Hi, On Wed, Mar 27, 2002 at 11:34:34AM -0800, Andrew Morton wrote: > > Stephen, I'm half-inclined to paint this as a core kernel > > bug. We shouldn't be diving into write_dquot for a > > GFP_NOFS allocation. > > Sorry, I misread the trace. It is: > > ext3_new_inode > ->DQUOT_ALLOC_INODE > ->read(quota file) > ->alloc_page > ->shrink_caches > ->prune_icache > ->dq_drop > ->write_dquot Yep, and hence we're using the quota inode's gfp_mask which is GFP_HIGHUSER. Alan's comment on this is basically to the effect that as far as 2.4 is concerned, Marcelo's tree is still using the old, known-broken quota code --- all the important fixes are in -ac and the mainline is full of holes. I'm fixing some other stuff right now but I'll check -ac in a bit to see if this error route still exists there. If so, ext3 should still be able to munge the quota file address_space->gfp_mask manually to work around the core problem. Cheers, Stephen
