Niv Sardi [Fri, 26 Oct 2007 16:08:10 +0000 (16:08 +0000)]
Refresh golden output for test 177
Some changes in the bulkstat binary test (see former commits)
change the output of 177, here is the refreshed one.
SGI-Git-Id: 21f80cf802b520f713d8fc78a871af623acd77f1
SGI-Git-Author: Niv Sardi-Altivanik <xaiki@chook.melbourne.sgi.com>
SGI-Git-Date: Wed, 24 Oct 2007 15:54:46 +1000
Merge of master-melb:xfs-cmds:29969a by kenmcd.
Niv Sardi [Fri, 26 Oct 2007 16:06:38 +0000 (16:06 +0000)]
Refresh golden output for test 177
Some changes in the bulkstat binary test (see former commits)
change the output of 177, here is the refreshed one.
SGI-Git-Id: b1579451b2454e6a9893ad43c7d3de5f6a8a336c
SGI-Git-author: Niv Sardi-Altivanik <xaiki@chook.melbourne.sgi.com>
SGI-Git-date: Wed, 24 Oct 2007 15:54:46 +1000
Merge of master-melb:xfs-cmds:29966a by kenmcd.
Niv Sardi [Fri, 26 Oct 2007 16:05:51 +0000 (16:05 +0000)]
Add QA test 183
(whitespace cleanups)
This test calls the new bulkstat_unlink_test with the -r switch
The patch contains the test, as well as the golden output.
SGI-Git-Id: a0d3341757e6013feaed7f630f3afc061f743059
SGI-Git-author: Niv Sardi-Altivanik <xaiki@chook.melbourne.sgi.com>
SGI-Git-date: Wed, 24 Oct 2007 15:52:25 +1000
Merge of master-melb:xfs-cmds:29965a by kenmcd.
Niv Sardi [Fri, 26 Oct 2007 16:05:04 +0000 (16:05 +0000)]
Add -r switch to src/bulkstat_unlink_test.c needed for QA test 183
(whitespace cleanups)
This patch adds a -r switch (using getopt as suggested by dchinner)
to src/bulkstat_unlink_test.c, that will implement the additional checks
described by vlad in PV 972128.
Basically we:
* Save the inode count in scount.
* Create nfiles new files (passed as argument).
* Open those nfiles.
* Unlink the files.
* Check that the inode count is greater than scount.
* close the files.
* Check that the inode count is the same as scount.
This patch also introduces more info on failure (Line number and file) before the exit()s, as well
as comments describing what we are doing.
This patch is to be used by QA test 183
SGI-Git-Id: 4ab8fa354f79070b4ba87cb09c4326d62da824fd
SGI-Git-author: Niv Sardi-Altivanik <xaiki@chook.melbourne.sgi.com>
SGI-Git-date: Wed, 24 Oct 2007 13:24:27 +1000
Merge of master-melb:xfs-cmds:29964a by kenmcd.
Tim Shimmin [Mon, 17 Sep 2007 06:14:40 +0000 (06:14 +0000)]
test out xfs_inactive_attrs in log replay for iunlink processing.
Merge of master-melb:xfs-cmds:29701a by kenmcd.
Update args to use command options for multi_open_unlink.
Change a sleep to a wait as we were just sleeping to account
for the death of a process - should really wait for its death.
David Disseldorp [Fri, 17 Aug 2007 16:09:29 +0000 (16:09 +0000)]
new common.punch file for exercising various hole punch functions,
_test_punch the main test function follows the procedure:
- mkfs & mount scratch
- optionally set files extent size hint
- truncate and optionally resvsp
- write or punch to file in specified order at given offsets & lengths
- display the file state (bmap) after each operation
All sizes, offsets and lengths are specified in FS blocks.
Problems may arise on non bash shells due to array usage.
Merge of master-melb:xfs-cmds:29434a by kenmcd.
David Disseldorp [Fri, 20 Jul 2007 04:11:09 +0000 (04:11 +0000)]
- OS common filestreams timeout change function
- use xfs_io instead of dd (working direct io)
- no filestreams directory flags, just use the mount option
Merge of master-melb:xfs-cmds:29185a by kenmcd.
A recent change to XFSQA to resolve mkfs option conflicts means mkfs can be run twice per _scratch_mkfs_xfs:
$MKFS_XFS_PROG $SCRATCH_OPTIONS $MKFS_OPTIONS $extra_mkfs_options $SCRATCH_DEV
local mkfs_status=$?
# a mkfs failure may be caused by conflicts between
# $MKFS_OPTIONS and $extra_mkfs_options
if [ $mkfs_status -ne 0 -a ! -z "$extra_mkfs_options" ]; then
echo "** mkfs failed with extra mkfs options added to \"$MKFS_OPTIONS\" by test $seq **" \
>>$here/$seq.full
echo "** attempting to mkfs using only test $seq options: $extra_mkfs_options **" \
>>$here/$seq.full
$MKFS_XFS_PROG $SCRATCH_OPTIONS $extra_mkfs_options $SCRATCH_DEV
mkfs_status=$?
fi
In the case of a failure caused by conflict between $MKFS_OPTIONS and $extra_mkfs_options,
the mkfs output is shown twice, first off is the failed mkfs output, secondly the resolved mkfs output.
This can cause QA failures for a number of tests, particularly those that filter the mkfs output.
This change shows only the applied (final) mkfs output from _scratch_mkfs_xfs.
Merge of master-melb:xfs-cmds:29041a by kenmcd.
David Disseldorp [Fri, 29 Jun 2007 16:06:04 +0000 (16:06 +0000)]
For XFSQA test 018 081 082 095 the test specifies test wide mkfs options using MKFS_OPTIONS env var.
Further mkfs options are then appended by _mkfs_log at mkfs time with _scratch_mkfs_xfs.
Currently there are a few ways mkfs options are specified in XFSQA:
1) suite wide MKFS_OPTIONS are specified before tests are run
e.g.
vimes:/home/fsgqa/kali/xfsqa/xfstests # export MKFS_OPTIONS="-l size=100m"
2) test wide MKFS_OPTIONS are specified during a particular test
e.g. 119
export MKFS_OPTIONS="-l version=2,size=1200b,su=64k"
3) mkfs options are appended by a test at mkfs time
e.g. 083
_scratch_mkfs_xfs -dsize=$fsz,agcount=$ags >>$seq.full
(4) Another form of appending is just appending to MKFS_OPTIONS directly
e.g. 114:export MKFS_OPTIONS="$MKFS_OPTIONS -i parent=1"
115:export MKFS_OPTIONS="$MKFS_OPTIONS -i paths=1"
e.g. Methods (2) *and* (3) for specifiying mkfs options are used in these tests.
This causes problems, because the mkfs options fallback code assumes the
_mkfs_log _scratch_mkfs_xfs appended options are the only test specified options.
This change causes tests using _mkfs_log to only use option (2).
~
~
Merge of master-melb:xfs-cmds:29031a by kenmcd.
XFSQA 073 is still not correctly cleaning up $tmp,
this is because $tmp.source_dir is not an empty dir.
This change recursively deletes everything in $tmp.source_dir.
Not required for $tmp.loop as it is used as a loopback mountpoint.
the ignore external MKFS_OPTIONS change is no longer required,
as _scratch_mkfs_xfs now handles conflicting options.
Merge of master-melb:xfs-cmds:28863a by kenmcd.
This test enables dmapi return_on_destroy and captures dmapi events
generated by two create/removes. One with file attibutes set, one without.
A dmapi destroy event should be generated in both cases, however PV960928
has shown that it may not be when attributes are not set.
Merge of master-melb:xfs-cmds:28771a by kenmcd.
David Disseldorp [Thu, 17 May 2007 15:58:12 +0000 (15:58 +0000)]
Currently there are a few ways mkfs options are specified in XFSQA:
1) suite wide MKFS_OPTIONS are specified before tests are run
e.g.
vimes:/home/fsgqa/kali/xfsqa/xfstests # export MKFS_OPTIONS="-l size=100m"
2) test wide MKFS_OPTIONS are specified during a particular test
e.g. 119
export MKFS_OPTIONS="-l version=2,size=1200b,su=64k"
3) mkfs options are appended by a test at mkfs time
e.g. 083
_scratch_mkfs_xfs -dsize=$fsz,agcount=$ags >>$seq.full
(4) Another form of appending is just appending to MKFS_OPTIONS directly
e.g. 114:export MKFS_OPTIONS="$MKFS_OPTIONS -i parent=1"
115:export MKFS_OPTIONS="$MKFS_OPTIONS -i paths=1"
Conflicts between mkfs options specified with method 1 & 3 are common.
