Re: [PATCH v2] Introduce a version6 of autofs interface, to fix design error.

[Linus Torvalds <torvalds@linux-foundation.org>]

On Sun, Apr 29, 2012 at 12:45 AM, Michael Tokarev <mjt@tls.msk.ru> wrote:
> On 29.04.2012 11:19, Linus Torvalds wrote:
>>
>> Gaah, it should, but it won't.
>>
>> I bet my original patch worked fine, because the pipe has only one
>> inode and pipe structure. But it has *two* 'struct file's associated
>> with it, and autofs only ever sees the writing side, and never gets to
>> mark the reading side O_DIRECT. So yeah, the reading side won't do the
>> proper packetized read.
>
> Can't we go - in kernel - from one struct file to pipe structure to
> another file structure and set O_DIRECT there?  Autofs kernel code
> checks if the file descriptor is a pipe, so it should be possible...

Nope. We don't have any back-pointers to the other "struct file". They
have the inode - and the "struct pipe_inode_info" in common, but
there's no way to find the other "struct file" from that.

And sure, a pipe in Linux could work with a single read-write "struct
file" that is used both for the reading and writing, but that's not
how the "pipe()" system call is defined - it gets two separate file
descriptors, one read-only, one write-only. So that's what autofs and
systemd hands the kernel - the write side of the pipe. And we can't
find the read-side one.

But I think I have another approach. We can make the *writer* be the
only one that cares about the packetized nature, and if it's a packet
write, it will set a PIPE_BUF_PACKET flag in the pipe buffer. The
reader can react to that. I think that actually has the potential to
make the code a bit prettier too..

I'll cook it up.

                     Linus