Re: [PATCH v3] zsmalloc: fix migrate_zspage-zs_free race condition

[Minchan Kim <minchan@kernel.org>]

On Mon, Jan 18, 2016 at 01:18:31PM +0100, Vlastimil Babka wrote:
> On 01/18/2016 09:20 AM, Minchan Kim wrote:
> >On Mon, Jan 18, 2016 at 08:54:07AM +0100, Vlastimil Babka wrote:
> >>On 18.1.2016 8:39, Sergey Senozhatsky wrote:
> >>>On (01/18/16 16:11), Minchan Kim wrote:
> >>>[..]
> >>>>>so, even if clear_bit_unlock/test_and_set_bit_lock do smp_mb or
> >>>>>barrier(), there is no corresponding barrier from record_obj()->WRITE_ONCE().
> >>>>>so I don't think WRITE_ONCE() will help the compiler, or am I missing
> >>>>>something?
> >>>>
> >>>>We need two things
> >>>>2. memory barrier.
> >>>>
> >>>>As compiler barrier, WRITE_ONCE works to prevent store tearing here
> >>>>by compiler.
> >>>>However, if we omit unpin_tag here, we lose memory barrier(e,g, smp_mb)
> >>>>so another CPU could see stale data caused CPU memory reordering.
> >>>
> >>>oh... good find! lost release semantic of unpin_tag()...
> >>
> >>Ah, release semantic, good point indeed. OK then we need the v2 approach again,
> >>with WRITE_ONCE() in record_obj(). Or some kind of record_obj_release() with
> >>release semantic, which would be a bit more effective, but I guess migration is
> >>not that critical path to be worth introducing it.
> >
> >WRITE_ONCE in record_obj would add more memory operations in obj_malloc
> 
> A simple WRITE_ONCE would just add a compiler barrier. What you
> suggests below does indeed add more operations, which are actually
> needed just in the migration. What I suggested is the v2 approach of
> adding the PIN bit before calling record_obj, *and* simply doing a
> WRITE_ONCE in record_obj() to make sure the PIN bit is indeed
> applied *before* writing to the handle, and not as two separate
> writes.
> 
> >but I don't feel it's too heavy in this phase so,
> 
> I'm afraid it's dangerous for the usage of record_obj() in
> zs_malloc() where the handle is freshly allocated by alloc_handle().
> Are we sure the bit is not set?
> 
> The code in alloc_handle() is:
>         return (unsigned long)kmem_cache_alloc(pool->handle_cachep,
>                 pool->flags & ~__GFP_HIGHMEM);
> 
> There's no explicit __GFP_ZERO, so the handles are not guaranteed to
> be allocated empty? And expecting all zpool users to include
> __GFP_ZERO in flags would be too subtle and error prone.

True.
Let's go with this. I hope it's the last.
Thanks, guys.

>From 389bbcbad9aba7d86a575b8c6ea3b8985cc801ea Mon Sep 17 00:00:00 2001
From: Junil Lee <junil0814.lee@lge.com>
Date: Mon, 18 Jan 2016 23:01:29 +0900
Subject: [PATCH v5] zsmalloc: fix migrate_zspage-zs_free race condition

record_obj() in migrate_zspage() does not preserve handle's
HANDLE_PIN_BIT, set by find_aloced_obj()->trypin_tag(), and implicitly
(accidentally) un-pins the handle, while migrate_zspage() still performs
an explicit unpin_tag() on the that handle.
This additional explicit unpin_tag() introduces a race condition with
zs_free(), which can pin that handle by this time, so the handle becomes
un-pinned.

Schematically, it goes like this:

CPU0					CPU1
migrate_zspage
  find_alloced_obj
    trypin_tag
      set HANDLE_PIN_BIT			zs_free()
						  pin_tag()
  obj_malloc() -- new object, no tag
  record_obj() -- remove HANDLE_PIN_BIT	    set HANDLE_PIN_BIT
  unpin_tag()  -- remove zs_free's HANDLE_PIN_BIT

The race condition may result in a NULL pointer dereference:
	Unable to handle kernel NULL pointer dereference at virtual address 00000000
	CPU: 0 PID: 19001 Comm: CookieMonsterCl Tainted:
	PC is at get_zspage_mapping+0x0/0x24
	LR is at obj_free.isra.22+0x64/0x128
	Call trace:
		[<ffffffc0001a3aa8>] get_zspage_mapping+0x0/0x24
		[<ffffffc0001a4918>] zs_free+0x88/0x114
		[<ffffffc00053ae54>] zram_free_page+0x64/0xcc
		[<ffffffc00053af4c>] zram_slot_free_notify+0x90/0x108
		[<ffffffc000196638>] swap_entry_free+0x278/0x294
		[<ffffffc000199008>] free_swap_and_cache+0x38/0x11c
		[<ffffffc0001837ac>] unmap_single_vma+0x480/0x5c8
		[<ffffffc000184350>] unmap_vmas+0x44/0x60
		[<ffffffc00018a53c>] exit_mmap+0x50/0x110
		[<ffffffc00009e408>] mmput+0x58/0xe0
		[<ffffffc0000a2854>] do_exit+0x320/0x8dc
		[<ffffffc0000a3cb4>] do_group_exit+0x44/0xa8
		[<ffffffc0000ae1bc>] get_signal+0x538/0x580
		[<ffffffc000087e44>] do_signal+0x98/0x4b8
		[<ffffffc00008843c>] do_notify_resume+0x14/0x5c

This patch keeps the lock bit in migration path and update
value atomically.

Signed-off-by: Junil Lee <junil0814.lee@lge.com>
Signed-off-by: Minchan Kim <minchan@kernel.org>
Cc: <stable@vger.kernel.org> [4.1+]
---
 mm/zsmalloc.c | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)

diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c
index e7414cec220b..2d7c4c11fc63 100644
--- a/mm/zsmalloc.c
+++ b/mm/zsmalloc.c
@@ -309,7 +309,12 @@ static void free_handle(struct zs_pool *pool, unsigned long handle)
 
 static void record_obj(unsigned long handle, unsigned long obj)
 {
-	*(unsigned long *)handle = obj;
+	/*
+	 * lsb of @obj represents handle lock while other bits
+	 * represent object value the handle is pointing so
+	 * updating shouldn't do store tearing.
+	 */
+	WRITE_ONCE(*(unsigned long *)handle, obj);
 }
 
 /* zpool driver */
@@ -1635,6 +1640,13 @@ static int migrate_zspage(struct zs_pool *pool, struct size_class *class,
 		free_obj = obj_malloc(d_page, class, handle);
 		zs_object_copy(free_obj, used_obj, class);
 		index++;
+		/*
+		 * record_obj updates handle's value to free_obj and it will
+		 * invalidate lock bit(ie, HANDLE_PIN_BIT) of handle, which
+		 * breaks synchronization using pin_tag(e,g, zs_free) so
+		 * let's keep the lock bit.
+		 */
+		free_obj |= BIT(HANDLE_PIN_BIT);
 		record_obj(handle, free_obj);
 		unpin_tag(handle);
 		obj_free(pool, class, used_obj);
-- 
1.9.1