Received-SPF: pass (google.com: domain of linux-kernel+bounces-106542-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Precedence: bulk
MIME-Version: 1.0
References: <20240317221522.896040-1-jsperbeck@google.com> <252d1ba7-0ae0-4616-9ee7-f7c204531eaa@infradead.org>
In-Reply-To: <252d1ba7-0ae0-4616-9ee7-f7c204531eaa@infradead.org>
From: John Sperbeck <jsperbeck@google.com>
Date: Mon, 18 Mar 2024 12:00:56 -0700
Message-ID: <CAFNjLiVws+mCrMp4wMZuWTjaryK+-cKtNR9yFzVdqGf2u44oLg@mail.gmail.com>
Subject: Re: [PATCH] init: open /initrd.image with O_LARGEFILE
To: Randy Dunlap <rdunlap@infradead.org>
Cc: Andrew Morton <akpm@linux-foundation.org>, Jens Axboe <axboe@kernel.dk>, 
	"Peter Zijlstra (Intel)" <peterz@infradead.org>, Thomas Gleixner <tglx@linutronix.de>, 
	"ndesaulniers@google.com" <ndesaulniers@google.com>, linux-kernel@vger.kernel.org, 
	Nazerke Turtayeva <nturtayeva@ucsb.edu>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Sun, Mar 17, 2024 at 4:47=E2=80=AFPM Randy Dunlap <rdunlap@infradead.org=
> wrote:
>
> Hi,
>
> On 3/17/24 15:15, John Sperbeck wrote:
> > If initrd data is larger than 2Gb, we'll eventually fail to write to
> > the /initrd.image file when we hit that limit, unless O_LARGEFILE is se=
t.
> >
>
> Could this be related to
> https://lore.kernel.org/lkml/CAHY78BqCpMQptPN0SMaXuRqHOhYi+wnMEUSTYt7OHDZ=
ih4e7yQ@mail.gmail.com/
> ?
>
> Thanks.
>

I think there's a separate problem that this patch doesn't address.
If an individual component of a CPIO file is larger than 2Gb, then
there will be an EFBIG error when processing it.  The backtrace where
this occurs would be:

    generic_write_checks_count
    generic_write_checks
    generic_file_write_iter
    __kernel_write_iter
    kernel_write
    xwrite
    do_copy
    unpack_to_rootfs
    do_populate_rootfs
    async_run_entry_fn
    process_scheduled_works
    worker_thread
    kthread
    ret_from_fork
    ret_from_fork_asm

I think that could be dealt with by a patch to add O_LARGEFILE to
'openflags' in initramfs:do_name().