we can append to it. Modify the code so that file_write is re-entrant,
even in the gnu long name/link hack.
The old code assumed that the buffer already contained the necessary
blocks to satisfy the read file request that contained the long
filename. This is not always the case, specially when we are dealing
with pipes which do shorter reads, thus having more probability
that a long file name will fall across a buffer boundary.
To reproduce, create a tar archive with a lot of gnu-long-names
(pkgsrc/devel/libsigc++2 is a good example), do a tar -tf to get
a list of filenames, compress it and do a tar -tzf to get another
list of the filenames. Notice that the two lists differ.
_NETBSD_SOURCE as this makes cross building from older/newer versions of
NetBSD harder, not easier (and also makes the resulting tools 'different')
Wrap all required code with the inclusion of nbtool_config.h, attempt to
only use POSIX code in all places (or when reasonable test w. configure and
provide definitions: ala u_int, etc).
Reviewed by lukem. Tested on FreeBSD 4.9, Redhat Linux ES3, NetBSD 1.6.2 x86
NetBSD current (x86 and amd64) and Solaris 9.
Fixes PR's: PR#17762 PR#25944
As mentioned in the previous commit, the switch statement in the longlink()
needed simplification and it was a bit incorrect. Only depend on the passed
type to determine what kind of gnu longlink to produce. Don't try to deduce
it from the archive file type.
- always put the @LongLink tag on the name, not the long-link name.
- pass in what type of long name record we want to create; one for long-name
or long-link name.
XXX: We should get rid of the switch too.
The problem is with the program that generates the tar file:
-rwxrwxrwx 1 root wheel 0 Feb 8 16:46 faad2/aacDECdrop/
It creates directory nodes without the 'd' bit set, so that pax thinks
they are files and does the temporary name and dance with them. Added
code to detect this condition, warn about it, and work around it.
because alternation (|) isn't available in "pax -s" RE's, we have to pass
four (yes 4!) different patterns:
.*\/<pattern>$
.*\/<pattern>\/.*
^<pattern>$
^<pattern>\/.*
instead of the more elegant
(^|.*\/)<pattern>($|\/.*)
fixes a problem reported by simonb.
once it has been instantiated correctly, rename it to desired name. This
prevents the problem of partially created files being accessed before they
are complete. If said file is a shared library, that can cause ramdon core
dumps.
(like executing "pax -Z" by itself), this caused a shr of 32 bits, which is
undefined behavior (C99) if the variable is 32 bits wide, too. Also solves
a problem where the flgch array could be indexed out of bounds.
Thanks to uwe@ and lha@ for their suggestions... I just found the bug :p
* Rename "config.h" to "nbtool_config.h" and
HAVE_CONFIG_H to HAVE_NBTOOL_CONFIG_H.
This makes in more obvious in the source when we're using
tools/compat/config.h versus "standard autoconf" config.h
* Consistently move the inclusion of nbtool_config.h to before
<sys/cdefs.h> so that the former can provide __RCSID() (et al),
and there's no need to protect those macros any more.
These changes should make it easier to "tool-ify" a program by adding:
#if HAVE_NBTOOL_CONFIG_H
#include "nbtool_config.h"
#endif
to the top of the source files (for the general case).
