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;
Date:   Sun, 18 Dec 2022 09:07:52 +0900
From:   asmadeus@codewreck.org
To:     Eric Van Hensbergen <evanhensbergen@icloud.com>
Cc:     v9fs-developer@lists.sourceforge.net, rminnich@gmail.com,
        lucho@ionkov.net, linux-kernel@vger.kernel.org,
        linux-fsdevel@vger.kernel.org, linux_oss@crudebyte.com
Subject: Re: [PATCH 2/6] Don't assume UID 0 attach
Message-ID: <Y55Z2DwZgRG+9zW3@codewreck.org>
References: <20221217185210.1431478-1-evanhensbergen@icloud.com>
 <20221217185210.1431478-3-evanhensbergen@icloud.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <20221217185210.1431478-3-evanhensbergen@icloud.com>
Precedence: bulk

Eric Van Hensbergen wrote on Sat, Dec 17, 2022 at 06:52:06PM +0000:
> The writeback_fid fallback code assumes a root uid fallback which
> breaks many server configurations (which don't run as root).  This
> patch switches to generic lookup which will follow argument
> guidence on access modes and default ids to use on failure.

Unfortunately this one will break writes to a file created as 400 I
think
That's the main reason we have this writeback fid afaik -- there are
cases where the user should be able to write to the file, but a plain
open/write won't work... I can't think of anything else than open 400
right now though

I'm sure there's an xfs_io command and xfstest for that, but for now:
python3 -c 'import os; f = os.open("testfile", os.O_CREAT + os.O_RDWR, 0o400); os.write(f, b"ok\n")'

iirc ganesha running as non-root just ignores root requests and opens as
current user-- this won't work for this particular case, but might be
good enough for you... With that said:

> There is a deeper underlying problem with writeback_fids in that
> this fallback is too standard and not an exception due to the way
> writeback mode works in the current implementation.  Subsequent
> patches will try to associate writeback fids from the original user
> either by flushing on close or by holding onto fid until writeback
> completes.

If we can address this problem though I agree we should stop using
wrieback fids as much as we do.
Now fids are refcounted, I think we could just use the normal fid as
writeback fid (getting a ref), and the regular close will not clunk it
so delayed IOs will pass.

Worth a try?
-- 
Dominique