Hi,
On 4 September 2017 at 16:23, <kusumi.tomohiro@xxxxxxxxx> wrote:
> From: Tomohiro Kusumi <tkusumi@xxxxxxxxxx>
>
> 8c43ba62('filesetup: align layout buffer') and
> 6e344dc3('filesetup: keep OS_O_DIRECT flag when pre-allocating file')
> need to keep the valid transfer size throughout the entire writes.
>
> The write(2) size may be truncated on the last write and break the
> dio requirement. This results in td_verror() in the output.
>
> --
> # mount | grep " on / "
> /dev/mapper/fedora-root on / type xfs (rw,relatime,attr2,inode64,noquota)
> # ./fio --name=xxx --kb_base=1000 --ioengine=sync --rw=read --bs=4KiB --size=10MiB --unlink=1 --direct=1 | grep "err="
> xxx: (groupid=0, jobs=1): err= 0: pid=3526: Mon Sep 4 18:03:57 2017
> # ./fio --name=xxx --kb_base=1000 --ioengine=sync --rw=read --bs=4KiB --size=10MB --unlink=1 --direct=1 | grep "err="
> fio: pid=3538, err=22/file:filesetup.c:238, func=write, error=Invalid argument
> xxx: (groupid=0, jobs=1): err=22 (file:filesetup.c:238, func=write, error=Invalid argument): pid=3538: Mon Sep 4 18:04:04 2017
>
> Signed-off-by: Tomohiro Kusumi <tkusumi@xxxxxxxxxx>
> ---
> filesetup.c | 13 ++++++++++---
> 1 file changed, 10 insertions(+), 3 deletions(-)
>
> diff --git a/filesetup.c b/filesetup.c
> index 5e8ea35..42d95db 100644
> --- a/filesetup.c
> +++ b/filesetup.c
> @@ -112,6 +112,7 @@ static int extend_file(struct thread_data *td, struct fio_file *f)
> unsigned long long left;
> unsigned int bs, alloc_size = 0;
> char *b = NULL;
> + bool done = false;
>
> if (read_only) {
> log_err("fio: refusing extend of file due to read-only\n");
> @@ -211,11 +212,15 @@ static int extend_file(struct thread_data *td, struct fio_file *f)
> goto err;
> }
>
> - while (left && !td->terminate) {
> + while (!done && !td->terminate) {
> ssize_t r;
>
> - if (bs > left)
> - bs = left;
> + /* If bs >= left this is the last write */
> + if (bs > left) {
> + done = true;
> + if (!td->o.odirect)
> + bs = left;
> + }
>
> fill_io_buffer(td, b, bs, bs);
>
Hmm. Won't this result in a file that is bigger than requested when
direct=1 and the maximum bs isn't an exact multiple of the extended
filesize? Should it start trying the minimum block size at this stage
with direct=1? If that goes on to not fit wouldn't it be better to do
less given that the main fio job won't be able to fill the gap either?
--
Sitsofe | http://sucs.org/~sits/
--
To unsubscribe from this list: send the line "unsubscribe fio" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
