211512 - PR_Connect() fails on panther beta

Status: RESOLVED FIXED

(NSPR :: NSPR, defect)

Description

[Brian Ryner (not reading)]

As wtc and I discussed, PR_Connect() fails on the Panther beta because the OS
connect() returns EINVAL.  The problem appears to be that the IPV6_V6ONLY socket
option is enabled by default for AF_INET6 sockets.

Comment 1

[Brian Ryner (not reading)]

Attachment: patch from wtc

This is a patch wtc sent to me.

Comment 2

[Brian Ryner (not reading)]

Comment on attachment 126972 [details] [diff] [review]
patch from wtc

r=bryner.  One minor thing, I noticed that the prototype for setsockopt has the
4th argument as a void*, not char*.

wtc, can you check this in on the trunk and client branch, if you think it's
ok?

Comment 3

[Simon Fraser [no longer active]]

Have we filed a bug with Apple on that option being on by default?

Comment 4

[Brian Ryner (not reading)]

Radar bug 3314011.

Comment 5

[Wan-Teh Chang]

Attachment: patch from wtc, v2

This patch performs a test during NSPR initialization
to find out whether we need to turn off the IPV6_V6ONLY
option.  This way we can save the setsockopt calls on
the Mac OS X releases that turn off the IPV6_V6ONLY option
by default per RFC 3493.

Please review and test this patch.

Comment 6

[Wan-Teh Chang]

Comment on attachment 127274 [details] [diff] [review]
patch from wtc, v2

The only concern I have about this patch is that
the use of IPV6_V6ONLY may break the build on Mac
OS X 10.1.  I don't know if that macro is defined
on 10.1.  Anyone knows?

Comment 7

[Brian Ryner (not reading)]

Comment on attachment 127274 [details] [diff] [review]
patch from wtc, v2

Could we possibly test this just once, instead of every time we create a
socket?

Comment 8

[Brian Ryner (not reading)]

Never mind my last comment, I saw the socket() nearby and figured this was
during socket creation.

Comment 9

[Brian Ryner (not reading)]

Comment on attachment 127274 [details] [diff] [review]
patch from wtc, v2

r=bryner.  I tested this and it works fine.  (The code indentation of the code
in _pr_test_ipv6_socket looks a little off though).

Comment 10

[Wan-Teh Chang]

Attachment: Patch that was checked in

This is the same as patch v2 except that I fixed the
indentation problem and defined what "Panther" is in
the comment.

Comment 11

[Wan-Teh Chang]

Comment on attachment 127321 [details] [diff] [review]
Patch that was checked in

This patch has been checked into the NSPR tip (NSPR 4.4.1)
and the NSPRPUB_PRE_4_2_CLIENT_BRANCH (Mozilla 1.5a).

Comment 12

[Dylan Foley]

As suspected the patch causes the build to fail on OS X 10.1.5.

From my build attempt:
...
ptio.c: In function `pt_Connect':
ptio.c:1530: warning: unused variable `addrp'
ptio.c: In function `pt_Bind':
ptio.c:1715: warning: unused variable `addrp'
ptio.c: In function `pt_SendTo':
ptio.c:1965: warning: unused variable `addrp'
ptio.c: In function `_pr_test_ipv6_socket':
ptio.c:3397: `IPV6_V6ONLY' undeclared (first use in this function)
ptio.c:3397: (Each undeclared identifier is reported only once
ptio.c:3397: for each function it appears in.)
ptio.c: In function `PR_Socket':
ptio.c:3459: `IPV6_V6ONLY' undeclared (first use in this function)
ptio.c: In function `_pr_poll_with_poll':
ptio.c:3723: warning: `start' might be used uninitialized in this function
ptio.c: In function `PR_Select':
ptio.c:4791: warning: `start' might be used uninitialized in this function
make[7]: *** [ptio.o] Error 1
make[7]: Leaving directory `/usr/local/cvsrep/moz/mozilla/nsprpub/pr/src/pthreads'
...

Comment 13

[Dylan Foley]

Attachment: patch by df for 10.1

This patch gets past the problem on my machine. (I know nothing of C/C++ so
it's probably horribly broken. Please forgive - thought it might still be
helpful.) It seems to do the trick, just adds a few checks to see if
defined(IPV6_V6ONLY). My build attempts still fails, though much later and in a
completely unrelated area.

Comment 14

[Wan-Teh Chang]

Attachment: Patch to define IPV6_V6ONLY on 10.1

Could you test this patch on Mac OS X 10.1?  Thanks.

Comment 15

[Dylan Foley]

This patch works for me on 10.1. Thanks!

Comment 16

[Wan-Teh Chang]

Attachment: Second patch that was checked in

This is the second patch that was checked in.
I also fixed the build problem for Mac OS X
deployment target 10.2 or higher.

Comment 17

[Wan-Teh Chang]

Both patches have been checked in.

Mozilla 1.5a only has the first patch.  So it
won't build on Mac OS X 10.1 or for Mac OS X
deployment target 10.2 or higher.

Mozilla 1.5b will have both patches.