[XFS] Return case-insensitive match for dentry cache

This implements the code to store the actual filename found during a
lookup in the dentry cache and to avoid multiple entries in the dcache
pointing to the same inode.

To avoid polluting the dcache, we implement a new directory inode
operations for lookup. xfs_vn_ci_lookup() stores the correct case name in
the dcache.

The "actual name" is only allocated and returned for a case- insensitive
match and not an actual match.

Another unusual interaction with the dcache is not storing negative
dentries like other filesystems doing a d_add(dentry, NULL) when an ENOENT
is returned. During the VFS lookup, if a dentry returned has no inode,
dput is called and ENOENT is returned. By not doing a d_add, this actually
removes it completely from the dcache to be reused. create/rename have to
be modified to support unhashed dentries being passed in.

SGI-PV: 981521
SGI-Modid: xfs-linux-melb:xfs-kern:31208a

Signed-off-by: Barry Naujok <bnaujok@sgi.com>
Signed-off-by: Christoph Hellwig <hch@infradead.org>

fs/xfs/linux-2.6/xfs_export.c

@@ -215,7 +215,7 @@ xfs_fs_get_parent(
 	struct xfs_inode	*cip;
 	struct dentry		*parent;
 
-	error = xfs_lookup(XFS_I(child->d_inode), &xfs_name_dotdot, &cip);
+	error = xfs_lookup(XFS_I(child->d_inode), &xfs_name_dotdot, &cip, NULL);
 	if (unlikely(error))
 		return ERR_PTR(-error);
 

fs/xfs/linux-2.6/xfs_iops.c

@@ -382,7 +382,7 @@ xfs_vn_lookup(
 		return ERR_PTR(-ENAMETOOLONG);
 
 	xfs_dentry_to_name(&name, dentry);
-	error = xfs_lookup(XFS_I(dir), &name, &cip);
+	error = xfs_lookup(XFS_I(dir), &name, &cip, NULL);
 	if (unlikely(error)) {
 		if (unlikely(error != ENOENT))
 			return ERR_PTR(-error);
@@ -393,6 +393,42 @@ xfs_vn_lookup(
 	return d_splice_alias(cip->i_vnode, dentry);
 }
 
+STATIC struct dentry *
+xfs_vn_ci_lookup(
+	struct inode	*dir,
+	struct dentry	*dentry,
+	struct nameidata *nd)
+{
+	struct xfs_inode *ip;
+	struct xfs_name	xname;
+	struct xfs_name ci_name;
+	struct qstr	dname;
+	int		error;
+
+	if (dentry->d_name.len >= MAXNAMELEN)
+		return ERR_PTR(-ENAMETOOLONG);
+
+	xfs_dentry_to_name(&xname, dentry);
+	error = xfs_lookup(XFS_I(dir), &xname, &ip, &ci_name);
+	if (unlikely(error)) {
+		if (unlikely(error != ENOENT))
+			return ERR_PTR(-error);
+		d_add(dentry, NULL);
+		return NULL;
+	}
+
+	/* if exact match, just splice and exit */
+	if (!ci_name.name)
+		return d_splice_alias(ip->i_vnode, dentry);
+
+	/* else case-insensitive match... */
+	dname.name = ci_name.name;
+	dname.len = ci_name.len;
+	dentry = d_add_ci(ip->i_vnode, dentry, &dname);
+	kmem_free(ci_name.name);
+	return dentry;
+}
+
 STATIC int
 xfs_vn_link(
 	struct dentry	*old_dentry,
@@ -892,6 +928,25 @@ const struct inode_operations xfs_dir_inode_operations = {
 	.removexattr		= xfs_vn_removexattr,
 };
 
+const struct inode_operations xfs_dir_ci_inode_operations = {
+	.create			= xfs_vn_create,
+	.lookup			= xfs_vn_ci_lookup,
+	.link			= xfs_vn_link,
+	.unlink			= xfs_vn_unlink,
+	.symlink		= xfs_vn_symlink,
+	.mkdir			= xfs_vn_mkdir,
+	.rmdir			= xfs_vn_rmdir,
+	.mknod			= xfs_vn_mknod,
+	.rename			= xfs_vn_rename,
+	.permission		= xfs_vn_permission,
+	.getattr		= xfs_vn_getattr,
+	.setattr		= xfs_vn_setattr,
+	.setxattr		= xfs_vn_setxattr,
+	.getxattr		= xfs_vn_getxattr,
+	.listxattr		= xfs_vn_listxattr,
+	.removexattr		= xfs_vn_removexattr,
+};
+
 const struct inode_operations xfs_symlink_inode_operations = {
 	.readlink		= generic_readlink,
 	.follow_link		= xfs_vn_follow_link,

