bug: tar doesnt understand pax extended headers

[Nick Owens <mischief@offblast.org>]

[-- Attachment #1: Type: text/plain, Size: 1183 bytes --]

tar(1) doesn't understand pax extended headers. attached is a patch
which will simply ignore pax headers.

see http://pubs.opengroup.org/onlinepubs/9699919799/utilities/pax.html
for a description of the headers and their format.

in the above documentation the type is referred to by 'typeflag', and
in tar.c, linkflag. pax extended headers have two types, 'x' and 'g'.

the main thing that these seem to be useful for are file names which
are longer than whats normally representable.

right now, tar(1) will treat these pax extended headers as normal
"files" and extract the headers to disk, cluttering the fs.

for now we could simply ignore them (easy), but it would be ideal to
read them to find the long path names. the only trouble with this is
that one needs to parse all of the pax 'size key=value\n' attributes
in the pax extended header, finding the 'path' key and using its value
as the new path. there's already some provision for this in tar.c's
"getname" function. there's also no fixed size for these pax
attributes, so filenames can (in theory) be of unlimited length.

if anyone wants to pick this up, the "archive/tar" package in go seems
like a decent reference.

[-- Attachment #2: tar.c.patch --]
[-- Type: text/x-patch, Size: 573 bytes --]

diff --git a/sys/src/cmd/tar.c b/sys/src/cmd/tar.c
--- a/sys/src/cmd/tar.c
+++ b/sys/src/cmd/tar.c
@@ -78,6 +78,8 @@ enum {
 
 	LF_LONGNAME =	'L',		/* GNU extenstion */
 	LF_LONGLINK = 	'K',
+	LF_PAXX		=	'x',
+	LF_PAXG		=	'g',
 };
 
 #define islink(lf)	(isreallink(lf) || issymlink(lf))
@@ -1061,6 +1063,9 @@ openfname(Hdr *hp, char *fname, int dir,
 	case LF_FIFO:
 		fprint(2, "%s: can't make fifo %s\n", argv0, fname);
 		break;
+	case LF_PAXX:
+	case LF_PAXG:
+		break;
 	default:
 		if (!keepexisting || access(fname, AEXIST) < 0) {
 			int rw = (dir? OREAD: OWRITE);