon_vma_prepare(vma)))
-               goto out_sig;
+               goto out;
 
        flush_dcache_page(page);
-       pgd = pgd_offset(mm, address);
-
-       spin_lock(&mm->page_table_lock);
-       pud = pud_alloc(mm, pgd, address);
-       if (!pud)
-               goto out;
-       pmd = pmd_alloc(mm, pud, address);
-       if (!pmd)
-               goto out;
-       pte = pte_alloc_map(mm, pmd, address);
+       pte = get_locked_pte(mm, address, &ptl);
        if (!pte)
                goto out;
        if (!pte_none(*pte)) {
-               pte_unmap(pte);
+               pte_unmap_unlock(pte, ptl);
                goto out;
        }
        inc_mm_counter(mm, anon_rss);
        lru_cache_add_active(page);
        set_pte_at(mm, address, pte, pte_mkdirty(pte_mkwrite(mk_pte(
                                        page, vma->vm_page_prot))));
-       page_add_anon_rmap(page, vma, address);
-       pte_unmap(pte);
-       spin_unlock(&mm->page_table_lock);
+       page_add_new_anon_rmap(page, vma, address);
+       pte_unmap_unlock(pte, ptl);
 
        /* no need for flush_tlb */
        return;
 out:
-       spin_unlock(&mm->page_table_lock);
-out_sig:
        __free_page(page);
        force_sig(SIGKILL, current);
 }
@@ -498,7 +485,7 @@ struct file *open_exec(const char *name)
                file = ERR_PTR(-EACCES);
                if (!(nd.mnt->mnt_flags & MNT_NOEXEC) &&
                    S_ISREG(inode->i_mode)) {
-                       int err = permission(inode, MAY_EXEC, &nd);
+                       int err = vfs_permission(&nd, MAY_EXEC);
                        if (!err && !(inode->i_mode & 0111))
                                err = -EACCES;
                        file = ERR_PTR(err);
@@ -593,6 +580,7 @@ static inline int de_thread(struct task_struct *tsk)
        struct signal_struct *sig = tsk->signal;
        struct sighand_struct *newsighand, *oldsighand = tsk->sighand;
        spinlock_t *lock = &oldsighand->siglock;
+       struct task_struct *leader = NULL;
        int count;
 
        /*
@@ -634,10 +622,9 @@ static inline int de_thread(struct task_struct *tsk)
        /*
         * Account for the thread group leader hanging around:
         */
-       count = 2;
-       if (thread_group_leader(current))
-               count = 1;
-       else {
+       count = 1;
+       if (!thread_group_leader(current)) {
+               count = 2;
                /*
                 * The SIGALRM timer survives the exec, but needs to point
                 * at us as the new group leader now.  We have a race with
@@ -646,8 +633,10 @@ static inline int de_thread(struct task_struct *tsk)
                 * before we can safely let the old group leader die.
                 */
                sig->real_timer.data = (unsigned long)current;
+               spin_unlock_irq(lock);
                if (del_timer_sync(&sig->real_timer))
                        add_timer(&sig->real_timer);
+               spin_lock_irq(lock);
        }
        while (atomic_read(&sig->count) > count) {
                sig->group_exit_task = current;
@@ -659,7 +648,6 @@ static inline int de_thread(struct task_struct *tsk)
        }
        sig->group_exit_task = NULL;
        sig->notify_count = 0;
-       sig->real_timer.data = (unsigned long)current;
        spin_unlock_irq(lock);
 
        /*
@@ -668,15 +656,16 @@ static inline int de_thread(struct task_struct *tsk)
         * and to assume its PID:
         */
        if (!thread_group_leader(current)) {
-               struct task_struct *leader = current->group_leader, *parent;
+               struct task_struct *parent;
                struct dentry *proc_dentry1, *proc_dentry2;
-               unsigned long exit_state, ptrace;
+               unsigned long ptrace;
 
                /*
                 * Wait for the thread group leader to be a zombie.
                 * It should already be zombie at this point, most
                 * of the time.
                 */
+               leader = current->group_leader;
                while (leader->exit_state != EXIT_ZOMBIE)
                        yield();
 
