getpwent bug?

[Jens Rehsack]

2010/7/16 Dan Nelson <dnelson at allantgroup.com>:
> In the last episode (Jul 16), Ashish SHUKLA said:
>> Dan Nelson writes:
>> > In the last episode (Jul 15), Jens Rehsack said:
>> >> Hi all,
>> >>
>> >> I detected an issue with getpwent on my FreeBSD test box:
>> >>
>> >> perl -MData::Dumper -e 'my @e = getpwent(); print Dumper(\@e); endpwent(); @e = getpwent(); print Dumper(\@e); endpwent(); @e = getpwent(); print Dumper(\@e); endpwent();'
>> >> $VAR1 = [ 'root', '', 0, 0, 0, '', 'Charlie &', '/root', '/bin/csh', 0 ];
>> >> $VAR1 = [ 'toor', '*', 0, 0, 0, '', 'Bourne-again Superuser', '/root', '', 0 ];
>> >> $VAR1 = [ 'daemon', '*', 1, 1, 0, '', 'Owner of many system processes', '/root', '/usr/sbin/nologin', 0 ];
>> >>
>> >> I'm using FreeBSD waldorf.muppets.liwing.de 7.3-PRERELEASE FreeBSD 7.3-PRERELEASE #0: Fri Mar 12 11:31:18 UTC 2010 root at waldorf.muppets.liwing.de:/usr/obj/usr/src/sys/WALDORF  amd64
>>
>> > The above output looks perfect, and should match the top three lines in
>> > /your etc/passwd files.
>>
>> Well, OP is also invoking 'endpwent()' after every 'getpwent()' invocation
>> which according to GNU/Linux's glibc and NetBSD's libc (as OP mentioned)
>> should rewind the position in passwd database to the beginning.
>
> Ah. I missed the endpwent calls.

Was difficult for me to format the single liner ;)

>> To me it definitely looks like a bug in FreeBSD's getpw*() family of
>> functions.
>>
>> As tested using sysutils/lsof, in the following program in FreeBSD, the
>> descriptor corresponding to '/etc/pwd.db' is closed on endpwent(3) but
>> position in database is never rewinded as shown in the output.
>
> It looks like the *pwent functions keep an internal counter that endpwent
> doesn't reset.

Could you please take a look to my other mail (getgrent related) - there seems
another bug ...

> Try the following patch:

Can I do this without a full world rebuild? (I do not develop in FBSD actively).
Otherwise I recommend (the test case was in OP) that someone with a
separate test box tries it out and commit it etc.

I had to develop a workaround for all other boxes anyway.

Thank you very much,
Jens

> Index: gen/getpwent.c
> ===================================================================
> --- gen/getpwent.c      (revision 210157)
> +++ gen/getpwent.c      (working copy)
> @@ -794,6 +794,7 @@ files_setpwent(void *retval, void *mdata, va_list
>                        (void)st->db->close(st->db);
>                        st->db = NULL;
>                }
> +               st->keynum = 0;
>                break;
>        default:
>                break;
>
>
> --
>        Dan Nelson
>        dnelson at allantgroup.com
>