I have encountered a problem with using OpenSSH 2.3.0p1 on 64-bit HP-UX 11.00 systems. This bug does not exhibit itself on any 32-bit HP-UX 11.00 or HP-UX 10.20 systems that I have built 2.3.0p1 on. OpenSSH 2.3.0p1 was built with HPs ANSI C compiler with OpenSSL 0.9.6 and zlib 1.1.3. The problem is with the call to vhangup(2) in sshd when interactive sessions are started. The problem does not occur for non-interactive sessions. When vhangup(2) is called, the following error is sent to syslog: Mar 13 21:14:37 hpux1100 vmunix: System call 76 (vhangup) was called in a kernel where the Mar 13 21:14:37 hpux1100 vmunix: type of at least one of its arguments is currently Mar 13 21:14:37 hpux1100 vmunix: unspecified. This is a problem that must be fixed by Mar 13 21:14:37 hpux1100 vmunix: the owner of the system call before the kernel can be Mar 13 21:14:37 hpux1100 vmunix: released. The process was pid 19386 (sshd). The child sshd receives as SIGSYS (signal 12), disconnects the client, dies, and then sends a SIGCHLD to the parent sshd. This prevents interactive sessions from beginning. I have reproduced this on two separate HP-UX 11.00 64-bit systems with locally built OpenSSH 2.3.0p1 packages. One of these systems has the latest March 2001 HP-UX 11.00 recommended bundle installed. Altering configure to set BROKEN_VHANGUP for HP-UX 11 prevents vhangup(2) from being used and builds usable OpenSSH daemons on my HP-UX 11.00 64-bit systems. Thanks for your hard work and dedication. -- Quentin C. Bracken UNIX Master Craftsman Application Development UNIX Services, Sabre Inc. quentin.bracken at sabre.com -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.mindrot.org/pipermail/openssh-unix-dev/attachments/20010313/5814077b/attachment.html
On Tue, 13 Mar 2001, Quentin Bracken wrote: : I have encountered a problem with using OpenSSH 2.3.0p1 on 64-bit HP-UX : 11.00 systems. This bug does not exhibit itself on any 32-bit HP-UX : 11.00 or HP-UX 10.20 systems that I have built 2.3.0p1 on. OpenSSH : 2.3.0p1 was built with HPs ANSI C compiler with OpenSSL 0.9.6 and zlib : 1.1.3. : : The problem is with the call to vhangup(2) in sshd when interactive : sessions are started. The problem does not occur for non-interactive : sessions. When vhangup(2) is called, the following error is sent to : syslog: : : Mar 13 21:14:37 hpux1100 vmunix: System call 76 (vhangup) was called in : a kernel where the : Mar 13 21:14:37 hpux1100 vmunix: type of at least one of its arguments : is currently : Mar 13 21:14:37 hpux1100 vmunix: unspecified. This is a problem that : must be fixed by : Mar 13 21:14:37 hpux1100 vmunix: the owner of the system call before the : kernel can be : Mar 13 21:14:37 hpux1100 vmunix: released. The process was pid 19386 : (sshd). i submitted a defect report on this in october. but i believe that for STREAMS ptys the vhangup equivalent is considered the grantpt(3C) and unlockpt(3C) pair. can someone confirm this? i propose the following: Index: configure.in ==================================================================RCS file: /var/cvs/openssh/configure.in,v retrieving revision 1.265 diff -u -r1.265 configure.in --- configure.in 2001/03/14 00:39:46 1.265 +++ configure.in 2001/03/14 17:20:09 @@ -71,7 +71,6 @@ AC_DEFINE(DISABLE_SHADOW) AC_DEFINE(IPV4_DEFAULT) AC_DEFINE(IP_TOS_IS_BROKEN) - AC_DEFINE(BROKEN_VHANGUP) AC_DEFINE(NO_X11_UNIX_SOCKETS) no_libsocket=1 no_libnsl=1 Index: defines.h ==================================================================RCS file: /var/cvs/openssh/defines.h,v retrieving revision 1.56 diff -u -r1.56 defines.h --- defines.h 2001/02/24 00:55:05 1.56 +++ defines.h 2001/03/14 17:20:15 @@ -406,9 +406,9 @@ # endif /* defined(HAVE_XATEXIT) */ #endif /* !defined(HAVE_ATEXIT) && defined(HAVE_ON_EXIT) */ -#if defined(HAVE_VHANGUP) && !defined(BROKEN_VHANGUP) +#if defined(HAVE_VHANGUP) && !defined(HAVE_DEV_PTMX) # define USE_VHANGUP -#endif /* defined(HAVE_VHANGUP) && !defined(BROKEN_VHANGUP) */ +#endif /* defined(HAVE_VHANGUP) && !defined(HAVE_DEV_PTMX) */ #ifndef GETPGRP_VOID # define getpgrp() getpgrp(0)
Kevin, HP-UX 11 does use /dev/ptmx and HAVE_DEV_PTMX is defined when
configure runs on an HP-UX 11.00 64-bit system. The suggested patch would
still result in a broken OpenSSH under 64-bit HP-UX 11.
