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;
MIME-Version: 1.0
References: <20220728114915.91021-1-zhangjiachen.jaycee@bytedance.com> <CAJfpegtxUrytqTFjVG0J03ZmEzzpFr8Ra1KP8HiZNPzTtgY3kw@mail.gmail.com>
In-Reply-To: <CAJfpegtxUrytqTFjVG0J03ZmEzzpFr8Ra1KP8HiZNPzTtgY3kw@mail.gmail.com>
From:   Amir Goldstein <amir73il@gmail.com>
Date:   Thu, 28 Jul 2022 17:50:09 +0200
Message-ID: <CAOQ4uxgbVi4F9hbsVAG9v1Ay8P_MB=QBTLv2-y9M2f9jcbq5NA@mail.gmail.com>
Subject: Re: [PATCH v2] ovl: drop WARN_ON() dentry is NULL in ovl_encode_fh()
To:     Miklos Szeredi <miklos@szeredi.hu>
Cc:     Jiachen Zhang <zhangjiachen.jaycee@bytedance.com>,
        overlayfs <linux-unionfs@vger.kernel.org>,
        linux-kernel <linux-kernel@vger.kernel.org>,
        songmuchun@bytedance.com, Hongbo Yin <yinhongbo@bytedance.com>,
        Tianci Zhang <zhangtianci.1997@bytedance.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, Jul 28, 2022 at 3:06 PM Miklos Szeredi <miklos@szeredi.hu> wrote:
>
> On Thu, 28 Jul 2022 at 13:49, Jiachen Zhang
> <zhangjiachen.jaycee@bytedance.com> wrote:
> >
> > Some code paths cannot guarantee the inode have any dentry alias. So
> > WARN_ON() all !dentry may flood the kernel logs.
> >
> > For example, when an overlayfs inode is watched by inotifywait (1), and
> > someone is trying to read the /proc/$(pidof inotifywait)/fdinfo/INOTIFY_FD,
> > at that time if the dentry has been reclaimed by kernel (such as
> > echo 2 > /proc/sys/vm/drop_caches), there will be a WARN_ON(). The
> > printed call stack would be like:
> >
> >     ? show_mark_fhandle+0xf0/0xf0
> >     show_mark_fhandle+0x4a/0xf0
> >     ? show_mark_fhandle+0xf0/0xf0
> >     ? seq_vprintf+0x30/0x50
> >     ? seq_printf+0x53/0x70
> >     ? show_mark_fhandle+0xf0/0xf0
> >     inotify_fdinfo+0x70/0x90
> >     show_fdinfo.isra.4+0x53/0x70
> >     seq_show+0x130/0x170
> >     seq_read+0x153/0x440
> >     vfs_read+0x94/0x150
> >     ksys_read+0x5f/0xe0
> >     do_syscall_64+0x59/0x1e0
> >     entry_SYSCALL_64_after_hwframe+0x44/0xa9
> >
> > So let's drop WARN_ON() to avoid kernel log flooding.
>
>
> Applied, thanks.

FWIW, no objection to this fix, but for the record, encode_fh
is basically an inode operation, so it shouldn't require an alias.

The only thing in the call chain down from ovl_encode_fh()
that needs the ovl dentry is ovl_connect_layer(). The rest of the
referenced to ovl dentry can use an ovl inode instead.

In some cases (e.g. non-dir or pure upper dir), ovl_connect_layer()
will not be called at all, but even if it would need to be called, it is
better to skip it and encode the lower inode if there is no ovl dentry
available.

The possible eventual outcome of an fh changing after disconnected
dir copy up is probably better than failing encode_fh out right.

No need to make any of those changes for this corner case IMO.
Just wanted to add this analysis to the thread.

Thanks,
Amir.