[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <Pine.LNX.4.64.0706081448330.5795@alien.or.mcafeemobile.com>
Date: Fri, 8 Jun 2007 14:59:10 -0700 (PDT)
From: Davide Libenzi <davidel@...ilserver.org>
To: Alan Cox <alan@...rguk.ukuu.org.uk>
cc: Ulrich Drepper <drepper@...hat.com>, Theodore Tso <tytso@....edu>,
Eric Dumazet <dada1@...mosbay.com>,
Kyle Moffett <mrmacman_g4@....com>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
Linus Torvalds <torvalds@...ux-foundation.org>,
Andrew Morton <akpm@...ux-foundation.org>,
Ingo Molnar <mingo@...e.hu>
Subject: Re: [patch 7/8] fdmap v2 - implement sys_socket2
On Fri, 8 Jun 2007, Alan Cox wrote:
> > > #1: Throw the whole thing away and accept its not a good idea anyway
> >
> > Unfortunately (exactly because of the same guarantees you're asking for
> > those handles), in order for userspace libraries to reliably internally
> > use fds to interact with the kernel, you need another kind of allocation
> > strategy.
>
> Unproven and dubious at best as a claim.
I really don't mean to be rude and pointing you to read the archives, but
the proof and the reason why claims are valid is inside there.
> > > #2: If I was really going this way and I wanted to use it for serious
> > > tricks for high performance I/O then I'd provide the handle from
> > > userspace so that the strategy for allocation is controlled by the caller
> > > who is the only one who can make the smart decisions
> >
> > It does not work. What if the main application, library A and library B
> > wants to implement their own allocation strategy?
>
> Its called "discipline". I would suggest that libc contains a default
> allocator. You might also want to assign library and application ranges
> for clarity.
That is really nice solution. Each library has to have each own allocator.
Then we'll have what, a committee that assigns fd ranges?
Ranges cannot be implemented with the current fdtable allocator, because
they'll be way far apart and they'll waste space. Multiple separate ranges
will require multiple fdtable structures (one for each range/allocator).
Instead of a two-way switch (legacy and non-sequential) you will have
multiple choices to select. This multiple choices will have to be
replicated all around the code that access directly the fdtables. I did
the fdmap consolidation patch, and I can tell you there are quite a few
places that access fdtables directly.
- Davide
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists