linux-kernel - Re: [PATCH] mm/gup: continue VM_FAULT_RETRY processing event for pre-faults

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <alpine.LSU.2.11.1905241429460.1141@eggly.anvils>
Date:   Fri, 24 May 2019 15:22:51 -0700 (PDT)
From:   Hugh Dickins <hughd@...gle.com>
To:     Sebastian Andrzej Siewior <bigeasy@...utronix.de>
cc:     Andrew Morton <akpm@...ux-foundation.org>,
        Mike Rapoport <rppt@...ux.ibm.com>,
        Andrea Arcangeli <aarcange@...hat.com>, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org, Borislav Petkov <bp@...e.de>,
        Pavel Machek <pavel@....cz>
Subject: Re: [PATCH] mm/gup: continue VM_FAULT_RETRY processing event for
 pre-faults

On Wed, 22 May 2019, Sebastian Andrzej Siewior wrote:
> On 2019-05-22 12:21:13 [-0700], Andrew Morton wrote:
> > On Tue, 14 May 2019 17:29:55 +0300 Mike Rapoport <rppt@...ux.ibm.com> wrote:
> > 
> > > When get_user_pages*() is called with pages = NULL, the processing of
> > > VM_FAULT_RETRY terminates early without actually retrying to fault-in all
> > > the pages.
> > > 
> > > If the pages in the requested range belong to a VMA that has userfaultfd
> > > registered, handle_userfault() returns VM_FAULT_RETRY *after* user space
> > > has populated the page, but for the gup pre-fault case there's no actual
> > > retry and the caller will get no pages although they are present.
> > > 
> > > This issue was uncovered when running post-copy memory restore in CRIU
> > > after commit d9c9ce34ed5c ("x86/fpu: Fault-in user stack if
> > > copy_fpstate_to_sigframe() fails").

I've been getting unexplained segmentation violations, and "make" giving
up early, when running kernel builds under swapping memory pressure: no
CRIU involved.

Bisected last night to that same x86/fpu commit, not itself guilty, but
suffering from the odd behavior of get_user_pages_unlocked() giving up
too early.

(I wondered at first if copy_fpstate_to_sigframe() ought to retry if
non-negative ret < nr_pages, but no, that would be wrong: a present page
followed by an invalid area would repeatedly return 1 for nr_pages 2.)

Cc'ing Pavel, who's been having segfault trouble in emacs: maybe same?

> > > 
> > > After this change, the copying of FPU state to the sigframe switched from
> > > copy_to_user() variants which caused a real page fault to get_user_pages()
> > > with pages parameter set to NULL.
...
> > Also, I wonder if copy_fpstate_to_sigframe() would be better using
> > fault_in_pages_writeable() rather than get_user_pages_unlocked().  That
> > seems like it operates at a more suitable level and I guess it will fix
> > this issue also.
> 
> It looks, like fault_in_pages_writeable() would work. If this is the
> recommendation from the MM department than I can switch to that.

I've now run a couple of hours of load successfully with Mike's patch
to GUP, no problem; but whatever the merits of that patch in general,
I agree with Andrew that fault_in_pages_writeable() seems altogether
more appropriate for copy_fpstate_to_sigframe(), and have now run a
couple of hours of load successfully with this instead (rewrite to taste):

--- 5.2-rc1/arch/x86/kernel/fpu/signal.c
+++ linux/arch/x86/kernel/fpu/signal.c
@@ -3,6 +3,7 @@
  * FPU signal frame handling routines.
  */
 
+#include <linux/pagemap.h>
 #include <linux/compat.h>
 #include <linux/cpu.h>
 
@@ -189,15 +190,7 @@ retry:
 	fpregs_unlock();
 
 	if (ret) {
-		int aligned_size;
-		int nr_pages;
-
-		aligned_size = offset_in_page(buf_fx) + fpu_user_xstate_size;
-		nr_pages = DIV_ROUND_UP(aligned_size, PAGE_SIZE);
-
-		ret = get_user_pages_unlocked((unsigned long)buf_fx, nr_pages,
-					      NULL, FOLL_WRITE);
-		if (ret == nr_pages)
+		if (!fault_in_pages_writeable(buf_fx, fpu_user_xstate_size))
 			goto retry;
 		return -EFAULT;
 	}

(I did wonder whether there needs to be an access_ok() check on buf_fx;
but if so, then I think it would already have been needed before the
earlier copy_fpregs_to_sigframe(); but I didn't get deep enough into
that to be sure, nor into whether access_ok() check on buf covers buf_fx.)

Hugh

> 
> > > In post-copy mode of CRIU, the destination memory is managed with
> > > userfaultfd and lack of the retry for pre-fault case in get_user_pages()
> > > causes a crash of the restored process.
> > > 
> > > Making the pre-fault behavior of get_user_pages() the same as the "normal"
> > > one fixes the issue.
> > 
> > Should this be backported into -stable trees?
> 
> Sebastian