[Pvfs2-developers] patch: error code bug fixes

[Phil Carns]

This patch corrects a variety of error code problems:

- several BMI error codes were not tagged with the BMI error class, 
which is important to allow client state machines to retry on network errors
- ditto above for a few flow errors
- ECONNRESET was not understood by BMI or included in the error code mapping
- error codes were printed in hex by some routines
- some kernel operation timeouts were translated as EINTR, which is 
misleading (these were changed to ETIMEDOUT)
- timeouts while waiting for a kernel buffer were reported as -1 (EPERM) 
rather than ETIMEDOUT

I also noticed quite a bit of fragility in how sockio.c handles error 
codes, but I didn't address that in this patch other than to work around 
one common case.  Basically, the issue is that sockio.c still sets errno 
and returns -1 when it has a problem (most of the PVFS2 source code APIs 
return -PVFS_ERRORs).  bmi_tcp.c then translates it.  This is fragile 
for a few reasons, but one that really stands out now is that only 
sockio.c knows how to translate h_errno values.  That means that now 
some of the sockio functions do translate error codes immediately, while 
others defer to let bmi_tcp.c do the work.  This is confusing :)

The BMI error handling and ECONNRESET parts of this patch are important 
for failover scenarios so that clients are able to pick back up without 
error.

The ones related to kernel timeouts are important if you are tuning 
kernel buffer sizes- at some point if your buffers are large enough you 
may end up with processes waiting for buffers long enough to exhaust the 
default timeouts.

-Phil
-------------- next part --------------
A non-text attachment was scrubbed...
Name: error-codes.patch
Type: text/x-patch
Size: 14304 bytes
Desc: not available
Url : http://www.beowulf-underground.org/pipermail/pvfs2-developers/attachments/20070320/0e984cef/error-codes.bin