Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Tue, 30 Apr 2019 15:59:39 +0100
From:   Mark Rutland <mark.rutland@arm.com>
To:     Matthew Wilcox <willy@infradead.org>
Cc:     Jens Axboe <axboe@kernel.dk>, linux-kernel@vger.kernel.org,
        Alexander Viro <viro@zeniv.linux.org.uk>,
        linux-fsdevel@vger.kernel.org, linux-block@vger.kernel.org
Subject: Re: [PATCH] io_uring: avoid page allocation warnings
Message-ID: <20190430145938.GA8314@lakrids.cambridge.arm.com>
References: <20190430132405.8268-1-mark.rutland@arm.com>
 <20190430141810.GF13796@bombadil.infradead.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190430141810.GF13796@bombadil.infradead.org>
User-Agent: Mutt/1.11.1+11 (2f07cb52) (2018-12-01)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Apr 30, 2019 at 07:18:10AM -0700, Matthew Wilcox wrote:
> On Tue, Apr 30, 2019 at 02:24:05PM +0100, Mark Rutland wrote:
> > In io_sqe_buffer_register() we allocate a number of arrays based on the
> > iov_len from the user-provided iov. While we limit iov_len to SZ_1G,
> > we can still attempt to allocate arrays exceeding MAX_ORDER.
> > 
> > On a 64-bit system with 4KiB pages, for an iov where iov_base = 0x10 and
> > iov_len = SZ_1G, we'll calculate that nr_pages = 262145. When we try to
> > allocate a corresponding array of (16-byte) bio_vecs, requiring 4194320
> > bytes, which is greater than 4MiB. This results in SLUB warning that
> > we're trying to allocate greater than MAX_ORDER, and failing the
> > allocation.
> > 
> > Avoid this by passing __GFP_NOWARN when allocating arrays for the
> > user-provided iov_len. We'll gracefully handle the failed allocation,
> > returning -ENOMEM to userspace.
> > 
> > We should probably consider lowering the limit below SZ_1G, or reworking
> > the array allocations.
> 
> I'd suggest that kvmalloc is probably our friend here ... we don't really
> want to return -ENOMEM to userspace for this case, I don't think.

Sure. I'll go verify that the uring code doesn't assume this memory is
physically contiguous.

I also guess we should be passing GFP_KERNEL_ACCOUNT rateh than a plain
GFP_KERNEL.

Thanks,
Mark.