smb: client: add support for WSL reparse points

Add support for creating special files via WSL reparse points when
using 'reparse=wsl' mount option.  They're faster than NFS reparse
points because they don't require extra roundtrips to figure out what
->d_type a specific dirent is as such information is already stored in
query dir responses and then making getdents() calls faster.

Signed-off-by: Paulo Alcantara <pc@manguebit.com>
Signed-off-by: Steve French <stfrench@microsoft.com>

Author: Paulo Alcantara

fs/smb/client/cifsglob.h

@@ -1374,6 +1374,7 @@ struct cifs_open_parms {
 	umode_t mode;
 	bool reconnect:1;
 	bool replay:1; /* indicates that this open is for a replay */
+	struct kvec *ea_cctx;
 };
 
 struct cifs_fid {

fs/smb/client/fs_context.c

@@ -317,8 +317,8 @@ static int parse_reparse_flavor(struct fs_context *fc, char *value,
 		ctx->reparse_type = CIFS_REPARSE_TYPE_NFS;
 		break;
 	case Opt_reparse_wsl:
-		cifs_errorf(fc, "unsupported reparse= option: %s\n", value);
-		return 1;
+		ctx->reparse_type = CIFS_REPARSE_TYPE_WSL;
+		break;
 	default:
 		cifs_errorf(fc, "bad reparse= option: %s\n", value);
 		return 1;

fs/smb/client/reparse.c

@@ -11,6 +11,7 @@
 #include "cifsproto.h"
 #include "cifs_unicode.h"
 #include "cifs_debug.h"
+#include "fs_context.h"
 #include "reparse.h"
 
 int smb2_create_reparse_symlink(const unsigned int xid, struct inode *inode,
@@ -68,7 +69,7 @@ int smb2_create_reparse_symlink(const unsigned int xid, struct inode *inode,
 	iov.iov_base = buf;
 	iov.iov_len = len;
 	new = smb2_get_reparse_inode(&data, inode->i_sb, xid,
-				     tcon, full_path, &iov);
+				     tcon, full_path, &iov, NULL);
 	if (!IS_ERR(new))
 		d_instantiate(dentry, new);
 	else
@@ -114,9 +115,9 @@ static int nfs_set_reparse_buf(struct reparse_posix_data *buf,
 	return 0;
 }
 
-int smb2_make_nfs_node(unsigned int xid, struct inode *inode,
-		       struct dentry *dentry, struct cifs_tcon *tcon,
-		       const char *full_path, umode_t mode, dev_t dev)
+static int mknod_nfs(unsigned int xid, struct inode *inode,
+		     struct dentry *dentry, struct cifs_tcon *tcon,
+		     const char *full_path, umode_t mode, dev_t dev)
 {
 	struct cifs_open_info_data data;
 	struct reparse_posix_data *p;
@@ -136,12 +137,171 @@ int smb2_make_nfs_node(unsigned int xid, struct inode *inode,
 	};
 
