getfacl gives octal output for Hebrew user/group names

Hello!

Running RHEL5 U2 x86_64, with Samba & Winbind configure to work with Active-Directory env. Users & Groups in the Active-Directory are in hebrew and locale settings are correct (Hebrew/UTF-8) Posix Acls on file has special characters ('\' & hebrew chars) for the group/user names. In circumstances like this, where the user/group accounts come from an AD
server, getfacl could be showing the "invalid" characters as octal.
Example:

[root]# getfacl a.a
# file: a.a
# owner: GTS\134\327\236\327\225\327\251\327\225\327\237 # group: root
user::rw-
group::r-- group:GTS\134\327\251\327\236\327\225\327\250:-w- group:GTS\134\327\251\327\250\327\252:r-x mask::rwx
other::r--

Now - we are also have GPFS fs and when we run the mmgetacl command the Posix acls in Hebrew showed OK.

I am trying to work out a patch which fixes this and the issue seems to be here const char *quote(const char *str)

{
     static char *quoted_str;
     static size_t quoted_str_len;
     const unsigned char *s;
     char *q;
     size_t nonpr;

     if (!str)
         return str;

     for (nonpr = 0, s = (unsigned char *)str; *s != '\0'; s++)
         if (!isprint(*s) || isspace(*s) || *s == '\\' || *s == '=') <========
             nonpr++;
     if (nonpr == 0)
         return str;

     if (high_water_alloc((void **)&quoted_str, &quoted_str_len,
                  (s - (unsigned char *)str) + nonpr * 3 + 1))
         return NULL;
     for (s = (unsigned char *)str, q = quoted_str; *s != '\0'; s++) {
         if (!isprint(*s) || isspace(*s) || *s == '\\' || *s == '=') { <========
             *q++ = '\\';
             *q++ = '0' + ((*s >> 6)    );
             *q++ = '0' + ((*s >> 3) & 7);
             *q++ = '0' + ((*s     ) & 7);
         } else
             *q++ = *s;
     }

Removing the checks isprint(*s) and *s == '\\' resolves the issue.
AD shares are associated with a doamin name and AD users/groups will be in the format Dom\user. Can someone explain why these checks are required?

PS. Please include me in the reply as I have not subscribed to this list.

Cheers!
- Anoop

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Hello!

Running RHEL5 U2 x86_64, with Samba & Winbind configure to work with Active-Directory env. Users & Groups in the Active-Directory are in hebrew and locale settings are correct (Hebrew/UTF-8) Posix Acls on file has special characters ('\' & hebrew chars) for the group/user names. In circumstances like this, where the user/group accounts come from an AD server, getfacl could be showing the "invalid" characters as octal. Example:

[root]# getfacl a.a
# file: a.a
# owner: GTS\134\327\236\327\225\327\251\327\225\327\237 # group: root
user::rw-
group::r-- group:GTS\134\327\251\327\236\327\225\327\250:-w- group:GTS\134\327\251\327\250\327\252:r-x mask::rwx
other::r--

Now - we are also have GPFS fs and when we run the mmgetacl command the Posix acls in Hebrew showed OK.

I am trying to work out a patch which fixes this and the issue seems to be here
const char *quote(const char *str)
[...]

Removing the checks isprint(*s) and *s == '\\' resolves the issue. AD shares are associated with a doamin name and AD users/groups will be in the format Dom\user. Can someone explain why these checks are required?

Quoted 3 lines

Removing the checks isprint(*s) and *s == '\\' resolves the issue. AD shares are associated with a doamin name and AD users/groups will be in the format Dom\user. Can someone explain why these checks are required?

The '\\' check is required so that backslashes will go through unquote(quote(string)) will come out correctly.

The isprint() check is a misguided attempt to do something reasonable with unprintable characters, but it leads to lots of problems with different encodings, so it should probably be removed. I'll fix that.

There also is a check for '=' in quote(). This is from the attr package which separates names and values with '=', but unnecessary in acl. quote() should probably be passed in a string of additional characters that need to be quoted so that it does the right thing in both packages.

Also, acl_entry_to_any_str() in libacl/__acl_to_any_text.c calls quote. I could not find the corresponding unquote calls for these.

If I wrap the above quote() calls with unquote(), the problem seems to be resolved. Attaching a patch for the same.