[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Bug-tar] xattr support on Linux

From: Tim Kientzle
Subject: Re: [Bug-tar] xattr support on Linux
Date: Fri, 5 Oct 2012 21:37:04 -0700

On Oct 5, 2012, at 5:30 AM, Joerg Schilling wrote:

> Joerg Schilling <address@hidden> wrote:
>>> As mentioned there, I'm also missing documentation for
>>> star's xattr format.  In particular, I'm not clear how it handles
>>> non-ASCII bytes in the attributes name.
>> Thank you for this hint, I thought that I did document everything.....
> Mmmmm, could you explain, what information you are missing in this text:
>     SCHILY.xattr.attr
>          A POSIX.1-2001 coded version of the Linux extended file
>          attributes.    Linux   extended   file  attributes  are
>          name/value pairs. Every attribute  name  results  in  a
>          SCHILY.xattr.name  tag  and  the  value of the extended
>          attribute is used as  the  value  of  the  POSIX.1-2001
>          header  tag.  Note that this way of coding is not port-
>          able across  platforms.   A  version  for  BSD  may  be
>          created  but  Solaris  includes  far more features with
>          extended attribute files than Linux does.
>          A future version  of  star  will  implement  a  similar
>          method  as  the  tar program on Solaris currently uses.
>          When    this    implementation    is     ready,     the
>          SCHILY.xattr.name  feature may be removed in favor of a
>          truly portable  implementation  that  supports  Solaris
>          also.

What do you do if the name includes byte 0x3d?

> BTW: converting name/value pairs with unknown meaning into something 
> different (as UTF-8) may cause problems …

This is why bsdtar URL-encodes the bytes of the xattr
property name when building the pax extended header
record name.  The result is ASCII and therefore won't
be damaged if the reader does any sort of character set

> … and is not needed as the meta data in extended 
> tar headers is binary clean due to it's length tag.

POSIX.1-2008 is very clear:
  "The <keyword> fields can be any UTF-8 characters."

The *value* is allowed to be binary in specific instances
(gname, uname, path, and link path in the presence of a
hdrcharset header), but the keyword should always be


reply via email to

[Prev in Thread] Current Thread [Next in Thread]