Re: [RFC v2 0/5] Non-blockling buffered fs read (page cache only)

From: Milosz Tanski
Date: Fri Sep 19 2014 - 13:33:21 EST


On Fri, Sep 19, 2014 at 10:42 AM, Jonathan Corbet <corbet@xxxxxxx> wrote:
> On Wed, 17 Sep 2014 22:20:45 +0000
> Milosz Tanski <milosz@xxxxxxxxx> wrote:
>
>> This patcheset introduces an ability to perform a non-blocking read from
>> regular files in buffered IO mode. This works by only for those filesystems
>> that have data in the page cache.
>>
>> It does this by introducing new syscalls new syscalls readv2/writev2 and
>> preadv2/pwritev2. These new syscalls behave like the network sendmsg, recvmsg
>> syscalls that accept an extra flag argument (O_NONBLOCK).
>
> So I'm trying to understand the reasoning behind this approach so I can
> explain it to others. When you decided to add these syscalls, you
> ruled out some other approaches that have been out there for a while.
> I assume that, before these syscalls can be merged, people will want to
> understand why you did that. So I'll ask the dumb questions:
>
> - Non-blocking I/O has long been supported with a well-understood set
> of operations - O_NONBLOCK and fcntl(). Why do we need a different
> mechanism here - one that's only understood in the context of
> buffered file I/O? I assume you didn't want to implement support
> for poll() and all that, but is that a good enough reason to add a
> new Linux-specific non-blocking I/O technique?

I realized that I didn't answer this question well in my other long
email. O_NONBLOCK doesn't work on files under any commonly used OS,
and people have gotten use to this behavior so I doubt we could change
that without breaking a lot of folks applications. If you want to
ignore my other long email, what I realized that I could solve a lot
of problems if I had a syscall like recvmsg that takes a MSG_NONBLOCK
argument that worked on regular files (not sockets) and thus
readv2/preadv2 was born.

>
> - Patches adding fincore() have been around since at least 2010; see,
> for example, https://lwn.net/Articles/371538/ or
> https://lwn.net/Articles/604640/. It seems this could be used in
> favor of four new read() syscalls; is there a reason it's not
> suitable for your use case?
>
> - Patches adding buffered support for AIO have been around since at
> least 2003 - https://lwn.net/Articles/24422/, for example. I guess
> I don't really have to ask why you don't want to take that
> approach! :)
>
> Apologies for my ignorance here; that's what I get for hanging around
> with the MM folks at LSFMM, I guess. Anyway, I suspect I'm not the
> only one who would appreciate any background you could give here.
>
> Thanks,
>
> jon



--
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-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/