Received-SPF: pass (google.com: domain of linux-kernel+bounces-8115-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Message-ID: <d4b227de-d609-aef2-888b-203dbcf06707@landley.net>
Date: Thu, 21 Dec 2023 03:30:37 -0600
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.10.0
Subject: Re: [PATCH v3] rootfs: Fix support for rootfstype= when root= is
 given
Content-Language: en-US
To: Askar Safin <safinaskar@gmail.com>, stefanb@linux.ibm.com
Cc: gregkh@linuxfoundation.org, initramfs@vger.kernel.org,
 linux-kernel@vger.kernel.org, stable@vger.kernel.org, zohar@linux.ibm.com
References: <CAPnZJGDcNwPLbzC99qNQ+bRMwxPU-Z0xe=TD6DWQU=0MNyeftA@mail.gmail.com>
From: Rob Landley <rob@landley.net>
In-Reply-To: <CAPnZJGDcNwPLbzC99qNQ+bRMwxPU-Z0xe=TD6DWQU=0MNyeftA@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit

On 12/19/23 20:19, Askar Safin wrote:
> On Sun, Nov 19, 2023 at 08:12:48PM -0500, Stefan Berger wrote:
>> Documentation/filesystems/ramfs-rootfs-initramfs.rst states:
>>
>>   If CONFIG_TMPFS is enabled, rootfs will use tmpfs instead of ramfs by
>>   default.  To force ramfs, add "rootfstype=ramfs" to the kernel command
>>   line.

I wrote that around 2005.

>> This currently does not work when root= is provided since then
>> saved_root_name contains a string and rootfstype= is ignored.

I wrote the code to populate a tmpfs from initramfs in 2013 (8 years later),
because nobody else had bothered to, and I've had a patch to fix the rootfstype=
issue for years, but I long ago lost my touch with linux-kernel proctology:

https://lkml.iu.edu/hypermail/linux/kernel/2302.2/05601.html

Last I heard it was going upstream via somebody else (who tripled the size of
the fix for no obvious reason, but that's modern linux-kernel for you), but I
lost track after the guy who'd said to send it to him instead did multiple
automated bounces from a bot that said "to save this one special guy's time,
we're going to spam linux-kernel with autogenerated content delivered to tens of
thousands of plebian mailboxes":

https://lkml.iu.edu/hypermail/linux/kernel/2311.1/05544.html
https://lkml.iu.edu/hypermail/linux/kernel/2311.1/06448.html

Last I saw the rewrite of the fix had been resubmitted to the Special Guy a
fourth time:

https://lkml.iu.edu/hypermail/linux/kernel/2311.2/02938.html

Maybe it'll show up in a git pull someday? Who knows...

> Therefore,
>> ramfs is currently always chosen when root= is provided.
> 
> Maybe it is a good idea to just fully remove ramfs?

"Let's delete a 20 year old filesystem that's been built into every system that
whole time, this can't POSSIBLY have any side effects."

We've already discussed this, more than once:

https://lkml.iu.edu/hypermail/linux/kernel/2311.0/00435.html

Can you build tmpfs on a nommu system? Last I checked the plumbing expects swap,
but it's been a while...

> initramfs will
> always be tmpfs. And tmpfs will always be enabled.
> 
> As well as I understand, ramfs was originally introduced, because
> tmpfs seemed too big.

No, ramfs came first, tmpfs was created later. Not just what initramfs uses, but
the order in which the filesystems were written.

I believe modern ramfs more or less dates to Al Viro factoring out "libfs":

https://lwn.net/Articles/57369/

Although the memory grows a bit hazy. (I vaguely remember what happened, but
exactly when it was, and what came before what... git log in my unified history
tree is showing fs/ramfs originating from 2.3.99pre4 in 2000 but it may be
confused by a rename. I remember somebody, I thought Linus, saying that ramfs
was an example user of the libfs code...)

> Here are my configs (x86_64). Just enough to run busybox in "qemu -serial stdio"

Back when I maintained busybox there were people in that space who cared about
14k, and that's not measuring runtime overhead.

There was a lovely ELC talk in 2015 about booting Linux in 256 kilobytes of SRAM
(everything else being XIP out of ROM-alikes), but alas the entire year's videos
got wiped by the Linux Foundation. The PDF is still online though:

https://elinux.org/images/9/90/Linux_for_Microcontrollers-_From_Marginal_to_Mainstream.pdf

Alas, XIP got replaced with DAX which I've never managed to get to work...

Rob