Re: [RFC PATCH] fs: block dev aio request priority support

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Mon 22-05-17 10:19:33, adam.manzanares@xxxxxxx wrote:
> From: Adam Manzanares <adam.manzanares@xxxxxxx>
> 
> Map the aio_reqprio to the bio priority field at
> the point the bio is created from the aio iocb.
> 
> The aio_reqprio field of iocb is used as a kernel IO class and priority
> iff the IOCB_FLAG_IOPRIO flag is set on the iocb.
> 
> Late last year device IO priority support was introduced to reduce application
> tail latency when iopriority information was set on the process [1]. This 
> patch mapped iopriority information to block io requests. This information 
> could be leveraged by device drivers to build device specific prioritized 
> commands.
> 
> The iopriority is set on the iocontext which is a structure associated with 
> a process. There exists a system call to set this iopriority information on 
> a process, but I believe it would be useful to also have a mechanism to set 
> priority on a per io command basis. 
> 
> The aio iocb has a field for the request priority which is currently not used 
> within the kernel. This patch leverages this field to pass a per command 
> iopriority value to devices. This work leverages the work in the previously 
> referenced patch [1]. When the bio is generated from the iocb we copy the 
> iocb iopriority information into the bio, which is eventually turned into a 
> request which also gets a copy of the iopriority information. 
> 
> To demonstrate how to use this feature I modified fio to use the new aio 
> feature. The modification to fio can be found at [2] and the new options 
> are cmndprioclass and cmndprio.
> 
> [1] https://lkml.org/lkml/2016/12/6/495
> [2] https://github.com/nmtadam/fio/tree/cmnd-prio.v2
> 
> Signed-off-by: Adam Manzanares <adam.manzanares@xxxxxxx>

Using aio_flags is problematic because we never checked this for containing
only expected flags. So userspace may be leaving this flag set
unintentionally and currently it doesn't have any adverse effects. So it
was decided to use a reserved word in struct iocb for new flags. And
Goldwyn already did this as a part of his series [1] together with other IO
flags. If you want, you can lobby for merging this particular patch earlier
:).

								Honza

[1] https://patchwork.kernel.org/patch/9722865/


> ---
>  fs/aio.c                     | 9 +++++++++
>  fs/block_dev.c               | 1 +
>  include/linux/fs.h           | 1 +
>  include/uapi/linux/aio_abi.h | 6 ++++++
>  4 files changed, 17 insertions(+)
> 
> diff --git a/fs/aio.c b/fs/aio.c
> index f52d925..a75a279 100644
> --- a/fs/aio.c
> +++ b/fs/aio.c
> @@ -1568,6 +1568,15 @@ static int io_submit_one(struct kioctx *ctx, struct iocb __user *user_iocb,
>  	req->common.ki_pos = iocb->aio_offset;
>  	req->common.ki_complete = aio_complete;
>  	req->common.ki_flags = iocb_flags(req->common.ki_filp);
> +	if (iocb->aio_flags & IOCB_FLAG_IOPRIO)
> +		/*
> +		 * If the IOCB_FLAG_IOPRIO flag of aio_flags is set,
> +		 * then the aio_reqprio is interpreted as a I/O
> +		 * scheduling class and priority. This is then set
> +		 * on the bio that is created from this request, which
> +		 * enables the priority to be passed to device drivers.
> +		 */
> +		req->common.ki_ioprio = iocb->aio_reqprio;
>  
>  	if (iocb->aio_flags & IOCB_FLAG_RESFD) {
>  		/*
> diff --git a/fs/block_dev.c b/fs/block_dev.c
> index 2eca00e..20d18db 100644
> --- a/fs/block_dev.c
> +++ b/fs/block_dev.c
> @@ -360,6 +360,7 @@ __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
>  		bio->bi_iter.bi_sector = pos >> 9;
>  		bio->bi_private = dio;
>  		bio->bi_end_io = blkdev_bio_end_io;
> +		bio->bi_ioprio = iocb->ki_ioprio;
>  
>  		ret = bio_iov_iter_get_pages(bio, iter);
>  		if (unlikely(ret)) {
> diff --git a/include/linux/fs.h b/include/linux/fs.h
> index 866c955..83135f0 100644
> --- a/include/linux/fs.h
> +++ b/include/linux/fs.h
> @@ -276,6 +276,7 @@ struct kiocb {
>  	void (*ki_complete)(struct kiocb *iocb, long ret, long ret2);
>  	void			*private;
>  	int			ki_flags;
> +	u16			ki_ioprio; /* See linux/ioprio.h */
>  };
>  
>  static inline bool is_sync_kiocb(struct kiocb *kiocb)
> diff --git a/include/uapi/linux/aio_abi.h b/include/uapi/linux/aio_abi.h
> index bb2554f..415980d 100644
> --- a/include/uapi/linux/aio_abi.h
> +++ b/include/uapi/linux/aio_abi.h
> @@ -54,6 +54,12 @@ enum {
>   */
>  #define IOCB_FLAG_RESFD		(1 << 0)
>  
> +/*
> + * IOCB_FLAG_IOPRIO - Set if the "aio_reqprio" member of the "struct iocb"
> + *                    is interpreted as an I/O scheduling class and priority
> + */
> +#define IOCB_FLAG_IOPRIO	(1 << 1)
> +
>  /* read() from /dev/aio returns these structures. */
>  struct io_event {
>  	__u64		data;		/* the data field from the iocb */
> -- 
> 2.7.4
> 
-- 
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR



[Index of Archives]     [Linux Ext4 Filesystem]     [Union Filesystem]     [Filesystem Testing]     [Ceph Users]     [Ecryptfs]     [AutoFS]     [Kernel Newbies]     [Share Photos]     [Security]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux Cachefs]     [Reiser Filesystem]     [Linux RAID]     [Samba]     [Device Mapper]     [CEPH Development]
  Powered by Linux