fat: relax the permission check of fat_setattr()

New chmod() allows only acceptable permission, and if not acceptable, it
returns -EPERM.  Old one allows even if it can't store permission to on
disk inode.  But it seems too strict for users.

E.g.  https://bugzilla.redhat.com/show_bug.cgi?id=449080: With new one,
rsync couldn't create the temporary file.

So, this patch allows like old one, but now it doesn't change the
permission if it can't store, and it returns 0.

Also, this patch fixes missing check.

Signed-off-by: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
diff --git a/fs/fat/file.c b/fs/fat/file.c
index 27cc116..771326b 100644
--- a/fs/fat/file.c
+++ b/fs/fat/file.c
@@ -257,26 +257,34 @@
 }
 EXPORT_SYMBOL_GPL(fat_getattr);
 
-static int fat_check_mode(const struct msdos_sb_info *sbi, struct inode *inode,
-			  mode_t mode)
+static int fat_sanitize_mode(const struct msdos_sb_info *sbi,
+			     struct inode *inode, umode_t *mode_ptr)
 {
-	mode_t mask, req = mode & ~S_IFMT;
+	mode_t mask, perm;
 
-	if (S_ISREG(mode))
+	/*
+	 * Note, the basic check is already done by a caller of
+	 * (attr->ia_mode & ~MSDOS_VALID_MODE)
+	 */
+
+	if (S_ISREG(inode->i_mode))
 		mask = sbi->options.fs_fmask;
 	else
 		mask = sbi->options.fs_dmask;
 
+	perm = *mode_ptr & ~(S_IFMT | mask);
+
 	/*
 	 * Of the r and x bits, all (subject to umask) must be present. Of the
 	 * w bits, either all (subject to umask) or none must be present.
 	 */
-	req &= ~mask;
-	if ((req & (S_IRUGO | S_IXUGO)) != (inode->i_mode & (S_IRUGO|S_IXUGO)))
+	if ((perm & (S_IRUGO | S_IXUGO)) != (inode->i_mode & (S_IRUGO|S_IXUGO)))
 		return -EPERM;
-	if ((req & S_IWUGO) && ((req & S_IWUGO) != (S_IWUGO & ~mask)))
+	if ((perm & S_IWUGO) && ((perm & S_IWUGO) != (S_IWUGO & ~mask)))
 		return -EPERM;
 
+	*mode_ptr &= S_IFMT | perm;
+
 	return 0;
 }
 
@@ -299,7 +307,7 @@
 {
 	struct msdos_sb_info *sbi = MSDOS_SB(dentry->d_sb);
 	struct inode *inode = dentry->d_inode;
-	int mask, error = 0;
+	int error = 0;
 	unsigned int ia_valid;
 
 	lock_kernel();
@@ -332,12 +340,13 @@
 			error = 0;
 		goto out;
 	}
+
 	if (((attr->ia_valid & ATTR_UID) &&
 	     (attr->ia_uid != sbi->options.fs_uid)) ||
 	    ((attr->ia_valid & ATTR_GID) &&
 	     (attr->ia_gid != sbi->options.fs_gid)) ||
 	    ((attr->ia_valid & ATTR_MODE) &&
-	     fat_check_mode(sbi, inode, attr->ia_mode) < 0))
+	     (attr->ia_mode & ~MSDOS_VALID_MODE)))
 		error = -EPERM;
 
 	if (error) {
@@ -346,15 +355,16 @@
 		goto out;
 	}
 
-	error = inode_setattr(inode, attr);
-	if (error)
-		goto out;
+	/*
+	 * We don't return -EPERM here. Yes, strange, but this is too
+	 * old behavior.
+	 */
+	if (attr->ia_valid & ATTR_MODE) {
+		if (fat_sanitize_mode(sbi, inode, &attr->ia_mode) < 0)
+			attr->ia_valid &= ~ATTR_MODE;
+	}
 
-	if (S_ISDIR(inode->i_mode))
-		mask = sbi->options.fs_dmask;
-	else
-		mask = sbi->options.fs_fmask;
-	inode->i_mode &= S_IFMT | (S_IRWXUGO & ~mask);
+	error = inode_setattr(inode, attr);
 out:
 	unlock_kernel();
 	return error;