ceph: document locking for ceph_set_dentry_offset
[linux-2.6.git] / fs / ext3 / dir.c
index 8528698..34f0a07 100644 (file)
@@ -42,14 +42,12 @@ const struct file_operations ext3_dir_operations = {
        .llseek         = generic_file_llseek,
        .read           = generic_read_dir,
        .readdir        = ext3_readdir,         /* we take BKL. needed?*/
-       .ioctl          = ext3_ioctl,           /* BKL held */
+       .unlocked_ioctl = ext3_ioctl,
 #ifdef CONFIG_COMPAT
        .compat_ioctl   = ext3_compat_ioctl,
 #endif
        .fsync          = ext3_sync_file,       /* BKL held */
-#ifdef CONFIG_EXT3_INDEX
        .release        = ext3_release_dir,
-#endif
 };
 
 
@@ -69,27 +67,28 @@ int ext3_check_dir_entry (const char * function, struct inode * dir,
                          unsigned long offset)
 {
        const char * error_msg = NULL;
-       const int rlen = le16_to_cpu(de->rec_len);
+       const int rlen = ext3_rec_len_from_disk(de->rec_len);
 
-       if (rlen < EXT3_DIR_REC_LEN(1))
+       if (unlikely(rlen < EXT3_DIR_REC_LEN(1)))
                error_msg = "rec_len is smaller than minimal";
-       else if (rlen % 4 != 0)
+       else if (unlikely(rlen % 4 != 0))
                error_msg = "rec_len % 4 != 0";
-       else if (rlen < EXT3_DIR_REC_LEN(de->name_len))
+       else if (unlikely(rlen < EXT3_DIR_REC_LEN(de->name_len)))
                error_msg = "rec_len is too small for name_len";
-       else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
+       else if (unlikely((((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)))
                error_msg = "directory entry across blocks";
-       else if (le32_to_cpu(de->inode) >
-                       le32_to_cpu(EXT3_SB(dir->i_sb)->s_es->s_inodes_count))
+       else if (unlikely(le32_to_cpu(de->inode) >
+                       le32_to_cpu(EXT3_SB(dir->i_sb)->s_es->s_inodes_count)))
                error_msg = "inode out of bounds";
 
-       if (error_msg != NULL)
+       if (unlikely(error_msg != NULL))
                ext3_error (dir->i_sb, function,
                        "bad entry in directory #%lu: %s - "
                        "offset=%lu, inode=%lu, rec_len=%d, name_len=%d",
                        dir->i_ino, error_msg, offset,
                        (unsigned long) le32_to_cpu(de->inode),
                        rlen, de->name_len);
+
        return error_msg == NULL ? 1 : 0;
 }
 
@@ -104,10 +103,10 @@ static int ext3_readdir(struct file * filp,
        int err;
        struct inode *inode = filp->f_path.dentry->d_inode;
        int ret = 0;
+       int dir_has_error = 0;
 
        sb = inode->i_sb;
 
-#ifdef CONFIG_EXT3_INDEX
        if (EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
                                    EXT3_FEATURE_COMPAT_DIR_INDEX) &&
            ((EXT3_I(inode)->i_flags & EXT3_INDEX_FL) ||
@@ -123,7 +122,6 @@ static int ext3_readdir(struct file * filp,
                 */
                EXT3_I(filp->f_path.dentry->d_inode)->i_flags &= ~EXT3_INDEX_FL;
        }
-#endif
        stored = 0;
        offset = filp->f_pos & (sb->s_blocksize - 1);
 
@@ -133,15 +131,16 @@ static int ext3_readdir(struct file * filp,
                struct buffer_head *bh = NULL;
 
                map_bh.b_state = 0;
-               err = ext3_get_blocks_handle(NULL, inode, blk, 1,
-                                               &map_bh, 0, 0);
+               err = ext3_get_blocks_handle(NULL, inode, blk, 1, &map_bh, 0);
                if (err > 0) {
-                       page_cache_readahead(sb->s_bdev->bd_inode->i_mapping,
-                               &filp->f_ra,
-                               filp,
-                               map_bh.b_blocknr >>
-                                       (PAGE_CACHE_SHIFT - inode->i_blkbits),
-                               1);
+                       pgoff_t index = map_bh.b_blocknr >>
+                                       (PAGE_CACHE_SHIFT - inode->i_blkbits);
+                       if (!ra_has_index(&filp->f_ra, index))
+                               page_cache_sync_readahead(
+                                       sb->s_bdev->bd_inode->i_mapping,
+                                       &filp->f_ra, filp,
+                                       index, 1);
+                       filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
                        bh = ext3_bread(NULL, inode, blk, 0, &err);
                }
 