fs/xfs/linux-2.6/xfs_iops.h

@@ -20,6 +20,7 @@
 
 extern const struct inode_operations xfs_inode_operations;
 extern const struct inode_operations xfs_dir_inode_operations;
+extern const struct inode_operations xfs_dir_ci_inode_operations;
 extern const struct inode_operations xfs_symlink_inode_operations;
 
 extern const struct file_operations xfs_file_operations;

fs/xfs/xfs_da_btree.h

@@ -143,6 +143,7 @@ typedef struct xfs_da_args {
 #define XFS_DA_OP_RENAME	0x0002	/* this is an atomic rename op */
 #define XFS_DA_OP_ADDNAME	0x0004	/* this is an add operation */
 #define XFS_DA_OP_OKNOENT	0x0008	/* lookup/add op, ENOENT ok, else die */
+#define XFS_DA_OP_CILOOKUP	0x0010	/* lookup to return CI name if found */
 
 /*
  * Structure to describe buffer(s) for a block.

fs/xfs/xfs_dir2.c

@@ -192,15 +192,44 @@ xfs_dir_createname(
 	return rval;
 }
 
+/*
+ * If doing a CI lookup and case-insensitive match, dup actual name into
+ * args.value. Return EEXIST for success (ie. name found) or an error.
+ */
+int
+xfs_dir_cilookup_result(
+	struct xfs_da_args *args,
+	const char	*name,
+	int		len)
+{
+	if (args->cmpresult == XFS_CMP_DIFFERENT)
+		return ENOENT;
+	if (args->cmpresult != XFS_CMP_CASE ||
+					!(args->op_flags & XFS_DA_OP_CILOOKUP))
+		return EEXIST;
+
+	args->value = kmem_alloc(len, KM_MAYFAIL);
+	if (!args->value)
+		return ENOMEM;
+
+	memcpy(args->value, name, len);
+	args->valuelen = len;
+	return EEXIST;
+}
+
 /*
  * Lookup a name in a directory, give back the inode number.
+ * If ci_name is not NULL, returns the actual name in ci_name if it differs
+ * to name, or ci_name->name is set to NULL for an exact match.
  */
+
 int
 xfs_dir_lookup(
 	xfs_trans_t	*tp,
 	xfs_inode_t	*dp,
 	struct xfs_name	*name,
-	xfs_ino_t	*inum)		/* out: inode number */
+	xfs_ino_t	*inum,		/* out: inode number */
+	struct xfs_name *ci_name)	/* out: actual name if CI match */
 {
 	xfs_da_args_t	args;
 	int		rval;
@@ -217,6 +246,8 @@ xfs_dir_lookup(
 	args.whichfork = XFS_DATA_FORK;
 	args.trans = tp;
 	args.op_flags = XFS_DA_OP_OKNOENT;
+	if (ci_name)
+		args.op_flags |= XFS_DA_OP_CILOOKUP;
 	args.cmpresult = XFS_CMP_DIFFERENT;
 
 	if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
@@ -233,8 +264,13 @@ xfs_dir_lookup(
 		rval = xfs_dir2_node_lookup(&args);
 	if (rval == EEXIST)
 		rval = 0;
-	if (rval == 0)
+	if (!rval) {
 		*inum = args.inumber;
+		if (ci_name) {
+			ci_name->name = args.value;
+			ci_name->len = args.valuelen;
+		}
+	}
 	return rval;
 }
 

fs/xfs/xfs_dir2.h

@@ -74,7 +74,8 @@ extern int xfs_dir_createname(struct xfs_trans *tp, struct xfs_inode *dp,
 				xfs_fsblock_t *first,
 				struct xfs_bmap_free *flist, xfs_extlen_t tot);
 extern int xfs_dir_lookup(struct xfs_trans *tp, struct xfs_inode *dp,
-				struct xfs_name *name, xfs_ino_t *inum);
+				struct xfs_name *name, xfs_ino_t *inum,
+				struct xfs_name *ci_name);
 extern int xfs_dir_removename(struct xfs_trans *tp, struct xfs_inode *dp,
 				struct xfs_name *name, xfs_ino_t ino,
 				xfs_fsblock_t *first,
@@ -99,4 +100,7 @@ extern int xfs_dir2_isleaf(struct xfs_trans *tp, struct xfs_inode *dp,
 extern int xfs_dir2_shrink_inode(struct xfs_da_args *args, xfs_dir2_db_t db,
 				struct xfs_dabuf *bp);
 
