Hi Miklos,
On 8/27/24 3:12 AM, Miklos Szeredi wrote:
> On Mon, 26 Aug 2024 at 15:07, yangyun <yangyun50@xxxxxxxxxx> wrote:
>>
>> Commit 23c94e1cdcbf ("fuse: Switch to using async direct IO
>> for FOPEN_DIRECT_IO") gave the async direct IO code path in the
>> fuse_direct_read_iter() and fuse_direct_write_iter(). But since
>> these two functions are only called under FOPEN_DIRECT_IO is set,
>> it seems that we can also use the async direct IO even the flag
>> IOCB_DIRECT is not set to enjoy the async direct IO method. Also
>> move the definition of fuse_io_priv to where it is used in fuse_
>> direct_write_iter.
>
> I'm interested in the motivation for this patch.
>
> There's a minor risk of regressions when introducing such a behavior
> change, so there should also be a strong supporting argument, which
> seems to be missing in this case.
>
I'm not sure what yangyun's use case is, but we indeed also observed a
potential performance optimization for FOPEN_DIRECT_IO path. When the
buffer IO is submitted to a file flagged with FOPEN_DIRECT_IO, the code
path is like:
fuse_direct_read_iter
__fuse_direct_read
fuse_direct_io
# split the request to multiple fuse requests according to
# max_read and max_pages constraint, for each split request:
fuse_send_read
fuse_simple_request
When the size of the user requested IO is greater than max_read and
max_pages constraint, it's split into multiple requests and these split
requests can not be sent to the fuse server until the previous split
request *completes* (since fuse_simple_request()), even when the user
request is submitted from async IO e.g. io-uring.
--
Thanks,
Jingbo
