| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Fri, 3 Jan 2020 13:31:14 +0100
From: Pali Rohár <pali.rohar@...il.com>
To: Namjae Jeon <namjae.jeon@...sung.com>
Cc: linux-kernel@...r.kernel.org, linux-fsdevel@...r.kernel.org,
gregkh@...uxfoundation.org, valdis.kletnieks@...edu, hch@....de,
sj1557.seo@...sung.com, linkinjeon@...il.com
Subject: Re: [PATCH v9 10/13] exfat: add nls operations
On Friday 03 January 2020 10:40:30 Pali Rohár wrote:
> On Thursday 02 January 2020 16:20:33 Namjae Jeon wrote:
> > This adds the implementation of nls operations for exfat.
> >
> > Signed-off-by: Namjae Jeon <namjae.jeon@...sung.com>
> > Signed-off-by: Sungjong Seo <sj1557.seo@...sung.com>
> > ---
> > fs/exfat/nls.c | 809 +++++++++++++++++++++++++++++++++++++++++++++++++
> > 1 file changed, 809 insertions(+)
> > create mode 100644 fs/exfat/nls.c
> >
> > diff --git a/fs/exfat/nls.c b/fs/exfat/nls.c
> > new file mode 100644
> > index 000000000000..af52328e28ff
> > --- /dev/null
> > +++ b/fs/exfat/nls.c
>
> ...
>
> > +static int exfat_convert_uni_to_ch(struct nls_table *nls, unsigned short uni,
> > + unsigned char *ch, int *lossy)
> > +{
> > + int len;
> > +
> > + ch[0] = 0x0;
> > +
> > + if (uni < 0x0080) {
> > + ch[0] = uni;
> > + return 1;
> > + }
> > +
> > + len = nls->uni2char(uni, ch, MAX_CHARSET_SIZE);
> > + if (len < 0) {
> > + /* conversion failed */
> > + if (lossy != NULL)
> > + *lossy |= NLS_NAME_LOSSY;
> > + ch[0] = '_';
> > + return 1;
> > + }
> > + return len;
> > +}
>
> Hello! This function takes one UCS-2 character in host endianity and
> converts it to one byte (via specified 8bit encoding).
>
> > +static int __exfat_nls_uni16s_to_vfsname(struct super_block *sb,
> > + struct exfat_uni_name *p_uniname, unsigned char *p_cstring,
> > + int buflen)
> > +{
> > + int i, j, len, out_len = 0;
> > + unsigned char buf[MAX_CHARSET_SIZE];
> > + const unsigned short *uniname = p_uniname->name;
> > + struct nls_table *nls = EXFAT_SB(sb)->nls_io;
> > +
> > + i = 0;
> > + while (i < MAX_NAME_LENGTH && out_len < (buflen - 1)) {
> > + if (*uniname == '\0')
> > + break;
> > +
> > + len = exfat_convert_uni_to_ch(nls, *uniname, buf, NULL);
> > + if (out_len + len >= buflen)
> > + len = buflen - 1 - out_len;
> > + out_len += len;
> > +
> > + if (len > 1) {
> > + for (j = 0; j < len; j++)
> > + *p_cstring++ = buf[j];
> > + } else { /* len == 1 */
> > + *p_cstring++ = *buf;
> > + }
> > +
> > + uniname++;
> > + i++;
> > + }
> > +
> > + *p_cstring = '\0';
> > + return out_len;
> > +}
> > +
>
> This function takes UCS-2 buffer in host endianity and converts it to
> string in specified 8bit encoding.
>
> > +
> > +int exfat_nls_uni16s_to_vfsname(struct super_block *sb,
> > + struct exfat_uni_name *uniname, unsigned char *p_cstring,
> > + int buflen)
> > +{
>
> Looking at the code and this function is called from dir.c to translate
> exfat filename buffer stored in filesystem to format expected by VFS
> layer.
>
> On exfat filesystem file names are always stored in UTF-16LE...
