[PATCH v2] ksmbd: make utf-8 file name comparison work in __caseless_lookup()

[PATCH v2] ksmbd: make utf-8 file name comparison work in __caseless_lookup()

[Atte Heikkilä]

Case-insensitive file name lookups with __caseless_lookup() use
strncasecmp() for file name comparison. strncasecmp() assumes an
ISO8859-1-compatible encoding, which is not the case here as UTF-8
is always used. As such, use of strncasecmp() here produces correct
results only if both strings use characters in the ASCII range only.
Fix this by using utf8_strncasecmp() if CONFIG_UNICODE is set. On
failure or if CONFIG_UNICODE is not set, fallback to strncasecmp().
Also, as we are adding an include for `linux/unicode.h', include it
in `fs/ksmbd/connection.h' as well since it should be explicit there.

Signed-off-by: Atte Heikkilä <atteh.mailbox@gmail.com>
---
 v2:
  - simplified fallback condition by initializing `cmp'

 fs/ksmbd/connection.h |  1 +
 fs/ksmbd/vfs.c        | 20 +++++++++++++++++---
 fs/ksmbd/vfs.h        |  2 ++
 3 files changed, 20 insertions(+), 3 deletions(-)

diff --git a/fs/ksmbd/connection.h b/fs/ksmbd/connection.h
index 41d96f5cef06..3643354a3fa7 100644
--- a/fs/ksmbd/connection.h
+++ b/fs/ksmbd/connection.h
@@ -14,6 +14,7 @@
 #include <net/request_sock.h>
 #include <linux/kthread.h>
 #include <linux/nls.h>
+#include <linux/unicode.h>
 
 #include "smb_common.h"
 #include "ksmbd_work.h"
diff --git a/fs/ksmbd/vfs.c b/fs/ksmbd/vfs.c
index 4fcf96a01c16..8d8af724df70 100644
--- a/fs/ksmbd/vfs.c
+++ b/fs/ksmbd/vfs.c
@@ -1145,12 +1145,23 @@ static int __caseless_lookup(struct dir_context *ctx, const char *name,
 			     unsigned int d_type)
 {
 	struct ksmbd_readdir_data *buf;
+	int cmp = -EINVAL;
 
 	buf = container_of(ctx, struct ksmbd_readdir_data, ctx);
 
 	if (buf->used != namlen)
 		return 0;
-	if (!strncasecmp((char *)buf->private, name, namlen)) {
+	if (IS_ENABLED(CONFIG_UNICODE) && buf->um) {
+		const struct qstr q_buf = {.name = buf->private,
+					   .len = buf->used};
+		const struct qstr q_name = {.name = name,
+					    .len = namlen};
+
+		cmp = utf8_strncasecmp(buf->um, &q_buf, &q_name);
+	}
+	if (cmp < 0)
+		cmp = strncasecmp((char *)buf->private, name, namlen);
+	if (!cmp) {
 		memcpy((char *)buf->private, name, namlen);
 		buf->dirent_count = 1;
 		return -EEXIST;
@@ -1166,7 +1177,8 @@ static int __caseless_lookup(struct dir_context *ctx, const char *name,
  *
  * Return:	0 on success, otherwise error
  */
-static int ksmbd_vfs_lookup_in_dir(const struct path *dir, char *name, size_t namelen)
+static int ksmbd_vfs_lookup_in_dir(const struct path *dir, char *name,
+				   size_t namelen, struct unicode_map *um)
 {
 	int ret;
 	struct file *dfilp;
@@ -1176,6 +1188,7 @@ static int ksmbd_vfs_lookup_in_dir(const struct path *dir, char *name, size_t na
 		.private	= name,
 		.used		= namelen,
 		.dirent_count	= 0,
+		.um		= um,
 	};
 
 	dfilp = dentry_open(dir, flags, current_cred());
@@ -1238,7 +1251,8 @@ int ksmbd_vfs_kern_path(struct ksmbd_work *work, char *name,
 				break;
 
 			err = ksmbd_vfs_lookup_in_dir(&parent, filename,
-						      filename_len);
+						      filename_len,
+						      work->conn->um);
 			path_put(&parent);
 			if (err)
 				goto out;
diff --git a/fs/ksmbd/vfs.h b/fs/ksmbd/vfs.h
index d7542a2dab52..593059ca8511 100644
--- a/fs/ksmbd/vfs.h
+++ b/fs/ksmbd/vfs.h
@@ -12,6 +12,7 @@
 #include <linux/namei.h>
 #include <uapi/linux/xattr.h>
 #include <linux/posix_acl.h>
+#include <linux/unicode.h>
 
 #include "smbacl.h"
 #include "xattr.h"
@@ -60,6 +61,7 @@ struct ksmbd_readdir_data {
 	unsigned int		used;
 	unsigned int		dirent_count;
 	unsigned int		file_attr;
+	struct unicode_map	*um;
 };
 
 /* ksmbd kstat wrapper to get valid create time when reading dir entry */
-- 
2.37.3

Re: [PATCH v2] ksmbd: make utf-8 file name comparison work in __caseless_lookup()

[Namjae Jeon]

2022-09-28 6:57 GMT+09:00, Atte Heikkilä <atteh.mailbox@gmail.com>:
> Case-insensitive file name lookups with __caseless_lookup() use
> strncasecmp() for file name comparison. strncasecmp() assumes an
> ISO8859-1-compatible encoding, which is not the case here as UTF-8
> is always used. As such, use of strncasecmp() here produces correct
> results only if both strings use characters in the ASCII range only.
> Fix this by using utf8_strncasecmp() if CONFIG_UNICODE is set. On
> failure or if CONFIG_UNICODE is not set, fallback to strncasecmp().
> Also, as we are adding an include for `linux/unicode.h', include it
> in `fs/ksmbd/connection.h' as well since it should be explicit there.
>
> Signed-off-by: Atte Heikkilä <atteh.mailbox@gmail.com>
Acked-by: Namjae Jeon <linkinjeon@kernel.org>

Thanks for your work!