This patch replaces the INSTALL macro with LTINSTALL so that libtool is used
to install the applications. Libtool will install the binary from the proper
location and display a warning if any shared library dependancies are not
properly installed. This ensures that a libtool wrapper shell script is not
installed in place of application when libtool wrappers are being used.
Signed-off-by: Rich Johnston <rjohnston@sgi.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Ben Myer <bpm@sgi.com>
This explicitly sends MS_SYNC as the flag to msync() in
fsx.c, in domapwrite(). Without this flag, the memory
written to the mmap'ed region will not be flushed to disk;
in fact, on Linux, calling msync() with a '0' or MS_ASYNC
flag is a no-op.
Signed-off-by: Curt Wohlgemuth <curtw@google.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
I was running 068 to test freeze changes, and realized that
"sync" is not in the op list when "-w" (write ops) is specified,
although fsync & fdatasync are. It seems to me that sync should
be a default write op as well.
Signed-off-by: Eric Sandeen <sandeen@sandeen.net>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Add two new operations:
- getattr: ioctl(fd, FS_IOC_GETFLAGS, &fl)
- setattr: ioctl(fd, FS_IOC_SETFLAGS, &random_flags)
Attribute mask may be passed via -M opt, by default is (~0).
By default FS_IOC_SETFLAGS has zero probability because
it may produce inodes with APPEND or IMMUTABLE flags which
are not deletable by default. Let's assumes that one who
enable it knows how to delete such inodes.
For example like follows:
find $TEST_PATH -exec chattr -i -a {} \;
rm -rf $TEST_PATH
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Add tests for fallocate(2) syscall
- fallocate: reserve the disk space
- punch: de-allocates the disk space
Since FALLOC_FL_PUNCH_HOLE is relatively new it's value defined
explicitly if not yet defined. Later we may clear that define.
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Once some combination of seed+fs_ops result in regression it is
reasonable to document that combination. It is usefull to dump
that configuration in command line style. Later this line may be
simply hardcoded in to regression test.
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
It is very hard to predict runtime for fsstress. In many cases it
is useful to give test to run a reasonable time, and then kill it.
But currently there is no reliable way to kill test without leaving
running children.
This patch add sanity cleanup logic which looks follow:
- On sigterm received by parent, it resend signal to it's children
- Wait for each child to terminates
- EXTRA_SANITY: Even if parent was killed by other signal, children
will be terminated with SIGKILL to preven staled children.
So now one can simply run fsstress like this:
./fsstress -p 1000 -n999999999 -d $TEST_DIR &
PID=$!
sleep 300
kill $PID
wait $PID
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Currently the only way to log fsstress's output is to redirect it's shared
stdout to pipe which is very painfull because:
1) Pipe writers are serialized via i_mutex so we waste cpu-cores power on stupid
sinchronization for loging purpose, instead of hunting real race conditions,
and bugs inside file system.
2) Usually output is corrupted due to luck of sychronization on shared stdout.
Since fsstress's children operate on independend paths, let's just open didicated
log file for each child and simply avoid useless sycnhronization.
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Fsstress exec behaviour is not completely determinated in case of
low resources mode due to ENOMEM, ENOSPC, etc. In some places we
call stat(2). This information may be halpfull for future
investigations purposes. Let's dump stat info where possible.
Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Before punching a hole in a file, TRIM_OFF_LEN() calls
TRIM_OFF_LEN() in order to make sure the offset and size
used are in a reasonable range. But currently the range
it's limited to is maxfilelen, which allows the offset
(and therefore offset + len) to be beyond EOF.
Later, do_punch_hole() ignores any request that starts beyond
EOF, so we might as well limit requests to the file size.
It appears that a hole punch request that starts within a
file but whose length extends beyond it is treated simply
as a hole punch up to EOF. So there's no harm in limiting
the end of a hole punch request to the file size either.
Therefore, use TRIM_OFF_LEN() to put both the the offset
and length of a request within the file size for hole
punch requests.
Signed-off-by: Alex Elder <aelder@sgi.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
A recent commit added a TRIM_OFF_LEN() macro in "ltp/fsx.c":
5843147e xfstests: fsx fallocate support is b0rked
A later commit fixed a problem with that macro:
c47d7a51 xfstests: fix modulo-by-zero error in fsx
There is an extra flag parameter in that macro that I didn't like
in either version. When looking at it the second time around I
concluded that there was no need for the flag after all.
Going back to the first commit, the code that TRIM_OFF_LEN()
replaced had one of two forms:
- For OP_READ and OP_MAP_READ:
if (file_size)
offset %= file_size;
else
offset = 0;
if (offset + size > file_size)
size = file_size - offset;
- For all other cases (except OP_TRUNCATE):
offset %= maxfilelen;
if (offset + size > maxfilelen)
size = maxfilelen - offset;
There's no harm in ensuring maxfilelen is non-zero (and doing so
is safer than what's done above). So both of the above can be
generalized this way:
if (SIZE_LIMIT)
offset %= SIZE_LIMIT;
else
offset = 0;
if (offset + size > SIZE_LIMIT)
size = SIZE_LIMIT - offset;
In other words, there is no need for the extra flag in the macro.
