Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Matt Whitlock <kernel@mattwhitlock.name>
To:     Linus Torvalds <torvalds@linux-foundation.org>
Cc:     Matthew Wilcox <willy@infradead.org>,
        Miklos Szeredi <miklos@szeredi.hu>,
        David Howells <dhowells@redhat.com>, <netdev@vger.kernel.org>,
        Dave Chinner <david@fromorbit.com>,
        Jens Axboe <axboe@kernel.dk>, <linux-fsdevel@kvack.org>,
        <linux-mm@kvack.org>, <linux-kernel@vger.kernel.org>,
        Christoph Hellwig <hch@lst.de>, <linux-fsdevel@vger.kernel.org>
Subject: Re: [RFC PATCH 1/4] splice: Fix corruption of spliced data after =?iso-8859-1?Q?splice()_returns?=
Date:   Wed, 19 Jul 2023 17:02:04 -0400
MIME-Version: 1.0
Message-ID: <6609f1b8-3264-4017-ac3c-84a01ea12690@mattwhitlock.name>
In-Reply-To: <CAHk-=wiq95bWiWLyz96ombPfpy=PNrc2KKyzJ2d+WMrxi6=OVA@mail.gmail.com>
References: <20230629155433.4170837-1-dhowells@redhat.com>
 <20230629155433.4170837-2-dhowells@redhat.com>
 <CAJfpegsJuvXJDcXpo9T19Gw0tDuvyOJdv44Y2bt04MEf1JLxGg@mail.gmail.com>
 <c634a18e-9f2b-4746-bd8f-aa1d41e6ddf7@mattwhitlock.name>
 <CAJfpegvq4M_Go7fHiWVBBkrK6h4ChLqQTd0+EOKbRWZDcVerWA@mail.gmail.com>
 <ZLg9HbhOVnLk1ogA@casper.infradead.org>
 <CAHk-=wiq95bWiWLyz96ombPfpy=PNrc2KKyzJ2d+WMrxi6=OVA@mail.gmail.com>
User-Agent: Trojita/v0.7-595-g7738cd47; Qt/5.15.10; xcb; Linux; Gentoo Linux
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Wednesday, 19 July 2023 16:16:07 EDT, Linus Torvalds wrote:
> The *ONLY* reason for splice() existing is for zero-copy.

The very first sentence of splice(2) reads: "splice() moves data between=20
two file descriptors without copying between kernel address space and user=20=

address space." Thus, it is not unreasonable to believe that the point of=20
splice is to avoid copying between user-space and kernel-space.

If you use read() and write(), then you're making two copies. If you use=20
splice(), then you're making one copy (or zero, but that's an optimization=20=

that should be invisible to the user).

> And no, we don't start some kind of crazy "versioned zero-copy with
> COW". That's a fundamental mistake.

Agreed. splice() should steal the reference if it can, copy the page data=20
if it must. Note that, even in the slow case where the page data must be=20
copied, this still gives a better-than-50% speedup over read()+write()=20
since an entire copy (and one syscall) is elided.

> IF YOU DON'T UNDERSTAND THE *POINT* OF SPLICE, DON'T USE SPLICE.

Thanks for being so condescending. Your reputation is deserved.