Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Peter Xu <peterx@redhat.com>
To:     linux-kernel@vger.kernel.org, linux-mm@kvack.org
Cc:     Andrea Arcangeli <aarcange@redhat.com>,
        Pengfei Xu <pengfei.xu@intel.com>, peterx@redhat.com,
        Nadav Amit <nadav.amit@gmail.com>,
        David Hildenbrand <david@redhat.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Miaohe Lin <linmiaohe@huawei.com>,
        Huang Ying <ying.huang@intel.com>
Subject: [PATCH 2/2] mm: Fix a few rare cases of using swapin error pte marker
Date:   Wed, 14 Dec 2022 15:04:53 -0500
Message-Id: <20221214200453.1772655-3-peterx@redhat.com>
In-Reply-To: <20221214200453.1772655-1-peterx@redhat.com>
References: <20221214200453.1772655-1-peterx@redhat.com>
MIME-Version: 1.0
Content-type: text/plain
Content-Transfer-Encoding: 8bit
Precedence: bulk

This patch should harden commit 15520a3f0469 ("mm: use pte markers for swap
errors") on using pte markers for swapin errors on a few corner cases.

1. Propagate swapin errors across fork()s: if there're swapin errors in
   the parent mm, after fork()s the child should sigbus too when an error
   page is accessed.

2. Fix a rare condition race in pte_marker_clear() where a uffd-wp pte
   marker can be quickly switched to a swapin error.

3. Explicitly ignore swapin error pte markers in change_protection().

I mostly don't worry on (2) or (3) at all, but we should still have them.
Case (1) is special because it can potentially cause silent data corrupt on
child when parent has swapin error triggered with swapoff, but since swapin
error is rare itself already it's probably not easy to trigger either.

Currently there is a priority difference between the uffd-wp bit and the
swapin error entry, in which the swapin error always has higher
priority (e.g. we don't need to wr-protect a swapin error pte marker).

If there will be a 3rd bit introduced, we'll probably need to consider a
more involved approach so we may need to start operate on the bits.  Let's
leave that for later.

This patch is tested with case (1) explicitly where we'll get corrupted
data before in the child if there's existing swapin error pte markers, and
after patch applied the child can be rightfully killed.

We don't need to copy stable for this one since 15520a3f0469 just landed as
part of v6.2-rc1, only "Fixes" applied.

Fixes: 15520a3f0469 ("mm: use pte markers for swap errors")
Signed-off-by: Peter Xu <peterx@redhat.com>
---
 mm/hugetlb.c  | 3 +++
 mm/memory.c   | 8 ++++++--
 mm/mprotect.c | 8 +++++++-
 3 files changed, 16 insertions(+), 3 deletions(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index f5f445c39dbc..1e8e4eb10328 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -4884,6 +4884,9 @@ int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src,
 				entry = huge_pte_clear_uffd_wp(entry);
 			set_huge_pte_at(dst, addr, dst_pte, entry);
 		} else if (unlikely(is_pte_marker(entry))) {
+			/* No swap on hugetlb */
+			WARN_ON_ONCE(
+			    is_swapin_error_entry(pte_to_swp_entry(entry)));
 			/*
 			 * We copy the pte marker only if the dst vma has
 			 * uffd-wp enabled.
diff --git a/mm/memory.c b/mm/memory.c
index 032ef700c3e8..3e836fecd035 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -828,7 +828,7 @@ copy_nonpresent_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
 			return -EBUSY;
 		return -ENOENT;
 	} else if (is_pte_marker_entry(entry)) {
-		if (userfaultfd_wp(dst_vma))
+		if (is_swapin_error_entry(entry) || userfaultfd_wp(dst_vma))
 			set_pte_at(dst_mm, addr, dst_pte, pte);
 		return 0;
 	}
@@ -3625,8 +3625,12 @@ static vm_fault_t pte_marker_clear(struct vm_fault *vmf)
 	/*
 	 * Be careful so that we will only recover a special uffd-wp pte into a
 	 * none pte.  Otherwise it means the pte could have changed, so retry.
+	 *
+	 * This should also cover the case where e.g. the pte changed
+	 * quickly from a PTE_MARKER_UFFD_WP into PTE_MARKER_SWAPIN_ERROR.
+	 * So is_pte_marker() check is not enough to safely drop the pte.
 	 */
-	if (is_pte_marker(*vmf->pte))
+	if (pte_same(vmf->orig_pte, *vmf->pte))
 		pte_clear(vmf->vma->vm_mm, vmf->address, vmf->pte);
 	pte_unmap_unlock(vmf->pte, vmf->ptl);
 	return 0;
diff --git a/mm/mprotect.c b/mm/mprotect.c
index 093cb50f2fc4..a6f905211327 100644
--- a/mm/mprotect.c
+++ b/mm/mprotect.c
@@ -245,7 +245,13 @@ static unsigned long change_pte_range(struct mmu_gather *tlb,
 					newpte = pte_swp_mksoft_dirty(newpte);
 				if (pte_swp_uffd_wp(oldpte))
 					newpte = pte_swp_mkuffd_wp(newpte);
-			} else if (pte_marker_entry_uffd_wp(entry)) {
+			} else if (is_pte_marker_entry(entry)) {
+				/*
+				 * Ignore swapin errors unconditionally,
+				 * because any access should sigbus anyway.
+				 */
+				if (is_swapin_error_entry(entry))
+					continue;
 				/*
 				 * If this is uffd-wp pte marker and we'd like
 				 * to unprotect it, drop it; the next page
-- 
2.37.3