linux-kernel - [PATCH 2/2] mm/hugetlb.c: Fix UAF of vma in hugetlb fault pathway

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] [day] [month] [year] [list]

Message-ID: <20240914194243.245-2-vishal.moola@gmail.com>
Date: Sat, 14 Sep 2024 12:41:19 -0700
From: "Vishal Moola (Oracle)" <vishal.moola@...il.com>
To: Muchun Song <muchun.song@...ux.dev>,
	syzbot <syzbot+2dab93857ee95f2eeb08@...kaller.appspotmail.com>,
	Andrew Morton <akpm@...ux-foundation.org>,
	LKML <linux-kernel@...r.kernel.org>,
	Linux Memory Management List <linux-mm@...ck.org>,
	syzkaller-bugs@...glegroups.com
Cc: "Vishal Moola (Oracle)" <vishal.moola@...il.com>,
	stable@...r.kernel.org
Subject: [PATCH 2/2] mm/hugetlb.c: Fix UAF of vma in hugetlb fault pathway

Syzbot reports a UAF in hugetlb_fault(). This happens because
vmf_anon_prepare() could drop the per-VMA lock and allow the current VMA
to be freed before hugetlb_vma_unlock_read() is called.

We can fix this by using a modified version of vmf_anon_prepare() that
doesn't release the VMA lock on failure, and then release it ourselves
after hugetlb_vma_unlock_read().

Reported-by: syzbot+2dab93857ee95f2eeb08@...kaller.appspotmail.com
Closes: https://lore.kernel.org/linux-mm/00000000000067c20b06219fbc26@google.com/
Fixes: 9acad7ba3e25 ("hugetlb: use vmf_anon_prepare() instead of anon_vma_prepare()")
Signed-off-by: Vishal Moola (Oracle) <vishal.moola@...il.com>
Cc: <stable@...r.kernel.org>
---
 mm/hugetlb.c | 20 ++++++++++++++++++--
 1 file changed, 18 insertions(+), 2 deletions(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 5c77defad295..190fa05635f4 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -5915,7 +5915,7 @@ static vm_fault_t hugetlb_wp(struct folio *pagecache_folio,
 	 * When the original hugepage is shared one, it does not have
 	 * anon_vma prepared.
 	 */
-	ret = vmf_anon_prepare(vmf);
+	ret = __vmf_anon_prepare(vmf);
 	if (unlikely(ret))
 		goto out_release_all;
 
@@ -6114,7 +6114,7 @@ static vm_fault_t hugetlb_no_page(struct address_space *mapping,
 		}
 
 		if (!(vma->vm_flags & VM_MAYSHARE)) {
-			ret = vmf_anon_prepare(vmf);
+			ret = __vmf_anon_prepare(vmf);
 			if (unlikely(ret))
 				goto out;
 		}
@@ -6245,6 +6245,14 @@ static vm_fault_t hugetlb_no_page(struct address_space *mapping,
 	folio_unlock(folio);
 out:
 	hugetlb_vma_unlock_read(vma);
+
+	/*
+	 * We must check to release the per-VMA lock. __vmf_anon_prepare() is
+	 * the only way ret can be set to VM_FAULT_RETRY.
+	 */
+	if (unlikely(ret & VM_FAULT_RETRY))
+		vma_end_read(vma);
+
 	mutex_unlock(&hugetlb_fault_mutex_table[hash]);
 	return ret;
 
@@ -6466,6 +6474,14 @@ vm_fault_t hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 	}
 out_mutex:
 	hugetlb_vma_unlock_read(vma);
+
+	/*
+	 * We must check to release the per-VMA lock. __vmf_anon_prepare() in
+	 * hugetlb_wp() is the only way ret can be set to VM_FAULT_RETRY.
+	 */
+	if (unlikely(ret & VM_FAULT_RETRY))
+		vma_end_read(vma);
+
 	mutex_unlock(&hugetlb_fault_mutex_table[hash]);
 	/*
 	 * Generally it's safe to hold refcount during waiting page lock. But
-- 
2.45.0