On Tue 20-12-16 08:17:11, Dave Chinner wrote:
> On Mon, Dec 19, 2016 at 08:53:02PM +0100, Jan Kara wrote:
> > On Sat 17-12-16 09:04:50, Dave Chinner wrote:
> > > On Fri, Dec 16, 2016 at 09:19:16AM -0700, Ross Zwisler wrote:
> > > > On Fri, Dec 16, 2016 at 12:07:30PM +1100, Dave Chinner wrote:
> > > > > On Thu, Dec 15, 2016 at 04:40:41PM -0700, Dave Jiang wrote:
> > > > > > The caller into dax needs to clear __GFP_FS mask bit since it's
> > > > > > responsible for acquiring locks / transactions that blocks __GFP_FS
> > > > > > allocation. The caller will restore the original mask when dax function
> > > > > > returns.
> > > > >
> > > > > What's the allocation problem you're working around here? Can you
> > > > > please describe the call chain that is the problem?
> > > > >
> > > > > > xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
> > > > > >
> > > > > > if (IS_DAX(inode)) {
> > > > > > + gfp_t old_gfp = vmf->gfp_mask;
> > > > > > +
> > > > > > + vmf->gfp_mask &= ~__GFP_FS;
> > > > > > ret = dax_iomap_fault(vma, vmf, &xfs_iomap_ops);
> > > > > > + vmf->gfp_mask = old_gfp;
> > > > >
> > > > > I really have to say that I hate code that clears and restores flags
> > > > > without any explanation of why the code needs to play flag tricks. I
> > > > > take one look at the XFS fault handling code and ask myself now "why
> > > > > the hell do we need to clear those flags?" Especially as the other
> > > > > paths into generic fault handlers /don't/ require us to do this.
> > > > > What does DAX do that require us to treat memory allocation contexts
> > > > > differently to the filemap_fault() path?
> > > >
> > > > This was done in response to Jan Kara's concern:
> > > >
> > > > The gfp_mask that propagates from __do_fault() or do_page_mkwrite() is fine
> > > > because at that point it is correct. But once we grab filesystem locks which
> > > > are not reclaim safe, we should update vmf->gfp_mask we pass further down
> > > > into DAX code to not contain __GFP_FS (that's a bug we apparently have
> > > > there). And inside DAX code, we definitely are not generally safe to add
> > > > __GFP_FS to mapping_gfp_mask(). Maybe we'd be better off propagating struct
> > > > vm_fault into this function, using passed gfp_mask there and make sure
> > > > callers update gfp_mask as appropriate.
> > > >
> > > > https://lkml.org/lkml/2016/10/4/37
> > > >
> > > > IIUC I think the concern is that, for example, in xfs_filemap_page_mkwrite()
> > > > we take a read lock on the struct inode.i_rwsem before we call
> > > > dax_iomap_fault().
> > >
> > > That, my friends, is exactly the problem that mapping_gfp_mask() is
> > > meant to solve. This:
> > >
> > > > > > + vmf.gfp_mask = mapping_gfp_mask(mapping) | __GFP_FS | __GFP_IO;
> > >
> > > Is just so wrong it's not funny.
> >
> > You mean like in mm/memory.c: __get_fault_gfp_mask()?
> >
> > Which was introduced by commit c20cd45eb017 "mm: allow GFP_{FS,IO} for
> > page_cache_read page cache allocation" by Michal (added to CC) and you were
> > even on CC ;).
>
> Sure, I was on the cc list, but that doesn't mean I /liked/ the
> patch. It also doesn't mean I had the time or patience to argue
> whether it was the right way to address whatever whacky OOM/reclaim
> deficiency was being reported....
>
> Oh, and this is a write fault, not a read fault. There's a big
> difference in filesystem behaviour between those two types of
> faults, so what might be fine for a page cache read (i.e. no
> transactions) isn't necessarily correct for a write operation...
>
> > The code here was replicating __get_fault_gfp_mask() and in fact the idea
> > of the cleanup is to get rid of this code and take whatever is in
> > vmf.gfp_mask and mask off __GFP_FS in the filesystem if it deems it is
> > needed (e.g. ext4 really needs this as inode reclaim is depending on being
> > able to force a transaction commit).
>
> And so now we add a flag to the fault that the filesystem says not
> to add to mapping masks, and now the filesystem has to mask off
> thati flag /again/ because it's mapping gfp mask guidelines are
> essentially being ignored.
>
> Remind me again why we even have the mapping gfp_mask if we just
> ignore it like this?
mapping mask still serves its _main_ purpose - the allocation
placement/movability properties. This is something only the owner of
the mapping knows. The (ab)use of the mapping gfp_mask to drop GFP_FS
was imho a bad decision. As the above mentioned commit has mentioned
we were doing a lot of GFP_NOFS allocations from the paths which are
inherently GFP_KERNEL so they couldn't prevent from recursion problems
while they still affected the direct relaim behavior. On the other hand
I do understand why mapping's mask has been used at the time. We simply
lacked a better api back then. But I believe that with the scope nofs
[1] api we can do much better and get rid of ~__GFP_FS in mapping's mask
finally. c20cd45eb017 was an intermediate step until we get there.
I am not fully familiar with the DAX changes which started this
discussion but if there is a reclaim recursion problem from within the
fault path then the scope api sounds like a good fit here.
[1] http://lkml.kernel.org/r/20161215140715.12732-1-mhocko@xxxxxxxxxx
--
Michal Hocko
SUSE Labs
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>