I believe that the problem is actually in the vhangup(2) implementation in
the HP-UX 11.00 64-bit kernel. It runs without fail on HP-UX 11.00 32-bit
and 10.20. Reviewing the vmunix error message:
System call 76 (vhangup) was called in a kernel where the type of at
least one of its arguments is currently unspecified.
leads me to believe that there is a deficiency in the vhangup implementation
in the HP-UX 64-bit kernel itself. It is also worth mentioning that the
vhangup(2) system call is completely undocumented in any man page on 11.00 or
10.20 which I find somewhat suspicious.
Kevin Steves wrote:
> On Tue, 13 Mar 2001, Quentin Bracken wrote:
> : I have encountered a problem with using OpenSSH 2.3.0p1 on 64-bit HP-UX
> : 11.00 systems. This bug does not exhibit itself on any 32-bit HP-UX
> : 11.00 or HP-UX 10.20 systems that I have built 2.3.0p1 on. OpenSSH
> : 2.3.0p1 was built with HPs ANSI C compiler with OpenSSL 0.9.6 and zlib
> : 1.1.3.
> :
> : The problem is with the call to vhangup(2) in sshd when interactive
> : sessions are started. The problem does not occur for non-interactive
> : sessions. When vhangup(2) is called, the following error is sent to
> : syslog:
> :
> : Mar 13 21:14:37 hpux1100 vmunix: System call 76 (vhangup) was called in
> : a kernel where the
> : Mar 13 21:14:37 hpux1100 vmunix: type of at least one of its arguments
> : is currently
> : Mar 13 21:14:37 hpux1100 vmunix: unspecified. This is a problem that
> : must be fixed by
> : Mar 13 21:14:37 hpux1100 vmunix: the owner of the system call before the
> : kernel can be
> : Mar 13 21:14:37 hpux1100 vmunix: released. The process was pid 19386
> : (sshd).
>
> i submitted a defect report on this in october.
>
> but i believe that for STREAMS ptys the vhangup equivalent is considered
> the grantpt(3C) and unlockpt(3C) pair. can someone confirm this?
>
> i propose the following:
>
> Index: configure.in
> ==================================================================> RCS
file: /var/cvs/openssh/configure.in,v
> retrieving revision 1.265
> diff -u -r1.265 configure.in
> --- configure.in 2001/03/14 00:39:46 1.265
> +++ configure.in 2001/03/14 17:20:09
> @@ -71,7 +71,6 @@
> AC_DEFINE(DISABLE_SHADOW)
> AC_DEFINE(IPV4_DEFAULT)
> AC_DEFINE(IP_TOS_IS_BROKEN)
> - AC_DEFINE(BROKEN_VHANGUP)
> AC_DEFINE(NO_X11_UNIX_SOCKETS)
> no_libsocket=1
> no_libnsl=1
> Index: defines.h
> ==================================================================> RCS
file: /var/cvs/openssh/defines.h,v
> retrieving revision 1.56
> diff -u -r1.56 defines.h
> --- defines.h 2001/02/24 00:55:05 1.56
> +++ defines.h 2001/03/14 17:20:15
> @@ -406,9 +406,9 @@
> # endif /* defined(HAVE_XATEXIT) */
> #endif /* !defined(HAVE_ATEXIT) && defined(HAVE_ON_EXIT) */
>
> -#if defined(HAVE_VHANGUP) && !defined(BROKEN_VHANGUP)
> +#if defined(HAVE_VHANGUP) && !defined(HAVE_DEV_PTMX)
> # define USE_VHANGUP
> -#endif /* defined(HAVE_VHANGUP) && !defined(BROKEN_VHANGUP) */
> +#endif /* defined(HAVE_VHANGUP) && !defined(HAVE_DEV_PTMX) */
>
> #ifndef GETPGRP_VOID
> # define getpgrp() getpgrp(0)
--
Quentin C. Bracken
UNIX Master Craftsman
Application Development UNIX Services, Sabre Inc.
quentin.bracken at sabre.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.mindrot.org/pipermail/openssh-unix-dev/attachments/20010314/c0adaf3b/attachment.html
On Wed, 14 Mar 2001, Quentin Bracken wrote: : Kevin, HP-UX 11 does use /dev/ptmx and HAVE_DEV_PTMX is defined when : configure runs on an HP-UX 11.00 64-bit system. The suggested patch would : still result in a broken OpenSSH under 64-bit HP-UX 11. yes, this means USE_VHANGUP would not be defined. this is what we want. : I believe that the problem is actually in the vhangup(2) implementation in : the HP-UX 11.00 64-bit kernel. It runs without fail on HP-UX 11.00 32-bit : and 10.20. Reviewing the vmunix error message: : : System call 76 (vhangup) was called in a kernel where the type of at : least one of its arguments is currently unspecified. : : leads me to believe that there is a deficiency in the vhangup implementation : in the HP-UX 64-bit kernel itself. It is also worth mentioning that the : vhangup(2) system call is completely undocumented in any man page on 11.00 or : 10.20 which I find somewhat suspicious. i think it means that vhangup is obsolete on hp-ux, at least >9.X.