Re: Bug#645346: pax: FTBFS on hurd-i386
Hi!
On Fri, 2011-10-14 at 18:33:13 +0200, Svante Signell wrote:
> Package: pax
> Version: 20090728-2
> Severity: important
> Tags: patch
> Usertags: hurd
>
> Attached is a small patch to make pax-20090728 build under GNU/Hurd.
> The first part is using the already issued lstat(2) call to find out
> the size for a dynamic allocation of a character string, avoiding the
> MAXPATHLEN issue. Secondly the reaplath(3) call is conditionally
> replaced with canonicalize_file_name(3).
> diff -ur pax-20090728/file_subs.c pax-20090728.modified/file_subs.c
> --- pax-20090728/file_subs.c 2009-07-28 17:38:28.000000000 +0000
> +++ pax-20090728.modified/file_subs.c 2011-10-13 14:45:13.000000000 +0000
> @@ -374,8 +374,15 @@
> if (strcmp(NM_TAR, argv0) == 0 && Lflag) {
> while (lstat(nm, &sb) == 0 &&
> S_ISLNK(sb.st_mode)) {
> + target = malloc(sb.st_size + 1);
> + if (target == NULL) {
> + oerrno = ENOMEM;
> + syswarn(1, oerrno,
> + "Insufficient memory");
> + return(-1);
> + }
> len = readlink(nm, target,
> - sizeof target - 1);
> + strlen(target));
This is not right, target can contain garbage so strlen(target) could
return anything, or segfault. The correct value is the same used in
the malloc call.
> if (len == -1) {
> syswarn(0, errno,
> "cannot follow symlink %s in chain for %s",
[...]
You forgot a free on the return(0) case.
> diff -ur pax-20090728/tables.c pax-20090728.modified/tables.c
> --- pax-20090728/tables.c 2009-07-28 17:38:28.000000000 +0000
> +++ pax-20090728.modified/tables.c 2011-10-13 14:20:10.000000000 +0000
> @@ -1126,13 +1126,21 @@
> add_dir(char *name, struct stat *psb, int frc_mode)
> {
> DIRDATA *dblk;
> +#ifdef __GNU__
> + char *rp = NULL;
> +#else
> char realname[MAXPATHLEN], *rp;
> +#endif
After considering this, I think it would be a better option and way more
portable to use realpath(path, NULL) when _POSIX_VERSION >= 200809L.
regards,
guillem
Reply to: