From mboxrd@z Thu Jan 1 00:00:00 1970 From: Al Viro Subject: Re: [GIT PULL] Orangefs (text only resend) Date: Tue, 8 Sep 2015 00:22:06 +0100 Message-ID: <20150907232206.GL22011@ZenIV.linux.org.uk> References: <20150906063552.GA7224@lst.de> <20150906090838.GI22011@ZenIV.linux.org.uk> <20150906202019.GJ22011@ZenIV.linux.org.uk> <20150907063704.GK22011@ZenIV.linux.org.uk> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Christoph Hellwig , Linus Torvalds , linux-fsdevel , Andrew Morton , Stephen Rothwell , Boaz Harrosh , Greg Kroah-Hartman To: Mike Marshall Return-path: Received: from zeniv.linux.org.uk ([195.92.253.2]:46749 "EHLO ZenIV.linux.org.uk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751162AbbIGXWN (ORCPT ); Mon, 7 Sep 2015 19:22:13 -0400 Content-Disposition: inline In-Reply-To: Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Mon, Sep 07, 2015 at 05:10:11PM -0400, Mike Marshall wrote: > I have failed so far today to figure out what you mean by GrepTFS or GTFS... grep the fucking source, of course... > and https://lwn.net/lwn/kernel/LDD2/ch07.html... I don't yet see where we're > doing GFP_KERNEL allocations within held locks... could you describe in more > detail one of the ones you saw? You don't want e.g. to have allocation request triggering an attempt to write a dirty page on a shared mapping of a file from your fs while you are holding a mutex that would block that attempt *and* waiting for a allocation to succeed. Same story as when a block driver needs to allocate something to process a write request - we do _not_ want that to trigger writes. That's what GFP_NOFS and GFP_NOIO are about; LDD2 is old and the analogue back then used to be called GFP_BUFFER. Which (if any) is needed depends on the locks you are holding; something like ->i_mutex on a directory is not a problem, but things like your request mutex almost certainly *are*.