Re: BUG: might_sleep in /proc/swaps code

[viro@parcelfarce.linux.theplanet.co.uk]

On Wed, Apr 28, 2004 at 04:24:58PM -0700, Tim Hockin wrote:
> * /proc/swaps uses seq_file code, calling seq_path() with swaplock held
> * seq_path() calls d_path()
> * d_path() calls mntput() which might_sleep()
> 
> Is this worth trying to solve?  

Hrm...  Yes, it is - we could have chroot(2) called from another thread
while traversing the path to current root and have e.g. umount -l trigger
final umount when d_path() is finished.

Note that we have another blocking function called there anyway - dput()
will happily block under similar conditions (s/umount -l/rm/).

Lovely...  So we need yet another semaphore in swapfile.c (we can't turn
swaplock into semaphore *and* can't reuse swaps_bdev_sem, since dput()
et.al. are not just blocking but can cause any IO and memory allocations).

I'll try to put together something not too revolting; will post the patch
in a followup...