linux-kernel - Re: general protection fault in find

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

Date:   Mon, 18 Jun 2018 15:32:28 +0200
From:   David Sterba <dsterba@...e.cz>
To:     Nikolay Borisov <nborisov@...e.com>
Cc:     clm@...com, dsterba@...e.com, jbacik@...com,
        linux-btrfs@...r.kernel.org, linux-kernel@...r.kernel.org,
        syzkaller-bugs@...glegroups.com, anand.jain@...cle.com
Subject: Re: general protection fault in find_device

On Mon, Jun 18, 2018 at 10:03:18AM +0300, Nikolay Borisov wrote:
> So this suggests some inconsistency on fs_devices->devices list. On a
> quick look indeed it doesn't seem clear what the locking rules for this
> list are. In device_list_add in the !device case a device is added with
> fs_devices->device_list_Mutex held and using list_add_rcu. In the same
> function if we want to read the list ie invoke find_devices (because we
> have found an fsid) we are using plain list_for_each_entry (ie not the
> _rcu version and i don't see device_list_mutex being held while
> iterating the list). Additionally in btrfs_free_extra_devids the
> fs_devices->devices list is iterated with uuid_mutex being held and not
> device_list_mutex. In open_fs_devices we don't get any protection
> whatsoever while reading the list.

The uuid_mutex or device_list_mutex is provided by a caller up the
stack.

> Same thing in
> btrfs_find_next_active_device. If the list is supposed to be
> RCU-protected then the rules are:
> 
> 1. There needs to be an out of band (ie not RCU) mutual exclusion of
> modifiers

that's device_list_mutex for fs_devices::devices

> 2. Iterating the list should use _rcu list primitives.
> 
> Currently I don't see those 2 invariants being enforced in every code path.

Where is it not enforced for example?

If the device_list_mutex is held, list traversal does not use
list_for_each_entry_rcu, otherwise it does (eg the DEV_INFO ioctl or
btrfs_show_devname).

The problem that triggers this report is IMO in device_list_add that
uses the device list unprotected. Anand sent patches for that, but they
were titled as 'cleanups' so I skipped them for the merge window.

Candidate fixes are:

https://patchwork.kernel.org/patch/10437705/
https://patchwork.kernel.org/patch/10437713/