 	new = smb2_get_reparse_inode(&data, inode->i_sb, xid,
-				     tcon, full_path, &iov);
+				     tcon, full_path, &iov, NULL);
+	if (!IS_ERR(new))
+		d_instantiate(dentry, new);
+	else
+		rc = PTR_ERR(new);
+	cifs_free_open_info(&data);
+	return rc;
+}
+
+static int wsl_set_reparse_buf(struct reparse_data_buffer *buf,
+			       mode_t mode, struct kvec *iov)
+{
+	u32 tag;
+
+	switch ((tag = reparse_mode_wsl_tag(mode))) {
+	case IO_REPARSE_TAG_LX_BLK:
+	case IO_REPARSE_TAG_LX_CHR:
+	case IO_REPARSE_TAG_LX_FIFO:
+	case IO_REPARSE_TAG_AF_UNIX:
+		break;
+	default:
+		return -EOPNOTSUPP;
+	}
+
+	buf->ReparseTag = cpu_to_le32(tag);
+	buf->Reserved = 0;
+	buf->ReparseDataLength = 0;
+	iov->iov_base = buf;
+	iov->iov_len = sizeof(*buf);
+	return 0;
+}
+
+static struct smb2_create_ea_ctx *ea_create_context(u32 dlen, size_t *cc_len)
+{
+	struct smb2_create_ea_ctx *cc;
+
+	*cc_len = round_up(sizeof(*cc) + dlen, 8);
+	cc = kzalloc(*cc_len, GFP_KERNEL);
+	if (!cc)
+		return ERR_PTR(-ENOMEM);
+
+	cc->ctx.NameOffset = cpu_to_le16(offsetof(struct smb2_create_ea_ctx,
+						  name));
+	cc->ctx.NameLength = cpu_to_le16(4);
+	memcpy(cc->name, SMB2_CREATE_EA_BUFFER, strlen(SMB2_CREATE_EA_BUFFER));
+	cc->ctx.DataOffset = cpu_to_le16(offsetof(struct smb2_create_ea_ctx, ea));
+	cc->ctx.DataLength = cpu_to_le32(dlen);
+	return cc;
+}
+
+struct wsl_xattr {
+	const char	*name;
+	__le64		value;
+	u16		size;
+	u32		next;
+};
+
+static int wsl_set_xattrs(struct inode *inode, umode_t _mode,
+			  dev_t _dev, struct kvec *iov)
+{
+	struct smb2_file_full_ea_info *ea;
+	struct smb2_create_ea_ctx *cc;
+	struct smb3_fs_context *ctx = CIFS_SB(inode->i_sb)->ctx;
+	__le64 uid = cpu_to_le64(from_kuid(current_user_ns(), ctx->linux_uid));
+	__le64 gid = cpu_to_le64(from_kgid(current_user_ns(), ctx->linux_gid));
+	__le64 dev = cpu_to_le64(((u64)MINOR(_dev) << 32) | MAJOR(_dev));
+	__le64 mode = cpu_to_le64(_mode);
+	struct wsl_xattr xattrs[] = {
+		{ .name = "$LXUID", .value = uid, .size = 4, },
+		{ .name = "$LXGID", .value = gid, .size = 4, },
+		{ .name = "$LXMOD", .value = mode, .size = 4, },
+		{ .name = "$LXDEV", .value = dev, .size = 8, },
+	};
+	size_t cc_len;
+	u32 dlen = 0, next = 0;
+	int i, num_xattrs;
+	u8 name_size = strlen(xattrs[0].name) + 1;
+
+	memset(iov, 0, sizeof(*iov));
+
+	/* Exclude $LXDEV xattr for sockets and fifos */
+	if (S_ISSOCK(_mode) || S_ISFIFO(_mode))
+		num_xattrs = ARRAY_SIZE(xattrs) - 1;
+	else
+		num_xattrs = ARRAY_SIZE(xattrs);
+
+	for (i = 0; i < num_xattrs; i++) {
+		xattrs[i].next = ALIGN(sizeof(*ea) + name_size +
+				       xattrs[i].size, 4);
+		dlen += xattrs[i].next;
+	}
+
+	cc = ea_create_context(dlen, &cc_len);
+	if (!cc)
+		return PTR_ERR(cc);
+
+	ea = &cc->ea;
+	for (i = 0; i < num_xattrs; i++) {
+		ea = (void *)((u8 *)ea + next);
+		next = xattrs[i].next;
+		ea->next_entry_offset = cpu_to_le32(next);
+
+		ea->ea_name_length = name_size - 1;
+		ea->ea_value_length = cpu_to_le16(xattrs[i].size);
+		memcpy(ea->ea_data, xattrs[i].name, name_size);
+		memcpy(&ea->ea_data[name_size],
+		       &xattrs[i].value, xattrs[i].size);
+	}
+	ea->next_entry_offset = 0;
+
+	iov->iov_base = cc;
+	iov->iov_len = cc_len;
+	return 0;
+}
+
+static int mknod_wsl(unsigned int xid, struct inode *inode,
+		     struct dentry *dentry, struct cifs_tcon *tcon,
+		     const char *full_path, umode_t mode, dev_t dev)
+{
+	struct cifs_open_info_data data;
+	struct reparse_data_buffer buf;
+	struct inode *new;
+	struct kvec reparse_iov, xattr_iov;
+	int rc;
+
+	rc = wsl_set_reparse_buf(&buf, mode, &reparse_iov);
+	if (rc)
+		return rc;
+
+	rc = wsl_set_xattrs(inode, mode, dev, &xattr_iov);
+	if (rc)
+		return rc;
+
+	data = (struct cifs_open_info_data) {
+		.reparse_point = true,
+		.reparse = { .tag = le32_to_cpu(buf.ReparseTag), .buf = &buf, },
+	};
+
+	new = smb2_get_reparse_inode(&data, inode->i_sb,
+				     xid, tcon, full_path,
+				     &reparse_iov, &xattr_iov);
 	if (!IS_ERR(new))
 		d_instantiate(dentry, new);
 	else
 		rc = PTR_ERR(new);
 	cifs_free_open_info(&data);
+	kfree(xattr_iov.iov_base);
+	return rc;
+}
+
+int smb2_mknod_reparse(unsigned int xid, struct inode *inode,
+		       struct dentry *dentry, struct cifs_tcon *tcon,
+		       const char *full_path, umode_t mode, dev_t dev)
+{
+	struct smb3_fs_context *ctx = CIFS_SB(inode->i_sb)->ctx;
+	int rc = -EOPNOTSUPP;
+
+	switch (ctx->reparse_type) {
+	case CIFS_REPARSE_TYPE_NFS:
+		rc = mknod_nfs(xid, inode, dentry, tcon, full_path, mode, dev);
+		break;
+	case CIFS_REPARSE_TYPE_WSL:
+		rc = mknod_wsl(xid, inode, dentry, tcon, full_path, mode, dev);
+		break;
+	}
 	return rc;
 }
 