+extern int xfs_dir_cilookup_result(struct xfs_da_args *args, const char *name,
+				int len);
+
 #endif	/* __XFS_DIR2_H__ */

fs/xfs/xfs_dir2_block.c

@@ -610,14 +610,15 @@ xfs_dir2_block_lookup(
 	/*
 	 * Get the offset from the leaf entry, to point to the data.
 	 */
-	dep = (xfs_dir2_data_entry_t *)
-	      ((char *)block + xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
+	dep = (xfs_dir2_data_entry_t *)((char *)block +
+		xfs_dir2_dataptr_to_off(mp, be32_to_cpu(blp[ent].address)));
 	/*
-	 * Fill in inode number, release the block.
+	 * Fill in inode number, CI name if appropriate, release the block.
 	 */
 	args->inumber = be64_to_cpu(dep->inumber);
+	error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
 	xfs_da_brelse(args->trans, bp);
-	return XFS_ERROR(EEXIST);
+	return XFS_ERROR(error);
 }
 
 /*

fs/xfs/xfs_dir2_leaf.c

@@ -1299,12 +1299,13 @@ xfs_dir2_leaf_lookup(
 	      ((char *)dbp->data +
 	       xfs_dir2_dataptr_to_off(dp->i_mount, be32_to_cpu(lep->address)));
 	/*
-	 * Return the found inode number.
+	 * Return the found inode number & CI name if appropriate
 	 */
 	args->inumber = be64_to_cpu(dep->inumber);
+	error = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
 	xfs_da_brelse(tp, dbp);
 	xfs_da_brelse(tp, lbp);
-	return XFS_ERROR(EEXIST);
+	return XFS_ERROR(error);
 }
 
 /*

fs/xfs/xfs_dir2_node.c

@@ -549,7 +549,7 @@ xfs_dir2_leafn_lookup_for_entry(
 	xfs_dir2_data_entry_t	*dep;		/* data block entry */
 	xfs_inode_t		*dp;		/* incore directory inode */
 	int			error;		/* error return value */
-	int			di;		/* data entry index */
+	int			di = -1;	/* data entry index */
 	int			index;		/* leaf entry index */
 	xfs_dir2_leaf_t		*leaf;		/* leaf structure */
 	xfs_dir2_leaf_entry_t	*lep;		/* leaf entry */
@@ -577,6 +577,7 @@ xfs_dir2_leafn_lookup_for_entry(
 	if (state->extravalid) {
 		curbp = state->extrablk.bp;
 		curdb = state->extrablk.blkno;
+		di = state->extrablk.index;
 	}
 	/*
 	 * Loop over leaf entries with the right hash value.
@@ -637,7 +638,6 @@ xfs_dir2_leafn_lookup_for_entry(
 	}
 	/* Didn't find an exact match. */
 	error = ENOENT;
-	di = -1;
 	ASSERT(index == be16_to_cpu(leaf->hdr.count) ||
 					(args->op_flags & XFS_DA_OP_OKNOENT));
 out:
@@ -652,7 +652,7 @@ out:
 		state->extravalid = 0;
 	}
 	/*
-	 * Return the index, that will be the insertion point.
+	 * Return the index, that will be the deletion point for remove/replace.
 	 */
 	*indexp = index;
 	return XFS_ERROR(error);
@@ -1820,8 +1820,14 @@ xfs_dir2_node_lookup(
 	error = xfs_da_node_lookup_int(state, &rval);
 	if (error)
 		rval = error;
-	else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE)
-		rval = EEXIST;	/* a case-insensitive match was found */
+	else if (rval == ENOENT && args->cmpresult == XFS_CMP_CASE) {
+		/* If a CI match, dup the actual name and return EEXIST */
+		xfs_dir2_data_entry_t	*dep;
+
+		dep = (xfs_dir2_data_entry_t *)((char *)state->extrablk.bp->
+						data + state->extrablk.index);
+		rval = xfs_dir_cilookup_result(args, dep->name, dep->namelen);
+	}
 	/*
 	 * Release the btree blocks and leaf block.
 	 */

