This file documents known bugs in the given release.

Version 1.5.6
-------------

- mmap may not be consistent across multiple client nodes

- "cp -pr" does not preserve creation time for directories (tar
  may not either).  We need the pvfs mgr to handle utime requests
  for files correctly to do this right.

- The kernel "highmem" option is incompatible with our "mapped" transfer
  mechanism.  Please do not use the mapped transfer mechanisms on
  systems that use this option and actually have > 2G of memory.

- O_EXCL is not reliable across seperate clients.

Version 0.9.3
-------------

- The kernel "highmem" option is incompatible with our "mapped" transfer
  mechanism.  Please do not use the mapped transfer mechanisms on
  systems that use this option and actually have > 2G of memory.

- We can still trigger kernel oopses with 2.4 SMP on Chiba City

- We don't have a configure test to detect the alloc_kiovec_sz function
  present in 2.4.9 -ac (or later) kernels.  If you wish to use this
  package with one of the newer -ac kernels you must manually define
  HAVE_ALLOC_KIOVEC_SZ. 

  ( fixed in 1.5.3 )

- pvfs kernel module always uses the major number 60- this may cause
  problems on systems in which this number is already taken.

  ( fixed in 1.5.4 )


Version 0.9.2
-------------
- Still problems with mount occasionally causing a panic -- seen on
  Chiba.

- Kernel "highmem" option gets us panics from the pvfsd.

  DON'T USE HIGHMEM FOR NOW.

- Kernel "ipchains" option seems to be getting us lots of panics.
  However, this might be a "highmem" problem in actuality.  Not sure
  yet.

- Unlink returns "Resource temporarily unavailable" error on large
  files.  Perhaps a timeout-related issue?

Version 0.9.1
-------------
- RH7.1 has a problem with "rm" giving out "XXX is not a directory"
  errors.  This is due to a problem in our use of the filldir() function.

  Fixed in 0.9.2-pre1 (or so).

- configure some times appeared to detect a 2.4 kernel as a 2.2 kernel.
  This had to do with autoconf deciding not to figure out how to run the
  C preprocessor.

  Fixed in 0.9.2-pre4.

- lseek() some times failed.  This had to do with a new structure (to
  2.4) in the superblock which gave a max. file size, which we weren't
  setting.

  Fixed in 0.9.2-pre3.

- open() some times failed.  This had to do with dcache consistency
  problems.

  Fixed in 0.9.2-pre1.
