[systemd-devel] BTRFS bug with parallel mounts?

[Gustavo Sverzut Barbieri]

On Wed, Nov 10, 2010 at 2:11 PM, Greg KH <greg at kroah.com> wrote:
> On Wed, Nov 10, 2010 at 01:02:51PM -0200, Gustavo Sverzut Barbieri wrote:
>> On Wed, Nov 10, 2010 at 9:33 AM, Gustavo Sverzut Barbieri
>> <barbieri at profusion.mobi> wrote:
>> > On Wednesday, November 10, 2010, Gustavo Sverzut Barbieri
>> > <barbieri at profusion.mobi> wrote:
>> >> Hi all,
>> >>
>> >> First, I'm not typing much or send the kernel oops since I'm at a
>> >> phone without access to my lap ATM
>> >>
>> >> So yesterday I got my new laptop with quad core i7 and SSD. gentoo's
>> >> openrc works nice, but systemd triggers a nasty kernel oops with btrfs
>> >>
>> >> The oops is while mounting my /home it get a null pointer at
>> >> btrfs_test_super. After the backtrace (which I had will send later) I
>> >> get a "btrfs detected SSD devices, enabling SSD mode"
>> >>
>> >> So really looks like a race. Ideas?
>> >>
>> >> That also bring us to the safety question: should we provide a way to
>> >> force /etc/fstab entries to be mounted serially and forcefully? This
>> >> would be helpful for buggy FS while you can't get the fix (or to
>> >> install and apply the fix -- as I'm using openrc but in future I'll
>> >> not)
>> >>
>> >
>> > Worths mentioning that all my fs are btrfs: /, /home and /var
>>
>> ----8<--------------------
>>
>> Part of the log (sorry, I had no way but to type it all):
>>
>> BUG: unable to handle kernel NULL pointer dereference at 0000000000000128
>> IP: [<ffffffff811329fb>] btrfs_test_super+0xb/0x30
>> PGD: 26b5b1067 PUD 26a465067 PMD 0
>> last sysfs file: /sys/devices/virtual/bdi/btrfs-2/uevent
>> CPU 0
>> Modules linked in: nvidia(P) lib80211_crypt_tkip wl(P)
>> Pid: 434, comm: mount Tainted: P     2.6.36 #3 Mac-F22586C8/MacBookPro6,2
>> RIP: ... btrfs_test_super+0xb/0x30
>
> Sounds like a btrfs issue, have you contacted those developers?

No, I just mailed systemd as people (kay?) seems to be using it, at
least I saw people here advertising its usage, even the readahead have
specific code, so maybe people had (or not) issues with it.

Let's face that mounting multiple points simultaneously is not what
one would expect.


> Also, you really are going to have to be able to reproduce this without
> the closed source kernel modules linked in (broadcom and nvidia) for
> anyone to even be able to help you out.

yeah, I'm pretty sure about that ;-)


-- 
Gustavo Sverzut Barbieri
http://profusion.mobi embedded systems
--------------------------------------
MSN: barbieri at gmail.com
Skype: gsbarbieri
Mobile: +55 (19) 9225-2202