On Mon, Mar 16, 2015 at 6:02 PM, Dave Chinner <david@xxxxxxxxxxxxx> wrote:
> On Mon, Mar 16, 2015 at 02:34:22PM -0400, Milosz Tanski wrote:
>> preadv2 is a new syscall introduced that is like preadv2 but with flag
>> argument. The first use case of this is to let us add a flag to perform a
>> non-blocking file using the page cache.
>> ---
>> src/Makefile | 2 +-
>> src/preadv2-pwritev2.h | 52 +++++++++++++++++
>> src/preadv2.c | 150 +++++++++++++++++++++++++++++++++++++++++++++++++
>
> You should add this syscall to support to xfs_io (in the xfsprogs
> package) rather than write a new helper for it. Mainly because:
>
>> +void
>> +usage(char *prog)
>> +{
>> + fprintf(stderr, "Usage: %s [-v] [-ctdw] [-n] -p POS -l LEN <filename>\n\n", prog);
>> + fprintf(stderr, "General arguments:\n");
>> + fprintf(stderr, " -v Verify that the syscall is supported and quit:\n");
>> + fprintf(stderr, "\n");
>> + fprintf(stderr, "Open arguments:\n");
>> + fprintf(stderr, " -c Open file with O_CREAT flag\n");
>> + fprintf(stderr, " -t Open file with O_TRUNC flag\n");
>> + fprintf(stderr, " -d Open file with O_DIRECT flag\n");
>> + fprintf(stderr, " -w Open file with O_RDWR flag vs O_RDONLY (default)\n");
>> + fprintf(stderr, "\n");
>> + fprintf(stderr, "preadv2 arguments:\n");
>> + fprintf(stderr, " -n use RWF_NONBLOCK when performing read\n");
>> + fprintf(stderr, " -p POS offset file to read at\n");
>> + fprintf(stderr, " -l LEN length of file data to read\n");
>
> The xfs_io pread command already supports all of these functions
> except for the RWF_NONBLOCK flag, and anyone testing bleeding edge
> functionality is also using a bleeding edge xfs_io binary.
>
> Then you test for whether the functionality is available via
> _require_xfs_io_command "preadv -n"
>
> .....
>> +# test file we'll be using
>> +file=$SCRATCH_MNT/067.preadv2.$$
>> +
>> +# Create a file:
>> +# two regions of data and a hole in the middle
>> +# use O_DIRECT so it's not in the page cache
>> +echo "create file"
>> +$XFS_IO_PROG -t -f -d \
>> + -c "pwrite 0 1024" \
>> + -c "pwrite 2048 1024" \
>> + $file > /dev/null
>
> This does not create holes on most filesystems. You'll need to leave
> holes of up 64k so that 64k block size filesystem end up with single
> block holes in them.
Noted and I shall fix this in the next round.
>
>> +# Make sure it returns EAGAIN on uncached data
>> +echo "uncached"
>> +$here/src/preadv2 -n -p 0 -l 1024 $file
>
> $XFS_IO_PROG -c "pread -n 0 1024" $file | _filter_xfs_io
>
>> +
>> +# Make sure we read in the whole file, after that RWF_NONBLOCK should return us all the data
>> +echo "cached"
>> +$XFS_IO_PROG -f $file -c "pread 0 4096" $file > /dev/null
>> +$here/src/preadv2 -n -p 0 -l 1024 $file
>
> $XFS_IO_PROG -c "pread 0 4096" -c "pread -n 0 1024" $file | _filter_xfs_io
>
>> +
>> +# O_DIRECT and RWF_NONBLOCK should return EAGAIN always
>> +echo "O_DIRECT"
>> +$here/src/preadv2 -d -n -p 0 -l 1024 $file
>
> $XFS_IO_PROG -d -c "pread -n 0 1024" $file | _filter_xfs_io
>
> And so on....
>
> Cheers,
>
> Dave.
> --
> Dave Chinner
> david@xxxxxxxxxxxxx
Dave,
My plan is/was to wait till the main patch makes it into the upstream
linux kernel with the syscall numbers are set in stone. Possibly after
till glibc adds support for them. After that I was going remove my
preadv2 application from xfs_tests and add that functionality to
xfs_io.
With xfs_io living in separate repository I wanted to the case when/if
syscall numbers change (there's a bunch of new syscalls queued around
epoll) of having somebody test against xfs_io that has preadv2 but bad
ids.
As a side note, I did add mlock / munlock support to xfs_io that I'll
send in another patch.
--
Milosz Tanski
CTO
16 East 34th Street, 15th floor
New York, NY 10016
p: 646-253-9055
e: milosz@xxxxxxxxx
--
To unsubscribe from this list: send the line "unsubscribe linux-api" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
