[dm-devel] Re: [PATCH resubmit] do_mount: reduce stack consumption

[Andrew Morton]

Adrian Bunk <bunk at stusta.de> wrote:
>
> On Fri, Nov 04, 2005 at 10:27:42PM +0100, Heiko Carstens wrote:
> > > > See original stack back trace below and Andreas' patch and analysis
> > > > here:
> > > > http://www.ussg.iu.edu/hypermail/linux/kernel/0410.3/1844.html
> > 
> > I probably should add that with "original" stack back trace a trace of
> > a 2.6.10 kernel was meant, if that wasn't clear, but the DM code is
> > still the same in 2.6.14.
> > 
> > > >     <4>Call Trace:
> > ...
> > > >     <4> [<0000000010831380>] __map_bio+0x70/0x160 [dm_mod]
> > > >     <4> [<000000001083173e>] __split_bio+0x1e6/0x538 [dm_mod]
> > > >     <4> [<0000000010831ba8>] dm_request+0x118/0x25c [dm_mod]
> > > >     <4> [<0000000000241074>] generic_make_request+0xf0/0x21c
> > > >     <4> [<0000000010831380>] __map_bio+0x70/0x160 [dm_mod]
> > > >     <4> [<000000001083173e>] __split_bio+0x1e6/0x538 [dm_mod]
> > > >     <4> [<0000000010831ba8>] dm_request+0x118/0x25c [dm_mod]
> > > >     <4> [<0000000000241074>] generic_make_request+0xf0/0x21c
> > > >     <4> [<0000000010831380>] __map_bio+0x70/0x160 [dm_mod]
> > > >     <4> [<000000001083173e>] __split_bio+0x1e6/0x538 [dm_mod]
> > > >     <4> [<0000000010831ba8>] dm_request+0x118/0x25c [dm_mod]
> > > >     <4> [<0000000000241074>] generic_make_request+0xf0/0x21c
> > ...
> > 
> > This part of the call trace is actually good for >1500 bytes of stack
> > usage and is what kills us and should be fixed.
> > I'm surprised that there are no other bug reports regarding DM and
> > stack overflow with 4k stacks.
> >...
> 
> There were some reports of dm+xfs overflows with 4k stacks on i386.
> 
> The xfs side was sorted out, but I son't know the state of the dm part.
> 

The state is Very Bad, IMO.  At the very least, DM should struggle to the
utmost to reduce the stack utilisation around that recursion point.