question about kcvs

question about kcvs

[Tom Tromey]

I'm getting an error from cvs.  I use kerberos.  I've gotten this
before but I forgot what it means and how it was fixed.

creche. cvs update -A
cvs: ../../../devo/cvssrc/src/client.c:3937: connect_to_gserver: Assertion `need <= sizeof buf' failed.


Any clues?

Tom

Re: question about kcvs

[Jason Molenda]

On Fri, May 19, 2000 at 04:05:13PM -0600, Tom Tromey wrote:
> I'm getting an error from cvs.  I use kerberos.  I've gotten this
> before but I forgot what it means and how it was fixed.
> 
> creche. cvs update -A
> cvs: ../../../devo/cvssrc/src/client.c:3937: connect_to_gserver: Assertion `need <= sizeof buf' failed.


Isn't this CVS's way of saying "invalid ticket"?  Maybe your clock
slew is too high? (or maybe sourceware's is?)

I forget exactly when I've seen these in the past, but it's something
along these lines.  I'd start by destroying your current ticket
and getting a new one.

J

Re: question about kcvs

[Tom Tromey]

Jason> Isn't this CVS's way of saying "invalid ticket"?  Maybe your clock
Jason> slew is too high? (or maybe sourceware's is?)

That's what I thought.
But the skew is about 2 seconds.
And "rsh sourceware -x" works.

Jason> I forget exactly when I've seen these in the past, but it's
Jason> something along these lines.  I'd start by destroying your
Jason> current ticket and getting a new one.

Didn't work.

Tom

Re: question about kcvs

[Ian Lance Taylor]

   From: Tom Tromey <tromey@cygnus.com>
   Date: 19 May 2000 16:05:13 -0600

   I'm getting an error from cvs.  I use kerberos.  I've gotten this
   before but I forgot what it means and how it was fixed.

   creche. cvs update -A
   cvs: ../../../devo/cvssrc/src/client.c:3937: connect_to_gserver: Assertion `need <= sizeof buf' failed.


   Any clues?

The CVS server is printing an error message.  The CVS client thinks
the first two characters of the error message are the number of bytes
it needs to read.  It is complaining because its static buffer isn't
large enough.

Put the CVS client under gdb and step through the function and frob
the need variable to read in the rest of the message to see what the
server is saying.

Usually you need to kinit again for some reason.

Ian

Re: question about kcvs

[Ian Lance Taylor]

   From: Tom Tromey <tromey@cygnus.com>
   Date: 19 May 2000 16:05:13 -0600

   I'm getting an error from cvs.  I use kerberos.  I've gotten this
   before but I forgot what it means and how it was fixed.

   creche. cvs update -A
   cvs: ../../../devo/cvssrc/src/client.c:3937: connect_to_gserver: Assertion `need <= sizeof buf' failed.


   Any clues?

This started happening to me too, so I looked at it.  The CVS server
on sourceware is printing
    cvs [pserver aborted]: could not acquire GSSAPI server credentials

This means that the gss_acquire_cred function run by the CVS server is
failing.  I believe this means that the cvs/egcs.cygnus.com ticket has
disappeared from the Kerberos database, or that it has disappeared
from the ticket file on sourceware.

I took a look at /etc/v5srvtab.  I see that it has an entry for
cvs/sourceware.cygnus.com.  I've never quite followed the
canonicalization strategy for Kerberos.  However, my vague
understanding is that Kerberos will canonicalize the host name.
Currently the canonical name for the sourceware machine is
egcs.cygnus.com.  Is there any chance that this changed recently?  Was
the canonical name sourceware.cygnus.com until recently?  If it was,
then the fix is to change the canonical name back to
sourceware.cygnus.com.

If that did not change recently, then somebody who knows the
CYGNUS.COM Kerberos administrator password needs to confirm that the
cvs/sourceware.cygnus.com Kerberos ticket still exists, and that it is
version number 2.

Ian

Re: question about kcvs

[Jason Molenda]

On Fri, May 19, 2000 at 08:10:18PM -0700, Ian Lance Taylor wrote:

> This started happening to me too, so I looked at it.  

Thanks Ian,

> Currently the canonical name for the sourceware machine is
> egcs.cygnus.com.  Is there any chance that this changed recently?  

Hum, that seems wrong.  Jeff/JimK, did one of you change anything
like this?  Could it have been the IS staff in SV?  I guess there
are contractors doing the Unix work in Sunnyvale now - could one
of them tweaked something?

When we set up the new box in August, it was configured from the
start as "sourceware.cygnus.com", with egcs.cygnus.com and gcc.gnu.org
accepted as aliases.

