Re: [patch 2.6.13-rc4] fix get_user_pages bug

[Hugh Dickins]

On Mon, 1 Aug 2005, Nick Piggin wrote:
> Ingo Molnar wrote:
> > Hugh's posting said:
> > 
> > "it's trying to avoid an endless loop of finding the pte not writable 
> > when ptrace is modifying a page which the user is currently protected
> > against writing to (setting a breakpoint in readonly text, perhaps?)"
> > 
> > i'm wondering, why should that case generate an infinite fault? The first
> > write access should copy the shared-library page into a private page and
> > map it into the task's MM, writable. If this make-writable 
> 
> It will be mapped readonly.

Yes.  Sorry to leave you so long pondering over my words!

> > operation races with a read access then we return a minor fault and the
> > page is still readonly, but retrying the write should then break up the
> > COW protection and generate a writable page, and a subsequent
> > follow_page() success. If the page cannot be made writable, shouldnt the
> > vma flags reflect this fact by not having the VM_MAYWRITE flag, and hence
> > get_user_pages() should have returned with -EFAULT earlier?
> 
> If it cannot be written to, then yes. If it can be written to
> but is mapped readonly then you have the problem.

Yes.  The problem case is the one where "maybe_mkwrite" finds !VM_WRITE
and so doesn't set writable (but as Linus has observed, dirty is set).

I'm no expert on that case, was just guessing its use, I think Robin
determined that Roland is the expert on it; but what's very clear is
that it's intentional, allowing strace to write where the user cannot
currently write.

> Aside, that brings up an interesting question - why should readonly
> mappings of writeable files (with VM_MAYWRITE set) disallow ptrace
> write access while readonly mappings of readonly files not? Or am I
> horribly confused?

Either you or I.  You'll have to spell that out to me in more detail,
I don't see it that way.

What I do see and am slightly disturbed by, is that do_wp_page on a
shared maywrite but not currently writable area, will go the break
cow route in mainline, and has done so forever; whereas my page_mkwrite
fixes in -mm inadvertently change that to go the the reuse route.

I think my inadvertent change is correct, and the current behaviour
(putting anonymous pages into a shared vma) is surprising, and may
have bad consequences (would at least get the overcommit accounting
wrong).

Hugh
-
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/