Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To:     linux-kernel@vger.kernel.org
Cc:     Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        stable@vger.kernel.org, Vlastimil Babka <vbabka@suse.com>,
        Gerald Schaefer <gerald.schaefer@linux.ibm.com>,
        Alexander Gordeev <agordeev@linux.ibm.com>,
        Heiko Carstens <hca@linux.ibm.com>
Subject: [PATCH 5.4 255/320] s390/mm: fix 2KB pgtable release race
Date:   Mon, 24 Jan 2022 19:43:59 +0100
Message-Id: <20220124184002.666556675@linuxfoundation.org>
In-Reply-To: <20220124183953.750177707@linuxfoundation.org>
References: <20220124183953.750177707@linuxfoundation.org>
User-Agent: quilt/0.66
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Precedence: bulk

From: Alexander Gordeev <agordeev@linux.ibm.com>

commit c2c224932fd0ee6854d6ebfc8d059c2bcad86606 upstream.

There is a race on concurrent 2KB-pgtables release paths when
both upper and lower halves of the containing parent page are
freed, one via page_table_free_rcu() + __tlb_remove_table(),
and the other via page_table_free(). The race might lead to a
corruption as result of remove of list item in page_table_free()
concurrently with __free_page() in __tlb_remove_table().

Let's assume first the lower and next the upper 2KB-pgtables are
freed from a page. Since both halves of the page are allocated
the tracking byte (bits 24-31 of the page _refcount) has value
of 0x03 initially:

CPU0				CPU1
----				----

page_table_free_rcu() // lower half
{
	// _refcount[31..24] == 0x03
	...
	atomic_xor_bits(&page->_refcount,
			0x11U << (0 + 24));
	// _refcount[31..24] <= 0x12
	...
	table = table | (1U << 0);
	tlb_remove_table(tlb, table);
}
...
__tlb_remove_table()
{
	// _refcount[31..24] == 0x12
	mask = _table & 3;
	// mask <= 0x01
	...

				page_table_free() // upper half
				{
					// _refcount[31..24] == 0x12
					...
					atomic_xor_bits(
						&page->_refcount,
						1U << (1 + 24));
					// _refcount[31..24] <= 0x10
					// mask <= 0x10
					...
	atomic_xor_bits(&page->_refcount,
			mask << (4 + 24));
	// _refcount[31..24] <= 0x00
	// mask <= 0x00
	...
	if (mask != 0) // == false
		break;
	fallthrough;
	...
					if (mask & 3) // == false
						...
					else
	__free_page(page);			list_del(&page->lru);
	^^^^^^^^^^^^^^^^^^	RACE!		^^^^^^^^^^^^^^^^^^^^^
}					...
				}

The problem is page_table_free() releases the page as result of
lower nibble unset and __tlb_remove_table() observing zero too
early. With this update page_table_free() will use the similar
logic as page_table_free_rcu() + __tlb_remove_table(), and mark
the fragment as pending for removal in the upper nibble until
after the list_del().

In other words, the parent page is considered as unreferenced and
safe to release only when the lower nibble is cleared already and
unsetting a bit in upper nibble results in that nibble turned zero.

Cc: stable@vger.kernel.org
Suggested-by: Vlastimil Babka <vbabka@suse.com>
Reviewed-by: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Signed-off-by: Alexander Gordeev <agordeev@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
---
 arch/s390/mm/pgalloc.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

--- a/arch/s390/mm/pgalloc.c
+++ b/arch/s390/mm/pgalloc.c
@@ -255,13 +255,15 @@ void page_table_free(struct mm_struct *m
 		/* Free 2K page table fragment of a 4K page */
 		bit = (__pa(table) & ~PAGE_MASK)/(PTRS_PER_PTE*sizeof(pte_t));
 		spin_lock_bh(&mm->context.lock);
-		mask = atomic_xor_bits(&page->_refcount, 1U << (bit + 24));
+		mask = atomic_xor_bits(&page->_refcount, 0x11U << (bit + 24));
 		mask >>= 24;
 		if (mask & 3)
 			list_add(&page->lru, &mm->context.pgtable_list);
 		else
 			list_del(&page->lru);
 		spin_unlock_bh(&mm->context.lock);
+		mask = atomic_xor_bits(&page->_refcount, 0x10U << (bit + 24));
+		mask >>= 24;
 		if (mask != 0)
 			return;
 	} else {