Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:8 as permitted sender) client-ip=2620:137:e000::3:8;
From:   Peter Xu <peterx@redhat.com>
To:     linux-mm@kvack.org, linux-kernel@vger.kernel.org
Cc:     peterx@redhat.com,
        Muhammad Usama Anjum <usama.anjum@collabora.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        David Hildenbrand <david@redhat.com>,
        Andrei Vagin <avagin@gmail.com>,
        syzbot+e94c5aaf7890901ebf9b@syzkaller.appspotmail.com
Subject: [PATCH 1/3] mm/pagemap: Fix ioctl(PAGEMAP_SCAN) on vma check
Date:   Thu, 16 Nov 2023 15:15:45 -0500
Message-ID: <20231116201547.536857-2-peterx@redhat.com>
In-Reply-To: <20231116201547.536857-1-peterx@redhat.com>
References: <20231116201547.536857-1-peterx@redhat.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

The new ioctl(PAGEMAP_SCAN) relies on vma wr-protect capability provided by
userfault, however in the vma test it didn't explicitly require the vma to
have wr-protect function enabled, even if PM_SCAN_WP_MATCHING flag is set.

It means the pagemap code can now apply uffd-wp bit to a page in the vma
even if not registered to userfaultfd at all.

Then in whatever way as long as the pte got written and page fault
resolved, we'll apply the write bit even if uffd-wp bit is set.  We'll see
a pte that has both UFFD_WP and WRITE bit set.  Anything later that looks
up the pte for uffd-wp bit will trigger the warning:

WARNING: CPU: 1 PID: 5071 at arch/x86/include/asm/pgtable.h:403 pte_uffd_wp arch/x86/include/asm/pgtable.h:403 [inline]

Fix it by doing proper check over the vma attributes when
PM_SCAN_WP_MATCHING is specified.

Fixes: 52526ca7fdb9 ("fs/proc/task_mmu: implement IOCTL to get and optionally clear info about PTEs")
Reported-by: syzbot+e94c5aaf7890901ebf9b@syzkaller.appspotmail.com
Signed-off-by: Peter Xu <peterx@redhat.com>
---
 fs/proc/task_mmu.c | 24 ++++++++++++++++++++----
 1 file changed, 20 insertions(+), 4 deletions(-)

diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
index 51e0ec658457..e91085d79926 100644
--- a/fs/proc/task_mmu.c
+++ b/fs/proc/task_mmu.c
@@ -1994,15 +1994,31 @@ static int pagemap_scan_test_walk(unsigned long start, unsigned long end,
 	struct pagemap_scan_private *p = walk->private;
 	struct vm_area_struct *vma = walk->vma;
 	unsigned long vma_category = 0;
+	bool wp_allowed = userfaultfd_wp_async(vma) &&
+	    userfaultfd_wp_use_markers(vma);
 
-	if (userfaultfd_wp_async(vma) && userfaultfd_wp_use_markers(vma))
-		vma_category |= PAGE_IS_WPALLOWED;
-	else if (p->arg.flags & PM_SCAN_CHECK_WPASYNC)
-		return -EPERM;
+	if (!wp_allowed) {
+		/* User requested explicit failure over wp-async capability */
+		if (p->arg.flags & PM_SCAN_CHECK_WPASYNC)
+			return -EPERM;
+		/*
+		 * User requires wr-protect, and allows silently skipping
+		 * unsupported vmas.
+		 */
+		if (p->arg.flags & PM_SCAN_WP_MATCHING)
+			return 1;
+		/*
+		 * Then the request doesn't involve wr-protects at all,
+		 * fall through to the rest checks, and allow vma walk.
+		 */
+	}
 
 	if (vma->vm_flags & VM_PFNMAP)
 		return 1;
 
+	if (wp_allowed)
+		vma_category |= PAGE_IS_WPALLOWED;
+
 	if (vma->vm_flags & VM_SOFTDIRTY)
 		vma_category |= PAGE_IS_SOFT_DIRTY;
 
-- 
2.41.0