960377 XFSQA 041, 042 - mkfs fails with large log size MKFS_OPTIONS.
This change means if a mkfs fails where mkfs options have been appended by a test (method 3),
The mkfs is retried using only those options defined by the test.
occurances of method 4 (in test 114 & 115) are also changed to use method 3.
$seq.full logs the fact that a mkfs options conflict has occured.
Merge of master-melb:xfs-cmds:28601a by kenmcd.
append -i parent mkfs option using _scratch_mkfs_xfs -i parent...
rather than export MKFS_OPTIONS="$MKFS_OPTIONS -i parent=1.
this avoids mkfs option conflicts
Tim Shimmin [Thu, 5 Apr 2007 15:45:28 +0000 (15:45 +0000)]
Output wasn't deterministic. Remove bmap output from stdout,
just have going to seq.full.
Merge of master-melb:xfs-cmds:28376a by kenmcd.
Execute bits changed from --- to x--
Fix up bmap output. Meant to remove the testfile on every
test but forgot to. Also needed to truncate the files
in the 2nd group of tests which used resvsp so that
the length/eof was correct for the test.
Tim Shimmin [Wed, 4 Apr 2007 03:54:24 +0000 (03:54 +0000)]
Some experimentation for pv#962014 doing prealloc, direct write, buffered read etc...
Add more here if come up with other ideas.
Merge of master-melb:xfs-cmds:28351a by kenmcd.
Some experimentation for pv#962014 doing prealloc, direct write, buffered read etc...
Add more here if come up with other ideas.
This change makes test 041 function correctly other blocks sizes, not just 4096.
The FS bsize is extracted from _filter_mkfs and then used to calculate growfs "destination" size.
Merge of master-melb:xfs-cmds:28222a by kenmcd.
calculate one meg in blocks, then use that value to calculate grow-to values.
Lachlan McIlroy [Thu, 8 Mar 2007 03:02:20 +0000 (03:02 +0000)]
fsx was calling vfprintf twice without resetting the va_list argument
Merge of master-melb:xfs-cmds:28212a by kenmcd.
fsx was calling vfprintf twice without resetting the va_list argument and
this caused a segfault on the second call. Now uses one call to vsnprintf
to print to a buffer and uses that multiple times.
Tim Shimmin [Mon, 12 Feb 2007 05:08:57 +0000 (05:08 +0000)]
Put the output of fsx into seq.full and output all of seq.full on failure.
Just to give more context around the error if we get one.
Merge of master-melb:xfs-cmds:28068a by kenmcd.
Put the output of fsx into seq.full and output all of seq.full on failure.
Just to give more context around the error if we get one.
_check_sids previously grepped the entire output of dm_getall_sessions for sids, which is:
ret=dm_getall_sessions return value
rnelem=number of sessions
sids="sid1 sid2 ... sidn "
If the session id of a previously destroyed session matches the current number of sessions (rnelem),
then _check_sids() $PREV_DESTROYED_SID will _not_ flag ERROR: SID(s) not created.
This fix also makes _check_sids() not rely on sids being consecutive. DMF may be running on a FS in background.
Tested on Linux and Irix.
Merge of master-melb:xfs-cmds:28043a by kenmcd.
change _check_sids to go through sids line 1 by 1 checking for matches
XFSQA 122 on machines with xfs_lazy_sb.patch have failed due to the addition of __be32 agf_btreeblks to xfs_agf_t.
This change causes 122 detect whether the patch is installed by greping for lazy count in mkfs output, the xfs_agf_t size is then filtered based on the patch status.
The original intention was to check /lib/modules/$(uname -r)/source/fs/xfs/xfs_ag.h for btreeblks and filter on this, however the kernel source is not a prereq for XFSQA.
Merge of master-melb:xfs-cmds:28014a by kenmcd.
pipe get_structs.c output through _type_size_filter to filter out known changes to xfs type sizes
projects / xfstests-dev.git / log