fs/smb/client/reparse.h

@@ -28,6 +28,17 @@ static inline u64 reparse_mode_nfs_type(mode_t mode)
 	return 0;
 }
 
+static inline u32 reparse_mode_wsl_tag(mode_t mode)
+{
+	switch (mode & S_IFMT) {
+	case S_IFBLK: return IO_REPARSE_TAG_LX_BLK;
+	case S_IFCHR: return IO_REPARSE_TAG_LX_CHR;
+	case S_IFIFO: return IO_REPARSE_TAG_LX_FIFO;
+	case S_IFSOCK: return IO_REPARSE_TAG_AF_UNIX;
+	}
+	return 0;
+}
+
 /*
  * Match a reparse point inode if reparse tag and ctime haven't changed.
  *
@@ -64,7 +75,7 @@ bool cifs_reparse_point_to_fattr(struct cifs_sb_info *cifs_sb,
 int smb2_create_reparse_symlink(const unsigned int xid, struct inode *inode,
 				struct dentry *dentry, struct cifs_tcon *tcon,
 				const char *full_path, const char *symname);
-int smb2_make_nfs_node(unsigned int xid, struct inode *inode,
+int smb2_mknod_reparse(unsigned int xid, struct inode *inode,
 		       struct dentry *dentry, struct cifs_tcon *tcon,
 		       const char *full_path, umode_t mode, dev_t dev);
 int smb2_parse_reparse_point(struct cifs_sb_info *cifs_sb, struct kvec *rsp_iov,

fs/smb/client/smb2inode.c

@@ -1060,7 +1060,8 @@ struct inode *smb2_get_reparse_inode(struct cifs_open_info_data *data,
 				     const unsigned int xid,
 				     struct cifs_tcon *tcon,
 				     const char *full_path,
-				     struct kvec *iov)
+				     struct kvec *reparse_iov,
+				     struct kvec *xattr_iov)
 {
 	struct cifs_open_parms oparms;
 	struct cifs_sb_info *cifs_sb = CIFS_SB(sb);
@@ -1077,8 +1078,11 @@ struct inode *smb2_get_reparse_inode(struct cifs_open_info_data *data,
 			     FILE_CREATE,
 			     CREATE_NOT_DIR | OPEN_REPARSE_POINT,
 			     ACL_NO_MODE);
+	if (xattr_iov)
+		oparms.ea_cctx = xattr_iov;
+
 	cmds[0] = SMB2_OP_SET_REPARSE;
-	in_iov[0] = *iov;
+	in_iov[0] = *reparse_iov;
 	in_iov[1].iov_base = data;
 	in_iov[1].iov_len = sizeof(*data);
 

