43519 - Classic NSPR: socket test asserts.

Status: RESOLVED WONTFIX

(NSPR :: NSPR, defect, P3)

Description

[Wan-Teh Chang]

NSPR 4.0
Classic NSPR
Red Hat Linux 6.0 or 6.1

The socket test asserts in UDP Client/Server test.  Here is
some info from the debugger.
---------------------------------------
UDP Client/Server Test - IPv4/Ipv4
Assertion failure: IsValidNetAddr(addr) == PR_TRUE, at prsocket.c:677

Program received signal SIGABRT, Aborted.
0x4008e4e1 in __kill () from /lib/libc.so.6
(gdb) where
#0  0x4008e4e1 in __kill () from /lib/libc.so.6
#1  0x4008e156 in raise (sig=6) at ../sysdeps/posix/raise.c:27
#2  0x4008f868 in abort () at ../sysdeps/generic/abort.c:88
#3  0x4002b560 in PR_Assert (s=0x40065280 "IsValidNetAddr(addr) == PR_TRUE",
    file=0x40065240 "prsocket.c", ln=677) at prlog.c:448
#4  0x4004be0c in SocketSendTo (fd=0x8053c60, buf=0x8058000, amount=128,
    flags=0, addr=0x50381de4, timeout=4294967295) at prsocket.c:677
#5  0x400282d3 in PR_SendTo (fd=0x8053c60, buf=0x8058000, amount=128, flags=0,
    addr=0x50381de4, timeout=4294967295) at priometh.c:223
#6  0x804a199 in UDP_Server (arg=0x8053c40) at socket.c:581
#7  0x400521ce in _PR_UserRunThread () at pruthr.c:489
(gdb) f 4
#4  0x4004be0c in SocketSendTo (fd=0x8053c60, buf=0x8058000, amount=128,
    flags=0, addr=0x50381de4, timeout=4294967295) at prsocket.c:677
677             PR_ASSERT(IsValidNetAddr(addr) == PR_TRUE);
(gdb) print *addr
$1 = {raw = {family = 0, data = '\000' <repeats 13 times>}, inet = {
    family = 0, port = 0, ip = 0, pad = "\000\000\000\000\000\000\000"},
  ipv6 = {family = 0, port = 0, flowinfo = 0, ip = {_S6_un = {
        _S6_u8 = '\000' <repeats 15 times>, _S6_u16 = {0, 0, 0, 0, 0, 0, 0,
          0}, _S6_u32 = {0, 0, 0, 0}, _S6_u64 = {0, 0}}}, scope_id = 0},
  local = {family = 0, path = '\000' <repeats 103 times>}}
(gdb)
---------------------------------------------

The 'addr' was just stored by a prior PR_RecvFrom call.

This is a Linux-only problem and only happens with classic NSPR.

Comment 1

[castor]

The problem here is that accept(2) cannot necessarily provide valid
a valid sockaddr if the connection is closed before the accept.

On NetBSD and FreeBSD, they decided to return a valid descriptor to avoid
race conditions when doing a select on the descriptor.

I think NetBSD returns garbage in the sockaddr, and in new releases of
FreeBSD I think they decided to set the sockaddr_len to 0 to signal
that the data is not valid.  Unfortunately, that will still cause
the assertion to fail in NSPR

Comment 2

[Wan-Teh Chang]

[e-mail]:

Were you commenting on this bug?  This bug is
about an assertion failure in one of our UDP
test, which does not use accept().

Comment 3

[leger]

Adding classic keyword.

Comment 4

[Wan-Teh Chang]

This bug has nothing to do with the classic Skin/Theme
of Mozilla.

Comment 5

[Thomas]

Hey Wan-Teh,
Does this only occur with Red Hat? 
(Was this tried with a more recent version?)

Comment 6

[Wan-Teh Chang]

Marked the bug WONTFIX.  On current and future Linux
releases we will only use the pthreads version of NSPR.