[Bug 20368] New: sockaddr_in6 wrong size; causes chromium's net_unittests HostResolverImplTest.NumericIPv6Address test case to fail

http://bugs.winehq.org/show_bug.cgi?id=20368

Summary: sockaddr_in6 wrong size; causes chromium's
                    net_unittests HostResolverImplTest.NumericIPv6Address
                    test case to fail
           Product: Wine
           Version: 1.1.31
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Keywords: download, source, testcase
          Severity: normal
          Priority: P2
         Component: winsock
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]

Running chromium's base_unittests.exe fails all over the place
unless it can find the source tree, but even if you put the
source tree where it wants it, it fails various tests.  The
first one in the list is 
HostResolverImplTest.NumericIPv6Address 
which fails with
[11:71:1014/112257:84976129:FATAL:address_list.cc(76)] Check failed:
sizeof(sockaddr_in6) == info->ai_addrlen (28 vs. 24)

Here's the failing bit of the testcase, extracted into a nearly plain C
program:

#undef NDEBUG
#include <ws2tcpip.h>
#include <stdio.h>
#include <assert.h>
#pragma comment(lib, "ws2_32.lib")
int main(int argc, char **argv)
{
    WSADATA wsaData;
    int err;
    struct addrinfo *ai = NULL;
    struct addrinfo hints;
    err = WSAStartup(MAKEWORD(2, 2), &wsaData);
    assert(err == 0);
    ZeroMemory( &hints, sizeof(hints) );
    hints.ai_family = AF_UNSPEC;
    hints.ai_socktype = SOCK_STREAM;
    hints.ai_protocol = IPPROTO_TCP;
    err = getaddrinfo("2001:db8::1", "10000", &hints, &ai);
    assert(err == 0);
    assert(ai->ai_addrlen == sizeof(struct sockaddr_in6));
    freeaddrinfo(ai);
    WSACleanup();
    return 0;
}
Compiled on Windows, this succeeds on Windows but fails on Wine.

[Bug 20368] sockaddr_in6 wrong size; causes chromium's net_unittests HostResolverImplTest.NumericIPv6Address test case to fail

http://bugs.winehq.org/show_bug.cgi?id=20368

--- Comment #1 from Dan Kegel   2009-10-14 14:24:54 ---
Created an attachment (id=24130)
 --> (http://bugs.winehq.org/attachment.cgi?id=24130)
Compiled test case.  Aborts on Wine.

[Bug 20368] sockaddr_in6 wrong size; causes chromium's net_unittests HostResolverImplTest.NumericIPv6Address test case to fail

http://bugs.winehq.org/show_bug.cgi?id=20368

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

[Bug 20368] sockaddr_in6 wrong size; causes chromium's net_unittests HostResolverImplTest.NumericIPv6Address test case to fail

http://bugs.winehq.org/show_bug.cgi?id=20368

--- Comment #2 from Juan Lang   2009-10-15 11:41:49 ---
(In reply to comment #0)
> [11:71:1014/112257:84976129:FATAL:address_list.cc(76)] Check failed:
> sizeof(sockaddr_in6) == info->ai_addrlen (28 vs. 24)

It's not clear from this which one it is expected to be, 24 or 28.  24 is
possible for struct sockaddr_in6 on Windows if it's compiled a certain way with
MS headers, I believe, as this is the size of struct sockaddr_in6_old (see
ws2ipdef.h).  24 is also a possible size for struct sockaddr_in6 on *ix, see in
configure.ac:
dnl Check for sin6_scope_id
AC_CHECK_MEMBERS([struct sockaddr_in6.sin6_scope_id],,,

It's a bug in either case, but I'm curious where it lies.  Could you look at
config.log to see whether your struct sockaddr_in6 has a sin6_scope_id?

[Bug 20368] sockaddr_in6 wrong size; causes chromium's net_unittests HostResolverImplTest.NumericIPv6Address test case to fail

http://bugs.winehq.org/show_bug.cgi?id=20368

--- Comment #3 from Juan Lang   2009-10-15 22:35:40 ---
This fails to start for me on Vista too, btw, complaining about the assembly.

[Bug 20368] sockaddr_in6 wrong size; causes chromium's net_unittests HostResolverImplTest.NumericIPv6Address test case to fail

http://bugs.winehq.org/show_bug.cgi?id=20368

--- Comment #4 from Juan Lang   2009-10-16 13:37:19 ---
Okay, I just tried it myself and it fails on Wine.  My config.log shows that
struct sockaddr_in6 has the sin6_scope_id field, and sizeof(struct
sockaddr_in6) == 28 in Linux.  Therefore, the app is compiled not to expect the
scope id field.

[Bug 20368] sockaddr_in6 wrong size; causes chromium's net_unittests HostResolverImplTest.NumericIPv6Address test case to fail

http://bugs.winehq.org/show_bug.cgi?id=20368

--- Comment #5 from Juan Lang   2009-10-16 16:09:55 ---
What's odd is that sizeof(struct sockaddr_in6) on Wine, compiled with our own
headers, is also 28.  So the question is, how is winsock supposed to know the
size the app was compiled with?  I suspect this bug may well be invalid.  What
systems have you tried it on?