@@ -150,9 +149,12 @@ static int ext3_readdir(struct file * filp,
                 * of recovering data when there's a bad sector
                 */
                if (!bh) {
-                       ext3_error (sb, "ext3_readdir",
-                               "directory #%lu contains a hole at offset %lu",
-                               inode->i_ino, (unsigned long)filp->f_pos);
+                       if (!dir_has_error) {
+                               ext3_error(sb, __func__, "directory #%lu "
+                                       "contains a hole at offset %lld",
+                                       inode->i_ino, filp->f_pos);
+                               dir_has_error = 1;
+                       }
                        /* corrupt size?  Maybe no more blocks to read */
                        if (filp->f_pos > inode->i_blocks << 9)
                                break;
@@ -175,10 +177,10 @@ revalidate:
                                 * least that it is non-zero.  A
                                 * failure will be detected in the
                                 * dirent test below. */
-                               if (le16_to_cpu(de->rec_len) <
+                               if (ext3_rec_len_from_disk(de->rec_len) <
                                                EXT3_DIR_REC_LEN(1))
                                        break;
-                               i += le16_to_cpu(de->rec_len);
+                               i += ext3_rec_len_from_disk(de->rec_len);
                        }
                        offset = i;
                        filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
@@ -199,7 +201,7 @@ revalidate:
                                ret = stored;
                                goto out;
                        }
-                       offset += le16_to_cpu(de->rec_len);
+                       offset += ext3_rec_len_from_disk(de->rec_len);
                        if (le32_to_cpu(de->inode)) {
                                /* We might block in the next section
                                 * if the data destination is
@@ -208,7 +210,7 @@ revalidate:
                                 * not the directory has been modified
                                 * during the copy operation.
                                 */
-                               unsigned long version = filp->f_version;
+                               u64 version = filp->f_version;
 
                                error = filldir(dirent, de->name,
                                                de->name_len,
@@ -221,7 +223,7 @@ revalidate:
                                        goto revalidate;
                                stored ++;
                        }
-                       filp->f_pos += le16_to_cpu(de->rec_len);
+                       filp->f_pos += ext3_rec_len_from_disk(de->rec_len);
                }
                offset = 0;
                brelse (bh);
@@ -230,7 +232,6 @@ out:
        return ret;
 }
 
-#ifdef CONFIG_EXT3_INDEX
 /*
  * These functions convert from the major/minor hash to an f_pos
  * value.
@@ -275,7 +276,7 @@ static void free_rb_tree_fname(struct rb_root *root)
 
        while (n) {
                /* Do the node's children first */
-               if ((n)->rb_left) {
+               if (n->rb_left) {
                        n = n->rb_left;
                        continue;
                }
@@ -297,31 +298,25 @@ static void free_rb_tree_fname(struct rb_root *root)
                        kfree (old);
                }
                if (!parent)
-                       root->rb_node = NULL;
+                       *root = RB_ROOT;
                else if (parent->rb_left == n)
                        parent->rb_left = NULL;
                else if (parent->rb_right == n)
                        parent->rb_right = NULL;
                n = parent;
        }
-       root->rb_node = NULL;
 }
 
 
-static struct dir_private_info *create_dir_info(loff_t pos)
+static struct dir_private_info *ext3_htree_create_dir_info(loff_t pos)
 {
        struct dir_private_info *p;
 
-       p = kmalloc(sizeof(struct dir_private_info), GFP_KERNEL);
+       p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
        if (!p)
                return NULL;
-       p->root.rb_node = NULL;
-       p->curr_node = NULL;
-       p->extra_fname = NULL;
-       p->last_pos = 0;
        p->curr_hash = pos2maj_hash(pos);
        p->curr_minor_hash = pos2min_hash(pos);
-       p->next_hash = 0;
        return p;
 }
 
@@ -419,7 +414,7 @@ static int call_filldir(struct file * filp, void * dirent,
                                get_dtype(sb, fname->file_type));
                if (error) {
                        filp->f_pos = curr_pos;
-                       info->extra_fname = fname->next;
+                       info->extra_fname = fname;
                        return error;
                }
                fname = fname->next;
@@ -436,7 +431,7 @@ static int ext3_dx_readdir(struct file * filp,
        int     ret;
 
        if (!info) {
-               info = create_dir_info(filp->f_pos);
+               info = ext3_htree_create_dir_info(filp->f_pos);
                if (!info)
                        return -ENOMEM;
                filp->private_data = info;
@@ -458,11 +453,12 @@ static int ext3_dx_readdir(struct file * filp,
         * If there are any leftover names on the hash collision
         * chain, return them first.
         */
-       if (info->extra_fname &&
-           call_filldir(filp, dirent, filldir, info->extra_fname))
-               goto finished;
-
-       if (!info->curr_node)
+       if (info->extra_fname) {
+               if (call_filldir(filp, dirent, filldir, info->extra_fname))
+                       goto finished;
+               info->extra_fname = NULL;
+               goto next_node;
+       } else if (!info->curr_node)
                info->curr_node = rb_first(&info->root);
 
        while (1) {
@@ -493,9 +489,14 @@ static int ext3_dx_readdir(struct file * filp,
                info->curr_minor_hash = fname->minor_hash;
                if (call_filldir(filp, dirent, filldir, fname))
                        break;
-
+       next_node:
                info->curr_node = rb_next(info->curr_node);
-               if (!info->curr_node) {
+               if (info->curr_node) {
+                       fname = rb_entry(info->curr_node, struct fname,
+                                        rb_hash);
+                       info->curr_hash = fname->hash;
+                       info->curr_minor_hash = fname->minor_hash;
+               } else {
                        if (info->next_hash == ~0) {
                                filp->f_pos = EXT3_HTREE_EOF;
                                break;
@@ -516,5 +517,3 @@ static int ext3_release_dir (struct inode * inode, struct file * filp)
 
        return 0;
 }
-
-#endif