Jason

Re: question about kcvs

[Chris Faylor]

On Fri, May 19, 2000 at 08:59:47PM -0700, Jason Molenda wrote:
>On Fri, May 19, 2000 at 08:10:18PM -0700, Ian Lance Taylor wrote:
>> This started happening to me too, so I looked at it.  
>
>Thanks Ian,
>
>> Currently the canonical name for the sourceware machine is
>> egcs.cygnus.com.  Is there any chance that this changed recently?  
>
>Hum, that seems wrong.  Jeff/JimK, did one of you change anything
>like this?  Could it have been the IS staff in SV?  I guess there
>are contractors doing the Unix work in Sunnyvale now - could one
>of them tweaked something?
>
>When we set up the new box in August, it was configured from the
>start as "sourceware.cygnus.com", with egcs.cygnus.com and gcc.gnu.org
>accepted as aliases.

Rumor has it that people are receiving calls from sysadmins asking for
details on five month old clarify CRs.  I don't suppose that anyone
filed a case asking to change the name to egcs.cygnus.com or something
like that at some point in the distant past?

Maybe one of the sourceware sysadmins should contact the Sunnyvale
sysadmin staff or the new head of IS in Durham.

cgf

Re: question about kcvs

[Jeffrey A Law]

  In message < 20000519205947.A26661@shell17.ba.best.com >you write:
  > > Currently the canonical name for the sourceware machine is
  > > egcs.cygnus.com.  Is there any chance that this changed recently?  
  > 
  > Hum, that seems wrong.  Jeff/JimK, did one of you change anything
  > like this?  Could it have been the IS staff in SV?  I guess there
  > are contractors doing the Unix work in Sunnyvale now - could one
  > of them tweaked something?
Twasn't me said the pig.


  > When we set up the new box in August, it was configured from the
  > start as "sourceware.cygnus.com", with egcs.cygnus.com and gcc.gnu.org
  > accepted as aliases.
Right.
jeff

Re: question about kcvs

[Jim Kingdon]

So, Ian, care to submit a patch to CVS to fix the error reporting bug
in CVS?

I'll even check it in if you don't want to (although if memory serves
you still do have checkin access to CVS).

That is just embarrassing...  "Use GDB to see the real error message"...
Kind of cool in a geek way, but....

Re: question about kcvs

[Ian Lance Taylor]

   Date: Sat, 20 May 2000 13:23:05 -0400
   From: Jim Kingdon <kingdon@redhat.com>

   So, Ian, care to submit a patch to CVS to fix the error reporting bug
   in CVS?

   I'll even check it in if you don't want to (although if memory serves
   you still do have checkin access to CVS).

   That is just embarrassing...  "Use GDB to see the real error message"...
   Kind of cool in a geek way, but....

Geeze, I can't believe people are just too lazy to drop into gdb.
What is the world coming to?

I don't know if I still have cvs checkin access.  I'm too lazy to find
out.

Ian


2000-05-20  Ian Lance Taylor  <ian@zembu.com>

	* client.c (connect_to_gserver): Handle server error messages
	reasonably.


--- client.c.~1~	Fri Aug 14 14:04:09 1998
+++ client.c	Sat May 20 14:27:32 2000
@@ -3927,7 +3927,29 @@
 
 	    recv_bytes (sock, cbuf, 2);
 	    need = ((cbuf[0] & 0xff) << 8) | (cbuf[1] & 0xff);
-	    assert (need <= sizeof buf);
+
+	    if (need > sizeof buf)
+	    {
+		int got;
+
+		/* This usually means that the server sent us an error
+		   message.  Read it byte by byte and print it out.
+		   FIXME: This is a terrible error handling strategy.
+		   However, even if we fix the server, we will still
+		   want to do this to work with older servers.  */
+		buf[0] = cbuf[0];
+		buf[1] = cbuf[1];
+		got = recv (sock, buf + 2, sizeof buf - 2, 0);
+		if (got < 0)
+		    error (1, 0, "recv() from server %s: %s",
+			   CVSroot_hostname, SOCK_STRERROR (SOCK_ERRNO));
+		buf[got + 2] = '\0';
+		if (buf[got + 1] == '\n')
+		    buf[got + 1] = '\0';
+		error (1, 0, "error from server %s: %s", CVSroot_hostname,
+		       buf);
+	    }
+
 	    recv_bytes (sock, buf, need);
 	    tok_in.length = need;
 	}

Re: question about kcvs

[Jim Kingdon]

> Geeze, I can't believe people are just too lazy to drop into gdb.
> What is the world coming to?

Yeah, I know, next thing you know people will be writing IDE's for
Unix or other heresies ;-).

I've checked in your CVS patch (and yes, it is a terrible error
handling strategy.  Hopefully Someone(TM) will fix this right but
until then I'm not going to say no).

