[Spread-users] fix for spread-4.4.0 crash in mixed endian environment
John Lane Schultz
jschultz at spreadconcepts.com
Wed Sep 10 09:53:16 EDT 2014
Yes, that is definitely a bug and your fix looks correct. I apologize, we need to do better mixed endian testing on our end.
With this fix, does Spread now function as expected for you?
John Lane Schultz
Spread Concepts LLC
Cell: 443 838 2200
On Sep 10, 2014, at 8:52 AM, Martin Schu <martin.sc11111 at gmail.com> wrote:
We have some SPARC and some Intel hosts running spread-4.1. Everything was working fine with spread-4.1 in that mixed endian environment.
After upgrading from spread-4.1 to spread 4.4.0 we observed problems when spread-4.4 daemons are started on both SPARC(Solaris) and Intel(RHEL). Everything is working fine as long as the spread-4.4 daemons are started *only* on SPARC hosts or *only* on Intel hosts.
Now we first start all spread daemons on Intel. When the first daemon of the “other endian” architecture is run up problems begin. We found the following alarm message in a spread logfile. And a spread daemon was crashing with core dump.
2014-09-08 17:43:59 GMT Prot_handle_bcast: invalid packet with seq -1062731160 from 21, processed bytes not equal data_len 16905 1395
The program code emitting that alarm is new since spread-4.3. It is only relevant in mixed endian environments.
Obviously something is wrong with the length calculation. The length reported in the alarm message (16905) is far away from the desired value 1395.
We believe there is a bug in the pointer calculation of the frag_ptr in protocol.c. Below a proposed patch for that.
Can you confirm this?
We would be glad if that could be included in a future release of spread.
--- daemon/protocol.c.orig 2014-05-15 17:04:35.000000000 +0200
+++ daemon/protocol.c 2014-09-09 14:37:40.845240310 +0200
@@ -409,7 +409,7 @@
pack_ptr->transmiter_id, pack_ptr->seq, processed_bytes, pack_ptr->data_len );
- frag_ptr = (fragment_header *) &pack_body_ptr[processed_bytes];
+ frag_ptr = (fragment_header *) (((char*)pack_body_ptr) + processed_bytes);
processed_bytes += sizeof(fragment_header) + frag_ptr->fragment_len;
More information about the Spread-users