[Gaim-patches] [ gaim-Patches-1030243 ] DNS child bad IPC on 64-bit Big endian

[SourceForge.net <no...@so...>]

Patches item #1030243, was opened at 2004-09-18 01:46
Message generated for change (Comment added) made by nosnilmot
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=300235&aid=1030243&group_id=235

Category: segfault
Group: None
>Status: Pending
Resolution: None
Priority: 5
Submitted By: rian hunter (chrono86)
Assigned to: Ethan Blanton (eblanton)
Summary: DNS child bad IPC on 64-bit Big endian

Initial Comment:
in the host_resolved function (parent) of src/proxy.c,
it receives the dns information like this
---------
int rc, err; (4 bytes on LP64)
size_t addrlen; (8 bytes on some LP64 unices)

operation, size
read(childfd, &err, sizeof(err)); 4
read(childfd, &addrlen, sizeof(addrlen)); 8
read(childfd, &addr, addrlen); varies
read(childfd, &addrlen, sizeof(addrlen)); 8
if addrlen == 0
 then stop;

--------
the child works like this:
-----
const int zero = 0; (4 bytes LP64)
size_t addrlen; (8 bytes on some LP64 unices);

operation, size
write(childfd, &zero, sizeof(zero)); 4
write(childfd, &addrlen, sizeof(addrlen)); 8
write(childfd, &addr, addrlen); varies
write(childfd, &zero, sizeof(zero)); 4


--------
in the last read of host_resolved only the first 4
bytes are read, and in big endian, these are the MSB
then the lower four bytes aren't cleared and addrlen
will retain it's lower 32-bit value, and not evaluate
to zero when it should.

my patch changes err (in host resolved) and zero (in
the child) to be size_t datatype. because addrlen in
the child must be size_t bytes, so zero must be size_t,
and if zero is size_t, then err must be size_t.

----------------------------------------------------------------------

>Comment By: Stu Tomlinson (nosnilmot)
Date: 2005-03-21 10:00

Message:
Logged In: YES 
user_id=309779

I think this is the same as what patch 1162827 fixed, right?

----------------------------------------------------------------------

Comment By: rian hunter (chrono86)
Date: 2004-10-02 14:52

Message:
Logged In: YES 
user_id=801658

This bug doesn't only affect Big Endian 64-Bit machines, but
all 64-Bit machines if that was unclear. The bug is just
more obvious on a 64-bit Big Endian machine.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=300235&aid=1030243&group_id=235