Re: question about kcvs

[Chris Faylor]

On Fri, May 19, 2000 at 08:59:47PM -0700, Jason Molenda wrote:
>On Fri, May 19, 2000 at 08:10:18PM -0700, Ian Lance Taylor wrote:
>> This started happening to me too, so I looked at it.  
>
>Thanks Ian,
>
>> Currently the canonical name for the sourceware machine is
>> egcs.cygnus.com.  Is there any chance that this changed recently?  
>
>Hum, that seems wrong.  Jeff/JimK, did one of you change anything
>like this?  Could it have been the IS staff in SV?  I guess there
>are contractors doing the Unix work in Sunnyvale now - could one
>of them tweaked something?
>
>When we set up the new box in August, it was configured from the
>start as "sourceware.cygnus.com", with egcs.cygnus.com and gcc.gnu.org
>accepted as aliases.

Rumor has it that people are receiving calls from sysadmins asking for
details on five month old clarify CRs.  I don't suppose that anyone
filed a case asking to change the name to egcs.cygnus.com or something
like that at some point in the distant past?

Maybe one of the sourceware sysadmins should contact the Sunnyvale
sysadmin staff or the new head of IS in Durham.

cgf

Re: question about kcvs

[Jim Kingdon]

> Geeze, I can't believe people are just too lazy to drop into gdb.
> What is the world coming to?

Yeah, I know, next thing you know people will be writing IDE's for
Unix or other heresies ;-).

I've checked in your CVS patch (and yes, it is a terrible error
handling strategy.  Hopefully Someone(TM) will fix this right but
until then I'm not going to say no).

Re: question about kcvs

[Jason Molenda]

On Fri, May 19, 2000 at 04:05:13PM -0600, Tom Tromey wrote:
> I'm getting an error from cvs.  I use kerberos.  I've gotten this
> before but I forgot what it means and how it was fixed.
> 
> creche. cvs update -A
> cvs: ../../../devo/cvssrc/src/client.c:3937: connect_to_gserver: Assertion `need <= sizeof buf' failed.


Isn't this CVS's way of saying "invalid ticket"?  Maybe your clock
slew is too high? (or maybe sourceware's is?)

I forget exactly when I've seen these in the past, but it's something
along these lines.  I'd start by destroying your current ticket
and getting a new one.

J

Re: question about kcvs

[Jim Kingdon]

So, Ian, care to submit a patch to CVS to fix the error reporting bug
in CVS?

I'll even check it in if you don't want to (although if memory serves
you still do have checkin access to CVS).

That is just embarrassing...  "Use GDB to see the real error message"...
Kind of cool in a geek way, but....

Re: question about kcvs

[Ian Lance Taylor]

   Date: Sat, 20 May 2000 13:23:05 -0400
   From: Jim Kingdon <kingdon@redhat.com>

   So, Ian, care to submit a patch to CVS to fix the error reporting bug
   in CVS?

   I'll even check it in if you don't want to (although if memory serves
   you still do have checkin access to CVS).

   That is just embarrassing...  "Use GDB to see the real error message"...
   Kind of cool in a geek way, but....

Geeze, I can't believe people are just too lazy to drop into gdb.
What is the world coming to?

I don't know if I still have cvs checkin access.  I'm too lazy to find
out.

Ian


2000-05-20  Ian Lance Taylor  <ian@zembu.com>

	* client.c (connect_to_gserver): Handle server error messages
	reasonably.


--- client.c.~1~	Fri Aug 14 14:04:09 1998
+++ client.c	Sat May 20 14:27:32 2000
@@ -3927,7 +3927,29 @@
 
 	    recv_bytes (sock, cbuf, 2);
 	    need = ((cbuf[0] & 0xff) << 8) | (cbuf[1] & 0xff);
-	    assert (need <= sizeof buf);
+
+	    if (need > sizeof buf)
+	    {
+		int got;
+
+		/* This usually means that the server sent us an error
+		   message.  Read it byte by byte and print it out.
+		   FIXME: This is a terrible error handling strategy.
+		   However, even if we fix the server, we will still
+		   want to do this to work with older servers.  */
+		buf[0] = cbuf[0];
+		buf[1] = cbuf[1];
+		got = recv (sock, buf + 2, sizeof buf - 2, 0);
+		if (got < 0)
+		    error (1, 0, "recv() from server %s: %s",
+			   CVSroot_hostname, SOCK_STRERROR (SOCK_ERRNO));
+		buf[got + 2] = '\0';
+		if (buf[got + 1] == '\n')
+		    buf[got + 1] = '\0';
+		error (1, 0, "error from server %s: %s", CVSroot_hostname,
+		       buf);
+	    }
+
 	    recv_bytes (sock, buf, need);
 	    tok_in.length = need;
 	}

Re: question about kcvs

[Ian Lance Taylor]

   From: Tom Tromey <tromey@cygnus.com>
   Date: 19 May 2000 16:05:13 -0600

   I'm getting an error from cvs.  I use kerberos.  I've gotten this
   before but I forgot what it means and how it was fixed.

   creche. cvs update -A
   cvs: ../../../devo/cvssrc/src/client.c:3937: connect_to_gserver: Assertion `need <= sizeof buf' failed.


