Re: kernel BUG at fs/dcache.c:648! with v3.11-7890-ge5c832d

[Josh Boyer]

On Tue, Sep 10, 2013 at 1:33 PM, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Tue, Sep 10, 2013 at 10:14 AM, Josh Boyer <jwboyer@xxxxxxxxxxxxxxxxx> wrote:
>>
>> We've had a user report a backtrace from hitting the
>> BUG_ON(!ret->d_lockref.count) added with the lockref infrastructure
>> (commit 98474236f72) on rawhide today[1].  I've grabbed the backtrace
>> below.  The user has btrfs, NFS, and sshfs in usage with this oops.
>>
>> I've not seen anything similar, but I could have missed it.  Does this
>> look familiar to anyone?
>
> Nope. And the dget_parent() case itself hasn't even changed - that
> BUG_ON() wasn't really added by the lockref code, it's just a
> search-and-replace change of a BUG_ON(!d_count) to
> BUG_ON(!d_lockref.count). The BUG_ON() existed before.
>
> That whole "dget_parent()" thing is also in the _simple_ case (not RCU
> mode), and the BUG_ON is for when the dentry is properly locked, so
> that's all "safe" code. The refcount must have gotten corrupted
> earlier.
>
> Do you have the mainline git ID of that rawhide kernel? Because there
> *was* a real bug in d_rcu_to_refcount. I don't see how it could
> trigger that particular issue, but it could trigger scheduling while
> in the rcu-protected region and that in turn could result in odd
> things down the line, so..
>
> That particular bug exists between commits 15570086b590 ("vfs:
> reimplement d_rcu_to_refcount() using lockref_get_or_lock()") that
> introduced it, and e5c832d55588 ("vfs: fix dentry RCU to refcounting
> possibly sleeping dput()") that should have fixed it. But I don't know
> what mainline kernel that "kernel-3.12.0-0.rc0.git16.2.fc21.x86_64" is
> based on. I'm sure that information exists somewhere..

The subject says v3.11-7890-ge5c832d, which is the git-describe output
of the mainline kernel for that Fedora build.  Sorry, I should have
made that clearer.  So according to that, it should be based on the
actual commit you identified as the fix.

I have your latest tree as of this morning (v3.11-8716-g26b0332)
building right now and have asked Moneta to test it.

josh
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/