On Thu, Apr 14, 2005 at 11:11:56AM +0200, Jan Kopriva wrote:
> Well, Venus itself check for 1000, but if somebody sets the ACL (using cfs)
> the kernel that accepts buffer from pioctl counts only with 2KB of data.

Huh, may be it is another reason to switch from ioctls to reads and writes?

Then it would be also possible to replace ifdefs around the mount operation
in venus by a runtime flag or a separate binary - and voila, one venus binary
would work on all platforms which share an ABI.
Now it is ioctls and mount() which force me to compile three versions of Coda
client, for Linux, FreeBSD and NetBSD. Otherwise it would be just one.

It is expected that Solaris 10 will provide Linux ABI soon - so we could profit
even more from making our kernel interface platform-independent.

The hard part is possibly to get the modified kernel module
into mainstream kernels.

> Do you think that a check should be included(I can make it) in cfs.cc that
> does not allow to set a list bigger than CFS_PIOBUFFSIZE
> (2KB - length of standard pioctl buffer), to avoid the hypothetical 
> sitution?

Sure, for the moment it should check for the limitation,
though at some time the limitation should be probably removed/relaxed...

My 0.0002 c
--
Ivan