]> git-server-git.apps.pok.os.sepia.ceph.com Git - qemu-iotests.git/log
qemu-iotests.git
14 years agoUpdate rbd support
Josh Durgin [Wed, 13 Apr 2011 00:47:45 +0000 (17:47 -0700)]
Update rbd support

rbd implements bdrv_truncate, so test 025 will work.

Signed-off-by: Josh Durgin <josh.durgin@dreamhost.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
14 years agocommon.config: Fix no $TEST_DIR directory
Mitnick Lyu [Mon, 11 Apr 2011 20:05:44 +0000 (04:05 +0800)]
common.config: Fix no $TEST_DIR directory

mkdir $TEST_DIR on common.config first run

Signed-off-by: Mitnick Lyu <mitnick.lyu@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
14 years agoonly run 016 for file and sheepdog protocols
Christoph Hellwig [Thu, 21 Apr 2011 06:26:04 +0000 (08:26 +0200)]
only run 016 for file and sheepdog protocols

016 writes past EOF which isn't support by most protocols, so limit
it to file and sheepdog, which explicitly support it.

Pointed out by Josh Durgin <josh.durgin@dreamhost.com>.

Signed-off-by: Christoph Hellwig <hch@lst.de>
15 years agoUse zero-based offsets for IO patterns
Stefan Hajnoczi [Fri, 4 Feb 2011 12:55:02 +0000 (12:55 +0000)]
Use zero-based offsets for IO patterns

The io_pattern style functions have the following loop:

  for i in `seq 1 $count`; do
      echo ... $(( start + i * step )) ...
  done

Offsets are 1-based so start=1024, step=512, count=4 yields:
1536, 2048, 2560, 3072

Normally we expect:
1024, 1536, 2048, 2560

Most tests ignore this detail, which means that they perform I/O to a
slightly different range than expected by the test author.

Later on things got less innocent and tests started trying to compensate
for the 1-based indexing.  This included negative start values in test
024 and my own attempt with count-1 in test 028!

The end result is that tests that use io_pattern are hard to reason
about and don't work the way you'd expect.  It's time to clean this mess
up.

This patch switches io_pattern to 0-based offsets.  This requires
adjusting the golden outputs since I/O ranges are now shifted and output
differs.

Verifying these output diffs is easy, however.  Each diff hunk moves one
I/O from beyond the end of the pattern range to the beginning.

Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>