>
> > + if (EXFAT_SB(sb)->options.utf8)
> > + return __exfat_nls_utf16s_to_vfsname(sb, uniname, p_cstring,
> > + buflen);
> > + return __exfat_nls_uni16s_to_vfsname(sb, uniname, p_cstring, buflen);
>
> ... and therefore above "__exfat_nls_uni16s_to_vfsname" function must
> expect UTF-16LE buffer and not just UCS-2 buffer in host endianity.
>
> So two other things needs to be done: Convert character from little
> endian to host endianity and then process UTF-16 buffer and not only
> UCS-2.
>
> I see that in kernel NLS module is missing a function for converting
> UTF-16 string to UTF-32 (encoding in which every code point is
> represented just by one u32 variable). Kernel has only utf16s_to_utf8s()
> and utf8_to_utf32().
What about just filtering two u16 (one surrogate pair)? Existing NLS
modules do not support code points above U+FFFF so two u16 (one
surrogate pair) just needs to be converted to one replacement character.
diff --git a/fs/exfat/nls.c b/fs/exfat/nls.c
index 81d75aed9..f626a0a89 100644
--- a/fs/exfat/nls.c
+++ b/fs/exfat/nls.c
@@ -545,7 +545,10 @@ static int __exfat_nls_vfsname_to_utf16s(struct super_block *sb,
return unilen;
}
-static int __exfat_nls_uni16s_to_vfsname(struct super_block *sb,
+#define SURROGATE_PAIR 0x0000d800
+#define SURROGATE_LOW 0x00000400
+
+static int __exfat_nls_utf16s_to_vfsname(struct super_block *sb,
struct exfat_uni_name *p_uniname, unsigned char *p_cstring,
int buflen)
{
@@ -559,7 +562,23 @@ static int __exfat_nls_uni16s_to_vfsname(struct super_block *sb,
if (*uniname == '\0')
break;
- len = exfat_convert_uni_to_ch(nls, *uniname, buf, NULL);
+ if ((*uniname & SURROGATE_MASK) != SURROGATE_PAIR) {
+ len = exfat_convert_uni_to_ch(nls, *uniname, buf, NULL);
+ } else {
+ /* Process UTF-16 surrogate pair as one character */
+ if (!(*uniname & SURROGATE_LOW) && i+1 < MAX_NAME_LENGTH &&
+ (*(uniname+1) & SURROGATE_MASK) == SURROGATE_PAIR &&
+ (*(uniname+1) & SURROGATE_LOW)) {
+ uniname++;
+ i++;
+ }
+ /* UTF-16 surrogate pair encodes code points above Ux+FFFF.
+ * Code points above U+FFFF are not supported by kernel NLS
+ * framework therefore use replacement character */
+ len = 1;
+ buf[0] = '_';
+ }
+
if (out_len + len >= buflen)
len = buflen - 1 - out_len;
out_len += len;
@@ -623,7 +642,7 @@ int exfat_nls_uni16s_to_vfsname(struct super_block *sb,
if (EXFAT_SB(sb)->options.utf8)
return __exfat_nls_utf16s_to_vfsname(sb, uniname, p_cstring,
buflen);
- return __exfat_nls_uni16s_to_vfsname(sb, uniname, p_cstring, buflen);
+ return __exfat_nls_utf16s_to_vfsname(sb, uniname, p_cstring, buflen);
}
int exfat_nls_vfsname_to_uni16s(struct super_block *sb,
I have not tested this code, it is just an idea how to quick & dirty
solve this problem that NLS framework works with UCS-2 encoding and
UCS-4/UTF-32 or UTF-16.
> > +}
>
> Btw, have you tested this exfat implementation on some big endian
> system? I think it cannot work because of missing conversion from
> UTF-16LE to UTF-16 in host endianity (therefore UTF-16BE).
Now I figured out that conversion from UTF-16LE to UTF-16 host endianity
is already done in exfat_extract_uni_name() function, called from
exfat_get_uniname_from_ext_entry() function. exfat_nls_uni16s_to_vfsname
is then called on result from exfat_get_uniname_from_ext_entry(), so
UTF-16LE processing on big endian systems should work. Sorry for that.
--
Pali Rohár
pali.rohar@...il.com
Powered by blists - more mailing lists