linux-kernel - overlayfs: WARN_ONCE(Can't encode file handler for inotify: 255)

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAGrbwDTLt6drB9eaUagnQVgdPBmhLfqqxAf3F+Juqy_o6oP8uw@mail.gmail.com>
Date: Wed, 18 Dec 2024 00:23:08 +0000
From: Dmitry Safonov <dima@...sta.com>
To: Amir Goldstein <amir73il@...il.com>, Miklos Szeredi <miklos@...redi.hu>
Cc: linux-unionfs@...r.kernel.org, LKML <linux-kernel@...r.kernel.org>, 
	Linux FS Devel <linux-fsdevel@...r.kernel.org>, Dmitry Safonov <0x7f454c46@...il.com>, 
	Sahil Gupta <s.gupta@...sta.com>
Subject: overlayfs: WARN_ONCE(Can't encode file handler for inotify: 255)

Hi Amir and Miklos, linux-unionfs,

On v6.9.0 kernel we stepped over the WARN_ON() in show_mark_fhandle():

> ------------[ cut here ]------------
> Can't encode file handler for inotify: 255
> WARNING: CPU: 0 PID: 11136 at fs/notify/fdinfo.c:55 show_mark_fhandle+0xfa/0x110
> CPU: 0 PID: 11136 Comm: lsof Kdump: loaded Tainted: P        W  O       6.9.0 #1
> RIP: 0010:show_mark_fhandle+0xfa/0x110
> Code: 00 00 00 5b 41 5c 5d e9 44 21 97 00 80 3d 0d af 99 01 00 75 d8 89 ce 48 c7 c7 68 ad 4a 82 c6 05 fb ae 99 01 01 e8 f6 98 cc ff <0f> 0b eb bf e8 4d        29 96 00 66 66 2e 0f 1f 84 00 00 00 00 00 66 90
...
> Call Trace:
>  <TASK>
>  inotify_show_fdinfo+0x124/0x170
>  seq_show+0x188/0x1f0
>  seq_read_iter+0x115/0x4a0
>  seq_read+0xf9/0x130
>  vfs_read+0xb6/0x330
>  ksys_read+0x6b/0xf0
>  __do_fast_syscall_32+0x80/0x110
>  do_fast_syscall_32+0x37/0x80
>  entry_SYSCALL_compat_after_hwframe+0x75/0x75
> RIP: 0023:0xf7f93569

it later reproduced on v6.12.0. With some debug, it was narrowed down
to the way overlayfs encodes file handlers in ovl_encode_fh(). It
seems that currently it calculates them with the help of dentries.
Straight away from that, the reproducer becomes an easy drop_caches +
lsof (which parses procfs and finds some pid(s) that utilize inotify,
reading their correspondent fdinfo(s)).

So, my questions are: is a dentry actually needed for
ovl_dentry_to_fid()? Can't it just encode fh based on an inode? It
seems that the only reason it "needs" a dentry is to find the origin
layer in ovl_check_encode_origin(), is it so?

I guess, the potential solution here would be either to populate the
dentry back (likely racy and ugh) or just encode file handles based on
the lower-inode? IIUC, old file handles will become stale anyway after
dropping the caches?

As a rare visitor to fs code, not sure I described the problem or
understood it well enough.
FWIW, reverting commit 16aac5ad1fa9 ("ovl: support encoding
non-decodable file handles") seems to "fix" the warning locally (not
considering it a long-term fix).

Thanks,
            Dmitry