fs/xfs/xfs_dir2_sf.c

@@ -812,9 +812,11 @@ xfs_dir2_sf_lookup(
 {
 	xfs_inode_t		*dp;		/* incore directory inode */
 	int			i;		/* entry index */
+	int			error;
 	xfs_dir2_sf_entry_t	*sfep;		/* shortform directory entry */
 	xfs_dir2_sf_t		*sfp;		/* shortform structure */
 	enum xfs_dacmp		cmp;		/* comparison result */
+	xfs_dir2_sf_entry_t	*ci_sfep;	/* case-insens. entry */
 
 	xfs_dir2_trace_args("sf_lookup", args);
 	xfs_dir2_sf_check(args);
@@ -852,6 +854,7 @@ xfs_dir2_sf_lookup(
 	/*
 	 * Loop over all the entries trying to match ours.
 	 */
+	ci_sfep = NULL;
 	for (i = 0, sfep = xfs_dir2_sf_firstentry(sfp); i < sfp->hdr.count;
 				i++, sfep = xfs_dir2_sf_nextentry(sfp, sfep)) {
 		/*
@@ -867,19 +870,19 @@ xfs_dir2_sf_lookup(
 						xfs_dir2_sf_inumberp(sfep));
 			if (cmp == XFS_CMP_EXACT)
 				return XFS_ERROR(EEXIST);
+			ci_sfep = sfep;
 		}
 	}
 	ASSERT(args->op_flags & XFS_DA_OP_OKNOENT);
 	/*
 	 * Here, we can only be doing a lookup (not a rename or replace).
-	 * If a case-insensitive match was found earlier, return "found".
+	 * If a case-insensitive match was not found, return ENOENT.
 	 */
-	if (args->cmpresult == XFS_CMP_CASE)
-		return XFS_ERROR(EEXIST);
-	/*
-	 * Didn't find it.
-	 */
-	return XFS_ERROR(ENOENT);
+	if (!ci_sfep)
+		return XFS_ERROR(ENOENT);
+	/* otherwise process the CI match as required by the caller */
+	error = xfs_dir_cilookup_result(args, ci_sfep->name, ci_sfep->namelen);
+	return XFS_ERROR(error);
 }
 
 /*

fs/xfs/xfs_vnodeops.c

@@ -1610,12 +1610,18 @@ xfs_inactive(
 	return VN_INACTIVE_CACHE;
 }
 
-
+/*
+ * Lookups up an inode from "name". If ci_name is not NULL, then a CI match
+ * is allowed, otherwise it has to be an exact match. If a CI match is found,
+ * ci_name->name will point to a the actual name (caller must free) or
+ * will be set to NULL if an exact match is found.
+ */
 int
 xfs_lookup(
 	xfs_inode_t		*dp,
 	struct xfs_name		*name,
-	xfs_inode_t		**ipp)
+	xfs_inode_t		**ipp,
+	struct xfs_name		*ci_name)
 {
 	xfs_ino_t		inum;
 	int			error;
@@ -1627,7 +1633,7 @@ xfs_lookup(
 		return XFS_ERROR(EIO);
 
 	lock_mode = xfs_ilock_map_shared(dp);
-	error = xfs_dir_lookup(NULL, dp, name, &inum);
+	error = xfs_dir_lookup(NULL, dp, name, &inum, ci_name);
 	xfs_iunlock_map_shared(dp, lock_mode);
 
 	if (error)
@@ -1635,12 +1641,15 @@ xfs_lookup(
 
 	error = xfs_iget(dp->i_mount, NULL, inum, 0, 0, ipp, 0);
 	if (error)
-		goto out;
+		goto out_free_name;
 
 	xfs_itrace_ref(*ipp);
 	return 0;
 
- out:
+out_free_name:
+	if (ci_name)
+		kmem_free(ci_name->name);
+out:
 	*ipp = NULL;
 	return error;
 }

fs/xfs/xfs_vnodeops.h

@@ -22,7 +22,7 @@ int xfs_fsync(struct xfs_inode *ip);
 int xfs_release(struct xfs_inode *ip);
 int xfs_inactive(struct xfs_inode *ip);
 int xfs_lookup(struct xfs_inode *dp, struct xfs_name *name,
-		struct xfs_inode **ipp);
+		struct xfs_inode **ipp, struct xfs_name *ci_name);
 int xfs_create(struct xfs_inode *dp, struct xfs_name *name, mode_t mode,
 		xfs_dev_t rdev, struct xfs_inode **ipp, struct cred *credp);
 int xfs_remove(struct xfs_inode *dp, struct xfs_name *name,