The following patch just does away with it. It uses the value of
the "size" parameter directly in avoiding a divide-by-zero, and in
the process avoids referencing the global "file_size" within the
macro expansion.
Signed-off-by: Alex Elder <aelder@sgi.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
The libxfs resync brought in a new round_up macro that conflicts with the
round_up function in fsx. Rename the latter to allow building against the
new headers.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
The recent fsx fixes has a logic error in the offset trimming code.
If a read is done when the file size is zero, then the logic error
causes a offset % 0 opertaion to occur. This causes fsx to get a
SIGFPE and die.
This was not discovered during my testing because I was using a
random seed that didn't trip this condition. Changing the seed to
that which test 091 uses (the default of 1) causes such an operation
to occur....
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
fsx segvs when dumping fallocate log entries. Fix magic string
array index parameters to be zero based rather than one based.
While touching log string related stuff, make the format consistent
with read and write operations so the log dump is easier to look at
with the human eye.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Alex Elder <aelder@sgi.com>
The recent fallocate/fpunch additions to fsx have not actually be
executing fallocate/fpunch operations. The logic to select what
operation to run is broken in such a way that fsx has been executing
mapped writes and truncates instead of fallocate and fpunch
operations.
Remove all the (b0rken) smarty-pants selection logic from the test()
function. Replace it with a clearly defined set of operations for
each mode and use understandable fallback logic when various
operation types have been disabled. Then use a simple switch
statement to execute each of the different operations, removing the
tortured nesting of if/else statements that only serve to obfuscate
the code.
As a result, fsx uses fallocate/fpunch appropriately during
operations and uses/disableѕ the operations as defined on the
command line correctly.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Alex Elder <aelder@sgi.com>
The operation flags parameter to fallocate is the second parameter,
not the last. Hence the fpunch test is actually testing for falloc
support, not fpunch. Somebody needs a brown paper bag.
Also, add a ftruncate call whenthe fpunch succeeds just in case the
file was not already zero sized. Failing to ensure we start with a
zero length file can cause read ops to fail size checks if they
occur before the file is written to be the main test loop.
While there, observe the quiet flag the same as the falloc test
does and have them both emit the warning at the same error level.
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Alex Elder <aelder@sgi.com>
Current the Makefile only builds the .c files and installs
them in the $INSTALL path. However, the ltp directory contains a shell script,
which doen't get copied over (installed) when a "make install"is invoked.
This behaviour causes test 080 to fail since it requires rwtest.sh to be
present in the ltp/ directory.
Tested: Updated the Makefile and ran a static build script & test
080.
Signed-off-by: Akshay Lal <akshaylal@google.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
This patch adds punch hole tests to the fsx stress test.
Signed-off-by: Allison Henderson <achender@us.ibm.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
ltp/fsx.c tests whether the filesystem it is run on supports fallocate.
If it is not supported the fsx will print warning to stderr. This leads
to fails of tests 075, 112, 127 for the filesystems that do not support
fallocate. The tests use ltp/fsx but do not filter out stderr. Since
ltp/fsx.c can work without fallocate support I propose to move this
message to stdout unless quiet output is not requested. Previous patch
printed the message even if -q flag was used. This patch honours the flag.
This simple patch fixes the issue for me, tested on all the mentioned tests:
Signed-off-by: Boris Ranto <branto@redhat.com>
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Make it so "make depend" is a generic target, like "make clean".
Each Makefile has a "depend" target that indicates whether making
dependencies means creating ".dep" or creating ".ltdep" (or, I
suppose, both, though none do that right now). Both files get
created even if there are no CFILES to scan (to ensure the target
up-to-date). The "default" target now depends on "depend" (there is
no "ltdepend" any more).
Remove the "depend" and "ltdepend" definitions from the "buildrules"
file; only the actual generated files (".dep" and ".ltdep") remain
as generic targets. The "depend' target is still defined as phony.
Do a shell trick when expanding the value of CFILES, to avoid a
problem that occurs if it is created by "make" by concatentating two
empty strings. The problem was that in that case CFILES will
contain a space, and that wasn't getting treated as empty as
desired.
Make the rule for tool/lib dependencies more generic, to reflect the
general desire that "lib" subdirectories need to be built before
things in the "tool" subdirectories.
Signed-off-by: Alex Elder <aelder@sgi.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Add random runtime fallocate calls to fsx (vs. the existing
preallocate file at start of run).
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Alex Elder <aelder@sgi.com>
Rich Johnston
Curt Wohlgemuth
Eric Sandeen
Dmitry Monakhov
Alex Elder
Christoph Hellwig
Dave Chinner
Akshay Lal
Allison Henderson
Boris Ranto