fs/smb/client/smb2ops.c

@@ -5043,7 +5043,7 @@ static int smb2_make_node(unsigned int xid, struct inode *inode,
 		rc = cifs_sfu_make_node(xid, inode, dentry, tcon,
 					full_path, mode, dev);
 	} else {
-		rc = smb2_make_nfs_node(xid, inode, dentry, tcon,
+		rc = smb2_mknod_reparse(xid, inode, dentry, tcon,
 					full_path, mode, dev);
 	}
 	return rc;

fs/smb/client/smb2pdu.c

@@ -2720,6 +2720,17 @@ add_query_id_context(struct kvec *iov, unsigned int *num_iovec)
 	return 0;
 }
 
+static void add_ea_context(struct cifs_open_parms *oparms,
+			   struct kvec *rq_iov, unsigned int *num_iovs)
+{
+	struct kvec *iov = oparms->ea_cctx;
+
+	if (iov && iov->iov_base && iov->iov_len) {
+		rq_iov[(*num_iovs)++] = *iov;
+		memset(iov, 0, sizeof(*iov));
+	}
+}
+
 static int
 alloc_path_with_tree_prefix(__le16 **out_path, int *out_size, int *out_len,
 			    const char *treename, const __le16 *path)
@@ -3086,6 +3097,7 @@ SMB2_open_init(struct cifs_tcon *tcon, struct TCP_Server_Info *server,
 	}
 
 	add_query_id_context(iov, &n_iov);
+	add_ea_context(oparms, iov, &n_iov);
 
 	if (n_iov > 2) {
 		/*

fs/smb/client/smb2pdu.h

@@ -117,9 +117,10 @@ struct share_redirect_error_context_rsp {
  * [4] : posix context
  * [5] : time warp context
  * [6] : query id context
- * [7] : compound padding
+ * [7] : create ea context
+ * [8] : compound padding
  */
-#define SMB2_CREATE_IOV_SIZE 8
+#define SMB2_CREATE_IOV_SIZE 9
 
 /*
  * Maximum size of a SMB2_CREATE response is 64 (smb2 header) +
@@ -413,4 +414,10 @@ struct smb2_posix_info_parsed {
 	const u8 *name;
 };
 
+struct smb2_create_ea_ctx {
+	struct create_context ctx;
+	__u8 name[8];
+	struct smb2_file_full_ea_info ea;
+} __packed;
+
 #endif				/* _SMB2PDU_H */

fs/smb/client/smb2proto.h

@@ -61,7 +61,8 @@ struct inode *smb2_get_reparse_inode(struct cifs_open_info_data *data,
 				     const unsigned int xid,
 				     struct cifs_tcon *tcon,
 				     const char *full_path,
-				     struct kvec *iov);
+				     struct kvec *reparse_iov,
+				     struct kvec *xattr_iov);
 int smb2_query_reparse_point(const unsigned int xid,
 			     struct cifs_tcon *tcon,
 			     struct cifs_sb_info *cifs_sb,

fs/smb/common/smbfsctl.h

@@ -158,12 +158,6 @@
 #define IO_REPARSE_TAG_LX_CHR	     0x80000025
 #define IO_REPARSE_TAG_LX_BLK	     0x80000026
 
-#define IO_REPARSE_TAG_LX_SYMLINK_LE	cpu_to_le32(0xA000001D)
-#define IO_REPARSE_TAG_AF_UNIX_LE	cpu_to_le32(0x80000023)
-#define IO_REPARSE_TAG_LX_FIFO_LE	cpu_to_le32(0x80000024)
-#define IO_REPARSE_TAG_LX_CHR_LE	cpu_to_le32(0x80000025)
-#define IO_REPARSE_TAG_LX_BLK_LE	cpu_to_le32(0x80000026)
-
 /* fsctl flags */
 /* If Flags is set to this value, the request is an FSCTL not ioctl request */
 #define SMB2_0_IOCTL_IS_FSCTL		0x00000001