   Any clues?

This started happening to me too, so I looked at it.  The CVS server
on sourceware is printing
    cvs [pserver aborted]: could not acquire GSSAPI server credentials

This means that the gss_acquire_cred function run by the CVS server is
failing.  I believe this means that the cvs/egcs.cygnus.com ticket has
disappeared from the Kerberos database, or that it has disappeared
from the ticket file on sourceware.

I took a look at /etc/v5srvtab.  I see that it has an entry for
cvs/sourceware.cygnus.com.  I've never quite followed the
canonicalization strategy for Kerberos.  However, my vague
understanding is that Kerberos will canonicalize the host name.
Currently the canonical name for the sourceware machine is
egcs.cygnus.com.  Is there any chance that this changed recently?  Was
the canonical name sourceware.cygnus.com until recently?  If it was,
then the fix is to change the canonical name back to
sourceware.cygnus.com.

If that did not change recently, then somebody who knows the
CYGNUS.COM Kerberos administrator password needs to confirm that the
cvs/sourceware.cygnus.com Kerberos ticket still exists, and that it is
version number 2.

Ian

Re: question about kcvs

[Jeffrey A Law]

  In message < 20000519205947.A26661@shell17.ba.best.com >you write:
  > > Currently the canonical name for the sourceware machine is
  > > egcs.cygnus.com.  Is there any chance that this changed recently?  
  > 
  > Hum, that seems wrong.  Jeff/JimK, did one of you change anything
  > like this?  Could it have been the IS staff in SV?  I guess there
  > are contractors doing the Unix work in Sunnyvale now - could one
  > of them tweaked something?
Twasn't me said the pig.


  > When we set up the new box in August, it was configured from the
  > start as "sourceware.cygnus.com", with egcs.cygnus.com and gcc.gnu.org
  > accepted as aliases.
Right.
jeff

Re: question about kcvs

[Tom Tromey]

Jason> Isn't this CVS's way of saying "invalid ticket"?  Maybe your clock
Jason> slew is too high? (or maybe sourceware's is?)

That's what I thought.
But the skew is about 2 seconds.
And "rsh sourceware -x" works.

Jason> I forget exactly when I've seen these in the past, but it's
Jason> something along these lines.  I'd start by destroying your
Jason> current ticket and getting a new one.

Didn't work.

Tom

Re: question about kcvs

[Ian Lance Taylor]

   From: Tom Tromey <tromey@cygnus.com>
   Date: 19 May 2000 16:05:13 -0600

   I'm getting an error from cvs.  I use kerberos.  I've gotten this
   before but I forgot what it means and how it was fixed.

   creche. cvs update -A
   cvs: ../../../devo/cvssrc/src/client.c:3937: connect_to_gserver: Assertion `need <= sizeof buf' failed.


   Any clues?

The CVS server is printing an error message.  The CVS client thinks
the first two characters of the error message are the number of bytes
it needs to read.  It is complaining because its static buffer isn't
large enough.

Put the CVS client under gdb and step through the function and frob
the need variable to read in the rest of the message to see what the
server is saying.

Usually you need to kinit again for some reason.

Ian

question about kcvs

[Tom Tromey]

I'm getting an error from cvs.  I use kerberos.  I've gotten this
before but I forgot what it means and how it was fixed.

creche. cvs update -A
cvs: ../../../devo/cvssrc/src/client.c:3937: connect_to_gserver: Assertion `need <= sizeof buf' failed.


Any clues?

Tom

Re: question about kcvs

[Jason Molenda]

On Fri, May 19, 2000 at 08:10:18PM -0700, Ian Lance Taylor wrote:

> This started happening to me too, so I looked at it.  

Thanks Ian,

> Currently the canonical name for the sourceware machine is
> egcs.cygnus.com.  Is there any chance that this changed recently?  

Hum, that seems wrong.  Jeff/JimK, did one of you change anything
like this?  Could it have been the IS staff in SV?  I guess there
are contractors doing the Unix work in Sunnyvale now - could one
of them tweaked something?

When we set up the new box in August, it was configured from the
start as "sourceware.cygnus.com", with egcs.cygnus.com and gcc.gnu.org
accepted as aliases.

Jason