[PATCH] "ext[34]: EA block reference count racing fix" performance fix

[mv-sheeva.git] / fs / ext3 / namei.c

diff --git a/fs/ext3/namei.c b/fs/ext3/namei.c
index 235e77b52ea59b2610c2ed5cf4af464f4cd12359..49159f13cc1f900d4b514a265ea3e97cbaf811a4 100644 (file)
--- a/fs/ext3/namei.c
+++ b/fs/ext3/namei.c
@@ -552,6 +552,15 @@ static int htree_dirblock_to_tree(struct file *dir_file,
                                           dir->i_sb->s_blocksize -
                                           EXT3_DIR_REC_LEN(0));
        for (; de < top; de = ext3_next_entry(de)) {
+               if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
+                                       (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
+                                               +((char *)de - bh->b_data))) {
+                       /* On error, skip the f_pos to the next block. */
+                       dir_file->f_pos = (dir_file->f_pos |
+                                       (dir->i_sb->s_blocksize - 1)) + 1;
+                       brelse (bh);
+                       return count;
+               }
                ext3fs_dirhash(de->name, de->name_len, hinfo);
                if ((hinfo->hash < start_hash) ||
                    ((hinfo->hash == start_hash) &&
@@ -593,7 +602,7 @@ int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
 
        dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
                       start_minor_hash));
-       dir = dir_file->f_dentry->d_inode;
+       dir = dir_file->f_path.dentry->d_inode;
        if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
                hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
                hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
@@ -604,7 +613,7 @@ int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
        }
        hinfo.hash = start_hash;
        hinfo.minor_hash = 0;
-       frame = dx_probe(NULL, dir_file->f_dentry->d_inode, &hinfo, frames, &err);
+       frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
        if (!frame)
                return err;
 
@@ -1609,21 +1618,6 @@ static int ext3_delete_entry (handle_t *handle,
        return -ENOENT;
 }
 
-/*
- * ext3_mark_inode_dirty is somewhat expensive, so unlike ext2 we
- * do not perform it in these functions.  We perform it at the call site,
- * if it is needed.
- */
-static inline void ext3_inc_count(handle_t *handle, struct inode *inode)
-{
-       inode->i_nlink++;
-}
-
-static inline void ext3_dec_count(handle_t *handle, struct inode *inode)
-{
-       inode->i_nlink--;
-}
-
 static int ext3_add_nondir(handle_t *handle,
                struct dentry *dentry, struct inode *inode)
 {
@@ -1633,7 +1627,7 @@ static int ext3_add_nondir(handle_t *handle,
                d_instantiate(dentry, inode);
                return 0;
        }
-       ext3_dec_count(handle, inode);
+       drop_nlink(inode);
        iput(inode);
        return err;
 }
@@ -1743,7 +1737,7 @@ retry:
        inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
        dir_block = ext3_bread (handle, inode, 0, 1, &err);
        if (!dir_block) {
-               inode->i_nlink--; /* is this nlink == 0? */
+               drop_nlink(inode); /* is this nlink == 0? */
                ext3_mark_inode_dirty(handle, inode);
                iput (inode);
                goto out_stop;
@@ -1775,7 +1769,7 @@ retry:
                iput (inode);
                goto out_stop;
        }
-       dir->i_nlink++;
+       inc_nlink(dir);
        ext3_update_dx_flag(dir);
        ext3_mark_inode_dirty(handle, dir);
        d_instantiate(dentry, inode);
@@ -2045,7 +2039,7 @@ static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
                              "empty directory has nlink!=2 (%d)",
                              inode->i_nlink);
        inode->i_version++;
-       inode->i_nlink = 0;
+       clear_nlink(inode);
        /* There's no need to set i_disksize: the fact that i_nlink is
         * zero will ensure that the right thing happens during any
         * recovery. */
@@ -2053,7 +2047,7 @@ static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
        ext3_orphan_add(handle, inode);
        inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
        ext3_mark_inode_dirty(handle, inode);
-       dir->i_nlink--;
+       drop_nlink(dir);
        ext3_update_dx_flag(dir);
        ext3_mark_inode_dirty(handle, dir);
 
@@ -2104,7 +2098,7 @@ static int ext3_unlink(struct inode * dir, struct dentry *dentry)
        dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
        ext3_update_dx_flag(dir);
        ext3_mark_inode_dirty(handle, dir);
-       inode->i_nlink--;
+       drop_nlink(inode);
        if (!inode->i_nlink)
                ext3_orphan_add(handle, inode);
        inode->i_ctime = dir->i_ctime;
@@ -2154,7 +2148,7 @@ retry:
                err = __page_symlink(inode, symname, l,
                                mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
                if (err) {
-                       ext3_dec_count(handle, inode);
+                       drop_nlink(inode);
                        ext3_mark_inode_dirty(handle, inode);
                        iput (inode);
                        goto out_stop;
@@ -2182,6 +2176,12 @@ static int ext3_link (struct dentry * old_dentry,
 
        if (inode->i_nlink >= EXT3_LINK_MAX)
                return -EMLINK;
+       /*
+        * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
+        * otherwise has the potential to corrupt the orphan inode list.
+        */
+       if (inode->i_nlink == 0)
+               return -ENOENT;
 
 retry:
        handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
@@ -2193,7 +2193,7 @@ retry:
                handle->h_sync = 1;
 
        inode->i_ctime = CURRENT_TIME_SEC;
-       ext3_inc_count(handle, inode);
+       inc_nlink(inode);
        atomic_inc(&inode->i_count);
 
        err = ext3_add_nondir(handle, dentry, inode);
@@ -2326,7 +2326,7 @@ static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
        }
 
        if (new_inode) {
-               new_inode->i_nlink--;
+               drop_nlink(new_inode);
                new_inode->i_ctime = CURRENT_TIME_SEC;
        }
        old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
@@ -2337,11 +2337,11 @@ static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
                PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
                BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
                ext3_journal_dirty_metadata(handle, dir_bh);
-               old_dir->i_nlink--;
+               drop_nlink(old_dir);
                if (new_inode) {
-                       new_inode->i_nlink--;
+                       drop_nlink(new_inode);
                } else {
-                       new_dir->i_nlink++;
+                       inc_nlink(new_dir);
                        ext3_update_dx_flag(new_dir);
                        ext3_mark_inode_dirty(handle, new_dir);
                }
@@ -2365,7 +2365,7 @@ end_rename:
 /*
  * directories can handle most operations...
  */
-struct inode_operations ext3_dir_inode_operations = {
+const struct inode_operations ext3_dir_inode_operations = {
        .create         = ext3_create,
        .lookup         = ext3_lookup,
        .link           = ext3_link,
@@ -2385,7 +2385,7 @@ struct inode_operations ext3_dir_inode_operations = {
        .permission     = ext3_permission,
 };
 
-struct inode_operations ext3_special_inode_operations = {
+const struct inode_operations ext3_special_inode_operations = {
        .setattr        = ext3_setattr,
 #ifdef CONFIG_EXT3_FS_XATTR
        .setxattr       = generic_setxattr,