@@ -727,16 +716,15 @@ static inline int de_thread(struct task_struct *tsk)
                list_del(&current->tasks);
                list_add_tail(&current->tasks, &init_task.tasks);
                current->exit_signal = SIGCHLD;
-               exit_state = leader->exit_state;
+
+               BUG_ON(leader->exit_state != EXIT_ZOMBIE);
+               leader->exit_state = EXIT_DEAD;
 
                write_unlock_irq(&tasklist_lock);
                spin_unlock(&leader->proc_lock);
                spin_unlock(&current->proc_lock);
                proc_pid_flush(proc_dentry1);
                proc_pid_flush(proc_dentry2);
-
-               BUG_ON(exit_state != EXIT_ZOMBIE);
-               release_task(leader);
         }
 
        /*
@@ -746,8 +734,11 @@ static inline int de_thread(struct task_struct *tsk)
        sig->flags = 0;
 
 no_thread_group:
-       BUG_ON(atomic_read(&sig->count) != 1);
        exit_itimers(sig);
+       if (leader)
+               release_task(leader);
+
+       BUG_ON(atomic_read(&sig->count) != 1);
 
        if (atomic_read(&oldsighand->count) == 1) {
                /*
@@ -895,7 +886,7 @@ int flush_old_exec(struct linux_binprm * bprm)
        flush_thread();
 
        if (bprm->e_uid != current->euid || bprm->e_gid != current->egid || 
-           permission(bprm->file->f_dentry->d_inode,MAY_READ, NULL) ||
+           file_permission(bprm->file, MAY_READ) ||
            (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
                suid_keys(current);
                current->mm->dumpable = suid_dumpable;
@@ -1100,6 +1091,7 @@ int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
                                        fput(bprm->file);
                                bprm->file = NULL;
                                current->did_exec = 1;
+                               proc_exec_connector(current);
                                return retval;
                        }
                        read_lock(&binfmt_lock);
@@ -1421,19 +1413,16 @@ static void zap_threads (struct mm_struct *mm)
 static void coredump_wait(struct mm_struct *mm)
 {
        DECLARE_COMPLETION(startup_done);
+       int core_waiters;
 
-       mm->core_waiters++; /* let other threads block */
        mm->core_startup_done = &startup_done;
 
-       /* give other threads a chance to run: */
-       yield();
-
        zap_threads(mm);
-       if (--mm->core_waiters) {
-               up_write(&mm->mmap_sem);
+       core_waiters = mm->core_waiters;
+       up_write(&mm->mmap_sem);
+
+       if (core_waiters)
                wait_for_completion(&startup_done);
-       } else
-               up_write(&mm->mmap_sem);
        BUG_ON(mm->core_waiters);
 }
 
@@ -1467,11 +1456,21 @@ int do_coredump(long signr, int exit_code, struct pt_regs * regs)
                current->fsuid = 0;     /* Dump root private */
        }
        mm->dumpable = 0;
-       init_completion(&mm->core_done);
+
+       retval = -EAGAIN;
        spin_lock_irq(&current->sighand->siglock);
-       current->signal->flags = SIGNAL_GROUP_EXIT;
-       current->signal->group_exit_code = exit_code;
+       if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
+               current->signal->flags = SIGNAL_GROUP_EXIT;
+               current->signal->group_exit_code = exit_code;
+               retval = 0;
+       }
        spin_unlock_irq(&current->sighand->siglock);
+       if (retval) {
+               up_write(&mm->mmap_sem);
+               goto fail;
+       }
+
+       init_completion(&mm->core_done);
        coredump_wait(mm);
 
        /*
@@ -1506,7 +1505,7 @@ int do_coredump(long signr, int exit_code, struct pt_regs * regs)
                goto close_fail;
        if (!file->f_op->write)
                goto close_fail;
-       if (do_truncate(file->f_dentry, 0) != 0)
+       if (do_truncate(file->f_dentry, 0, file) != 0)
                goto close_fail;
 
        retval = binfmt->core_dump(signr, regs, file);