[Bug 20368] sockaddr_in6 wrong size; causes chromium's net_unittests HostResolverImplTest.NumericIPv6Address test case to fail

http://bugs.winehq.org/show_bug.cgi?id=20368

--- Comment #6 from Dan Kegel   2009-10-16 16:45:16 ---
The test was compiled on vista home with vs 2008 express, nothing special.

[Bug 20368] sockaddr_in6 wrong size; causes chromium's net_unittests HostResolverImplTest.NumericIPv6Address test case to fail

http://bugs.winehq.org/show_bug.cgi?id=20368

--- Comment #7 from Juan Lang   2009-10-16 17:44:28 ---
Well, it doesn't run for me on Windows XP SP3, either, so I think this test is
invalid.

[Bug 20368] sockaddr_in6 wrong size; causes chromium's net_unittests HostResolverImplTest.NumericIPv6Address test case to fail

http://bugs.winehq.org/show_bug.cgi?id=20368

--- Comment #8 from Juan Lang   2009-10-16 17:54:13 ---
Oh, right, it fails on XP because I don't have IPv6 enabled.  At least the
first assert needs to be changed to an if.  I modified it to check the size of
sockaddr_in6, and it's 28 on XP.

So, I *believe* this is invalid.  I'll upload my own compiled version in a sec.
 It fails for me on XP because getaddrinfo fails, but I'll have another look
with Wine.

[Bug 20368] sockaddr_in6 wrong size; causes chromium's net_unittests HostResolverImplTest.NumericIPv6Address test case to fail

http://bugs.winehq.org/show_bug.cgi?id=20368

--- Comment #9 from Juan Lang   2009-10-16 17:56:10 ---
Created an attachment (id=24175)
 --> (http://bugs.winehq.org/attachment.cgi?id=24175)
Another compiled version

Compiled with VS2008 express.  I also added a couple printfs to the source:
    printf("sizeof(struct sockaddr_in6) = %d\n", sizeof(struct sockaddr_in6));
and:
    printf("ai_addrlen = %d\n", ai->ai_addrlen);
The second is never reached on WinXP because of lacking IPv6 support, but the
first prints out:
sizeof(struct sockaddr_in6) = 28

[Bug 20368] sockaddr_in6 wrong size; causes chromium's net_unittests HostResolverImplTest.NumericIPv6Address test case to fail

http://bugs.winehq.org/show_bug.cgi?id=20368

Juan Lang  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #24175|0                           |1
        is obsolete|                            |

--- Comment #10 from Juan Lang   2009-10-16 18:48:43 ---
Created an attachment (id=24176)
 --> (http://bugs.winehq.org/attachment.cgi?id=24176)
Another compiled version (release build)

Oops, the last one was a debug build, which wouldn't run on Wine.  The attached
release build does, providing you run 'winetricks vcrun2008' first.  This one
produces the following output for me under Wine:

sizeof(struct sockaddr_in6) = 28
ai_addrlen = 24

I finally understand the issue:  sizeof(struct sockaddr_in6) is 28, that is, it
includes the scope_id, yet getaddrinfo is returning an address of size 24, even
though sizeof(struct sockaddr_in6) is 28 on my Linux system too.

This bug is valid, pardon my noise from earlier.

[Bug 20368] sockaddr_in6 wrong size; causes chromium's net_unittests HostResolverImplTest.NumericIPv6Address test case to fail

http://bugs.winehq.org/show_bug.cgi?id=20368

--- Comment #11 from Juan Lang   2009-10-16 18:59:45 ---
Patch sent:
http://www.winehq.org/pipermail/wine-patches/2009-October/080114.html

Like the existing socket.c code, the app will still fail when run on systems
that don't have a sin6_scope_id field as part of struct sockaddr_in6.  I'm not
sure whether this is incorrect, and a default scope id should be returned, or
if the chromium test is invalid.  I'll defer that because it appears to be a
corner case anyway.

[Bug 20368] sockaddr_in6 wrong size; causes chromium's net_unittests HostResolverImplTest.NumericIPv6Address test case to fail

http://bugs.winehq.org/show_bug.cgi?id=20368

Juan Lang  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED

--- Comment #12 from Juan Lang   2009-10-19 09:59:29 ---
Fixed by commit d1963ee2f19c931cbd6d7c039081440d08cb9011.

[Bug 20368] sockaddr_in6 wrong size; causes chromium's net_unittests HostResolverImplTest.NumericIPv6Address test case to fail

http://bugs.winehq.org/show_bug.cgi?id=20368

Alexandre Julliard  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |CLOSED

--- Comment #13 from Alexandre Julliard   2009-10-23 13:19:39 ---
Closing bugs fixed in 1.1.32.