segfault on copy

[Matt McLeod]

Currently testing copy with three hosts:  (1)  Solaris 8 on Intel
(faramir);  (2)  Solaris 8 on UltraSPARC (bruce);  (3)  Debian Linux 
'testing' branch on Intel (buffy).  cfengine 2.0.2, also tested and got
the same results with 2.0.1.

With the following rule in place on faramir:

copy:
  faramir.copytest::
    /local
      dest=/sw/solaris/local
      recurse=inf
      server=bruce
      trustkey=true

I consistently get a segfault+core dump on the same file:

cfengine:faramir: Server returned error:  Host authentication failed. Did
you forget the domain name?
cfengine:faramir: (Can't stat /local/bin/ris/local/lib/uSystem/bin/moncc)
zsh: segmentation fault (core dumped)  /var/cfengine/bin/cfagent -v
--no-splay -D copytest

The source file is /local/bin/moncc, which is a symlink;

lrwxrwxrwx   1 root     other         24 Oct  4  2001 moncc -> 
../lib/uSystem/bin/moncc

The symlink has been created on the destination host before the segfault.

Restarting cfagent, again segfaulting after a while.  The file is
consistently the same one, another symlink:

cfengine:faramir: Server returned error:  Host authentication failed. Did
you forget the domain name?
cfengine:faramir: (Can't stat 
/local/bin/ris/local/lib/java-1.3/bin/native2ascii)
zsh: segmentation fault (core dumped)  /var/cfengine/bin/cfagent -v
--no-splay -D copytest

lrwxrwxrwx   1 root     other         32 May 29 12:37
/sw/solaris/local/bin/native2ascii -> ../lib/java-1.3/bin/native2ascii

And so it goes.

Running the same rule on buffy (the Linux box) gets the same result.

Running a similar rule on buffy, grabbing a different set of files (similar
amount of data though) from faramir runs into a new and exciting problem:

copy:
  do_irix:: /sw/irix/local
    dest=/sw/irix/local
    recurse=inf
    server=faramir
    trustkey=true

cfengine:buffy: 
/sw/irix/local/lib/Acrobat4/Reader/mipsirix/lib/libreadcore.so.4.0 wasn't at 
destination (copying)
cfengine:buffy: Copying from 
faramir:/sw/irix/local/lib/Acrobat4/Reader/mipsirix/lib/libreadcore.so.4.0
Segmentation fault

Only this time it isn't a symlink, and the file is copied across to
leadreadcore.so.4.0.cfnew prior to the segfault.  This is again consistent
across runs -- same file, same result.

Adding "purge-true" prompts a much quicker segfault in all cases.

Unfortunately I don't know enough about using gdb to track this down
quickly, but the first one does rather scream "problem with handling of
pointers to char" at me.

I'd be happy to run any tests anyone wants to try and resolve this,
as I'd really like to use cfengine to replace rdist in addition to
the stuff I've already got it doing.

Matt